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 : AXDI5<(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:$addr)))]>;
141 let mayLoad = 1, neverHasSideEffects = 1 in {
142 // Use vld1 to load a Q register as a D register pair.
143 // This alternative to VLDMQ allows an alignment to be specified.
144 // This is equivalent to VLD1q64 except that it has a Q register operand.
146 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
147 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
148 } // mayLoad = 1, neverHasSideEffects = 1
150 // Use vstmia to store a Q register as a D register pair.
151 // This is equivalent to VSTMD except that it has a Q register operand
152 // instead of a pair of D registers.
154 : AXDI5<(outs), (ins QPR:$src, addrmode4:$addr, pred:$p),
155 IndexModeNone, IIC_fpStorem,
156 "vstm${addr:submode}${p}\t$addr, ${src:dregpair}", "",
157 [(store (v2f64 QPR:$src), addrmode4:$addr)]>;
159 let mayStore = 1, neverHasSideEffects = 1 in {
160 // Use vst1 to store a Q register as a D register pair.
161 // This alternative to VSTMQ allows an alignment to be specified.
162 // This is equivalent to VST1q64 except that it has a Q register operand.
164 : NLdSt<0,0b00,0b1010,0b1100, (outs), (ins addrmode6:$addr, QPR:$src),
165 IIC_VST, "vst1", "64", "${src:dregpair}, $addr", "", []>;
166 } // mayStore = 1, neverHasSideEffects = 1
168 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
170 // Classes for VLD* pseudo-instructions with multi-register operands.
171 // These are expanded to real instructions after register allocation.
173 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), IIC_VST, "">;
175 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
176 (ins addrmode6:$addr, am6offset:$offset), IIC_VST,
179 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), IIC_VST, "">;
181 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
182 (ins addrmode6:$addr, am6offset:$offset), IIC_VST,
185 // VLD1 : Vector Load (multiple single elements)
186 class VLD1D<bits<4> op7_4, string Dt>
187 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
188 (ins addrmode6:$addr), IIC_VLD1,
189 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
190 class VLD1Q<bits<4> op7_4, string Dt>
191 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
192 (ins addrmode6:$addr), IIC_VLD1,
193 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
195 def VLD1d8 : VLD1D<0b0000, "8">;
196 def VLD1d16 : VLD1D<0b0100, "16">;
197 def VLD1d32 : VLD1D<0b1000, "32">;
198 def VLD1d64 : VLD1D<0b1100, "64">;
200 def VLD1q8 : VLD1Q<0b0000, "8">;
201 def VLD1q16 : VLD1Q<0b0100, "16">;
202 def VLD1q32 : VLD1Q<0b1000, "32">;
203 def VLD1q64 : VLD1Q<0b1100, "64">;
205 def VLD1q8Pseudo : VLDQPseudo;
206 def VLD1q16Pseudo : VLDQPseudo;
207 def VLD1q32Pseudo : VLDQPseudo;
208 def VLD1q64Pseudo : VLDQPseudo;
210 // ...with address register writeback:
211 class VLD1DWB<bits<4> op7_4, string Dt>
212 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
213 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
214 "vld1", Dt, "\\{$dst\\}, $addr$offset",
215 "$addr.addr = $wb", []>;
216 class VLD1QWB<bits<4> op7_4, string Dt>
217 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
218 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
219 "vld1", Dt, "${dst:dregpair}, $addr$offset",
220 "$addr.addr = $wb", []>;
222 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
223 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
224 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
225 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
227 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
228 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
229 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
230 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
232 def VLD1q8Pseudo_UPD : VLDQWBPseudo;
233 def VLD1q16Pseudo_UPD : VLDQWBPseudo;
234 def VLD1q32Pseudo_UPD : VLDQWBPseudo;
235 def VLD1q64Pseudo_UPD : VLDQWBPseudo;
237 // ...with 3 registers (some of these are only for the disassembler):
238 class VLD1D3<bits<4> op7_4, string Dt>
239 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
240 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
241 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
242 class VLD1D3WB<bits<4> op7_4, string Dt>
243 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
244 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
245 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
247 def VLD1d8T : VLD1D3<0b0000, "8">;
248 def VLD1d16T : VLD1D3<0b0100, "16">;
249 def VLD1d32T : VLD1D3<0b1000, "32">;
250 def VLD1d64T : VLD1D3<0b1100, "64">;
252 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
253 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
254 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
255 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
257 def VLD1d64TPseudo : VLDQQPseudo;
258 def VLD1d64TPseudo_UPD : VLDQQWBPseudo;
260 // ...with 4 registers (some of these are only for the disassembler):
261 class VLD1D4<bits<4> op7_4, string Dt>
262 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
263 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
264 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
265 class VLD1D4WB<bits<4> op7_4, string Dt>
266 : NLdSt<0,0b10,0b0010,op7_4,
267 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
268 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
269 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
272 def VLD1d8Q : VLD1D4<0b0000, "8">;
273 def VLD1d16Q : VLD1D4<0b0100, "16">;
274 def VLD1d32Q : VLD1D4<0b1000, "32">;
275 def VLD1d64Q : VLD1D4<0b1100, "64">;
277 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
278 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
279 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
280 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
282 def VLD1d64QPseudo : VLDQQPseudo;
283 def VLD1d64QPseudo_UPD : VLDQQWBPseudo;
285 // VLD2 : Vector Load (multiple 2-element structures)
286 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
287 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
288 (ins addrmode6:$addr), IIC_VLD2,
289 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
290 class VLD2Q<bits<4> op7_4, string Dt>
291 : NLdSt<0, 0b10, 0b0011, op7_4,
292 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
293 (ins addrmode6:$addr), IIC_VLD2,
294 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
296 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
297 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
298 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
300 def VLD2q8 : VLD2Q<0b0000, "8">;
301 def VLD2q16 : VLD2Q<0b0100, "16">;
302 def VLD2q32 : VLD2Q<0b1000, "32">;
304 def VLD2d8Pseudo : VLDQPseudo;
305 def VLD2d16Pseudo : VLDQPseudo;
306 def VLD2d32Pseudo : VLDQPseudo;
308 def VLD2q8Pseudo : VLDQQPseudo;
309 def VLD2q16Pseudo : VLDQQPseudo;
310 def VLD2q32Pseudo : VLDQQPseudo;
312 // ...with address register writeback:
313 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
314 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
315 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
316 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
317 "$addr.addr = $wb", []>;
318 class VLD2QWB<bits<4> op7_4, string Dt>
319 : NLdSt<0, 0b10, 0b0011, op7_4,
320 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
321 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
322 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
323 "$addr.addr = $wb", []>;
325 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
326 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
327 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
329 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
330 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
331 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
333 def VLD2d8Pseudo_UPD : VLDQWBPseudo;
334 def VLD2d16Pseudo_UPD : VLDQWBPseudo;
335 def VLD2d32Pseudo_UPD : VLDQWBPseudo;
337 def VLD2q8Pseudo_UPD : VLDQQWBPseudo;
338 def VLD2q16Pseudo_UPD : VLDQQWBPseudo;
339 def VLD2q32Pseudo_UPD : VLDQQWBPseudo;
341 // ...with double-spaced registers (for disassembly only):
342 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
343 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
344 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
345 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
346 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
347 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
349 // VLD3 : Vector Load (multiple 3-element structures)
350 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
351 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
352 (ins addrmode6:$addr), IIC_VLD3,
353 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
355 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
356 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
357 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
359 // ...with address register writeback:
360 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
361 : NLdSt<0, 0b10, op11_8, op7_4,
362 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
363 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
364 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
365 "$addr.addr = $wb", []>;
367 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
368 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
369 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
371 // ...with double-spaced registers (non-updating versions for disassembly only):
372 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
373 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
374 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
375 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
376 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
377 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
379 // ...alternate versions to be allocated odd register numbers:
380 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
381 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
382 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
384 // VLD4 : Vector Load (multiple 4-element structures)
385 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
386 : NLdSt<0, 0b10, op11_8, op7_4,
387 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
388 (ins addrmode6:$addr), IIC_VLD4,
389 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
391 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
392 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
393 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
395 // ...with address register writeback:
396 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
397 : NLdSt<0, 0b10, op11_8, op7_4,
398 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
399 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
400 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
401 "$addr.addr = $wb", []>;
403 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
404 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
405 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
407 // ...with double-spaced registers (non-updating versions for disassembly only):
408 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
409 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
410 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
411 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
412 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
413 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
415 // ...alternate versions to be allocated odd register numbers:
416 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
417 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
418 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
420 // VLD1LN : Vector Load (single element to one lane)
421 // FIXME: Not yet implemented.
423 // VLD2LN : Vector Load (single 2-element structure to one lane)
424 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
425 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
426 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
427 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
428 "$src1 = $dst1, $src2 = $dst2", []>;
430 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
431 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
432 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
434 // ...with double-spaced registers:
435 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
436 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
438 // ...alternate versions to be allocated odd register numbers:
439 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
440 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
442 // ...with address register writeback:
443 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
444 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
445 (ins addrmode6:$addr, am6offset:$offset,
446 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
447 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
448 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
450 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
451 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
452 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
454 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
455 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
457 // VLD3LN : Vector Load (single 3-element structure to one lane)
458 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
459 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
460 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
461 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
462 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
463 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
465 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
466 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
467 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
469 // ...with double-spaced registers:
470 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
471 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
473 // ...alternate versions to be allocated odd register numbers:
474 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
475 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
477 // ...with address register writeback:
478 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
479 : NLdSt<1, 0b10, op11_8, op7_4,
480 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
481 (ins addrmode6:$addr, am6offset:$offset,
482 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
483 IIC_VLD3, "vld3", Dt,
484 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
485 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
488 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
489 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
490 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
492 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
493 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
495 // VLD4LN : Vector Load (single 4-element structure to one lane)
496 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
497 : NLdSt<1, 0b10, op11_8, op7_4,
498 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
499 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
500 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
501 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
502 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
504 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
505 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
506 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
508 // ...with double-spaced registers:
509 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
510 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
512 // ...alternate versions to be allocated odd register numbers:
513 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
514 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
516 // ...with address register writeback:
517 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
518 : NLdSt<1, 0b10, op11_8, op7_4,
519 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
520 (ins addrmode6:$addr, am6offset:$offset,
521 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
522 IIC_VLD4, "vld4", Dt,
523 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
524 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
527 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
528 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
529 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
531 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
532 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
534 // VLD1DUP : Vector Load (single element to all lanes)
535 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
536 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
537 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
538 // FIXME: Not yet implemented.
539 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
541 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
543 // Classes for VST* pseudo-instructions with multi-register operands.
544 // These are expanded to real instructions after register allocation.
546 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), IIC_VST, "">;
548 : PseudoNLdSt<(outs GPR:$wb),
549 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
552 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), IIC_VST, "">;
554 : PseudoNLdSt<(outs GPR:$wb),
555 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), IIC_VST,
557 class VSTQQQQWBPseudo
558 : PseudoNLdSt<(outs GPR:$wb),
559 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), IIC_VST,
562 // VST1 : Vector Store (multiple single elements)
563 class VST1D<bits<4> op7_4, string Dt>
564 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
565 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
566 class VST1Q<bits<4> op7_4, string Dt>
567 : NLdSt<0,0b00,0b1010,op7_4, (outs),
568 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
569 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
571 def VST1d8 : VST1D<0b0000, "8">;
572 def VST1d16 : VST1D<0b0100, "16">;
573 def VST1d32 : VST1D<0b1000, "32">;
574 def VST1d64 : VST1D<0b1100, "64">;
576 def VST1q8 : VST1Q<0b0000, "8">;
577 def VST1q16 : VST1Q<0b0100, "16">;
578 def VST1q32 : VST1Q<0b1000, "32">;
579 def VST1q64 : VST1Q<0b1100, "64">;
581 def VST1q8Pseudo : VSTQPseudo;
582 def VST1q16Pseudo : VSTQPseudo;
583 def VST1q32Pseudo : VSTQPseudo;
584 def VST1q64Pseudo : VSTQPseudo;
586 // ...with address register writeback:
587 class VST1DWB<bits<4> op7_4, string Dt>
588 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
589 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
590 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
591 class VST1QWB<bits<4> op7_4, string Dt>
592 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
593 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
594 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
596 def VST1d8_UPD : VST1DWB<0b0000, "8">;
597 def VST1d16_UPD : VST1DWB<0b0100, "16">;
598 def VST1d32_UPD : VST1DWB<0b1000, "32">;
599 def VST1d64_UPD : VST1DWB<0b1100, "64">;
601 def VST1q8_UPD : VST1QWB<0b0000, "8">;
602 def VST1q16_UPD : VST1QWB<0b0100, "16">;
603 def VST1q32_UPD : VST1QWB<0b1000, "32">;
604 def VST1q64_UPD : VST1QWB<0b1100, "64">;
606 def VST1q8Pseudo_UPD : VSTQWBPseudo;
607 def VST1q16Pseudo_UPD : VSTQWBPseudo;
608 def VST1q32Pseudo_UPD : VSTQWBPseudo;
609 def VST1q64Pseudo_UPD : VSTQWBPseudo;
611 // ...with 3 registers (some of these are only for the disassembler):
612 class VST1D3<bits<4> op7_4, string Dt>
613 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
614 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
615 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
616 class VST1D3WB<bits<4> op7_4, string Dt>
617 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
618 (ins addrmode6:$addr, am6offset:$offset,
619 DPR:$src1, DPR:$src2, DPR:$src3),
620 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
621 "$addr.addr = $wb", []>;
623 def VST1d8T : VST1D3<0b0000, "8">;
624 def VST1d16T : VST1D3<0b0100, "16">;
625 def VST1d32T : VST1D3<0b1000, "32">;
626 def VST1d64T : VST1D3<0b1100, "64">;
628 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
629 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
630 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
631 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
633 def VST1d64TPseudo : VSTQQPseudo;
634 def VST1d64TPseudo_UPD : VSTQQWBPseudo;
636 // ...with 4 registers (some of these are only for the disassembler):
637 class VST1D4<bits<4> op7_4, string Dt>
638 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
639 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
640 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
642 class VST1D4WB<bits<4> op7_4, string Dt>
643 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
644 (ins addrmode6:$addr, am6offset:$offset,
645 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
646 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
647 "$addr.addr = $wb", []>;
649 def VST1d8Q : VST1D4<0b0000, "8">;
650 def VST1d16Q : VST1D4<0b0100, "16">;
651 def VST1d32Q : VST1D4<0b1000, "32">;
652 def VST1d64Q : VST1D4<0b1100, "64">;
654 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
655 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
656 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
657 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
659 def VST1d64QPseudo : VSTQQPseudo;
660 def VST1d64QPseudo_UPD : VSTQQWBPseudo;
662 // VST2 : Vector Store (multiple 2-element structures)
663 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
664 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
665 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
666 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
667 class VST2Q<bits<4> op7_4, string Dt>
668 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
669 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
670 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
673 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
674 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
675 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
677 def VST2q8 : VST2Q<0b0000, "8">;
678 def VST2q16 : VST2Q<0b0100, "16">;
679 def VST2q32 : VST2Q<0b1000, "32">;
681 def VST2d8Pseudo : VSTQPseudo;
682 def VST2d16Pseudo : VSTQPseudo;
683 def VST2d32Pseudo : VSTQPseudo;
685 def VST2q8Pseudo : VSTQQPseudo;
686 def VST2q16Pseudo : VSTQQPseudo;
687 def VST2q32Pseudo : VSTQQPseudo;
689 // ...with address register writeback:
690 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
691 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
692 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
693 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
694 "$addr.addr = $wb", []>;
695 class VST2QWB<bits<4> op7_4, string Dt>
696 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
697 (ins addrmode6:$addr, am6offset:$offset,
698 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
699 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
700 "$addr.addr = $wb", []>;
702 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
703 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
704 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
706 def VST2q8_UPD : VST2QWB<0b0000, "8">;
707 def VST2q16_UPD : VST2QWB<0b0100, "16">;
708 def VST2q32_UPD : VST2QWB<0b1000, "32">;
710 def VST2d8Pseudo_UPD : VSTQWBPseudo;
711 def VST2d16Pseudo_UPD : VSTQWBPseudo;
712 def VST2d32Pseudo_UPD : VSTQWBPseudo;
714 def VST2q8Pseudo_UPD : VSTQQWBPseudo;
715 def VST2q16Pseudo_UPD : VSTQQWBPseudo;
716 def VST2q32Pseudo_UPD : VSTQQWBPseudo;
718 // ...with double-spaced registers (for disassembly only):
719 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
720 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
721 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
722 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
723 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
724 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
726 // VST3 : Vector Store (multiple 3-element structures)
727 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
728 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
729 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
730 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
732 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
733 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
734 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
736 def VST3d8Pseudo : VSTQQPseudo;
737 def VST3d16Pseudo : VSTQQPseudo;
738 def VST3d32Pseudo : VSTQQPseudo;
740 // ...with address register writeback:
741 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
742 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
743 (ins addrmode6:$addr, am6offset:$offset,
744 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
745 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
746 "$addr.addr = $wb", []>;
748 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
749 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
750 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
752 def VST3d8Pseudo_UPD : VSTQQWBPseudo;
753 def VST3d16Pseudo_UPD : VSTQQWBPseudo;
754 def VST3d32Pseudo_UPD : VSTQQWBPseudo;
756 // ...with double-spaced registers (non-updating versions for disassembly only):
757 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
758 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
759 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
760 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
761 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
762 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
764 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo;
765 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo;
766 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo;
768 // ...alternate versions to be allocated odd register numbers:
769 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo;
770 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo;
771 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo;
773 // VST4 : Vector Store (multiple 4-element structures)
774 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
775 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
776 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
777 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
780 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
781 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
782 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
784 def VST4d8Pseudo : VSTQQPseudo;
785 def VST4d16Pseudo : VSTQQPseudo;
786 def VST4d32Pseudo : VSTQQPseudo;
788 // ...with address register writeback:
789 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
790 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
791 (ins addrmode6:$addr, am6offset:$offset,
792 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
793 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
794 "$addr.addr = $wb", []>;
796 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
797 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
798 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
800 def VST4d8Pseudo_UPD : VSTQQWBPseudo;
801 def VST4d16Pseudo_UPD : VSTQQWBPseudo;
802 def VST4d32Pseudo_UPD : VSTQQWBPseudo;
804 // ...with double-spaced registers (non-updating versions for disassembly only):
805 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
806 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
807 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
808 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
809 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
810 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
812 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo;
813 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo;
814 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo;
816 // ...alternate versions to be allocated odd register numbers:
817 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo;
818 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo;
819 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo;
821 // VST1LN : Vector Store (single element from one lane)
822 // FIXME: Not yet implemented.
824 // VST2LN : Vector Store (single 2-element structure from one lane)
825 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
826 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
827 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
828 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
831 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
832 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
833 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
835 // ...with double-spaced registers:
836 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
837 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
839 // ...alternate versions to be allocated odd register numbers:
840 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
841 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
843 // ...with address register writeback:
844 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
845 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
846 (ins addrmode6:$addr, am6offset:$offset,
847 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
848 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
849 "$addr.addr = $wb", []>;
851 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
852 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
853 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
855 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
856 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
858 // VST3LN : Vector Store (single 3-element structure from one lane)
859 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
860 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
861 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
862 nohash_imm:$lane), IIC_VST, "vst3", Dt,
863 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
865 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
866 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
867 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
869 // ...with double-spaced registers:
870 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
871 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
873 // ...alternate versions to be allocated odd register numbers:
874 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
875 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
877 // ...with address register writeback:
878 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
879 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
880 (ins addrmode6:$addr, am6offset:$offset,
881 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
883 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
884 "$addr.addr = $wb", []>;
886 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
887 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
888 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
890 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
891 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
893 // VST4LN : Vector Store (single 4-element structure from one lane)
894 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
895 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
896 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
897 nohash_imm:$lane), IIC_VST, "vst4", Dt,
898 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
901 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
902 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
903 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
905 // ...with double-spaced registers:
906 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
907 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
909 // ...alternate versions to be allocated odd register numbers:
910 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
911 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
913 // ...with address register writeback:
914 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
915 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
916 (ins addrmode6:$addr, am6offset:$offset,
917 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
919 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
920 "$addr.addr = $wb", []>;
922 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
923 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
924 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
926 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
927 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
929 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
932 //===----------------------------------------------------------------------===//
933 // NEON pattern fragments
934 //===----------------------------------------------------------------------===//
936 // Extract D sub-registers of Q registers.
937 def DSubReg_i8_reg : SDNodeXForm<imm, [{
938 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
939 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
941 def DSubReg_i16_reg : SDNodeXForm<imm, [{
942 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
943 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
945 def DSubReg_i32_reg : SDNodeXForm<imm, [{
946 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
947 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
949 def DSubReg_f64_reg : SDNodeXForm<imm, [{
950 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
951 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
954 // Extract S sub-registers of Q/D registers.
955 def SSubReg_f32_reg : SDNodeXForm<imm, [{
956 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
957 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
960 // Translate lane numbers from Q registers to D subregs.
961 def SubReg_i8_lane : SDNodeXForm<imm, [{
962 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
964 def SubReg_i16_lane : SDNodeXForm<imm, [{
965 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
967 def SubReg_i32_lane : SDNodeXForm<imm, [{
968 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
971 //===----------------------------------------------------------------------===//
972 // Instruction Classes
973 //===----------------------------------------------------------------------===//
975 // Basic 2-register operations: single-, double- and quad-register.
976 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
977 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
978 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
979 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
980 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
981 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
982 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
983 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
984 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
985 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
986 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
987 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
988 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
989 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
990 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
991 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
992 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
993 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
995 // Basic 2-register intrinsics, both double- and quad-register.
996 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
997 bits<2> op17_16, bits<5> op11_7, bit op4,
998 InstrItinClass itin, string OpcodeStr, string Dt,
999 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1000 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1001 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1002 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1003 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1004 bits<2> op17_16, bits<5> op11_7, bit op4,
1005 InstrItinClass itin, string OpcodeStr, string Dt,
1006 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1007 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1008 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1009 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1011 // Narrow 2-register operations.
1012 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1013 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1014 InstrItinClass itin, string OpcodeStr, string Dt,
1015 ValueType TyD, ValueType TyQ, SDNode OpNode>
1016 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1017 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1018 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1020 // Narrow 2-register intrinsics.
1021 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1022 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1023 InstrItinClass itin, string OpcodeStr, string Dt,
1024 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1025 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1026 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1027 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1029 // Long 2-register operations (currently only used for VMOVL).
1030 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1031 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1032 InstrItinClass itin, string OpcodeStr, string Dt,
1033 ValueType TyQ, ValueType TyD, SDNode OpNode>
1034 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1035 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1036 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1038 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1039 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1040 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1041 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1042 OpcodeStr, Dt, "$dst1, $dst2",
1043 "$src1 = $dst1, $src2 = $dst2", []>;
1044 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1045 InstrItinClass itin, string OpcodeStr, string Dt>
1046 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1047 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1048 "$src1 = $dst1, $src2 = $dst2", []>;
1050 // Basic 3-register operations: single-, double- and quad-register.
1051 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1052 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1053 SDNode OpNode, bit Commutable>
1054 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1055 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1056 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1057 let isCommutable = Commutable;
1060 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1061 InstrItinClass itin, string OpcodeStr, string Dt,
1062 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1063 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1064 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1065 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1066 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1067 let isCommutable = Commutable;
1069 // Same as N3VD but no data type.
1070 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1071 InstrItinClass itin, string OpcodeStr,
1072 ValueType ResTy, ValueType OpTy,
1073 SDNode OpNode, bit Commutable>
1074 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1075 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1076 OpcodeStr, "$dst, $src1, $src2", "",
1077 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1078 let isCommutable = Commutable;
1081 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1082 InstrItinClass itin, string OpcodeStr, string Dt,
1083 ValueType Ty, SDNode ShOp>
1084 : N3V<0, 1, op21_20, op11_8, 1, 0,
1085 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1086 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1087 [(set (Ty DPR:$dst),
1088 (Ty (ShOp (Ty DPR:$src1),
1089 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1090 let isCommutable = 0;
1092 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1093 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1094 : N3V<0, 1, op21_20, op11_8, 1, 0,
1095 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1096 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1097 [(set (Ty DPR:$dst),
1098 (Ty (ShOp (Ty DPR:$src1),
1099 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1100 let isCommutable = 0;
1103 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1104 InstrItinClass itin, string OpcodeStr, string Dt,
1105 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1106 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1107 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1108 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1109 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1110 let isCommutable = Commutable;
1112 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1113 InstrItinClass itin, string OpcodeStr,
1114 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1115 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1116 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1117 OpcodeStr, "$dst, $src1, $src2", "",
1118 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1119 let isCommutable = Commutable;
1121 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1122 InstrItinClass itin, string OpcodeStr, string Dt,
1123 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1124 : N3V<1, 1, op21_20, op11_8, 1, 0,
1125 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1126 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1127 [(set (ResTy QPR:$dst),
1128 (ResTy (ShOp (ResTy QPR:$src1),
1129 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1131 let isCommutable = 0;
1133 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1134 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1135 : N3V<1, 1, op21_20, op11_8, 1, 0,
1136 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1137 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1138 [(set (ResTy QPR:$dst),
1139 (ResTy (ShOp (ResTy QPR:$src1),
1140 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1142 let isCommutable = 0;
1145 // Basic 3-register intrinsics, both double- and quad-register.
1146 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1147 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1148 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1149 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1150 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1151 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1152 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1153 let isCommutable = Commutable;
1155 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1156 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1157 : N3V<0, 1, op21_20, op11_8, 1, 0,
1158 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1159 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1160 [(set (Ty DPR:$dst),
1161 (Ty (IntOp (Ty DPR:$src1),
1162 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1164 let isCommutable = 0;
1166 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1167 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1168 : N3V<0, 1, op21_20, op11_8, 1, 0,
1169 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1170 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1171 [(set (Ty DPR:$dst),
1172 (Ty (IntOp (Ty DPR:$src1),
1173 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1174 let isCommutable = 0;
1177 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1178 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1179 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1180 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1181 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1182 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1183 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1184 let isCommutable = Commutable;
1186 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1187 string OpcodeStr, string Dt,
1188 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1189 : N3V<1, 1, op21_20, op11_8, 1, 0,
1190 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1191 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1192 [(set (ResTy QPR:$dst),
1193 (ResTy (IntOp (ResTy QPR:$src1),
1194 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1196 let isCommutable = 0;
1198 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1199 string OpcodeStr, string Dt,
1200 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1201 : N3V<1, 1, op21_20, op11_8, 1, 0,
1202 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1203 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1204 [(set (ResTy QPR:$dst),
1205 (ResTy (IntOp (ResTy QPR:$src1),
1206 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1208 let isCommutable = 0;
1211 // Multiply-Add/Sub operations: single-, double- and quad-register.
1212 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1213 InstrItinClass itin, string OpcodeStr, string Dt,
1214 ValueType Ty, SDNode MulOp, SDNode OpNode>
1215 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1216 (outs DPR_VFP2:$dst),
1217 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1218 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1220 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1221 InstrItinClass itin, string OpcodeStr, string Dt,
1222 ValueType Ty, SDNode MulOp, SDNode OpNode>
1223 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1224 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1225 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1226 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1227 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1228 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1229 string OpcodeStr, string Dt,
1230 ValueType Ty, SDNode MulOp, SDNode ShOp>
1231 : N3V<0, 1, op21_20, op11_8, 1, 0,
1233 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1235 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1236 [(set (Ty DPR:$dst),
1237 (Ty (ShOp (Ty DPR:$src1),
1238 (Ty (MulOp DPR:$src2,
1239 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1241 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1242 string OpcodeStr, string Dt,
1243 ValueType Ty, SDNode MulOp, SDNode ShOp>
1244 : N3V<0, 1, op21_20, op11_8, 1, 0,
1246 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1248 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1249 [(set (Ty DPR:$dst),
1250 (Ty (ShOp (Ty DPR:$src1),
1251 (Ty (MulOp DPR:$src2,
1252 (Ty (NEONvduplane (Ty DPR_8:$src3),
1255 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1256 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1257 SDNode MulOp, SDNode OpNode>
1258 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1259 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1260 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1261 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1262 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1263 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1264 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1265 SDNode MulOp, SDNode ShOp>
1266 : N3V<1, 1, op21_20, op11_8, 1, 0,
1268 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1270 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1271 [(set (ResTy QPR:$dst),
1272 (ResTy (ShOp (ResTy QPR:$src1),
1273 (ResTy (MulOp QPR:$src2,
1274 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1276 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1277 string OpcodeStr, string Dt,
1278 ValueType ResTy, ValueType OpTy,
1279 SDNode MulOp, SDNode ShOp>
1280 : N3V<1, 1, op21_20, op11_8, 1, 0,
1282 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1284 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1285 [(set (ResTy QPR:$dst),
1286 (ResTy (ShOp (ResTy QPR:$src1),
1287 (ResTy (MulOp QPR:$src2,
1288 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1291 // Neon 3-argument intrinsics, both double- and quad-register.
1292 // The destination register is also used as the first source operand register.
1293 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1294 InstrItinClass itin, string OpcodeStr, string Dt,
1295 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1296 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1297 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1298 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1299 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1300 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1301 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1302 InstrItinClass itin, string OpcodeStr, string Dt,
1303 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1304 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1305 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1306 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1307 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1308 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1310 // Long Multiply-Add/Sub operations.
1311 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1312 InstrItinClass itin, string OpcodeStr, string Dt,
1313 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1314 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1315 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1316 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1317 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1318 (TyQ (MulOp (TyD DPR:$src2),
1319 (TyD DPR:$src3)))))]>;
1320 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1321 InstrItinClass itin, string OpcodeStr, string Dt,
1322 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1323 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1324 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1326 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1328 (OpNode (TyQ QPR:$src1),
1329 (TyQ (MulOp (TyD DPR:$src2),
1330 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1332 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1333 InstrItinClass itin, string OpcodeStr, string Dt,
1334 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1335 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1336 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1338 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1340 (OpNode (TyQ QPR:$src1),
1341 (TyQ (MulOp (TyD DPR:$src2),
1342 (TyD (NEONvduplane (TyD DPR_8:$src3),
1346 // Neon Long 3-argument intrinsic. The destination register is
1347 // a quad-register and is also used as the first source operand register.
1348 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1349 InstrItinClass itin, string OpcodeStr, string Dt,
1350 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1351 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1352 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1353 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1355 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1356 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1357 string OpcodeStr, string Dt,
1358 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1359 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1361 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1363 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1364 [(set (ResTy QPR:$dst),
1365 (ResTy (IntOp (ResTy QPR:$src1),
1367 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1369 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1370 InstrItinClass itin, string OpcodeStr, string Dt,
1371 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1372 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1374 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1376 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1377 [(set (ResTy QPR:$dst),
1378 (ResTy (IntOp (ResTy QPR:$src1),
1380 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1383 // Narrowing 3-register intrinsics.
1384 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1385 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1386 Intrinsic IntOp, bit Commutable>
1387 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1388 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1389 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1390 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1391 let isCommutable = Commutable;
1394 // Long 3-register operations.
1395 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1396 InstrItinClass itin, string OpcodeStr, string Dt,
1397 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1398 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1399 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1400 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1401 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1402 let isCommutable = Commutable;
1404 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1405 InstrItinClass itin, string OpcodeStr, string Dt,
1406 ValueType TyQ, ValueType TyD, SDNode OpNode>
1407 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1408 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1409 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1411 (TyQ (OpNode (TyD DPR:$src1),
1412 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1413 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1414 InstrItinClass itin, string OpcodeStr, string Dt,
1415 ValueType TyQ, ValueType TyD, SDNode OpNode>
1416 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1417 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1418 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1420 (TyQ (OpNode (TyD DPR:$src1),
1421 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1423 // Long 3-register operations with explicitly extended operands.
1424 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1425 InstrItinClass itin, string OpcodeStr, string Dt,
1426 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1428 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1429 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1430 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1431 [(set QPR:$dst, (OpNode (TyQ (ExtOp (TyD DPR:$src1))),
1432 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1433 let isCommutable = Commutable;
1436 // Long 3-register intrinsics.
1437 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1438 InstrItinClass itin, string OpcodeStr, string Dt,
1439 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1440 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1441 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1442 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1443 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1444 let isCommutable = Commutable;
1446 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1447 string OpcodeStr, string Dt,
1448 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1449 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1450 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1451 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1452 [(set (ResTy QPR:$dst),
1453 (ResTy (IntOp (OpTy DPR:$src1),
1454 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1456 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1457 InstrItinClass itin, string OpcodeStr, string Dt,
1458 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1459 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1460 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1461 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1462 [(set (ResTy QPR:$dst),
1463 (ResTy (IntOp (OpTy DPR:$src1),
1464 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1467 // Wide 3-register operations.
1468 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1469 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1470 SDNode OpNode, SDNode ExtOp, bit Commutable>
1471 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1472 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1473 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1474 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1475 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1476 let isCommutable = Commutable;
1479 // Pairwise long 2-register intrinsics, both double- and quad-register.
1480 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1481 bits<2> op17_16, bits<5> op11_7, bit op4,
1482 string OpcodeStr, string Dt,
1483 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1484 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1485 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1486 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1487 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1488 bits<2> op17_16, bits<5> op11_7, bit op4,
1489 string OpcodeStr, string Dt,
1490 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1491 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1492 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1493 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1495 // Pairwise long 2-register accumulate intrinsics,
1496 // both double- and quad-register.
1497 // The destination register is also used as the first source operand register.
1498 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1499 bits<2> op17_16, bits<5> op11_7, bit op4,
1500 string OpcodeStr, string Dt,
1501 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1502 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1503 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1504 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1505 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1506 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1507 bits<2> op17_16, bits<5> op11_7, bit op4,
1508 string OpcodeStr, string Dt,
1509 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1510 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1511 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1512 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1513 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1515 // Shift by immediate,
1516 // both double- and quad-register.
1517 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1518 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1519 ValueType Ty, SDNode OpNode>
1520 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1521 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1522 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1523 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1524 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1525 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1526 ValueType Ty, SDNode OpNode>
1527 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1528 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1529 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1530 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1532 // Long shift by immediate.
1533 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1534 string OpcodeStr, string Dt,
1535 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1536 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1537 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1538 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1539 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1540 (i32 imm:$SIMM))))]>;
1542 // Narrow shift by immediate.
1543 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1544 InstrItinClass itin, string OpcodeStr, string Dt,
1545 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1546 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1547 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1548 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1549 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1550 (i32 imm:$SIMM))))]>;
1552 // Shift right by immediate and accumulate,
1553 // both double- and quad-register.
1554 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1555 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1556 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1557 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1558 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1559 [(set DPR:$dst, (Ty (add DPR:$src1,
1560 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1561 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1562 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1563 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1564 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1565 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1566 [(set QPR:$dst, (Ty (add QPR:$src1,
1567 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1569 // Shift by immediate and insert,
1570 // both double- and quad-register.
1571 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1572 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1573 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1574 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1575 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1576 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1577 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1578 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1579 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1580 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1581 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1582 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1584 // Convert, with fractional bits immediate,
1585 // both double- and quad-register.
1586 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1587 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1589 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1590 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1591 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1592 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1593 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1594 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1596 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1597 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1598 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1599 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1601 //===----------------------------------------------------------------------===//
1603 //===----------------------------------------------------------------------===//
1605 // Abbreviations used in multiclass suffixes:
1606 // Q = quarter int (8 bit) elements
1607 // H = half int (16 bit) elements
1608 // S = single int (32 bit) elements
1609 // D = double int (64 bit) elements
1611 // Neon 2-register vector operations -- for disassembly only.
1613 // First with only element sizes of 8, 16 and 32 bits:
1614 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1615 bits<5> op11_7, bit op4, string opc, string Dt,
1617 // 64-bit vector types.
1618 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1619 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1620 opc, !strconcat(Dt, "8"), asm, "", []>;
1621 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1622 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1623 opc, !strconcat(Dt, "16"), asm, "", []>;
1624 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1625 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1626 opc, !strconcat(Dt, "32"), asm, "", []>;
1627 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1628 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1629 opc, "f32", asm, "", []> {
1630 let Inst{10} = 1; // overwrite F = 1
1633 // 128-bit vector types.
1634 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1635 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1636 opc, !strconcat(Dt, "8"), asm, "", []>;
1637 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1638 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1639 opc, !strconcat(Dt, "16"), asm, "", []>;
1640 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1641 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1642 opc, !strconcat(Dt, "32"), asm, "", []>;
1643 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1644 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1645 opc, "f32", asm, "", []> {
1646 let Inst{10} = 1; // overwrite F = 1
1650 // Neon 3-register vector operations.
1652 // First with only element sizes of 8, 16 and 32 bits:
1653 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1654 InstrItinClass itinD16, InstrItinClass itinD32,
1655 InstrItinClass itinQ16, InstrItinClass itinQ32,
1656 string OpcodeStr, string Dt,
1657 SDNode OpNode, bit Commutable = 0> {
1658 // 64-bit vector types.
1659 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1660 OpcodeStr, !strconcat(Dt, "8"),
1661 v8i8, v8i8, OpNode, Commutable>;
1662 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1663 OpcodeStr, !strconcat(Dt, "16"),
1664 v4i16, v4i16, OpNode, Commutable>;
1665 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1666 OpcodeStr, !strconcat(Dt, "32"),
1667 v2i32, v2i32, OpNode, Commutable>;
1669 // 128-bit vector types.
1670 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1671 OpcodeStr, !strconcat(Dt, "8"),
1672 v16i8, v16i8, OpNode, Commutable>;
1673 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1674 OpcodeStr, !strconcat(Dt, "16"),
1675 v8i16, v8i16, OpNode, Commutable>;
1676 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1677 OpcodeStr, !strconcat(Dt, "32"),
1678 v4i32, v4i32, OpNode, Commutable>;
1681 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1682 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1684 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1686 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1687 v8i16, v4i16, ShOp>;
1688 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1689 v4i32, v2i32, ShOp>;
1692 // ....then also with element size 64 bits:
1693 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1694 InstrItinClass itinD, InstrItinClass itinQ,
1695 string OpcodeStr, string Dt,
1696 SDNode OpNode, bit Commutable = 0>
1697 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1698 OpcodeStr, Dt, OpNode, Commutable> {
1699 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1700 OpcodeStr, !strconcat(Dt, "64"),
1701 v1i64, v1i64, OpNode, Commutable>;
1702 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1703 OpcodeStr, !strconcat(Dt, "64"),
1704 v2i64, v2i64, OpNode, Commutable>;
1708 // Neon Narrowing 2-register vector operations,
1709 // source operand element sizes of 16, 32 and 64 bits:
1710 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1711 bits<5> op11_7, bit op6, bit op4,
1712 InstrItinClass itin, string OpcodeStr, string Dt,
1714 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1715 itin, OpcodeStr, !strconcat(Dt, "16"),
1716 v8i8, v8i16, OpNode>;
1717 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1718 itin, OpcodeStr, !strconcat(Dt, "32"),
1719 v4i16, v4i32, OpNode>;
1720 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1721 itin, OpcodeStr, !strconcat(Dt, "64"),
1722 v2i32, v2i64, OpNode>;
1725 // Neon Narrowing 2-register vector intrinsics,
1726 // source operand element sizes of 16, 32 and 64 bits:
1727 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1728 bits<5> op11_7, bit op6, bit op4,
1729 InstrItinClass itin, string OpcodeStr, string Dt,
1731 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1732 itin, OpcodeStr, !strconcat(Dt, "16"),
1733 v8i8, v8i16, IntOp>;
1734 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1735 itin, OpcodeStr, !strconcat(Dt, "32"),
1736 v4i16, v4i32, IntOp>;
1737 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1738 itin, OpcodeStr, !strconcat(Dt, "64"),
1739 v2i32, v2i64, IntOp>;
1743 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1744 // source operand element sizes of 16, 32 and 64 bits:
1745 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1746 string OpcodeStr, string Dt, SDNode OpNode> {
1747 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1748 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
1749 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1750 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1751 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1752 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1756 // Neon 3-register vector intrinsics.
1758 // First with only element sizes of 16 and 32 bits:
1759 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1760 InstrItinClass itinD16, InstrItinClass itinD32,
1761 InstrItinClass itinQ16, InstrItinClass itinQ32,
1762 string OpcodeStr, string Dt,
1763 Intrinsic IntOp, bit Commutable = 0> {
1764 // 64-bit vector types.
1765 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1766 OpcodeStr, !strconcat(Dt, "16"),
1767 v4i16, v4i16, IntOp, Commutable>;
1768 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1769 OpcodeStr, !strconcat(Dt, "32"),
1770 v2i32, v2i32, IntOp, Commutable>;
1772 // 128-bit vector types.
1773 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1774 OpcodeStr, !strconcat(Dt, "16"),
1775 v8i16, v8i16, IntOp, Commutable>;
1776 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1777 OpcodeStr, !strconcat(Dt, "32"),
1778 v4i32, v4i32, IntOp, Commutable>;
1781 multiclass N3VIntSL_HS<bits<4> op11_8,
1782 InstrItinClass itinD16, InstrItinClass itinD32,
1783 InstrItinClass itinQ16, InstrItinClass itinQ32,
1784 string OpcodeStr, string Dt, Intrinsic IntOp> {
1785 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1786 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1787 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1788 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1789 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1790 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1791 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1792 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1795 // ....then also with element size of 8 bits:
1796 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1797 InstrItinClass itinD16, InstrItinClass itinD32,
1798 InstrItinClass itinQ16, InstrItinClass itinQ32,
1799 string OpcodeStr, string Dt,
1800 Intrinsic IntOp, bit Commutable = 0>
1801 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1802 OpcodeStr, Dt, IntOp, Commutable> {
1803 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1804 OpcodeStr, !strconcat(Dt, "8"),
1805 v8i8, v8i8, IntOp, Commutable>;
1806 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1807 OpcodeStr, !strconcat(Dt, "8"),
1808 v16i8, v16i8, IntOp, Commutable>;
1811 // ....then also with element size of 64 bits:
1812 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1813 InstrItinClass itinD16, InstrItinClass itinD32,
1814 InstrItinClass itinQ16, InstrItinClass itinQ32,
1815 string OpcodeStr, string Dt,
1816 Intrinsic IntOp, bit Commutable = 0>
1817 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1818 OpcodeStr, Dt, IntOp, Commutable> {
1819 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1820 OpcodeStr, !strconcat(Dt, "64"),
1821 v1i64, v1i64, IntOp, Commutable>;
1822 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1823 OpcodeStr, !strconcat(Dt, "64"),
1824 v2i64, v2i64, IntOp, Commutable>;
1827 // Neon Narrowing 3-register vector intrinsics,
1828 // source operand element sizes of 16, 32 and 64 bits:
1829 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1830 string OpcodeStr, string Dt,
1831 Intrinsic IntOp, bit Commutable = 0> {
1832 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1833 OpcodeStr, !strconcat(Dt, "16"),
1834 v8i8, v8i16, IntOp, Commutable>;
1835 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1836 OpcodeStr, !strconcat(Dt, "32"),
1837 v4i16, v4i32, IntOp, Commutable>;
1838 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1839 OpcodeStr, !strconcat(Dt, "64"),
1840 v2i32, v2i64, IntOp, Commutable>;
1844 // Neon Long 3-register vector operations.
1846 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1847 InstrItinClass itin16, InstrItinClass itin32,
1848 string OpcodeStr, string Dt,
1849 SDNode OpNode, bit Commutable = 0> {
1850 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
1851 OpcodeStr, !strconcat(Dt, "8"),
1852 v8i16, v8i8, OpNode, Commutable>;
1853 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
1854 OpcodeStr, !strconcat(Dt, "16"),
1855 v4i32, v4i16, OpNode, Commutable>;
1856 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
1857 OpcodeStr, !strconcat(Dt, "32"),
1858 v2i64, v2i32, OpNode, Commutable>;
1861 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
1862 InstrItinClass itin, string OpcodeStr, string Dt,
1864 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
1865 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1866 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
1867 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1870 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1871 InstrItinClass itin16, InstrItinClass itin32,
1872 string OpcodeStr, string Dt,
1873 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
1874 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
1875 OpcodeStr, !strconcat(Dt, "8"),
1876 v8i16, v8i8, OpNode, ExtOp, Commutable>;
1877 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
1878 OpcodeStr, !strconcat(Dt, "16"),
1879 v4i32, v4i16, OpNode, ExtOp, Commutable>;
1880 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
1881 OpcodeStr, !strconcat(Dt, "32"),
1882 v2i64, v2i32, OpNode, ExtOp, Commutable>;
1885 // Neon Long 3-register vector intrinsics.
1887 // First with only element sizes of 16 and 32 bits:
1888 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1889 InstrItinClass itin16, InstrItinClass itin32,
1890 string OpcodeStr, string Dt,
1891 Intrinsic IntOp, bit Commutable = 0> {
1892 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1893 OpcodeStr, !strconcat(Dt, "16"),
1894 v4i32, v4i16, IntOp, Commutable>;
1895 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1896 OpcodeStr, !strconcat(Dt, "32"),
1897 v2i64, v2i32, IntOp, Commutable>;
1900 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1901 InstrItinClass itin, string OpcodeStr, string Dt,
1903 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1904 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1905 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1906 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1909 // ....then also with element size of 8 bits:
1910 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1911 InstrItinClass itin16, InstrItinClass itin32,
1912 string OpcodeStr, string Dt,
1913 Intrinsic IntOp, bit Commutable = 0>
1914 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1915 IntOp, Commutable> {
1916 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1917 OpcodeStr, !strconcat(Dt, "8"),
1918 v8i16, v8i8, IntOp, Commutable>;
1922 // Neon Wide 3-register vector intrinsics,
1923 // source operand element sizes of 8, 16 and 32 bits:
1924 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1925 string OpcodeStr, string Dt,
1926 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
1927 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
1928 OpcodeStr, !strconcat(Dt, "8"),
1929 v8i16, v8i8, OpNode, ExtOp, Commutable>;
1930 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
1931 OpcodeStr, !strconcat(Dt, "16"),
1932 v4i32, v4i16, OpNode, ExtOp, Commutable>;
1933 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
1934 OpcodeStr, !strconcat(Dt, "32"),
1935 v2i64, v2i32, OpNode, ExtOp, Commutable>;
1939 // Neon Multiply-Op vector operations,
1940 // element sizes of 8, 16 and 32 bits:
1941 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1942 InstrItinClass itinD16, InstrItinClass itinD32,
1943 InstrItinClass itinQ16, InstrItinClass itinQ32,
1944 string OpcodeStr, string Dt, SDNode OpNode> {
1945 // 64-bit vector types.
1946 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1947 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1948 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1949 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1950 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1951 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1953 // 128-bit vector types.
1954 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1955 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1956 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1957 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1958 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1959 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1962 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1963 InstrItinClass itinD16, InstrItinClass itinD32,
1964 InstrItinClass itinQ16, InstrItinClass itinQ32,
1965 string OpcodeStr, string Dt, SDNode ShOp> {
1966 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1967 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1968 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1969 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1970 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1971 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1973 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1974 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1978 // Neon 3-argument intrinsics,
1979 // element sizes of 8, 16 and 32 bits:
1980 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1981 InstrItinClass itinD, InstrItinClass itinQ,
1982 string OpcodeStr, string Dt, Intrinsic IntOp> {
1983 // 64-bit vector types.
1984 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1985 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1986 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1987 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1988 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1989 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1991 // 128-bit vector types.
1992 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1993 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1994 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1995 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1996 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1997 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2001 // Neon Long Multiply-Op vector operations,
2002 // element sizes of 8, 16 and 32 bits:
2003 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2004 InstrItinClass itin16, InstrItinClass itin32,
2005 string OpcodeStr, string Dt, SDNode MulOp,
2007 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2008 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2009 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2010 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2011 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2012 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2015 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2016 string Dt, SDNode MulOp, SDNode OpNode> {
2017 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2018 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2019 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2020 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2024 // Neon Long 3-argument intrinsics.
2026 // First with only element sizes of 16 and 32 bits:
2027 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2028 InstrItinClass itin16, InstrItinClass itin32,
2029 string OpcodeStr, string Dt, Intrinsic IntOp> {
2030 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2031 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2032 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2033 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2036 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2037 string OpcodeStr, string Dt, Intrinsic IntOp> {
2038 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2039 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2040 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2041 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2044 // ....then also with element size of 8 bits:
2045 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2046 InstrItinClass itin16, InstrItinClass itin32,
2047 string OpcodeStr, string Dt, Intrinsic IntOp>
2048 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2049 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2050 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2054 // Neon 2-register vector intrinsics,
2055 // element sizes of 8, 16 and 32 bits:
2056 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2057 bits<5> op11_7, bit op4,
2058 InstrItinClass itinD, InstrItinClass itinQ,
2059 string OpcodeStr, string Dt, Intrinsic IntOp> {
2060 // 64-bit vector types.
2061 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2062 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2063 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2064 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2065 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2066 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2068 // 128-bit vector types.
2069 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2070 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2071 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2072 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2073 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2074 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2078 // Neon Pairwise long 2-register intrinsics,
2079 // element sizes of 8, 16 and 32 bits:
2080 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2081 bits<5> op11_7, bit op4,
2082 string OpcodeStr, string Dt, Intrinsic IntOp> {
2083 // 64-bit vector types.
2084 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2085 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2086 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2087 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2088 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2089 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2091 // 128-bit vector types.
2092 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2093 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2094 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2095 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2096 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2097 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2101 // Neon Pairwise long 2-register accumulate intrinsics,
2102 // element sizes of 8, 16 and 32 bits:
2103 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2104 bits<5> op11_7, bit op4,
2105 string OpcodeStr, string Dt, Intrinsic IntOp> {
2106 // 64-bit vector types.
2107 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2108 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2109 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2110 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2111 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2112 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2114 // 128-bit vector types.
2115 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2116 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2117 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2118 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2119 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2120 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2124 // Neon 2-register vector shift by immediate,
2125 // with f of either N2RegVShLFrm or N2RegVShRFrm
2126 // element sizes of 8, 16, 32 and 64 bits:
2127 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2128 InstrItinClass itin, string OpcodeStr, string Dt,
2129 SDNode OpNode, Format f> {
2130 // 64-bit vector types.
2131 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2132 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2133 let Inst{21-19} = 0b001; // imm6 = 001xxx
2135 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2136 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2137 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2139 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2140 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2141 let Inst{21} = 0b1; // imm6 = 1xxxxx
2143 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2144 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2147 // 128-bit vector types.
2148 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2149 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2150 let Inst{21-19} = 0b001; // imm6 = 001xxx
2152 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2153 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2154 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2156 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2157 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2158 let Inst{21} = 0b1; // imm6 = 1xxxxx
2160 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2161 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2165 // Neon Shift-Accumulate vector operations,
2166 // element sizes of 8, 16, 32 and 64 bits:
2167 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2168 string OpcodeStr, string Dt, SDNode ShOp> {
2169 // 64-bit vector types.
2170 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2171 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2172 let Inst{21-19} = 0b001; // imm6 = 001xxx
2174 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2175 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2176 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2178 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2179 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2180 let Inst{21} = 0b1; // imm6 = 1xxxxx
2182 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2183 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2186 // 128-bit vector types.
2187 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2188 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2189 let Inst{21-19} = 0b001; // imm6 = 001xxx
2191 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2192 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2193 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2195 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2196 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2197 let Inst{21} = 0b1; // imm6 = 1xxxxx
2199 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2200 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2205 // Neon Shift-Insert vector operations,
2206 // with f of either N2RegVShLFrm or N2RegVShRFrm
2207 // element sizes of 8, 16, 32 and 64 bits:
2208 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2209 string OpcodeStr, SDNode ShOp,
2211 // 64-bit vector types.
2212 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2213 f, OpcodeStr, "8", v8i8, ShOp> {
2214 let Inst{21-19} = 0b001; // imm6 = 001xxx
2216 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2217 f, OpcodeStr, "16", v4i16, ShOp> {
2218 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2220 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2221 f, OpcodeStr, "32", v2i32, ShOp> {
2222 let Inst{21} = 0b1; // imm6 = 1xxxxx
2224 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2225 f, OpcodeStr, "64", v1i64, ShOp>;
2228 // 128-bit vector types.
2229 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2230 f, OpcodeStr, "8", v16i8, ShOp> {
2231 let Inst{21-19} = 0b001; // imm6 = 001xxx
2233 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2234 f, OpcodeStr, "16", v8i16, ShOp> {
2235 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2237 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2238 f, OpcodeStr, "32", v4i32, ShOp> {
2239 let Inst{21} = 0b1; // imm6 = 1xxxxx
2241 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2242 f, OpcodeStr, "64", v2i64, ShOp>;
2246 // Neon Shift Long operations,
2247 // element sizes of 8, 16, 32 bits:
2248 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2249 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2250 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2251 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2252 let Inst{21-19} = 0b001; // imm6 = 001xxx
2254 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2255 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2256 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2258 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2259 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2260 let Inst{21} = 0b1; // imm6 = 1xxxxx
2264 // Neon Shift Narrow operations,
2265 // element sizes of 16, 32, 64 bits:
2266 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2267 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2269 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2270 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2271 let Inst{21-19} = 0b001; // imm6 = 001xxx
2273 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2274 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2275 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2277 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2278 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2279 let Inst{21} = 0b1; // imm6 = 1xxxxx
2283 //===----------------------------------------------------------------------===//
2284 // Instruction Definitions.
2285 //===----------------------------------------------------------------------===//
2287 // Vector Add Operations.
2289 // VADD : Vector Add (integer and floating-point)
2290 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2292 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2293 v2f32, v2f32, fadd, 1>;
2294 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2295 v4f32, v4f32, fadd, 1>;
2296 // VADDL : Vector Add Long (Q = D + D)
2297 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2298 "vaddl", "s", add, sext, 1>;
2299 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2300 "vaddl", "u", add, zext, 1>;
2301 // VADDW : Vector Add Wide (Q = Q + D)
2302 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2303 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2304 // VHADD : Vector Halving Add
2305 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2306 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2307 "vhadd", "s", int_arm_neon_vhadds, 1>;
2308 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2309 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2310 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2311 // VRHADD : Vector Rounding Halving Add
2312 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2313 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2314 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2315 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2316 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2317 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2318 // VQADD : Vector Saturating Add
2319 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2320 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2321 "vqadd", "s", int_arm_neon_vqadds, 1>;
2322 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2323 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2324 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2325 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2326 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2327 int_arm_neon_vaddhn, 1>;
2328 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2329 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2330 int_arm_neon_vraddhn, 1>;
2332 // Vector Multiply Operations.
2334 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2335 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2336 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2337 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2338 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2339 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2340 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2341 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2342 v2f32, v2f32, fmul, 1>;
2343 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2344 v4f32, v4f32, fmul, 1>;
2345 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2346 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2347 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2350 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2351 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2352 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2353 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2354 (DSubReg_i16_reg imm:$lane))),
2355 (SubReg_i16_lane imm:$lane)))>;
2356 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2357 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2358 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2359 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2360 (DSubReg_i32_reg imm:$lane))),
2361 (SubReg_i32_lane imm:$lane)))>;
2362 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2363 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2364 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2365 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2366 (DSubReg_i32_reg imm:$lane))),
2367 (SubReg_i32_lane imm:$lane)))>;
2369 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2370 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2371 IIC_VMULi16Q, IIC_VMULi32Q,
2372 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2373 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2374 IIC_VMULi16Q, IIC_VMULi32Q,
2375 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2376 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2377 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2379 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2380 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2381 (DSubReg_i16_reg imm:$lane))),
2382 (SubReg_i16_lane imm:$lane)))>;
2383 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2384 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2386 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2387 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2388 (DSubReg_i32_reg imm:$lane))),
2389 (SubReg_i32_lane imm:$lane)))>;
2391 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2392 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2393 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2394 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2395 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2396 IIC_VMULi16Q, IIC_VMULi32Q,
2397 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2398 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2399 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2401 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2402 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2403 (DSubReg_i16_reg imm:$lane))),
2404 (SubReg_i16_lane imm:$lane)))>;
2405 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2406 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2408 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2409 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2410 (DSubReg_i32_reg imm:$lane))),
2411 (SubReg_i32_lane imm:$lane)))>;
2413 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2414 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2415 "vmull", "s", NEONvmulls, 1>;
2416 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2417 "vmull", "u", NEONvmullu, 1>;
2418 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2419 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2420 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
2421 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
2423 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2424 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2425 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2426 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2427 "vqdmull", "s", int_arm_neon_vqdmull>;
2429 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2431 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2432 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2433 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2434 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2436 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2438 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2439 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2440 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2442 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2443 v4f32, v2f32, fmul, fadd>;
2445 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2446 (mul (v8i16 QPR:$src2),
2447 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2448 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2449 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2450 (DSubReg_i16_reg imm:$lane))),
2451 (SubReg_i16_lane imm:$lane)))>;
2453 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2454 (mul (v4i32 QPR:$src2),
2455 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2456 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2457 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2458 (DSubReg_i32_reg imm:$lane))),
2459 (SubReg_i32_lane imm:$lane)))>;
2461 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2462 (fmul (v4f32 QPR:$src2),
2463 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2464 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2466 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2467 (DSubReg_i32_reg imm:$lane))),
2468 (SubReg_i32_lane imm:$lane)))>;
2470 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2471 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2472 "vmlal", "s", NEONvmulls, add>;
2473 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2474 "vmlal", "u", NEONvmullu, add>;
2476 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
2477 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
2479 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2480 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2481 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2482 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2484 // VMLS : Vector Multiply Subtract (integer and floating-point)
2485 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2486 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2487 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2489 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2491 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2492 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2493 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2495 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2496 v4f32, v2f32, fmul, fsub>;
2498 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2499 (mul (v8i16 QPR:$src2),
2500 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2501 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2502 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2503 (DSubReg_i16_reg imm:$lane))),
2504 (SubReg_i16_lane imm:$lane)))>;
2506 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2507 (mul (v4i32 QPR:$src2),
2508 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2509 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2510 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2511 (DSubReg_i32_reg imm:$lane))),
2512 (SubReg_i32_lane imm:$lane)))>;
2514 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2515 (fmul (v4f32 QPR:$src2),
2516 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2517 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2518 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2519 (DSubReg_i32_reg imm:$lane))),
2520 (SubReg_i32_lane imm:$lane)))>;
2522 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2523 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2524 "vmlsl", "s", NEONvmulls, sub>;
2525 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2526 "vmlsl", "u", NEONvmullu, sub>;
2528 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
2529 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
2531 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2532 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2533 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2534 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2536 // Vector Subtract Operations.
2538 // VSUB : Vector Subtract (integer and floating-point)
2539 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2540 "vsub", "i", sub, 0>;
2541 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2542 v2f32, v2f32, fsub, 0>;
2543 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2544 v4f32, v4f32, fsub, 0>;
2545 // VSUBL : Vector Subtract Long (Q = D - D)
2546 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2547 "vsubl", "s", sub, sext, 0>;
2548 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2549 "vsubl", "u", sub, zext, 0>;
2550 // VSUBW : Vector Subtract Wide (Q = Q - D)
2551 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
2552 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
2553 // VHSUB : Vector Halving Subtract
2554 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2555 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2556 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2557 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2558 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2559 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2560 // VQSUB : Vector Saturing Subtract
2561 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2562 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2563 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2564 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2565 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2566 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2567 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2568 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2569 int_arm_neon_vsubhn, 0>;
2570 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2571 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2572 int_arm_neon_vrsubhn, 0>;
2574 // Vector Comparisons.
2576 // VCEQ : Vector Compare Equal
2577 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2578 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2579 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2581 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2583 // For disassembly only.
2584 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2587 // VCGE : Vector Compare Greater Than or Equal
2588 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2589 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2590 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2591 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2592 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2594 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2596 // For disassembly only.
2597 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2599 // For disassembly only.
2600 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2603 // VCGT : Vector Compare Greater Than
2604 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2605 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2606 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2607 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2608 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2610 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2612 // For disassembly only.
2613 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2615 // For disassembly only.
2616 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2619 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2620 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2621 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2622 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2623 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2624 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2625 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2626 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2627 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2628 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2629 // VTST : Vector Test Bits
2630 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2631 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2633 // Vector Bitwise Operations.
2635 def vnotd : PatFrag<(ops node:$in),
2636 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
2637 def vnotq : PatFrag<(ops node:$in),
2638 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
2641 // VAND : Vector Bitwise AND
2642 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2643 v2i32, v2i32, and, 1>;
2644 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2645 v4i32, v4i32, and, 1>;
2647 // VEOR : Vector Bitwise Exclusive OR
2648 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2649 v2i32, v2i32, xor, 1>;
2650 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2651 v4i32, v4i32, xor, 1>;
2653 // VORR : Vector Bitwise OR
2654 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2655 v2i32, v2i32, or, 1>;
2656 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2657 v4i32, v4i32, or, 1>;
2659 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2660 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2661 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2662 "vbic", "$dst, $src1, $src2", "",
2663 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2664 (vnotd DPR:$src2))))]>;
2665 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2666 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2667 "vbic", "$dst, $src1, $src2", "",
2668 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2669 (vnotq QPR:$src2))))]>;
2671 // VORN : Vector Bitwise OR NOT
2672 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2673 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2674 "vorn", "$dst, $src1, $src2", "",
2675 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2676 (vnotd DPR:$src2))))]>;
2677 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2678 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2679 "vorn", "$dst, $src1, $src2", "",
2680 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2681 (vnotq QPR:$src2))))]>;
2683 // VMVN : Vector Bitwise NOT (Immediate)
2685 let isReMaterializable = 1 in {
2686 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
2687 (ins nModImm:$SIMM), IIC_VMOVImm,
2688 "vmvn", "i16", "$dst, $SIMM", "",
2689 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]>;
2690 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
2691 (ins nModImm:$SIMM), IIC_VMOVImm,
2692 "vmvn", "i16", "$dst, $SIMM", "",
2693 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]>;
2695 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
2696 (ins nModImm:$SIMM), IIC_VMOVImm,
2697 "vmvn", "i32", "$dst, $SIMM", "",
2698 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]>;
2699 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
2700 (ins nModImm:$SIMM), IIC_VMOVImm,
2701 "vmvn", "i32", "$dst, $SIMM", "",
2702 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]>;
2705 // VMVN : Vector Bitwise NOT
2706 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2707 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2708 "vmvn", "$dst, $src", "",
2709 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
2710 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2711 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2712 "vmvn", "$dst, $src", "",
2713 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
2714 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
2715 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
2717 // VBSL : Vector Bitwise Select
2718 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2719 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2720 N3RegFrm, IIC_VCNTiD,
2721 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2723 (v2i32 (or (and DPR:$src2, DPR:$src1),
2724 (and DPR:$src3, (vnotd DPR:$src1)))))]>;
2725 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2726 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2727 N3RegFrm, IIC_VCNTiQ,
2728 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2730 (v4i32 (or (and QPR:$src2, QPR:$src1),
2731 (and QPR:$src3, (vnotq QPR:$src1)))))]>;
2733 // VBIF : Vector Bitwise Insert if False
2734 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2735 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2736 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2737 N3RegFrm, IIC_VBINiD,
2738 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2739 [/* For disassembly only; pattern left blank */]>;
2740 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2741 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2742 N3RegFrm, IIC_VBINiQ,
2743 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2744 [/* For disassembly only; pattern left blank */]>;
2746 // VBIT : Vector Bitwise Insert if True
2747 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2748 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2749 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2750 N3RegFrm, IIC_VBINiD,
2751 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2752 [/* For disassembly only; pattern left blank */]>;
2753 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2754 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2755 N3RegFrm, IIC_VBINiQ,
2756 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2757 [/* For disassembly only; pattern left blank */]>;
2759 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2760 // for equivalent operations with different register constraints; it just
2763 // Vector Absolute Differences.
2765 // VABD : Vector Absolute Difference
2766 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2767 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2768 "vabd", "s", int_arm_neon_vabds, 0>;
2769 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2770 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2771 "vabd", "u", int_arm_neon_vabdu, 0>;
2772 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2773 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2774 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2775 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2777 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2778 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2779 "vabdl", "s", int_arm_neon_vabdls, 0>;
2780 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2781 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2783 // VABA : Vector Absolute Difference and Accumulate
2784 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2785 "vaba", "s", int_arm_neon_vabas>;
2786 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2787 "vaba", "u", int_arm_neon_vabau>;
2789 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2790 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2791 "vabal", "s", int_arm_neon_vabals>;
2792 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2793 "vabal", "u", int_arm_neon_vabalu>;
2795 // Vector Maximum and Minimum.
2797 // VMAX : Vector Maximum
2798 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2799 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2800 "vmax", "s", int_arm_neon_vmaxs, 1>;
2801 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2802 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2803 "vmax", "u", int_arm_neon_vmaxu, 1>;
2804 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2806 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2807 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2809 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2811 // VMIN : Vector Minimum
2812 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2813 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2814 "vmin", "s", int_arm_neon_vmins, 1>;
2815 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2816 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2817 "vmin", "u", int_arm_neon_vminu, 1>;
2818 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2820 v2f32, v2f32, int_arm_neon_vmins, 1>;
2821 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2823 v4f32, v4f32, int_arm_neon_vmins, 1>;
2825 // Vector Pairwise Operations.
2827 // VPADD : Vector Pairwise Add
2828 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2830 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2831 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2833 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2834 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2836 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2837 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2838 IIC_VBIND, "vpadd", "f32",
2839 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2841 // VPADDL : Vector Pairwise Add Long
2842 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2843 int_arm_neon_vpaddls>;
2844 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2845 int_arm_neon_vpaddlu>;
2847 // VPADAL : Vector Pairwise Add and Accumulate Long
2848 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2849 int_arm_neon_vpadals>;
2850 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2851 int_arm_neon_vpadalu>;
2853 // VPMAX : Vector Pairwise Maximum
2854 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2855 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2856 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2857 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2858 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2859 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2860 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2861 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2862 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2863 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2864 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2865 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2866 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2867 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2869 // VPMIN : Vector Pairwise Minimum
2870 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2871 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2872 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2873 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2874 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2875 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2876 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2877 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2878 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2879 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2880 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2881 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2882 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2883 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2885 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2887 // VRECPE : Vector Reciprocal Estimate
2888 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2889 IIC_VUNAD, "vrecpe", "u32",
2890 v2i32, v2i32, int_arm_neon_vrecpe>;
2891 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2892 IIC_VUNAQ, "vrecpe", "u32",
2893 v4i32, v4i32, int_arm_neon_vrecpe>;
2894 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2895 IIC_VUNAD, "vrecpe", "f32",
2896 v2f32, v2f32, int_arm_neon_vrecpe>;
2897 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2898 IIC_VUNAQ, "vrecpe", "f32",
2899 v4f32, v4f32, int_arm_neon_vrecpe>;
2901 // VRECPS : Vector Reciprocal Step
2902 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2903 IIC_VRECSD, "vrecps", "f32",
2904 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2905 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2906 IIC_VRECSQ, "vrecps", "f32",
2907 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2909 // VRSQRTE : Vector Reciprocal Square Root Estimate
2910 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2911 IIC_VUNAD, "vrsqrte", "u32",
2912 v2i32, v2i32, int_arm_neon_vrsqrte>;
2913 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2914 IIC_VUNAQ, "vrsqrte", "u32",
2915 v4i32, v4i32, int_arm_neon_vrsqrte>;
2916 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2917 IIC_VUNAD, "vrsqrte", "f32",
2918 v2f32, v2f32, int_arm_neon_vrsqrte>;
2919 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2920 IIC_VUNAQ, "vrsqrte", "f32",
2921 v4f32, v4f32, int_arm_neon_vrsqrte>;
2923 // VRSQRTS : Vector Reciprocal Square Root Step
2924 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2925 IIC_VRECSD, "vrsqrts", "f32",
2926 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2927 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2928 IIC_VRECSQ, "vrsqrts", "f32",
2929 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2933 // VSHL : Vector Shift
2934 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2935 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2936 "vshl", "s", int_arm_neon_vshifts, 0>;
2937 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2938 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2939 "vshl", "u", int_arm_neon_vshiftu, 0>;
2940 // VSHL : Vector Shift Left (Immediate)
2941 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2943 // VSHR : Vector Shift Right (Immediate)
2944 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2946 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2949 // VSHLL : Vector Shift Left Long
2950 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2951 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2953 // VSHLL : Vector Shift Left Long (with maximum shift count)
2954 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2955 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2956 ValueType OpTy, SDNode OpNode>
2957 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2958 ResTy, OpTy, OpNode> {
2959 let Inst{21-16} = op21_16;
2961 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2962 v8i16, v8i8, NEONvshlli>;
2963 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2964 v4i32, v4i16, NEONvshlli>;
2965 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2966 v2i64, v2i32, NEONvshlli>;
2968 // VSHRN : Vector Shift Right and Narrow
2969 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2972 // VRSHL : Vector Rounding Shift
2973 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2974 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2975 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2976 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2977 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2978 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2979 // VRSHR : Vector Rounding Shift Right
2980 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2982 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2985 // VRSHRN : Vector Rounding Shift Right and Narrow
2986 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2989 // VQSHL : Vector Saturating Shift
2990 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2991 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2992 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2993 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2994 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2995 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2996 // VQSHL : Vector Saturating Shift Left (Immediate)
2997 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2999 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3001 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3002 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3005 // VQSHRN : Vector Saturating Shift Right and Narrow
3006 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3008 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3011 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3012 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3015 // VQRSHL : Vector Saturating Rounding Shift
3016 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
3017 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3018 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
3019 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
3020 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3021 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
3023 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3024 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3026 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3029 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3030 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3033 // VSRA : Vector Shift Right and Accumulate
3034 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3035 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3036 // VRSRA : Vector Rounding Shift Right and Accumulate
3037 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3038 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3040 // VSLI : Vector Shift Left and Insert
3041 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3042 // VSRI : Vector Shift Right and Insert
3043 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3045 // Vector Absolute and Saturating Absolute.
3047 // VABS : Vector Absolute Value
3048 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3049 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3051 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3052 IIC_VUNAD, "vabs", "f32",
3053 v2f32, v2f32, int_arm_neon_vabs>;
3054 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3055 IIC_VUNAQ, "vabs", "f32",
3056 v4f32, v4f32, int_arm_neon_vabs>;
3058 // VQABS : Vector Saturating Absolute Value
3059 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3060 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3061 int_arm_neon_vqabs>;
3065 def vnegd : PatFrag<(ops node:$in),
3066 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3067 def vnegq : PatFrag<(ops node:$in),
3068 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3070 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3071 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3072 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3073 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3074 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3075 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3076 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3077 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3079 // VNEG : Vector Negate (integer)
3080 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3081 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3082 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3083 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3084 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3085 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3087 // VNEG : Vector Negate (floating-point)
3088 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3089 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3090 "vneg", "f32", "$dst, $src", "",
3091 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3092 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3093 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3094 "vneg", "f32", "$dst, $src", "",
3095 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3097 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3098 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3099 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3100 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3101 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3102 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3104 // VQNEG : Vector Saturating Negate
3105 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3106 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3107 int_arm_neon_vqneg>;
3109 // Vector Bit Counting Operations.
3111 // VCLS : Vector Count Leading Sign Bits
3112 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3113 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3115 // VCLZ : Vector Count Leading Zeros
3116 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3117 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3119 // VCNT : Vector Count One Bits
3120 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3121 IIC_VCNTiD, "vcnt", "8",
3122 v8i8, v8i8, int_arm_neon_vcnt>;
3123 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3124 IIC_VCNTiQ, "vcnt", "8",
3125 v16i8, v16i8, int_arm_neon_vcnt>;
3127 // Vector Swap -- for disassembly only.
3128 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3129 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3130 "vswp", "$dst, $src", "", []>;
3131 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3132 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3133 "vswp", "$dst, $src", "", []>;
3135 // Vector Move Operations.
3137 // VMOV : Vector Move (Register)
3139 let neverHasSideEffects = 1 in {
3140 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3141 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3142 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3143 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3145 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3146 // be expanded after register allocation is completed.
3147 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3148 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3150 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3151 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3152 } // neverHasSideEffects
3154 // VMOV : Vector Move (Immediate)
3156 let isReMaterializable = 1 in {
3157 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3158 (ins nModImm:$SIMM), IIC_VMOVImm,
3159 "vmov", "i8", "$dst, $SIMM", "",
3160 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3161 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3162 (ins nModImm:$SIMM), IIC_VMOVImm,
3163 "vmov", "i8", "$dst, $SIMM", "",
3164 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3166 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3167 (ins nModImm:$SIMM), IIC_VMOVImm,
3168 "vmov", "i16", "$dst, $SIMM", "",
3169 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]>;
3170 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3171 (ins nModImm:$SIMM), IIC_VMOVImm,
3172 "vmov", "i16", "$dst, $SIMM", "",
3173 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]>;
3175 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3176 (ins nModImm:$SIMM), IIC_VMOVImm,
3177 "vmov", "i32", "$dst, $SIMM", "",
3178 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]>;
3179 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3180 (ins nModImm:$SIMM), IIC_VMOVImm,
3181 "vmov", "i32", "$dst, $SIMM", "",
3182 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]>;
3184 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3185 (ins nModImm:$SIMM), IIC_VMOVImm,
3186 "vmov", "i64", "$dst, $SIMM", "",
3187 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3188 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3189 (ins nModImm:$SIMM), IIC_VMOVImm,
3190 "vmov", "i64", "$dst, $SIMM", "",
3191 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3192 } // isReMaterializable
3194 // VMOV : Vector Get Lane (move scalar to ARM core register)
3196 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3197 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3198 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
3199 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
3201 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3202 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3203 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
3204 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
3206 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3207 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3208 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
3209 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
3211 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3212 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3213 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
3214 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
3216 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3217 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3218 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
3219 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
3221 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3222 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3223 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3224 (DSubReg_i8_reg imm:$lane))),
3225 (SubReg_i8_lane imm:$lane))>;
3226 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3227 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3228 (DSubReg_i16_reg imm:$lane))),
3229 (SubReg_i16_lane imm:$lane))>;
3230 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3231 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3232 (DSubReg_i8_reg imm:$lane))),
3233 (SubReg_i8_lane imm:$lane))>;
3234 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3235 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3236 (DSubReg_i16_reg imm:$lane))),
3237 (SubReg_i16_lane imm:$lane))>;
3238 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3239 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3240 (DSubReg_i32_reg imm:$lane))),
3241 (SubReg_i32_lane imm:$lane))>;
3242 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3243 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3244 (SSubReg_f32_reg imm:$src2))>;
3245 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3246 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3247 (SSubReg_f32_reg imm:$src2))>;
3248 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3249 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3250 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3251 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3254 // VMOV : Vector Set Lane (move ARM core register to scalar)
3256 let Constraints = "$src1 = $dst" in {
3257 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
3258 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3259 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
3260 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
3261 GPR:$src2, imm:$lane))]>;
3262 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
3263 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3264 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
3265 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
3266 GPR:$src2, imm:$lane))]>;
3267 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
3268 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3269 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
3270 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
3271 GPR:$src2, imm:$lane))]>;
3273 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3274 (v16i8 (INSERT_SUBREG QPR:$src1,
3275 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3276 (DSubReg_i8_reg imm:$lane))),
3277 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3278 (DSubReg_i8_reg imm:$lane)))>;
3279 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3280 (v8i16 (INSERT_SUBREG QPR:$src1,
3281 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3282 (DSubReg_i16_reg imm:$lane))),
3283 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3284 (DSubReg_i16_reg imm:$lane)))>;
3285 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3286 (v4i32 (INSERT_SUBREG QPR:$src1,
3287 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3288 (DSubReg_i32_reg imm:$lane))),
3289 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3290 (DSubReg_i32_reg imm:$lane)))>;
3292 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3293 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3294 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3295 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3296 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3297 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3299 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3300 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3301 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3302 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3304 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3305 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3306 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3307 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3308 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3309 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3311 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3312 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3313 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3314 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3315 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3316 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3318 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3319 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3320 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3322 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3323 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3324 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3326 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3327 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3328 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3331 // VDUP : Vector Duplicate (from ARM core register to all elements)
3333 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3334 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3335 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3336 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3337 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3338 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3339 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3340 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3342 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3343 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3344 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3345 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3346 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3347 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3349 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3350 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3351 [(set DPR:$dst, (v2f32 (NEONvdup
3352 (f32 (bitconvert GPR:$src)))))]>;
3353 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3354 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3355 [(set QPR:$dst, (v4f32 (NEONvdup
3356 (f32 (bitconvert GPR:$src)))))]>;
3358 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3360 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3362 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3363 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3364 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3366 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3367 ValueType ResTy, ValueType OpTy>
3368 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3369 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3370 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3373 // Inst{19-16} is partially specified depending on the element size.
3375 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3376 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3377 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3378 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3379 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3380 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3381 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3382 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3384 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3385 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3386 (DSubReg_i8_reg imm:$lane))),
3387 (SubReg_i8_lane imm:$lane)))>;
3388 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3389 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3390 (DSubReg_i16_reg imm:$lane))),
3391 (SubReg_i16_lane imm:$lane)))>;
3392 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3393 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3394 (DSubReg_i32_reg imm:$lane))),
3395 (SubReg_i32_lane imm:$lane)))>;
3396 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3397 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3398 (DSubReg_i32_reg imm:$lane))),
3399 (SubReg_i32_lane imm:$lane)))>;
3401 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3402 (outs DPR:$dst), (ins SPR:$src),
3403 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3404 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3406 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3407 (outs QPR:$dst), (ins SPR:$src),
3408 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3409 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3411 // VMOVN : Vector Narrowing Move
3412 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3413 "vmovn", "i", trunc>;
3414 // VQMOVN : Vector Saturating Narrowing Move
3415 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3416 "vqmovn", "s", int_arm_neon_vqmovns>;
3417 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3418 "vqmovn", "u", int_arm_neon_vqmovnu>;
3419 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3420 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3421 // VMOVL : Vector Lengthening Move
3422 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3423 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3425 // Vector Conversions.
3427 // VCVT : Vector Convert Between Floating-Point and Integers
3428 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3429 v2i32, v2f32, fp_to_sint>;
3430 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3431 v2i32, v2f32, fp_to_uint>;
3432 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3433 v2f32, v2i32, sint_to_fp>;
3434 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3435 v2f32, v2i32, uint_to_fp>;
3437 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3438 v4i32, v4f32, fp_to_sint>;
3439 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3440 v4i32, v4f32, fp_to_uint>;
3441 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3442 v4f32, v4i32, sint_to_fp>;
3443 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3444 v4f32, v4i32, uint_to_fp>;
3446 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3447 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3448 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3449 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3450 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3451 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3452 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3453 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3454 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3456 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3457 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3458 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3459 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3460 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3461 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3462 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3463 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3467 // VREV64 : Vector Reverse elements within 64-bit doublewords
3469 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3470 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3471 (ins DPR:$src), IIC_VMOVD,
3472 OpcodeStr, Dt, "$dst, $src", "",
3473 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3474 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3475 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3476 (ins QPR:$src), IIC_VMOVD,
3477 OpcodeStr, Dt, "$dst, $src", "",
3478 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3480 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3481 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3482 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3483 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3485 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3486 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3487 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3488 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3490 // VREV32 : Vector Reverse elements within 32-bit words
3492 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3493 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3494 (ins DPR:$src), IIC_VMOVD,
3495 OpcodeStr, Dt, "$dst, $src", "",
3496 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3497 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3498 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3499 (ins QPR:$src), IIC_VMOVD,
3500 OpcodeStr, Dt, "$dst, $src", "",
3501 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3503 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3504 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3506 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3507 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3509 // VREV16 : Vector Reverse elements within 16-bit halfwords
3511 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3512 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3513 (ins DPR:$src), IIC_VMOVD,
3514 OpcodeStr, Dt, "$dst, $src", "",
3515 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3516 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3517 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3518 (ins QPR:$src), IIC_VMOVD,
3519 OpcodeStr, Dt, "$dst, $src", "",
3520 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3522 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3523 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3525 // Other Vector Shuffles.
3527 // VEXT : Vector Extract
3529 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3530 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3531 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3532 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3533 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3534 (Ty DPR:$rhs), imm:$index)))]>;
3536 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3537 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3538 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3539 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3540 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3541 (Ty QPR:$rhs), imm:$index)))]>;
3543 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3544 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3545 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3546 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3548 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3549 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3550 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3551 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3553 // VTRN : Vector Transpose
3555 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3556 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3557 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3559 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3560 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3561 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3563 // VUZP : Vector Unzip (Deinterleave)
3565 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3566 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3567 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3569 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3570 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3571 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3573 // VZIP : Vector Zip (Interleave)
3575 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3576 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3577 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3579 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3580 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3581 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3583 // Vector Table Lookup and Table Extension.
3585 // VTBL : Vector Table Lookup
3587 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3588 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3589 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3590 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3591 let hasExtraSrcRegAllocReq = 1 in {
3593 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3594 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3595 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "", []>;
3597 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3598 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3599 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "", []>;
3601 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3602 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3604 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "", []>;
3605 } // hasExtraSrcRegAllocReq = 1
3607 // VTBX : Vector Table Extension
3609 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3610 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3611 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3612 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3613 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3614 let hasExtraSrcRegAllocReq = 1 in {
3616 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3617 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3618 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst", []>;
3620 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3621 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3622 NVTBLFrm, IIC_VTBX3,
3623 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src",
3624 "$orig = $dst", []>;
3626 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3627 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3628 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3629 "$orig = $dst", []>;
3630 } // hasExtraSrcRegAllocReq = 1
3632 //===----------------------------------------------------------------------===//
3633 // NEON instructions for single-precision FP math
3634 //===----------------------------------------------------------------------===//
3636 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3637 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3638 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3642 class N3VSPat<SDNode OpNode, NeonI Inst>
3643 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3644 (EXTRACT_SUBREG (v2f32
3645 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3647 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3651 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3652 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3653 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3655 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3657 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3661 // These need separate instructions because they must use DPR_VFP2 register
3662 // class which have SPR sub-registers.
3664 // Vector Add Operations used for single-precision FP
3665 let neverHasSideEffects = 1 in
3666 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3667 def : N3VSPat<fadd, VADDfd_sfp>;
3669 // Vector Sub Operations used for single-precision FP
3670 let neverHasSideEffects = 1 in
3671 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3672 def : N3VSPat<fsub, VSUBfd_sfp>;
3674 // Vector Multiply Operations used for single-precision FP
3675 let neverHasSideEffects = 1 in
3676 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3677 def : N3VSPat<fmul, VMULfd_sfp>;
3679 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3680 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3681 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3683 //let neverHasSideEffects = 1 in
3684 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3685 // v2f32, fmul, fadd>;
3686 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3688 //let neverHasSideEffects = 1 in
3689 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3690 // v2f32, fmul, fsub>;
3691 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3693 // Vector Absolute used for single-precision FP
3694 let neverHasSideEffects = 1 in
3695 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3696 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3697 "vabs", "f32", "$dst, $src", "", []>;
3698 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3700 // Vector Negate used for single-precision FP
3701 let neverHasSideEffects = 1 in
3702 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3703 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3704 "vneg", "f32", "$dst, $src", "", []>;
3705 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3707 // Vector Maximum used for single-precision FP
3708 let neverHasSideEffects = 1 in
3709 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3710 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3711 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3712 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3714 // Vector Minimum used for single-precision FP
3715 let neverHasSideEffects = 1 in
3716 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3717 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3718 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3719 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3721 // Vector Convert between single-precision FP and integer
3722 let neverHasSideEffects = 1 in
3723 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3724 v2i32, v2f32, fp_to_sint>;
3725 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3727 let neverHasSideEffects = 1 in
3728 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3729 v2i32, v2f32, fp_to_uint>;
3730 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3732 let neverHasSideEffects = 1 in
3733 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3734 v2f32, v2i32, sint_to_fp>;
3735 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3737 let neverHasSideEffects = 1 in
3738 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3739 v2f32, v2i32, uint_to_fp>;
3740 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3742 //===----------------------------------------------------------------------===//
3743 // Non-Instruction Patterns
3744 //===----------------------------------------------------------------------===//
3747 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3748 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3749 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3750 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3751 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3752 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3753 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3754 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3755 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3756 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3757 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3758 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3759 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3760 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3761 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3762 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3763 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3764 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3765 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3766 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3767 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3768 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3769 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3770 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3771 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3772 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3773 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3774 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3775 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3776 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3778 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3779 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3780 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3781 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3782 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3783 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3784 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3785 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3786 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3787 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3788 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3789 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3790 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3791 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3792 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3793 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3794 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3795 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3796 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3797 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3798 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3799 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3800 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3801 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3802 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3803 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3804 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3805 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3806 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3807 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;