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 SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
98 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
99 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
101 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 2>]>;
103 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
104 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
106 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
107 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
108 unsigned EltBits = 0;
109 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
110 return (EltBits == 32 && EltVal == 0);
113 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
114 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
115 unsigned EltBits = 0;
116 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
117 return (EltBits == 8 && EltVal == 0xff);
120 //===----------------------------------------------------------------------===//
121 // NEON operand definitions
122 //===----------------------------------------------------------------------===//
124 def nModImm : Operand<i32> {
125 let PrintMethod = "printNEONModImmOperand";
128 //===----------------------------------------------------------------------===//
129 // NEON load / store instructions
130 //===----------------------------------------------------------------------===//
132 // Use vldmia to load a Q register as a D register pair.
133 // This is equivalent to VLDMD except that it has a Q register operand
134 // instead of a pair of D registers.
136 : AXDI4<(outs QPR:$dst), (ins addrmode4:$addr, pred:$p),
137 IndexModeNone, IIC_fpLoadm,
138 "vldm${addr:submode}${p}\t$addr, ${dst:dregpair}", "",
139 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]>;
141 // Use vstmia to store a Q register as a D register pair.
142 // This is equivalent to VSTMD except that it has a Q register operand
143 // instead of a pair of D registers.
145 : AXDI4<(outs), (ins QPR:$src, addrmode4:$addr, pred:$p),
146 IndexModeNone, IIC_fpStorem,
147 "vstm${addr:submode}${p}\t$addr, ${src:dregpair}", "",
148 [(store (v2f64 QPR:$src), addrmode4:$addr)]>;
150 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
152 // Classes for VLD* pseudo-instructions with multi-register operands.
153 // These are expanded to real instructions after register allocation.
154 class VLDQPseudo<InstrItinClass itin>
155 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
156 class VLDQWBPseudo<InstrItinClass itin>
157 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
158 (ins addrmode6:$addr, am6offset:$offset), itin,
160 class VLDQQPseudo<InstrItinClass itin>
161 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
162 class VLDQQWBPseudo<InstrItinClass itin>
163 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
164 (ins addrmode6:$addr, am6offset:$offset), itin,
166 class VLDQQQQWBPseudo<InstrItinClass itin>
167 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
168 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
169 "$addr.addr = $wb, $src = $dst">;
171 // VLD1 : Vector Load (multiple single elements)
172 class VLD1D<bits<4> op7_4, string Dt>
173 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
174 (ins addrmode6:$addr), IIC_VLD1,
175 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
176 class VLD1Q<bits<4> op7_4, string Dt>
177 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
178 (ins addrmode6:$addr), IIC_VLD2,
179 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
181 def VLD1d8 : VLD1D<0b0000, "8">;
182 def VLD1d16 : VLD1D<0b0100, "16">;
183 def VLD1d32 : VLD1D<0b1000, "32">;
184 def VLD1d64 : VLD1D<0b1100, "64">;
186 def VLD1q8 : VLD1Q<0b0000, "8">;
187 def VLD1q16 : VLD1Q<0b0100, "16">;
188 def VLD1q32 : VLD1Q<0b1000, "32">;
189 def VLD1q64 : VLD1Q<0b1100, "64">;
191 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD2>;
192 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD2>;
193 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD2>;
194 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD2>;
196 // ...with address register writeback:
197 class VLD1DWB<bits<4> op7_4, string Dt>
198 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
199 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
200 "vld1", Dt, "\\{$dst\\}, $addr$offset",
201 "$addr.addr = $wb", []>;
202 class VLD1QWB<bits<4> op7_4, string Dt>
203 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
204 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
205 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
206 "$addr.addr = $wb", []>;
208 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
209 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
210 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
211 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
213 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
214 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
215 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
216 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
218 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2>;
219 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2>;
220 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2>;
221 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD2>;
223 // ...with 3 registers (some of these are only for the disassembler):
224 class VLD1D3<bits<4> op7_4, string Dt>
225 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
226 (ins addrmode6:$addr), IIC_VLD3, "vld1", Dt,
227 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
228 class VLD1D3WB<bits<4> op7_4, string Dt>
229 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
230 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3, "vld1", Dt,
231 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
233 def VLD1d8T : VLD1D3<0b0000, "8">;
234 def VLD1d16T : VLD1D3<0b0100, "16">;
235 def VLD1d32T : VLD1D3<0b1000, "32">;
236 def VLD1d64T : VLD1D3<0b1100, "64">;
238 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
239 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
240 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
241 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
243 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD3>;
244 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD3>;
246 // ...with 4 registers (some of these are only for the disassembler):
247 class VLD1D4<bits<4> op7_4, string Dt>
248 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
249 (ins addrmode6:$addr), IIC_VLD4, "vld1", Dt,
250 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
251 class VLD1D4WB<bits<4> op7_4, string Dt>
252 : NLdSt<0,0b10,0b0010,op7_4,
253 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
254 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4, "vld1", Dt,
255 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
258 def VLD1d8Q : VLD1D4<0b0000, "8">;
259 def VLD1d16Q : VLD1D4<0b0100, "16">;
260 def VLD1d32Q : VLD1D4<0b1000, "32">;
261 def VLD1d64Q : VLD1D4<0b1100, "64">;
263 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
264 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
265 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
266 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
268 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD4>;
269 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
271 // VLD2 : Vector Load (multiple 2-element structures)
272 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
273 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
274 (ins addrmode6:$addr), IIC_VLD2,
275 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
276 class VLD2Q<bits<4> op7_4, string Dt>
277 : NLdSt<0, 0b10, 0b0011, op7_4,
278 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
279 (ins addrmode6:$addr), IIC_VLD4,
280 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
282 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
283 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
284 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
286 def VLD2q8 : VLD2Q<0b0000, "8">;
287 def VLD2q16 : VLD2Q<0b0100, "16">;
288 def VLD2q32 : VLD2Q<0b1000, "32">;
290 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
291 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
292 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
294 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD4>;
295 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD4>;
296 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD4>;
298 // ...with address register writeback:
299 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
300 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
301 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
302 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
303 "$addr.addr = $wb", []>;
304 class VLD2QWB<bits<4> op7_4, string Dt>
305 : NLdSt<0, 0b10, 0b0011, op7_4,
306 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
307 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
308 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
309 "$addr.addr = $wb", []>;
311 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
312 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
313 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
315 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
316 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
317 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
319 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2>;
320 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2>;
321 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2>;
323 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
324 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
325 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
327 // ...with double-spaced registers (for disassembly only):
328 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
329 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
330 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
331 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
332 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
333 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
335 // VLD3 : Vector Load (multiple 3-element structures)
336 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
337 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
338 (ins addrmode6:$addr), IIC_VLD3,
339 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
341 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
342 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
343 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
345 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
346 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
347 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
349 // ...with address register writeback:
350 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
351 : NLdSt<0, 0b10, op11_8, op7_4,
352 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
353 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
354 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
355 "$addr.addr = $wb", []>;
357 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
358 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
359 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
361 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3>;
362 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3>;
363 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3>;
365 // ...with double-spaced registers (non-updating versions for disassembly only):
366 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
367 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
368 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
369 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
370 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
371 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
373 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
374 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
375 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
377 // ...alternate versions to be allocated odd register numbers:
378 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
379 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
380 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
382 // VLD4 : Vector Load (multiple 4-element structures)
383 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
384 : NLdSt<0, 0b10, op11_8, op7_4,
385 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
386 (ins addrmode6:$addr), IIC_VLD4,
387 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
389 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
390 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
391 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
393 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
394 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
395 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
397 // ...with address register writeback:
398 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
399 : NLdSt<0, 0b10, op11_8, op7_4,
400 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
401 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
402 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
403 "$addr.addr = $wb", []>;
405 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
406 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
407 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
409 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
410 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
411 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
413 // ...with double-spaced registers (non-updating versions for disassembly only):
414 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
415 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
416 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
417 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
418 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
419 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
421 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
422 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
423 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
425 // ...alternate versions to be allocated odd register numbers:
426 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
427 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
428 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
430 // Classes for VLD*LN pseudo-instructions with multi-register operands.
431 // These are expanded to real instructions after register allocation.
432 class VLDQLNPseudo<InstrItinClass itin>
433 : PseudoNLdSt<(outs QPR:$dst),
434 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
435 itin, "$src = $dst">;
436 class VLDQLNWBPseudo<InstrItinClass itin>
437 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
438 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
439 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
440 class VLDQQLNPseudo<InstrItinClass itin>
441 : PseudoNLdSt<(outs QQPR:$dst),
442 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
443 itin, "$src = $dst">;
444 class VLDQQLNWBPseudo<InstrItinClass itin>
445 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
446 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
447 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
448 class VLDQQQQLNPseudo<InstrItinClass itin>
449 : PseudoNLdSt<(outs QQQQPR:$dst),
450 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
451 itin, "$src = $dst">;
452 class VLDQQQQLNWBPseudo<InstrItinClass itin>
453 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
454 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
455 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
457 // VLD1LN : Vector Load (single element to one lane)
458 // FIXME: Not yet implemented.
460 // VLD2LN : Vector Load (single 2-element structure to one lane)
461 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
462 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
463 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
464 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
465 "$src1 = $dst1, $src2 = $dst2", []>;
467 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
468 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
469 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
471 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2>;
472 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2>;
473 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2>;
475 // ...with double-spaced registers:
476 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
477 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
479 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2>;
480 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2>;
482 // ...with address register writeback:
483 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
484 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
485 (ins addrmode6:$addr, am6offset:$offset,
486 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
487 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
488 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
490 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
491 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
492 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
494 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2>;
495 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2>;
496 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2>;
498 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
499 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
501 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2>;
502 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2>;
504 // VLD3LN : Vector Load (single 3-element structure to one lane)
505 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
506 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
507 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
508 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
509 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
510 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
512 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
513 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
514 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
516 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3>;
517 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3>;
518 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3>;
520 // ...with double-spaced registers:
521 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
522 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
524 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3>;
525 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3>;
527 // ...with address register writeback:
528 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
529 : NLdSt<1, 0b10, op11_8, op7_4,
530 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
531 (ins addrmode6:$addr, am6offset:$offset,
532 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
533 IIC_VLD3, "vld3", Dt,
534 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
535 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
538 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
539 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
540 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
542 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
543 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
544 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
546 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
547 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
549 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3>;
550 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3>;
552 // VLD4LN : Vector Load (single 4-element structure to one lane)
553 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
554 : NLdSt<1, 0b10, op11_8, op7_4,
555 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
556 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
557 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
558 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
559 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
561 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
562 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
563 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
565 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4>;
566 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4>;
567 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4>;
569 // ...with double-spaced registers:
570 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
571 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
573 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4>;
574 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4>;
576 // ...with address register writeback:
577 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
578 : NLdSt<1, 0b10, op11_8, op7_4,
579 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
580 (ins addrmode6:$addr, am6offset:$offset,
581 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
582 IIC_VLD4, "vld4", Dt,
583 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
584 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
587 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
588 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
589 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
591 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4>;
592 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4>;
593 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4>;
595 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
596 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
598 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4>;
599 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4>;
601 // VLD1DUP : Vector Load (single element to all lanes)
602 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
603 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
604 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
605 // FIXME: Not yet implemented.
606 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
608 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
610 // Classes for VST* pseudo-instructions with multi-register operands.
611 // These are expanded to real instructions after register allocation.
612 class VSTQPseudo<InstrItinClass itin>
613 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
614 class VSTQWBPseudo<InstrItinClass itin>
615 : PseudoNLdSt<(outs GPR:$wb),
616 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
618 class VSTQQPseudo<InstrItinClass itin>
619 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
620 class VSTQQWBPseudo<InstrItinClass itin>
621 : PseudoNLdSt<(outs GPR:$wb),
622 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
624 class VSTQQQQWBPseudo<InstrItinClass itin>
625 : PseudoNLdSt<(outs GPR:$wb),
626 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), IIC_VST,
629 // VST1 : Vector Store (multiple single elements)
630 class VST1D<bits<4> op7_4, string Dt>
631 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
632 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
633 class VST1Q<bits<4> op7_4, string Dt>
634 : NLdSt<0,0b00,0b1010,op7_4, (outs),
635 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
636 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
638 def VST1d8 : VST1D<0b0000, "8">;
639 def VST1d16 : VST1D<0b0100, "16">;
640 def VST1d32 : VST1D<0b1000, "32">;
641 def VST1d64 : VST1D<0b1100, "64">;
643 def VST1q8 : VST1Q<0b0000, "8">;
644 def VST1q16 : VST1Q<0b0100, "16">;
645 def VST1q32 : VST1Q<0b1000, "32">;
646 def VST1q64 : VST1Q<0b1100, "64">;
648 def VST1q8Pseudo : VSTQPseudo<IIC_VST>;
649 def VST1q16Pseudo : VSTQPseudo<IIC_VST>;
650 def VST1q32Pseudo : VSTQPseudo<IIC_VST>;
651 def VST1q64Pseudo : VSTQPseudo<IIC_VST>;
653 // ...with address register writeback:
654 class VST1DWB<bits<4> op7_4, string Dt>
655 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
656 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
657 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
658 class VST1QWB<bits<4> op7_4, string Dt>
659 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
660 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
661 IIC_VST, "vst1", Dt, "\\{$src1, $src2\\}, $addr$offset",
662 "$addr.addr = $wb", []>;
664 def VST1d8_UPD : VST1DWB<0b0000, "8">;
665 def VST1d16_UPD : VST1DWB<0b0100, "16">;
666 def VST1d32_UPD : VST1DWB<0b1000, "32">;
667 def VST1d64_UPD : VST1DWB<0b1100, "64">;
669 def VST1q8_UPD : VST1QWB<0b0000, "8">;
670 def VST1q16_UPD : VST1QWB<0b0100, "16">;
671 def VST1q32_UPD : VST1QWB<0b1000, "32">;
672 def VST1q64_UPD : VST1QWB<0b1100, "64">;
674 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST>;
675 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST>;
676 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST>;
677 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST>;
679 // ...with 3 registers (some of these are only for the disassembler):
680 class VST1D3<bits<4> op7_4, string Dt>
681 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
682 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
683 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
684 class VST1D3WB<bits<4> op7_4, string Dt>
685 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
686 (ins addrmode6:$addr, am6offset:$offset,
687 DPR:$src1, DPR:$src2, DPR:$src3),
688 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
689 "$addr.addr = $wb", []>;
691 def VST1d8T : VST1D3<0b0000, "8">;
692 def VST1d16T : VST1D3<0b0100, "16">;
693 def VST1d32T : VST1D3<0b1000, "32">;
694 def VST1d64T : VST1D3<0b1100, "64">;
696 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
697 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
698 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
699 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
701 def VST1d64TPseudo : VSTQQPseudo<IIC_VST>;
702 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST>;
704 // ...with 4 registers (some of these are only for the disassembler):
705 class VST1D4<bits<4> op7_4, string Dt>
706 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
707 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
708 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
710 class VST1D4WB<bits<4> op7_4, string Dt>
711 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
712 (ins addrmode6:$addr, am6offset:$offset,
713 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
714 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
715 "$addr.addr = $wb", []>;
717 def VST1d8Q : VST1D4<0b0000, "8">;
718 def VST1d16Q : VST1D4<0b0100, "16">;
719 def VST1d32Q : VST1D4<0b1000, "32">;
720 def VST1d64Q : VST1D4<0b1100, "64">;
722 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
723 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
724 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
725 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
727 def VST1d64QPseudo : VSTQQPseudo<IIC_VST>;
728 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST>;
730 // VST2 : Vector Store (multiple 2-element structures)
731 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
732 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
733 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
734 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
735 class VST2Q<bits<4> op7_4, string Dt>
736 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
737 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
738 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
741 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
742 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
743 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
745 def VST2q8 : VST2Q<0b0000, "8">;
746 def VST2q16 : VST2Q<0b0100, "16">;
747 def VST2q32 : VST2Q<0b1000, "32">;
749 def VST2d8Pseudo : VSTQPseudo<IIC_VST>;
750 def VST2d16Pseudo : VSTQPseudo<IIC_VST>;
751 def VST2d32Pseudo : VSTQPseudo<IIC_VST>;
753 def VST2q8Pseudo : VSTQQPseudo<IIC_VST>;
754 def VST2q16Pseudo : VSTQQPseudo<IIC_VST>;
755 def VST2q32Pseudo : VSTQQPseudo<IIC_VST>;
757 // ...with address register writeback:
758 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
759 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
760 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
761 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
762 "$addr.addr = $wb", []>;
763 class VST2QWB<bits<4> op7_4, string Dt>
764 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
765 (ins addrmode6:$addr, am6offset:$offset,
766 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
767 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
768 "$addr.addr = $wb", []>;
770 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
771 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
772 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
774 def VST2q8_UPD : VST2QWB<0b0000, "8">;
775 def VST2q16_UPD : VST2QWB<0b0100, "16">;
776 def VST2q32_UPD : VST2QWB<0b1000, "32">;
778 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST>;
779 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST>;
780 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST>;
782 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
783 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
784 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
786 // ...with double-spaced registers (for disassembly only):
787 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
788 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
789 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
790 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
791 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
792 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
794 // VST3 : Vector Store (multiple 3-element structures)
795 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
796 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
797 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
798 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
800 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
801 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
802 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
804 def VST3d8Pseudo : VSTQQPseudo<IIC_VST>;
805 def VST3d16Pseudo : VSTQQPseudo<IIC_VST>;
806 def VST3d32Pseudo : VSTQQPseudo<IIC_VST>;
808 // ...with address register writeback:
809 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
810 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
811 (ins addrmode6:$addr, am6offset:$offset,
812 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
813 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
814 "$addr.addr = $wb", []>;
816 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
817 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
818 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
820 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
821 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
822 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
824 // ...with double-spaced registers (non-updating versions for disassembly only):
825 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
826 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
827 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
828 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
829 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
830 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
832 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
833 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
834 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
836 // ...alternate versions to be allocated odd register numbers:
837 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
838 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
839 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
841 // VST4 : Vector Store (multiple 4-element structures)
842 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
843 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
844 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
845 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
848 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
849 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
850 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
852 def VST4d8Pseudo : VSTQQPseudo<IIC_VST>;
853 def VST4d16Pseudo : VSTQQPseudo<IIC_VST>;
854 def VST4d32Pseudo : VSTQQPseudo<IIC_VST>;
856 // ...with address register writeback:
857 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
858 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
859 (ins addrmode6:$addr, am6offset:$offset,
860 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
861 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
862 "$addr.addr = $wb", []>;
864 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
865 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
866 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
868 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
869 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
870 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST>;
872 // ...with double-spaced registers (non-updating versions for disassembly only):
873 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
874 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
875 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
876 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
877 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
878 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
880 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
881 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
882 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
884 // ...alternate versions to be allocated odd register numbers:
885 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
886 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
887 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST>;
889 // Classes for VST*LN pseudo-instructions with multi-register operands.
890 // These are expanded to real instructions after register allocation.
891 class VSTQLNPseudo<InstrItinClass itin>
892 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
894 class VSTQLNWBPseudo<InstrItinClass itin>
895 : PseudoNLdSt<(outs GPR:$wb),
896 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
897 nohash_imm:$lane), itin, "$addr.addr = $wb">;
898 class VSTQQLNPseudo<InstrItinClass itin>
899 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
901 class VSTQQLNWBPseudo<InstrItinClass itin>
902 : PseudoNLdSt<(outs GPR:$wb),
903 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
904 nohash_imm:$lane), itin, "$addr.addr = $wb">;
905 class VSTQQQQLNPseudo<InstrItinClass itin>
906 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
908 class VSTQQQQLNWBPseudo<InstrItinClass itin>
909 : PseudoNLdSt<(outs GPR:$wb),
910 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
911 nohash_imm:$lane), itin, "$addr.addr = $wb">;
913 // VST1LN : Vector Store (single element from one lane)
914 // FIXME: Not yet implemented.
916 // VST2LN : Vector Store (single 2-element structure from one lane)
917 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
918 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
919 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
920 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
923 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
924 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
925 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
927 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST>;
928 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST>;
929 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST>;
931 // ...with double-spaced registers:
932 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
933 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
935 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST>;
936 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST>;
938 // ...with address register writeback:
939 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
940 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
941 (ins addrmode6:$addr, am6offset:$offset,
942 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
943 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
944 "$addr.addr = $wb", []>;
946 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
947 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
948 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
950 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST>;
951 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST>;
952 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST>;
954 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
955 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
957 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
958 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
960 // VST3LN : Vector Store (single 3-element structure from one lane)
961 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
962 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
963 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
964 nohash_imm:$lane), IIC_VST, "vst3", Dt,
965 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
967 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
968 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
969 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
971 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST>;
972 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST>;
973 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST>;
975 // ...with double-spaced registers:
976 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
977 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
979 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST>;
980 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST>;
982 // ...with address register writeback:
983 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
984 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
985 (ins addrmode6:$addr, am6offset:$offset,
986 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
988 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
989 "$addr.addr = $wb", []>;
991 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
992 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
993 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
995 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
996 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
997 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
999 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
1000 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
1002 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1003 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1005 // VST4LN : Vector Store (single 4-element structure from one lane)
1006 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1007 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1008 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1009 nohash_imm:$lane), IIC_VST, "vst4", Dt,
1010 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
1013 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
1014 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
1015 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
1017 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST>;
1018 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST>;
1019 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST>;
1021 // ...with double-spaced registers:
1022 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
1023 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
1025 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST>;
1026 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST>;
1028 // ...with address register writeback:
1029 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1030 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1031 (ins addrmode6:$addr, am6offset:$offset,
1032 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1033 IIC_VST, "vst4", Dt,
1034 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
1035 "$addr.addr = $wb", []>;
1037 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
1038 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
1039 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
1041 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1042 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1043 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1045 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
1046 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
1048 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1049 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1051 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1054 //===----------------------------------------------------------------------===//
1055 // NEON pattern fragments
1056 //===----------------------------------------------------------------------===//
1058 // Extract D sub-registers of Q registers.
1059 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1060 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1061 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1063 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1064 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1065 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1067 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1068 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1069 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1071 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1072 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1073 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1076 // Extract S sub-registers of Q/D registers.
1077 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1078 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1079 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1082 // Translate lane numbers from Q registers to D subregs.
1083 def SubReg_i8_lane : SDNodeXForm<imm, [{
1084 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1086 def SubReg_i16_lane : SDNodeXForm<imm, [{
1087 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1089 def SubReg_i32_lane : SDNodeXForm<imm, [{
1090 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1093 //===----------------------------------------------------------------------===//
1094 // Instruction Classes
1095 //===----------------------------------------------------------------------===//
1097 // Basic 2-register operations: single-, double- and quad-register.
1098 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1099 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1100 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1101 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1102 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1103 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1104 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1105 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1106 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1107 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1108 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1109 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1110 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1111 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1112 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1113 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1114 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1115 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1117 // Basic 2-register intrinsics, both double- and quad-register.
1118 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1119 bits<2> op17_16, bits<5> op11_7, bit op4,
1120 InstrItinClass itin, string OpcodeStr, string Dt,
1121 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1122 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1123 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1124 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1125 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1126 bits<2> op17_16, bits<5> op11_7, bit op4,
1127 InstrItinClass itin, string OpcodeStr, string Dt,
1128 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1129 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1130 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1131 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1133 // Narrow 2-register operations.
1134 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1135 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1136 InstrItinClass itin, string OpcodeStr, string Dt,
1137 ValueType TyD, ValueType TyQ, SDNode OpNode>
1138 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1139 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1140 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1142 // Narrow 2-register intrinsics.
1143 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1144 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1145 InstrItinClass itin, string OpcodeStr, string Dt,
1146 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1147 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1148 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1149 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1151 // Long 2-register operations (currently only used for VMOVL).
1152 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1153 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1154 InstrItinClass itin, string OpcodeStr, string Dt,
1155 ValueType TyQ, ValueType TyD, SDNode OpNode>
1156 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1157 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1158 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1160 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1161 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1162 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1163 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1164 OpcodeStr, Dt, "$dst1, $dst2",
1165 "$src1 = $dst1, $src2 = $dst2", []>;
1166 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1167 InstrItinClass itin, string OpcodeStr, string Dt>
1168 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1169 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1170 "$src1 = $dst1, $src2 = $dst2", []>;
1172 // Basic 3-register operations: single-, double- and quad-register.
1173 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1174 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1175 SDNode OpNode, bit Commutable>
1176 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1177 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1178 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1179 let isCommutable = Commutable;
1182 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1183 InstrItinClass itin, string OpcodeStr, string Dt,
1184 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1185 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1186 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1187 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1188 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1189 let isCommutable = Commutable;
1191 // Same as N3VD but no data type.
1192 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1193 InstrItinClass itin, string OpcodeStr,
1194 ValueType ResTy, ValueType OpTy,
1195 SDNode OpNode, bit Commutable>
1196 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1197 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1198 OpcodeStr, "$dst, $src1, $src2", "",
1199 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1200 let isCommutable = Commutable;
1203 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1204 InstrItinClass itin, string OpcodeStr, string Dt,
1205 ValueType Ty, SDNode ShOp>
1206 : N3V<0, 1, op21_20, op11_8, 1, 0,
1207 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1208 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1209 [(set (Ty DPR:$dst),
1210 (Ty (ShOp (Ty DPR:$src1),
1211 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1212 let isCommutable = 0;
1214 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1215 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1216 : N3V<0, 1, op21_20, op11_8, 1, 0,
1217 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1218 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1219 [(set (Ty DPR:$dst),
1220 (Ty (ShOp (Ty DPR:$src1),
1221 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1222 let isCommutable = 0;
1225 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1226 InstrItinClass itin, string OpcodeStr, string Dt,
1227 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1228 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1229 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1230 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1231 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1232 let isCommutable = Commutable;
1234 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1235 InstrItinClass itin, string OpcodeStr,
1236 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1237 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1238 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1239 OpcodeStr, "$dst, $src1, $src2", "",
1240 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1241 let isCommutable = Commutable;
1243 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1244 InstrItinClass itin, string OpcodeStr, string Dt,
1245 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1246 : N3V<1, 1, op21_20, op11_8, 1, 0,
1247 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1248 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1249 [(set (ResTy QPR:$dst),
1250 (ResTy (ShOp (ResTy QPR:$src1),
1251 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1253 let isCommutable = 0;
1255 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1256 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1257 : N3V<1, 1, op21_20, op11_8, 1, 0,
1258 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1259 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1260 [(set (ResTy QPR:$dst),
1261 (ResTy (ShOp (ResTy QPR:$src1),
1262 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1264 let isCommutable = 0;
1267 // Basic 3-register intrinsics, both double- and quad-register.
1268 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1269 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1270 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1271 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1272 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1273 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1274 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1275 let isCommutable = Commutable;
1277 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1278 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1279 : N3V<0, 1, op21_20, op11_8, 1, 0,
1280 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1281 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1282 [(set (Ty DPR:$dst),
1283 (Ty (IntOp (Ty DPR:$src1),
1284 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1286 let isCommutable = 0;
1288 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1289 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1290 : N3V<0, 1, op21_20, op11_8, 1, 0,
1291 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1292 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1293 [(set (Ty DPR:$dst),
1294 (Ty (IntOp (Ty DPR:$src1),
1295 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1296 let isCommutable = 0;
1299 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1300 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1301 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1302 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1303 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1304 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1305 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1306 let isCommutable = Commutable;
1308 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1309 string OpcodeStr, string Dt,
1310 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1311 : N3V<1, 1, op21_20, op11_8, 1, 0,
1312 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1313 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1314 [(set (ResTy QPR:$dst),
1315 (ResTy (IntOp (ResTy QPR:$src1),
1316 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1318 let isCommutable = 0;
1320 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1321 string OpcodeStr, string Dt,
1322 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1323 : N3V<1, 1, op21_20, op11_8, 1, 0,
1324 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1325 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1326 [(set (ResTy QPR:$dst),
1327 (ResTy (IntOp (ResTy QPR:$src1),
1328 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1330 let isCommutable = 0;
1333 // Multiply-Add/Sub operations: single-, double- and quad-register.
1334 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1335 InstrItinClass itin, string OpcodeStr, string Dt,
1336 ValueType Ty, SDNode MulOp, SDNode OpNode>
1337 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1338 (outs DPR_VFP2:$dst),
1339 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1340 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1342 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1343 InstrItinClass itin, string OpcodeStr, string Dt,
1344 ValueType Ty, SDNode MulOp, SDNode OpNode>
1345 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1346 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1347 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1348 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1349 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1350 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1351 string OpcodeStr, string Dt,
1352 ValueType Ty, SDNode MulOp, SDNode ShOp>
1353 : N3V<0, 1, op21_20, op11_8, 1, 0,
1355 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1357 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1358 [(set (Ty DPR:$dst),
1359 (Ty (ShOp (Ty DPR:$src1),
1360 (Ty (MulOp DPR:$src2,
1361 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1363 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1364 string OpcodeStr, string Dt,
1365 ValueType Ty, SDNode MulOp, SDNode ShOp>
1366 : N3V<0, 1, op21_20, op11_8, 1, 0,
1368 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1370 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1371 [(set (Ty DPR:$dst),
1372 (Ty (ShOp (Ty DPR:$src1),
1373 (Ty (MulOp DPR:$src2,
1374 (Ty (NEONvduplane (Ty DPR_8:$src3),
1377 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1378 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1379 SDNode MulOp, SDNode OpNode>
1380 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1381 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1382 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1383 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1384 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1385 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1386 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1387 SDNode MulOp, SDNode ShOp>
1388 : N3V<1, 1, op21_20, op11_8, 1, 0,
1390 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1392 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1393 [(set (ResTy QPR:$dst),
1394 (ResTy (ShOp (ResTy QPR:$src1),
1395 (ResTy (MulOp QPR:$src2,
1396 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1398 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1399 string OpcodeStr, string Dt,
1400 ValueType ResTy, ValueType OpTy,
1401 SDNode MulOp, SDNode ShOp>
1402 : N3V<1, 1, op21_20, op11_8, 1, 0,
1404 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1406 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1407 [(set (ResTy QPR:$dst),
1408 (ResTy (ShOp (ResTy QPR:$src1),
1409 (ResTy (MulOp QPR:$src2,
1410 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1413 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1414 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1415 InstrItinClass itin, string OpcodeStr, string Dt,
1416 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1417 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1418 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1419 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1420 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1421 (Ty (IntOp (Ty DPR:$src2), (Ty DPR:$src3))))))]>;
1422 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1423 InstrItinClass itin, string OpcodeStr, string Dt,
1424 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1425 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1426 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1427 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1428 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1429 (Ty (IntOp (Ty QPR:$src2), (Ty QPR:$src3))))))]>;
1431 // Neon 3-argument intrinsics, both double- and quad-register.
1432 // The destination register is also used as the first source operand register.
1433 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1434 InstrItinClass itin, string OpcodeStr, string Dt,
1435 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1436 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1437 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1438 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1439 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1440 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1441 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1442 InstrItinClass itin, string OpcodeStr, string Dt,
1443 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1444 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1445 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1446 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1447 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1448 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1450 // Long Multiply-Add/Sub operations.
1451 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1452 InstrItinClass itin, string OpcodeStr, string Dt,
1453 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1454 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1455 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1456 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1457 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1458 (TyQ (MulOp (TyD DPR:$src2),
1459 (TyD DPR:$src3)))))]>;
1460 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1461 InstrItinClass itin, string OpcodeStr, string Dt,
1462 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1463 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1464 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1466 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1468 (OpNode (TyQ QPR:$src1),
1469 (TyQ (MulOp (TyD DPR:$src2),
1470 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1472 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1473 InstrItinClass itin, string OpcodeStr, string Dt,
1474 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1475 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1476 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1478 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1480 (OpNode (TyQ QPR:$src1),
1481 (TyQ (MulOp (TyD DPR:$src2),
1482 (TyD (NEONvduplane (TyD DPR_8:$src3),
1485 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1486 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1487 InstrItinClass itin, string OpcodeStr, string Dt,
1488 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1490 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1491 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1492 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1493 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1494 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src2),
1495 (TyD DPR:$src3)))))))]>;
1497 // Neon Long 3-argument intrinsic. The destination register is
1498 // a quad-register and is also used as the first source operand register.
1499 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1500 InstrItinClass itin, string OpcodeStr, string Dt,
1501 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1502 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1503 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1504 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1506 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1507 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1508 string OpcodeStr, string Dt,
1509 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1510 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1512 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1514 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1515 [(set (ResTy QPR:$dst),
1516 (ResTy (IntOp (ResTy QPR:$src1),
1518 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1520 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1521 InstrItinClass itin, string OpcodeStr, string Dt,
1522 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1523 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1525 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1527 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1528 [(set (ResTy QPR:$dst),
1529 (ResTy (IntOp (ResTy QPR:$src1),
1531 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1534 // Narrowing 3-register intrinsics.
1535 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1536 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1537 Intrinsic IntOp, bit Commutable>
1538 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1539 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1540 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1541 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1542 let isCommutable = Commutable;
1545 // Long 3-register operations.
1546 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1547 InstrItinClass itin, string OpcodeStr, string Dt,
1548 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1549 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1550 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1551 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1552 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1553 let isCommutable = Commutable;
1555 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1556 InstrItinClass itin, string OpcodeStr, string Dt,
1557 ValueType TyQ, ValueType TyD, SDNode OpNode>
1558 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1559 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1560 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1562 (TyQ (OpNode (TyD DPR:$src1),
1563 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1564 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1565 InstrItinClass itin, string OpcodeStr, string Dt,
1566 ValueType TyQ, ValueType TyD, SDNode OpNode>
1567 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1568 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1569 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1571 (TyQ (OpNode (TyD DPR:$src1),
1572 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1574 // Long 3-register operations with explicitly extended operands.
1575 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1576 InstrItinClass itin, string OpcodeStr, string Dt,
1577 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1579 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1580 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1581 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1582 [(set QPR:$dst, (OpNode (TyQ (ExtOp (TyD DPR:$src1))),
1583 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1584 let isCommutable = Commutable;
1587 // Long 3-register intrinsics with explicit extend (VABDL).
1588 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1589 InstrItinClass itin, string OpcodeStr, string Dt,
1590 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1592 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1593 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1594 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1595 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1596 (TyD DPR:$src2))))))]> {
1597 let isCommutable = Commutable;
1600 // Long 3-register intrinsics.
1601 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1602 InstrItinClass itin, string OpcodeStr, string Dt,
1603 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1604 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1605 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1606 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1607 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1608 let isCommutable = Commutable;
1610 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1611 string OpcodeStr, string Dt,
1612 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1613 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1614 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1615 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1616 [(set (ResTy QPR:$dst),
1617 (ResTy (IntOp (OpTy DPR:$src1),
1618 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1620 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1621 InstrItinClass itin, string OpcodeStr, string Dt,
1622 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1623 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1624 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1625 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1626 [(set (ResTy QPR:$dst),
1627 (ResTy (IntOp (OpTy DPR:$src1),
1628 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1631 // Wide 3-register operations.
1632 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1633 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1634 SDNode OpNode, SDNode ExtOp, bit Commutable>
1635 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1636 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1637 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1638 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1639 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1640 let isCommutable = Commutable;
1643 // Pairwise long 2-register intrinsics, both double- and quad-register.
1644 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1645 bits<2> op17_16, bits<5> op11_7, bit op4,
1646 string OpcodeStr, string Dt,
1647 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1648 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1649 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1650 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1651 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1652 bits<2> op17_16, bits<5> op11_7, bit op4,
1653 string OpcodeStr, string Dt,
1654 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1655 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1656 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1657 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1659 // Pairwise long 2-register accumulate intrinsics,
1660 // both double- and quad-register.
1661 // The destination register is also used as the first source operand register.
1662 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1663 bits<2> op17_16, bits<5> op11_7, bit op4,
1664 string OpcodeStr, string Dt,
1665 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1666 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1667 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1668 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1669 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1670 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1671 bits<2> op17_16, bits<5> op11_7, bit op4,
1672 string OpcodeStr, string Dt,
1673 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1674 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1675 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1676 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1677 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1679 // Shift by immediate,
1680 // both double- and quad-register.
1681 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1682 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1683 ValueType Ty, SDNode OpNode>
1684 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1685 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1686 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1687 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1688 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1689 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1690 ValueType Ty, SDNode OpNode>
1691 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1692 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1693 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1694 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1696 // Long shift by immediate.
1697 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1698 string OpcodeStr, string Dt,
1699 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1700 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1701 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1702 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1703 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1704 (i32 imm:$SIMM))))]>;
1706 // Narrow shift by immediate.
1707 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1708 InstrItinClass itin, string OpcodeStr, string Dt,
1709 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1710 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1711 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1712 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1713 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1714 (i32 imm:$SIMM))))]>;
1716 // Shift right by immediate and accumulate,
1717 // both double- and quad-register.
1718 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1719 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1720 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1721 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1722 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1723 [(set DPR:$dst, (Ty (add DPR:$src1,
1724 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1725 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1726 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1727 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1728 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1729 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1730 [(set QPR:$dst, (Ty (add QPR:$src1,
1731 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1733 // Shift by immediate and insert,
1734 // both double- and quad-register.
1735 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1736 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1737 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1738 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1739 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1740 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1741 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1742 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1743 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1744 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1745 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1746 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1748 // Convert, with fractional bits immediate,
1749 // both double- and quad-register.
1750 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1751 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1753 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1754 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1755 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1756 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1757 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1758 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1760 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1761 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1762 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1763 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1765 //===----------------------------------------------------------------------===//
1767 //===----------------------------------------------------------------------===//
1769 // Abbreviations used in multiclass suffixes:
1770 // Q = quarter int (8 bit) elements
1771 // H = half int (16 bit) elements
1772 // S = single int (32 bit) elements
1773 // D = double int (64 bit) elements
1775 // Neon 2-register vector operations -- for disassembly only.
1777 // First with only element sizes of 8, 16 and 32 bits:
1778 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1779 bits<5> op11_7, bit op4, string opc, string Dt,
1781 // 64-bit vector types.
1782 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1783 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1784 opc, !strconcat(Dt, "8"), asm, "", []>;
1785 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1786 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1787 opc, !strconcat(Dt, "16"), asm, "", []>;
1788 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1789 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1790 opc, !strconcat(Dt, "32"), asm, "", []>;
1791 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1792 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1793 opc, "f32", asm, "", []> {
1794 let Inst{10} = 1; // overwrite F = 1
1797 // 128-bit vector types.
1798 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1799 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1800 opc, !strconcat(Dt, "8"), asm, "", []>;
1801 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1802 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1803 opc, !strconcat(Dt, "16"), asm, "", []>;
1804 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1805 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1806 opc, !strconcat(Dt, "32"), asm, "", []>;
1807 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1808 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1809 opc, "f32", asm, "", []> {
1810 let Inst{10} = 1; // overwrite F = 1
1814 // Neon 3-register vector operations.
1816 // First with only element sizes of 8, 16 and 32 bits:
1817 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1818 InstrItinClass itinD16, InstrItinClass itinD32,
1819 InstrItinClass itinQ16, InstrItinClass itinQ32,
1820 string OpcodeStr, string Dt,
1821 SDNode OpNode, bit Commutable = 0> {
1822 // 64-bit vector types.
1823 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1824 OpcodeStr, !strconcat(Dt, "8"),
1825 v8i8, v8i8, OpNode, Commutable>;
1826 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1827 OpcodeStr, !strconcat(Dt, "16"),
1828 v4i16, v4i16, OpNode, Commutable>;
1829 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1830 OpcodeStr, !strconcat(Dt, "32"),
1831 v2i32, v2i32, OpNode, Commutable>;
1833 // 128-bit vector types.
1834 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1835 OpcodeStr, !strconcat(Dt, "8"),
1836 v16i8, v16i8, OpNode, Commutable>;
1837 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1838 OpcodeStr, !strconcat(Dt, "16"),
1839 v8i16, v8i16, OpNode, Commutable>;
1840 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1841 OpcodeStr, !strconcat(Dt, "32"),
1842 v4i32, v4i32, OpNode, Commutable>;
1845 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1846 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1848 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1850 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1851 v8i16, v4i16, ShOp>;
1852 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1853 v4i32, v2i32, ShOp>;
1856 // ....then also with element size 64 bits:
1857 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1858 InstrItinClass itinD, InstrItinClass itinQ,
1859 string OpcodeStr, string Dt,
1860 SDNode OpNode, bit Commutable = 0>
1861 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1862 OpcodeStr, Dt, OpNode, Commutable> {
1863 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1864 OpcodeStr, !strconcat(Dt, "64"),
1865 v1i64, v1i64, OpNode, Commutable>;
1866 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1867 OpcodeStr, !strconcat(Dt, "64"),
1868 v2i64, v2i64, OpNode, Commutable>;
1872 // Neon Narrowing 2-register vector operations,
1873 // source operand element sizes of 16, 32 and 64 bits:
1874 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1875 bits<5> op11_7, bit op6, bit op4,
1876 InstrItinClass itin, string OpcodeStr, string Dt,
1878 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1879 itin, OpcodeStr, !strconcat(Dt, "16"),
1880 v8i8, v8i16, OpNode>;
1881 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1882 itin, OpcodeStr, !strconcat(Dt, "32"),
1883 v4i16, v4i32, OpNode>;
1884 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1885 itin, OpcodeStr, !strconcat(Dt, "64"),
1886 v2i32, v2i64, OpNode>;
1889 // Neon Narrowing 2-register vector intrinsics,
1890 // source operand element sizes of 16, 32 and 64 bits:
1891 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1892 bits<5> op11_7, bit op6, bit op4,
1893 InstrItinClass itin, string OpcodeStr, string Dt,
1895 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1896 itin, OpcodeStr, !strconcat(Dt, "16"),
1897 v8i8, v8i16, IntOp>;
1898 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1899 itin, OpcodeStr, !strconcat(Dt, "32"),
1900 v4i16, v4i32, IntOp>;
1901 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1902 itin, OpcodeStr, !strconcat(Dt, "64"),
1903 v2i32, v2i64, IntOp>;
1907 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1908 // source operand element sizes of 16, 32 and 64 bits:
1909 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1910 string OpcodeStr, string Dt, SDNode OpNode> {
1911 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1912 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
1913 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1914 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1915 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1916 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1920 // Neon 3-register vector intrinsics.
1922 // First with only element sizes of 16 and 32 bits:
1923 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1924 InstrItinClass itinD16, InstrItinClass itinD32,
1925 InstrItinClass itinQ16, InstrItinClass itinQ32,
1926 string OpcodeStr, string Dt,
1927 Intrinsic IntOp, bit Commutable = 0> {
1928 // 64-bit vector types.
1929 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1930 OpcodeStr, !strconcat(Dt, "16"),
1931 v4i16, v4i16, IntOp, Commutable>;
1932 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1933 OpcodeStr, !strconcat(Dt, "32"),
1934 v2i32, v2i32, IntOp, Commutable>;
1936 // 128-bit vector types.
1937 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1938 OpcodeStr, !strconcat(Dt, "16"),
1939 v8i16, v8i16, IntOp, Commutable>;
1940 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1941 OpcodeStr, !strconcat(Dt, "32"),
1942 v4i32, v4i32, IntOp, Commutable>;
1945 multiclass N3VIntSL_HS<bits<4> op11_8,
1946 InstrItinClass itinD16, InstrItinClass itinD32,
1947 InstrItinClass itinQ16, InstrItinClass itinQ32,
1948 string OpcodeStr, string Dt, Intrinsic IntOp> {
1949 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1950 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1951 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1952 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1953 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1954 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1955 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1956 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1959 // ....then also with element size of 8 bits:
1960 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1961 InstrItinClass itinD16, InstrItinClass itinD32,
1962 InstrItinClass itinQ16, InstrItinClass itinQ32,
1963 string OpcodeStr, string Dt,
1964 Intrinsic IntOp, bit Commutable = 0>
1965 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1966 OpcodeStr, Dt, IntOp, Commutable> {
1967 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1968 OpcodeStr, !strconcat(Dt, "8"),
1969 v8i8, v8i8, IntOp, Commutable>;
1970 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1971 OpcodeStr, !strconcat(Dt, "8"),
1972 v16i8, v16i8, IntOp, Commutable>;
1975 // ....then also with element size of 64 bits:
1976 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1977 InstrItinClass itinD16, InstrItinClass itinD32,
1978 InstrItinClass itinQ16, InstrItinClass itinQ32,
1979 string OpcodeStr, string Dt,
1980 Intrinsic IntOp, bit Commutable = 0>
1981 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1982 OpcodeStr, Dt, IntOp, Commutable> {
1983 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1984 OpcodeStr, !strconcat(Dt, "64"),
1985 v1i64, v1i64, IntOp, Commutable>;
1986 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1987 OpcodeStr, !strconcat(Dt, "64"),
1988 v2i64, v2i64, IntOp, Commutable>;
1991 // Neon Narrowing 3-register vector intrinsics,
1992 // source operand element sizes of 16, 32 and 64 bits:
1993 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1994 string OpcodeStr, string Dt,
1995 Intrinsic IntOp, bit Commutable = 0> {
1996 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1997 OpcodeStr, !strconcat(Dt, "16"),
1998 v8i8, v8i16, IntOp, Commutable>;
1999 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2000 OpcodeStr, !strconcat(Dt, "32"),
2001 v4i16, v4i32, IntOp, Commutable>;
2002 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2003 OpcodeStr, !strconcat(Dt, "64"),
2004 v2i32, v2i64, IntOp, Commutable>;
2008 // Neon Long 3-register vector operations.
2010 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2011 InstrItinClass itin16, InstrItinClass itin32,
2012 string OpcodeStr, string Dt,
2013 SDNode OpNode, bit Commutable = 0> {
2014 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2015 OpcodeStr, !strconcat(Dt, "8"),
2016 v8i16, v8i8, OpNode, Commutable>;
2017 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2018 OpcodeStr, !strconcat(Dt, "16"),
2019 v4i32, v4i16, OpNode, Commutable>;
2020 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2021 OpcodeStr, !strconcat(Dt, "32"),
2022 v2i64, v2i32, OpNode, Commutable>;
2025 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2026 InstrItinClass itin, string OpcodeStr, string Dt,
2028 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2029 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2030 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2031 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2034 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2035 InstrItinClass itin16, InstrItinClass itin32,
2036 string OpcodeStr, string Dt,
2037 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2038 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2039 OpcodeStr, !strconcat(Dt, "8"),
2040 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2041 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2042 OpcodeStr, !strconcat(Dt, "16"),
2043 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2044 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2045 OpcodeStr, !strconcat(Dt, "32"),
2046 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2049 // Neon Long 3-register vector intrinsics.
2051 // First with only element sizes of 16 and 32 bits:
2052 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2053 InstrItinClass itin16, InstrItinClass itin32,
2054 string OpcodeStr, string Dt,
2055 Intrinsic IntOp, bit Commutable = 0> {
2056 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2057 OpcodeStr, !strconcat(Dt, "16"),
2058 v4i32, v4i16, IntOp, Commutable>;
2059 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2060 OpcodeStr, !strconcat(Dt, "32"),
2061 v2i64, v2i32, IntOp, Commutable>;
2064 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2065 InstrItinClass itin, string OpcodeStr, string Dt,
2067 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2068 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2069 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2070 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2073 // ....then also with element size of 8 bits:
2074 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2075 InstrItinClass itin16, InstrItinClass itin32,
2076 string OpcodeStr, string Dt,
2077 Intrinsic IntOp, bit Commutable = 0>
2078 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2079 IntOp, Commutable> {
2080 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2081 OpcodeStr, !strconcat(Dt, "8"),
2082 v8i16, v8i8, IntOp, Commutable>;
2085 // ....with explicit extend (VABDL).
2086 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2087 InstrItinClass itin, string OpcodeStr, string Dt,
2088 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2089 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2090 OpcodeStr, !strconcat(Dt, "8"),
2091 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2092 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2093 OpcodeStr, !strconcat(Dt, "16"),
2094 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2095 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2096 OpcodeStr, !strconcat(Dt, "32"),
2097 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2101 // Neon Wide 3-register vector intrinsics,
2102 // source operand element sizes of 8, 16 and 32 bits:
2103 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2104 string OpcodeStr, string Dt,
2105 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2106 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2107 OpcodeStr, !strconcat(Dt, "8"),
2108 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2109 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2110 OpcodeStr, !strconcat(Dt, "16"),
2111 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2112 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2113 OpcodeStr, !strconcat(Dt, "32"),
2114 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2118 // Neon Multiply-Op vector operations,
2119 // element sizes of 8, 16 and 32 bits:
2120 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2121 InstrItinClass itinD16, InstrItinClass itinD32,
2122 InstrItinClass itinQ16, InstrItinClass itinQ32,
2123 string OpcodeStr, string Dt, SDNode OpNode> {
2124 // 64-bit vector types.
2125 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2126 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2127 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2128 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2129 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2130 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2132 // 128-bit vector types.
2133 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2134 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2135 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2136 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2137 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2138 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2141 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2142 InstrItinClass itinD16, InstrItinClass itinD32,
2143 InstrItinClass itinQ16, InstrItinClass itinQ32,
2144 string OpcodeStr, string Dt, SDNode ShOp> {
2145 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2146 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2147 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2148 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2149 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2150 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2152 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2153 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2157 // Neon Intrinsic-Op vector operations,
2158 // element sizes of 8, 16 and 32 bits:
2159 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2160 InstrItinClass itinD, InstrItinClass itinQ,
2161 string OpcodeStr, string Dt, Intrinsic IntOp,
2163 // 64-bit vector types.
2164 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2165 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2166 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2167 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2168 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2169 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2171 // 128-bit vector types.
2172 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2173 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2174 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2175 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2176 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2177 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2180 // Neon 3-argument intrinsics,
2181 // element sizes of 8, 16 and 32 bits:
2182 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2183 InstrItinClass itinD, InstrItinClass itinQ,
2184 string OpcodeStr, string Dt, Intrinsic IntOp> {
2185 // 64-bit vector types.
2186 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2187 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2188 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2189 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2190 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2191 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2193 // 128-bit vector types.
2194 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2195 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2196 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2197 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2198 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2199 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2203 // Neon Long Multiply-Op vector operations,
2204 // element sizes of 8, 16 and 32 bits:
2205 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2206 InstrItinClass itin16, InstrItinClass itin32,
2207 string OpcodeStr, string Dt, SDNode MulOp,
2209 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2210 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2211 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2212 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2213 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2214 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2217 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2218 string Dt, SDNode MulOp, SDNode OpNode> {
2219 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2220 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2221 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2222 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2226 // Neon Long 3-argument intrinsics.
2228 // First with only element sizes of 16 and 32 bits:
2229 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2230 InstrItinClass itin16, InstrItinClass itin32,
2231 string OpcodeStr, string Dt, Intrinsic IntOp> {
2232 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2233 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2234 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2235 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2238 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2239 string OpcodeStr, string Dt, Intrinsic IntOp> {
2240 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2241 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2242 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2243 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2246 // ....then also with element size of 8 bits:
2247 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2248 InstrItinClass itin16, InstrItinClass itin32,
2249 string OpcodeStr, string Dt, Intrinsic IntOp>
2250 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2251 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2252 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2255 // ....with explicit extend (VABAL).
2256 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2257 InstrItinClass itin, string OpcodeStr, string Dt,
2258 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2259 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2260 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2261 IntOp, ExtOp, OpNode>;
2262 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2263 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2264 IntOp, ExtOp, OpNode>;
2265 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2266 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2267 IntOp, ExtOp, OpNode>;
2271 // Neon 2-register vector intrinsics,
2272 // element sizes of 8, 16 and 32 bits:
2273 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2274 bits<5> op11_7, bit op4,
2275 InstrItinClass itinD, InstrItinClass itinQ,
2276 string OpcodeStr, string Dt, Intrinsic IntOp> {
2277 // 64-bit vector types.
2278 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2279 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2280 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2281 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2282 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2283 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2285 // 128-bit vector types.
2286 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2287 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2288 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2289 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2290 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2291 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2295 // Neon Pairwise long 2-register intrinsics,
2296 // element sizes of 8, 16 and 32 bits:
2297 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2298 bits<5> op11_7, bit op4,
2299 string OpcodeStr, string Dt, Intrinsic IntOp> {
2300 // 64-bit vector types.
2301 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2302 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2303 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2304 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2305 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2306 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2308 // 128-bit vector types.
2309 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2310 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2311 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2312 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2313 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2314 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2318 // Neon Pairwise long 2-register accumulate intrinsics,
2319 // element sizes of 8, 16 and 32 bits:
2320 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2321 bits<5> op11_7, bit op4,
2322 string OpcodeStr, string Dt, Intrinsic IntOp> {
2323 // 64-bit vector types.
2324 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2325 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2326 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2327 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2328 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2329 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2331 // 128-bit vector types.
2332 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2333 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2334 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2335 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2336 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2337 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2341 // Neon 2-register vector shift by immediate,
2342 // with f of either N2RegVShLFrm or N2RegVShRFrm
2343 // element sizes of 8, 16, 32 and 64 bits:
2344 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2345 InstrItinClass itin, string OpcodeStr, string Dt,
2346 SDNode OpNode, Format f> {
2347 // 64-bit vector types.
2348 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2349 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2350 let Inst{21-19} = 0b001; // imm6 = 001xxx
2352 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2353 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2354 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2356 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2357 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2358 let Inst{21} = 0b1; // imm6 = 1xxxxx
2360 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2361 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2364 // 128-bit vector types.
2365 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2366 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2367 let Inst{21-19} = 0b001; // imm6 = 001xxx
2369 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2370 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2371 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2373 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2374 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2375 let Inst{21} = 0b1; // imm6 = 1xxxxx
2377 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2378 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2382 // Neon Shift-Accumulate vector operations,
2383 // element sizes of 8, 16, 32 and 64 bits:
2384 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2385 string OpcodeStr, string Dt, SDNode ShOp> {
2386 // 64-bit vector types.
2387 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2388 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2389 let Inst{21-19} = 0b001; // imm6 = 001xxx
2391 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2392 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2393 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2395 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2396 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2397 let Inst{21} = 0b1; // imm6 = 1xxxxx
2399 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2400 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2403 // 128-bit vector types.
2404 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2405 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2406 let Inst{21-19} = 0b001; // imm6 = 001xxx
2408 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2409 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2410 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2412 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2413 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2414 let Inst{21} = 0b1; // imm6 = 1xxxxx
2416 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2417 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2422 // Neon Shift-Insert vector operations,
2423 // with f of either N2RegVShLFrm or N2RegVShRFrm
2424 // element sizes of 8, 16, 32 and 64 bits:
2425 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2426 string OpcodeStr, SDNode ShOp,
2428 // 64-bit vector types.
2429 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2430 f, OpcodeStr, "8", v8i8, ShOp> {
2431 let Inst{21-19} = 0b001; // imm6 = 001xxx
2433 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2434 f, OpcodeStr, "16", v4i16, ShOp> {
2435 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2437 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2438 f, OpcodeStr, "32", v2i32, ShOp> {
2439 let Inst{21} = 0b1; // imm6 = 1xxxxx
2441 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2442 f, OpcodeStr, "64", v1i64, ShOp>;
2445 // 128-bit vector types.
2446 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2447 f, OpcodeStr, "8", v16i8, ShOp> {
2448 let Inst{21-19} = 0b001; // imm6 = 001xxx
2450 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2451 f, OpcodeStr, "16", v8i16, ShOp> {
2452 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2454 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2455 f, OpcodeStr, "32", v4i32, ShOp> {
2456 let Inst{21} = 0b1; // imm6 = 1xxxxx
2458 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2459 f, OpcodeStr, "64", v2i64, ShOp>;
2463 // Neon Shift Long operations,
2464 // element sizes of 8, 16, 32 bits:
2465 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2466 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2467 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2468 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2469 let Inst{21-19} = 0b001; // imm6 = 001xxx
2471 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2472 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2473 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2475 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2476 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2477 let Inst{21} = 0b1; // imm6 = 1xxxxx
2481 // Neon Shift Narrow operations,
2482 // element sizes of 16, 32, 64 bits:
2483 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2484 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2486 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2487 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2488 let Inst{21-19} = 0b001; // imm6 = 001xxx
2490 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2491 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2492 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2494 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2495 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2496 let Inst{21} = 0b1; // imm6 = 1xxxxx
2500 //===----------------------------------------------------------------------===//
2501 // Instruction Definitions.
2502 //===----------------------------------------------------------------------===//
2504 // Vector Add Operations.
2506 // VADD : Vector Add (integer and floating-point)
2507 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2509 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2510 v2f32, v2f32, fadd, 1>;
2511 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2512 v4f32, v4f32, fadd, 1>;
2513 // VADDL : Vector Add Long (Q = D + D)
2514 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2515 "vaddl", "s", add, sext, 1>;
2516 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2517 "vaddl", "u", add, zext, 1>;
2518 // VADDW : Vector Add Wide (Q = Q + D)
2519 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2520 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2521 // VHADD : Vector Halving Add
2522 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2523 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2524 "vhadd", "s", int_arm_neon_vhadds, 1>;
2525 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2526 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2527 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2528 // VRHADD : Vector Rounding Halving Add
2529 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2530 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2531 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2532 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2533 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2534 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2535 // VQADD : Vector Saturating Add
2536 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2537 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2538 "vqadd", "s", int_arm_neon_vqadds, 1>;
2539 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2540 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2541 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2542 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2543 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2544 int_arm_neon_vaddhn, 1>;
2545 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2546 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2547 int_arm_neon_vraddhn, 1>;
2549 // Vector Multiply Operations.
2551 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2552 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2553 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2554 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2555 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2556 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2557 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2558 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2559 v2f32, v2f32, fmul, 1>;
2560 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2561 v4f32, v4f32, fmul, 1>;
2562 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2563 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2564 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2567 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2568 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2569 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2570 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2571 (DSubReg_i16_reg imm:$lane))),
2572 (SubReg_i16_lane imm:$lane)))>;
2573 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2574 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2575 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2576 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2577 (DSubReg_i32_reg imm:$lane))),
2578 (SubReg_i32_lane imm:$lane)))>;
2579 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2580 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2581 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2582 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2583 (DSubReg_i32_reg imm:$lane))),
2584 (SubReg_i32_lane imm:$lane)))>;
2586 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2587 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2588 IIC_VMULi16Q, IIC_VMULi32Q,
2589 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2590 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2591 IIC_VMULi16Q, IIC_VMULi32Q,
2592 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2593 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2594 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2596 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2597 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2598 (DSubReg_i16_reg imm:$lane))),
2599 (SubReg_i16_lane imm:$lane)))>;
2600 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2601 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2603 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2604 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2605 (DSubReg_i32_reg imm:$lane))),
2606 (SubReg_i32_lane imm:$lane)))>;
2608 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2609 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2610 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2611 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2612 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2613 IIC_VMULi16Q, IIC_VMULi32Q,
2614 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2615 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2616 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2618 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2619 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2620 (DSubReg_i16_reg imm:$lane))),
2621 (SubReg_i16_lane imm:$lane)))>;
2622 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2623 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2625 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2626 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2627 (DSubReg_i32_reg imm:$lane))),
2628 (SubReg_i32_lane imm:$lane)))>;
2630 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2631 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2632 "vmull", "s", NEONvmulls, 1>;
2633 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2634 "vmull", "u", NEONvmullu, 1>;
2635 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2636 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2637 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
2638 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
2640 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2641 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2642 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2643 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2644 "vqdmull", "s", int_arm_neon_vqdmull>;
2646 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2648 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2649 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2650 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2651 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2653 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2655 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2656 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2657 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2659 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2660 v4f32, v2f32, fmul, fadd>;
2662 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2663 (mul (v8i16 QPR:$src2),
2664 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2665 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2666 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2667 (DSubReg_i16_reg imm:$lane))),
2668 (SubReg_i16_lane imm:$lane)))>;
2670 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2671 (mul (v4i32 QPR:$src2),
2672 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2673 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2674 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2675 (DSubReg_i32_reg imm:$lane))),
2676 (SubReg_i32_lane imm:$lane)))>;
2678 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2679 (fmul (v4f32 QPR:$src2),
2680 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2681 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2683 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2684 (DSubReg_i32_reg imm:$lane))),
2685 (SubReg_i32_lane imm:$lane)))>;
2687 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2688 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2689 "vmlal", "s", NEONvmulls, add>;
2690 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2691 "vmlal", "u", NEONvmullu, add>;
2693 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
2694 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
2696 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2697 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2698 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2699 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2701 // VMLS : Vector Multiply Subtract (integer and floating-point)
2702 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2703 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2704 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2706 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2708 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2709 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2710 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2712 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2713 v4f32, v2f32, fmul, fsub>;
2715 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2716 (mul (v8i16 QPR:$src2),
2717 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2718 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2719 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2720 (DSubReg_i16_reg imm:$lane))),
2721 (SubReg_i16_lane imm:$lane)))>;
2723 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2724 (mul (v4i32 QPR:$src2),
2725 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2726 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2727 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2728 (DSubReg_i32_reg imm:$lane))),
2729 (SubReg_i32_lane imm:$lane)))>;
2731 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2732 (fmul (v4f32 QPR:$src2),
2733 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2734 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2735 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2736 (DSubReg_i32_reg imm:$lane))),
2737 (SubReg_i32_lane imm:$lane)))>;
2739 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2740 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2741 "vmlsl", "s", NEONvmulls, sub>;
2742 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2743 "vmlsl", "u", NEONvmullu, sub>;
2745 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
2746 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
2748 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2749 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2750 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2751 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2753 // Vector Subtract Operations.
2755 // VSUB : Vector Subtract (integer and floating-point)
2756 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2757 "vsub", "i", sub, 0>;
2758 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2759 v2f32, v2f32, fsub, 0>;
2760 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2761 v4f32, v4f32, fsub, 0>;
2762 // VSUBL : Vector Subtract Long (Q = D - D)
2763 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2764 "vsubl", "s", sub, sext, 0>;
2765 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2766 "vsubl", "u", sub, zext, 0>;
2767 // VSUBW : Vector Subtract Wide (Q = Q - D)
2768 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
2769 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
2770 // VHSUB : Vector Halving Subtract
2771 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2772 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2773 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2774 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2775 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2776 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2777 // VQSUB : Vector Saturing Subtract
2778 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2779 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2780 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2781 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2782 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2783 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2784 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2785 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2786 int_arm_neon_vsubhn, 0>;
2787 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2788 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2789 int_arm_neon_vrsubhn, 0>;
2791 // Vector Comparisons.
2793 // VCEQ : Vector Compare Equal
2794 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2795 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2796 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2798 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2800 // For disassembly only.
2801 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2804 // VCGE : Vector Compare Greater Than or Equal
2805 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2806 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2807 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2808 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2809 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2811 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2813 // For disassembly only.
2814 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2816 // For disassembly only.
2817 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2820 // VCGT : Vector Compare Greater Than
2821 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2822 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2823 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2824 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2825 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2827 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2829 // For disassembly only.
2830 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2832 // For disassembly only.
2833 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2836 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2837 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2838 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2839 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2840 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2841 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2842 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2843 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2844 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2845 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2846 // VTST : Vector Test Bits
2847 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2848 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2850 // Vector Bitwise Operations.
2852 def vnotd : PatFrag<(ops node:$in),
2853 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
2854 def vnotq : PatFrag<(ops node:$in),
2855 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
2858 // VAND : Vector Bitwise AND
2859 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2860 v2i32, v2i32, and, 1>;
2861 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2862 v4i32, v4i32, and, 1>;
2864 // VEOR : Vector Bitwise Exclusive OR
2865 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2866 v2i32, v2i32, xor, 1>;
2867 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2868 v4i32, v4i32, xor, 1>;
2870 // VORR : Vector Bitwise OR
2871 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2872 v2i32, v2i32, or, 1>;
2873 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2874 v4i32, v4i32, or, 1>;
2876 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2877 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2878 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2879 "vbic", "$dst, $src1, $src2", "",
2880 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2881 (vnotd DPR:$src2))))]>;
2882 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2883 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2884 "vbic", "$dst, $src1, $src2", "",
2885 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2886 (vnotq QPR:$src2))))]>;
2888 // VORN : Vector Bitwise OR NOT
2889 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2890 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2891 "vorn", "$dst, $src1, $src2", "",
2892 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2893 (vnotd DPR:$src2))))]>;
2894 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2895 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2896 "vorn", "$dst, $src1, $src2", "",
2897 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2898 (vnotq QPR:$src2))))]>;
2900 // VMVN : Vector Bitwise NOT (Immediate)
2902 let isReMaterializable = 1 in {
2903 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
2904 (ins nModImm:$SIMM), IIC_VMOVImm,
2905 "vmvn", "i16", "$dst, $SIMM", "",
2906 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]>;
2907 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
2908 (ins nModImm:$SIMM), IIC_VMOVImm,
2909 "vmvn", "i16", "$dst, $SIMM", "",
2910 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]>;
2912 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
2913 (ins nModImm:$SIMM), IIC_VMOVImm,
2914 "vmvn", "i32", "$dst, $SIMM", "",
2915 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]>;
2916 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
2917 (ins nModImm:$SIMM), IIC_VMOVImm,
2918 "vmvn", "i32", "$dst, $SIMM", "",
2919 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]>;
2922 // VMVN : Vector Bitwise NOT
2923 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2924 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2925 "vmvn", "$dst, $src", "",
2926 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
2927 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2928 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2929 "vmvn", "$dst, $src", "",
2930 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
2931 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
2932 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
2934 // VBSL : Vector Bitwise Select
2935 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2936 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2937 N3RegFrm, IIC_VCNTiD,
2938 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2940 (v2i32 (or (and DPR:$src2, DPR:$src1),
2941 (and DPR:$src3, (vnotd DPR:$src1)))))]>;
2942 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2943 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2944 N3RegFrm, IIC_VCNTiQ,
2945 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2947 (v4i32 (or (and QPR:$src2, QPR:$src1),
2948 (and QPR:$src3, (vnotq QPR:$src1)))))]>;
2950 // VBIF : Vector Bitwise Insert if False
2951 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2952 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2953 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2954 N3RegFrm, IIC_VBINiD,
2955 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2956 [/* For disassembly only; pattern left blank */]>;
2957 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2958 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2959 N3RegFrm, IIC_VBINiQ,
2960 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2961 [/* For disassembly only; pattern left blank */]>;
2963 // VBIT : Vector Bitwise Insert if True
2964 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2965 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2966 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2967 N3RegFrm, IIC_VBINiD,
2968 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2969 [/* For disassembly only; pattern left blank */]>;
2970 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2971 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2972 N3RegFrm, IIC_VBINiQ,
2973 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2974 [/* For disassembly only; pattern left blank */]>;
2976 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2977 // for equivalent operations with different register constraints; it just
2980 // Vector Absolute Differences.
2982 // VABD : Vector Absolute Difference
2983 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2984 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2985 "vabd", "s", int_arm_neon_vabds, 1>;
2986 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2987 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2988 "vabd", "u", int_arm_neon_vabdu, 1>;
2989 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2990 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
2991 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2992 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
2994 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2995 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
2996 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
2997 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
2998 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3000 // VABA : Vector Absolute Difference and Accumulate
3001 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3002 "vaba", "s", int_arm_neon_vabds, add>;
3003 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3004 "vaba", "u", int_arm_neon_vabdu, add>;
3006 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3007 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3008 "vabal", "s", int_arm_neon_vabds, zext, add>;
3009 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3010 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3012 // Vector Maximum and Minimum.
3014 // VMAX : Vector Maximum
3015 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3016 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3017 "vmax", "s", int_arm_neon_vmaxs, 1>;
3018 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3019 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3020 "vmax", "u", int_arm_neon_vmaxu, 1>;
3021 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3023 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3024 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3026 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3028 // VMIN : Vector Minimum
3029 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3030 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3031 "vmin", "s", int_arm_neon_vmins, 1>;
3032 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3033 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3034 "vmin", "u", int_arm_neon_vminu, 1>;
3035 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3037 v2f32, v2f32, int_arm_neon_vmins, 1>;
3038 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3040 v4f32, v4f32, int_arm_neon_vmins, 1>;
3042 // Vector Pairwise Operations.
3044 // VPADD : Vector Pairwise Add
3045 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3047 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3048 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3050 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3051 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3053 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3054 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3055 IIC_VBIND, "vpadd", "f32",
3056 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3058 // VPADDL : Vector Pairwise Add Long
3059 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3060 int_arm_neon_vpaddls>;
3061 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3062 int_arm_neon_vpaddlu>;
3064 // VPADAL : Vector Pairwise Add and Accumulate Long
3065 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3066 int_arm_neon_vpadals>;
3067 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3068 int_arm_neon_vpadalu>;
3070 // VPMAX : Vector Pairwise Maximum
3071 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3072 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3073 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3074 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3075 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3076 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3077 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3078 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3079 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3080 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3081 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3082 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3083 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3084 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3086 // VPMIN : Vector Pairwise Minimum
3087 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3088 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3089 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3090 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3091 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3092 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3093 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3094 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3095 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3096 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3097 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3098 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3099 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
3100 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3102 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3104 // VRECPE : Vector Reciprocal Estimate
3105 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3106 IIC_VUNAD, "vrecpe", "u32",
3107 v2i32, v2i32, int_arm_neon_vrecpe>;
3108 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3109 IIC_VUNAQ, "vrecpe", "u32",
3110 v4i32, v4i32, int_arm_neon_vrecpe>;
3111 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3112 IIC_VUNAD, "vrecpe", "f32",
3113 v2f32, v2f32, int_arm_neon_vrecpe>;
3114 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3115 IIC_VUNAQ, "vrecpe", "f32",
3116 v4f32, v4f32, int_arm_neon_vrecpe>;
3118 // VRECPS : Vector Reciprocal Step
3119 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3120 IIC_VRECSD, "vrecps", "f32",
3121 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3122 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3123 IIC_VRECSQ, "vrecps", "f32",
3124 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3126 // VRSQRTE : Vector Reciprocal Square Root Estimate
3127 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3128 IIC_VUNAD, "vrsqrte", "u32",
3129 v2i32, v2i32, int_arm_neon_vrsqrte>;
3130 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3131 IIC_VUNAQ, "vrsqrte", "u32",
3132 v4i32, v4i32, int_arm_neon_vrsqrte>;
3133 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3134 IIC_VUNAD, "vrsqrte", "f32",
3135 v2f32, v2f32, int_arm_neon_vrsqrte>;
3136 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3137 IIC_VUNAQ, "vrsqrte", "f32",
3138 v4f32, v4f32, int_arm_neon_vrsqrte>;
3140 // VRSQRTS : Vector Reciprocal Square Root Step
3141 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3142 IIC_VRECSD, "vrsqrts", "f32",
3143 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3144 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3145 IIC_VRECSQ, "vrsqrts", "f32",
3146 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3150 // VSHL : Vector Shift
3151 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
3152 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3153 "vshl", "s", int_arm_neon_vshifts, 0>;
3154 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
3155 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3156 "vshl", "u", int_arm_neon_vshiftu, 0>;
3157 // VSHL : Vector Shift Left (Immediate)
3158 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3160 // VSHR : Vector Shift Right (Immediate)
3161 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3163 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3166 // VSHLL : Vector Shift Left Long
3167 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3168 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3170 // VSHLL : Vector Shift Left Long (with maximum shift count)
3171 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3172 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3173 ValueType OpTy, SDNode OpNode>
3174 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3175 ResTy, OpTy, OpNode> {
3176 let Inst{21-16} = op21_16;
3178 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3179 v8i16, v8i8, NEONvshlli>;
3180 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3181 v4i32, v4i16, NEONvshlli>;
3182 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3183 v2i64, v2i32, NEONvshlli>;
3185 // VSHRN : Vector Shift Right and Narrow
3186 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3189 // VRSHL : Vector Rounding Shift
3190 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
3191 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3192 "vrshl", "s", int_arm_neon_vrshifts, 0>;
3193 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
3194 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3195 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
3196 // VRSHR : Vector Rounding Shift Right
3197 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3199 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3202 // VRSHRN : Vector Rounding Shift Right and Narrow
3203 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3206 // VQSHL : Vector Saturating Shift
3207 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
3208 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3209 "vqshl", "s", int_arm_neon_vqshifts, 0>;
3210 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
3211 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3212 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
3213 // VQSHL : Vector Saturating Shift Left (Immediate)
3214 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3216 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3218 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3219 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3222 // VQSHRN : Vector Saturating Shift Right and Narrow
3223 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3225 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3228 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3229 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3232 // VQRSHL : Vector Saturating Rounding Shift
3233 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
3234 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3235 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
3236 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
3237 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3238 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
3240 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3241 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3243 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3246 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3247 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3250 // VSRA : Vector Shift Right and Accumulate
3251 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3252 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3253 // VRSRA : Vector Rounding Shift Right and Accumulate
3254 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3255 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3257 // VSLI : Vector Shift Left and Insert
3258 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3259 // VSRI : Vector Shift Right and Insert
3260 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3262 // Vector Absolute and Saturating Absolute.
3264 // VABS : Vector Absolute Value
3265 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3266 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3268 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3269 IIC_VUNAD, "vabs", "f32",
3270 v2f32, v2f32, int_arm_neon_vabs>;
3271 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3272 IIC_VUNAQ, "vabs", "f32",
3273 v4f32, v4f32, int_arm_neon_vabs>;
3275 // VQABS : Vector Saturating Absolute Value
3276 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3277 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3278 int_arm_neon_vqabs>;
3282 def vnegd : PatFrag<(ops node:$in),
3283 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3284 def vnegq : PatFrag<(ops node:$in),
3285 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3287 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3288 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3289 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3290 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3291 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3292 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3293 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3294 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3296 // VNEG : Vector Negate (integer)
3297 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3298 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3299 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3300 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3301 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3302 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3304 // VNEG : Vector Negate (floating-point)
3305 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3306 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3307 "vneg", "f32", "$dst, $src", "",
3308 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3309 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3310 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3311 "vneg", "f32", "$dst, $src", "",
3312 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3314 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3315 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3316 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3317 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3318 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3319 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3321 // VQNEG : Vector Saturating Negate
3322 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3323 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3324 int_arm_neon_vqneg>;
3326 // Vector Bit Counting Operations.
3328 // VCLS : Vector Count Leading Sign Bits
3329 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3330 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3332 // VCLZ : Vector Count Leading Zeros
3333 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3334 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3336 // VCNT : Vector Count One Bits
3337 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3338 IIC_VCNTiD, "vcnt", "8",
3339 v8i8, v8i8, int_arm_neon_vcnt>;
3340 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3341 IIC_VCNTiQ, "vcnt", "8",
3342 v16i8, v16i8, int_arm_neon_vcnt>;
3344 // Vector Swap -- for disassembly only.
3345 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3346 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3347 "vswp", "$dst, $src", "", []>;
3348 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3349 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3350 "vswp", "$dst, $src", "", []>;
3352 // Vector Move Operations.
3354 // VMOV : Vector Move (Register)
3356 let neverHasSideEffects = 1 in {
3357 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3358 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3359 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3360 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3362 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3363 // be expanded after register allocation is completed.
3364 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3365 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3367 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3368 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3369 } // neverHasSideEffects
3371 // VMOV : Vector Move (Immediate)
3373 let isReMaterializable = 1 in {
3374 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3375 (ins nModImm:$SIMM), IIC_VMOVImm,
3376 "vmov", "i8", "$dst, $SIMM", "",
3377 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3378 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3379 (ins nModImm:$SIMM), IIC_VMOVImm,
3380 "vmov", "i8", "$dst, $SIMM", "",
3381 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3383 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3384 (ins nModImm:$SIMM), IIC_VMOVImm,
3385 "vmov", "i16", "$dst, $SIMM", "",
3386 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]>;
3387 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3388 (ins nModImm:$SIMM), IIC_VMOVImm,
3389 "vmov", "i16", "$dst, $SIMM", "",
3390 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]>;
3392 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3393 (ins nModImm:$SIMM), IIC_VMOVImm,
3394 "vmov", "i32", "$dst, $SIMM", "",
3395 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]>;
3396 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3397 (ins nModImm:$SIMM), IIC_VMOVImm,
3398 "vmov", "i32", "$dst, $SIMM", "",
3399 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]>;
3401 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3402 (ins nModImm:$SIMM), IIC_VMOVImm,
3403 "vmov", "i64", "$dst, $SIMM", "",
3404 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3405 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3406 (ins nModImm:$SIMM), IIC_VMOVImm,
3407 "vmov", "i64", "$dst, $SIMM", "",
3408 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3409 } // isReMaterializable
3411 // VMOV : Vector Get Lane (move scalar to ARM core register)
3413 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3414 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3415 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
3416 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
3418 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3419 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3420 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
3421 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
3423 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3424 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3425 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
3426 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
3428 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3429 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3430 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
3431 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
3433 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3434 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3435 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
3436 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
3438 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3439 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3440 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3441 (DSubReg_i8_reg imm:$lane))),
3442 (SubReg_i8_lane imm:$lane))>;
3443 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3444 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3445 (DSubReg_i16_reg imm:$lane))),
3446 (SubReg_i16_lane imm:$lane))>;
3447 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3448 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3449 (DSubReg_i8_reg imm:$lane))),
3450 (SubReg_i8_lane imm:$lane))>;
3451 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3452 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3453 (DSubReg_i16_reg imm:$lane))),
3454 (SubReg_i16_lane imm:$lane))>;
3455 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3456 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3457 (DSubReg_i32_reg imm:$lane))),
3458 (SubReg_i32_lane imm:$lane))>;
3459 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3460 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3461 (SSubReg_f32_reg imm:$src2))>;
3462 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3463 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3464 (SSubReg_f32_reg imm:$src2))>;
3465 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3466 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3467 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3468 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3471 // VMOV : Vector Set Lane (move ARM core register to scalar)
3473 let Constraints = "$src1 = $dst" in {
3474 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
3475 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3476 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
3477 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
3478 GPR:$src2, imm:$lane))]>;
3479 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
3480 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3481 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
3482 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
3483 GPR:$src2, imm:$lane))]>;
3484 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
3485 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3486 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
3487 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
3488 GPR:$src2, imm:$lane))]>;
3490 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3491 (v16i8 (INSERT_SUBREG QPR:$src1,
3492 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3493 (DSubReg_i8_reg imm:$lane))),
3494 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3495 (DSubReg_i8_reg imm:$lane)))>;
3496 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3497 (v8i16 (INSERT_SUBREG QPR:$src1,
3498 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3499 (DSubReg_i16_reg imm:$lane))),
3500 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3501 (DSubReg_i16_reg imm:$lane)))>;
3502 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3503 (v4i32 (INSERT_SUBREG QPR:$src1,
3504 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3505 (DSubReg_i32_reg imm:$lane))),
3506 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3507 (DSubReg_i32_reg imm:$lane)))>;
3509 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3510 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3511 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3512 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3513 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3514 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3516 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3517 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3518 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3519 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3521 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3522 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3523 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3524 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3525 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3526 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3528 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3529 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3530 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3531 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3532 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3533 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3535 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3536 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3537 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3539 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3540 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3541 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3543 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3544 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3545 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3548 // VDUP : Vector Duplicate (from ARM core register to all elements)
3550 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3551 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3552 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3553 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3554 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3555 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3556 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3557 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3559 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3560 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3561 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3562 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3563 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3564 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3566 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3567 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3568 [(set DPR:$dst, (v2f32 (NEONvdup
3569 (f32 (bitconvert GPR:$src)))))]>;
3570 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3571 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3572 [(set QPR:$dst, (v4f32 (NEONvdup
3573 (f32 (bitconvert GPR:$src)))))]>;
3575 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3577 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3579 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3580 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3581 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3583 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3584 ValueType ResTy, ValueType OpTy>
3585 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3586 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3587 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3590 // Inst{19-16} is partially specified depending on the element size.
3592 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3593 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3594 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3595 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3596 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3597 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3598 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3599 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3601 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3602 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3603 (DSubReg_i8_reg imm:$lane))),
3604 (SubReg_i8_lane imm:$lane)))>;
3605 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3606 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3607 (DSubReg_i16_reg imm:$lane))),
3608 (SubReg_i16_lane imm:$lane)))>;
3609 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3610 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3611 (DSubReg_i32_reg imm:$lane))),
3612 (SubReg_i32_lane imm:$lane)))>;
3613 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3614 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3615 (DSubReg_i32_reg imm:$lane))),
3616 (SubReg_i32_lane imm:$lane)))>;
3618 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3619 (outs DPR:$dst), (ins SPR:$src),
3620 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3621 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3623 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3624 (outs QPR:$dst), (ins SPR:$src),
3625 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3626 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3628 // VMOVN : Vector Narrowing Move
3629 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3630 "vmovn", "i", trunc>;
3631 // VQMOVN : Vector Saturating Narrowing Move
3632 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3633 "vqmovn", "s", int_arm_neon_vqmovns>;
3634 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3635 "vqmovn", "u", int_arm_neon_vqmovnu>;
3636 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3637 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3638 // VMOVL : Vector Lengthening Move
3639 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3640 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3642 // Vector Conversions.
3644 // VCVT : Vector Convert Between Floating-Point and Integers
3645 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3646 v2i32, v2f32, fp_to_sint>;
3647 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3648 v2i32, v2f32, fp_to_uint>;
3649 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3650 v2f32, v2i32, sint_to_fp>;
3651 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3652 v2f32, v2i32, uint_to_fp>;
3654 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3655 v4i32, v4f32, fp_to_sint>;
3656 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3657 v4i32, v4f32, fp_to_uint>;
3658 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3659 v4f32, v4i32, sint_to_fp>;
3660 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3661 v4f32, v4i32, uint_to_fp>;
3663 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3664 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3665 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3666 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3667 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3668 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3669 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3670 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3671 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3673 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3674 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3675 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3676 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3677 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3678 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3679 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3680 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3684 // VREV64 : Vector Reverse elements within 64-bit doublewords
3686 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3687 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3688 (ins DPR:$src), IIC_VMOVD,
3689 OpcodeStr, Dt, "$dst, $src", "",
3690 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3691 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3692 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3693 (ins QPR:$src), IIC_VMOVD,
3694 OpcodeStr, Dt, "$dst, $src", "",
3695 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3697 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3698 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3699 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3700 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3702 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3703 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3704 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3705 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3707 // VREV32 : Vector Reverse elements within 32-bit words
3709 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3710 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3711 (ins DPR:$src), IIC_VMOVD,
3712 OpcodeStr, Dt, "$dst, $src", "",
3713 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3714 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3715 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3716 (ins QPR:$src), IIC_VMOVD,
3717 OpcodeStr, Dt, "$dst, $src", "",
3718 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3720 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3721 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3723 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3724 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3726 // VREV16 : Vector Reverse elements within 16-bit halfwords
3728 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3729 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3730 (ins DPR:$src), IIC_VMOVD,
3731 OpcodeStr, Dt, "$dst, $src", "",
3732 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3733 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3734 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3735 (ins QPR:$src), IIC_VMOVD,
3736 OpcodeStr, Dt, "$dst, $src", "",
3737 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3739 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3740 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3742 // Other Vector Shuffles.
3744 // VEXT : Vector Extract
3746 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3747 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3748 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3749 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3750 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3751 (Ty DPR:$rhs), imm:$index)))]>;
3753 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3754 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3755 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3756 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3757 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3758 (Ty QPR:$rhs), imm:$index)))]>;
3760 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3761 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3762 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3763 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3765 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3766 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3767 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3768 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3770 // VTRN : Vector Transpose
3772 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3773 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3774 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3776 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3777 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3778 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3780 // VUZP : Vector Unzip (Deinterleave)
3782 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3783 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3784 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3786 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3787 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3788 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3790 // VZIP : Vector Zip (Interleave)
3792 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3793 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3794 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3796 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3797 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3798 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3800 // Vector Table Lookup and Table Extension.
3802 // VTBL : Vector Table Lookup
3804 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3805 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3806 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3807 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3808 let hasExtraSrcRegAllocReq = 1 in {
3810 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3811 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3812 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "", []>;
3814 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3815 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3816 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "", []>;
3818 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3819 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3821 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "", []>;
3822 } // hasExtraSrcRegAllocReq = 1
3825 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "">;
3827 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "">;
3829 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "">;
3831 // VTBX : Vector Table Extension
3833 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3834 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3835 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3836 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3837 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3838 let hasExtraSrcRegAllocReq = 1 in {
3840 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3841 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3842 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst", []>;
3844 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3845 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3846 NVTBLFrm, IIC_VTBX3,
3847 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src",
3848 "$orig = $dst", []>;
3850 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3851 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3852 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3853 "$orig = $dst", []>;
3854 } // hasExtraSrcRegAllocReq = 1
3857 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
3858 IIC_VTBX2, "$orig = $dst">;
3860 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
3861 IIC_VTBX3, "$orig = $dst">;
3863 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
3864 IIC_VTBX4, "$orig = $dst">;
3866 //===----------------------------------------------------------------------===//
3867 // NEON instructions for single-precision FP math
3868 //===----------------------------------------------------------------------===//
3870 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3871 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3872 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3876 class N3VSPat<SDNode OpNode, NeonI Inst>
3877 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3878 (EXTRACT_SUBREG (v2f32
3879 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3881 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3885 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3886 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3887 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3889 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3891 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3895 // These need separate instructions because they must use DPR_VFP2 register
3896 // class which have SPR sub-registers.
3898 // Vector Add Operations used for single-precision FP
3899 let neverHasSideEffects = 1 in
3900 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3901 def : N3VSPat<fadd, VADDfd_sfp>;
3903 // Vector Sub Operations used for single-precision FP
3904 let neverHasSideEffects = 1 in
3905 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3906 def : N3VSPat<fsub, VSUBfd_sfp>;
3908 // Vector Multiply Operations used for single-precision FP
3909 let neverHasSideEffects = 1 in
3910 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3911 def : N3VSPat<fmul, VMULfd_sfp>;
3913 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3914 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3915 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3917 //let neverHasSideEffects = 1 in
3918 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3919 // v2f32, fmul, fadd>;
3920 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3922 //let neverHasSideEffects = 1 in
3923 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3924 // v2f32, fmul, fsub>;
3925 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3927 // Vector Absolute used for single-precision FP
3928 let neverHasSideEffects = 1 in
3929 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3930 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3931 "vabs", "f32", "$dst, $src", "", []>;
3932 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3934 // Vector Negate used for single-precision FP
3935 let neverHasSideEffects = 1 in
3936 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3937 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3938 "vneg", "f32", "$dst, $src", "", []>;
3939 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3941 // Vector Maximum used for single-precision FP
3942 let neverHasSideEffects = 1 in
3943 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3944 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3945 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3946 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3948 // Vector Minimum used for single-precision FP
3949 let neverHasSideEffects = 1 in
3950 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3951 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3952 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3953 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3955 // Vector Convert between single-precision FP and integer
3956 let neverHasSideEffects = 1 in
3957 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3958 v2i32, v2f32, fp_to_sint>;
3959 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3961 let neverHasSideEffects = 1 in
3962 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3963 v2i32, v2f32, fp_to_uint>;
3964 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3966 let neverHasSideEffects = 1 in
3967 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3968 v2f32, v2i32, sint_to_fp>;
3969 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3971 let neverHasSideEffects = 1 in
3972 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3973 v2f32, v2i32, uint_to_fp>;
3974 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3976 //===----------------------------------------------------------------------===//
3977 // Non-Instruction Patterns
3978 //===----------------------------------------------------------------------===//
3981 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3982 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3983 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3984 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3985 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3986 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3987 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3988 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3989 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3990 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3991 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3992 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3993 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3994 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3995 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3996 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3997 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3998 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3999 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4000 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4001 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4002 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4003 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4004 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4005 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4006 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4007 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4008 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4009 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4010 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4012 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4013 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4014 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4015 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4016 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4017 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4018 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4019 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4020 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4021 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4022 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4023 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4024 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4025 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4026 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4027 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4028 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4029 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4030 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4031 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4032 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4033 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4034 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4035 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4036 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4037 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4038 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4039 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4040 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4041 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;