1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
69 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
70 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
72 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
74 // VDUPLANE can produce a quad-register result from a double-register source,
75 // so the result is not constrained to match the source.
76 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
77 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
80 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
81 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
82 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
84 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
85 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
86 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
87 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
89 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
93 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
94 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
96 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
98 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
99 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
101 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 2>]>;
103 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
104 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
106 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
107 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
108 unsigned EltBits = 0;
109 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
110 return (EltBits == 32 && EltVal == 0);
113 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
114 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
115 unsigned EltBits = 0;
116 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
117 return (EltBits == 8 && EltVal == 0xff);
120 //===----------------------------------------------------------------------===//
121 // NEON operand definitions
122 //===----------------------------------------------------------------------===//
124 def nModImm : Operand<i32> {
125 let PrintMethod = "printNEONModImmOperand";
128 //===----------------------------------------------------------------------===//
129 // NEON load / store instructions
130 //===----------------------------------------------------------------------===//
132 // Use vldmia to load a Q register as a D register pair.
133 // This is equivalent to VLDMD except that it has a Q register operand
134 // instead of a pair of D registers.
136 : AXDI4<(outs QPR:$dst), (ins addrmode4:$addr, pred:$p),
137 IndexModeNone, IIC_fpLoadm,
138 "vldm${addr:submode}${p}\t$addr, ${dst:dregpair}", "",
139 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]>;
141 let mayLoad = 1, neverHasSideEffects = 1 in {
142 // Use vld1 to load a Q register as a D register pair.
143 // This alternative to VLDMQ allows an alignment to be specified.
144 // This is equivalent to VLD1q64 except that it has a Q register operand.
146 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
147 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
148 } // mayLoad = 1, neverHasSideEffects = 1
150 // Use vstmia to store a Q register as a D register pair.
151 // This is equivalent to VSTMD except that it has a Q register operand
152 // instead of a pair of D registers.
154 : AXDI4<(outs), (ins QPR:$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:$addr), IIC_VLD2, "">;
175 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
176 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
179 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), IIC_VLD4, "">;
181 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
182 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
184 class VLDQQQQWBPseudo
185 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
186 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), IIC_VLD4,
187 "$addr.addr = $wb, $src = $dst">;
189 // VLD1 : Vector Load (multiple single elements)
190 class VLD1D<bits<4> op7_4, string Dt>
191 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
192 (ins addrmode6:$addr), IIC_VLD1,
193 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
194 class VLD1Q<bits<4> op7_4, string Dt>
195 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
196 (ins addrmode6:$addr), IIC_VLD1,
197 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
199 def VLD1d8 : VLD1D<0b0000, "8">;
200 def VLD1d16 : VLD1D<0b0100, "16">;
201 def VLD1d32 : VLD1D<0b1000, "32">;
202 def VLD1d64 : VLD1D<0b1100, "64">;
204 def VLD1q8 : VLD1Q<0b0000, "8">;
205 def VLD1q16 : VLD1Q<0b0100, "16">;
206 def VLD1q32 : VLD1Q<0b1000, "32">;
207 def VLD1q64 : VLD1Q<0b1100, "64">;
209 def VLD1q8Pseudo : VLDQPseudo;
210 def VLD1q16Pseudo : VLDQPseudo;
211 def VLD1q32Pseudo : VLDQPseudo;
212 def VLD1q64Pseudo : VLDQPseudo;
214 // ...with address register writeback:
215 class VLD1DWB<bits<4> op7_4, string Dt>
216 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
217 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
218 "vld1", Dt, "\\{$dst\\}, $addr$offset",
219 "$addr.addr = $wb", []>;
220 class VLD1QWB<bits<4> op7_4, string Dt>
221 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
222 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
223 "vld1", Dt, "${dst:dregpair}, $addr$offset",
224 "$addr.addr = $wb", []>;
226 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
227 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
228 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
229 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
231 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
232 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
233 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
234 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
236 def VLD1q8Pseudo_UPD : VLDQWBPseudo;
237 def VLD1q16Pseudo_UPD : VLDQWBPseudo;
238 def VLD1q32Pseudo_UPD : VLDQWBPseudo;
239 def VLD1q64Pseudo_UPD : VLDQWBPseudo;
241 // ...with 3 registers (some of these are only for the disassembler):
242 class VLD1D3<bits<4> op7_4, string Dt>
243 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
244 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
245 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
246 class VLD1D3WB<bits<4> op7_4, string Dt>
247 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
248 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
249 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
251 def VLD1d8T : VLD1D3<0b0000, "8">;
252 def VLD1d16T : VLD1D3<0b0100, "16">;
253 def VLD1d32T : VLD1D3<0b1000, "32">;
254 def VLD1d64T : VLD1D3<0b1100, "64">;
256 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
257 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
258 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
259 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
261 def VLD1d64TPseudo : VLDQQPseudo;
262 def VLD1d64TPseudo_UPD : VLDQQWBPseudo;
264 // ...with 4 registers (some of these are only for the disassembler):
265 class VLD1D4<bits<4> op7_4, string Dt>
266 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
267 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
268 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
269 class VLD1D4WB<bits<4> op7_4, string Dt>
270 : NLdSt<0,0b10,0b0010,op7_4,
271 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
272 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
273 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
276 def VLD1d8Q : VLD1D4<0b0000, "8">;
277 def VLD1d16Q : VLD1D4<0b0100, "16">;
278 def VLD1d32Q : VLD1D4<0b1000, "32">;
279 def VLD1d64Q : VLD1D4<0b1100, "64">;
281 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
282 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
283 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
284 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
286 def VLD1d64QPseudo : VLDQQPseudo;
287 def VLD1d64QPseudo_UPD : VLDQQWBPseudo;
289 // VLD2 : Vector Load (multiple 2-element structures)
290 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
291 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
292 (ins addrmode6:$addr), IIC_VLD2,
293 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
294 class VLD2Q<bits<4> op7_4, string Dt>
295 : NLdSt<0, 0b10, 0b0011, op7_4,
296 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
297 (ins addrmode6:$addr), IIC_VLD2,
298 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
300 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
301 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
302 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
304 def VLD2q8 : VLD2Q<0b0000, "8">;
305 def VLD2q16 : VLD2Q<0b0100, "16">;
306 def VLD2q32 : VLD2Q<0b1000, "32">;
308 def VLD2d8Pseudo : VLDQPseudo;
309 def VLD2d16Pseudo : VLDQPseudo;
310 def VLD2d32Pseudo : VLDQPseudo;
312 def VLD2q8Pseudo : VLDQQPseudo;
313 def VLD2q16Pseudo : VLDQQPseudo;
314 def VLD2q32Pseudo : VLDQQPseudo;
316 // ...with address register writeback:
317 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
318 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
319 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
320 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
321 "$addr.addr = $wb", []>;
322 class VLD2QWB<bits<4> op7_4, string Dt>
323 : NLdSt<0, 0b10, 0b0011, op7_4,
324 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
325 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
326 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
327 "$addr.addr = $wb", []>;
329 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
330 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
331 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
333 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
334 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
335 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
337 def VLD2d8Pseudo_UPD : VLDQWBPseudo;
338 def VLD2d16Pseudo_UPD : VLDQWBPseudo;
339 def VLD2d32Pseudo_UPD : VLDQWBPseudo;
341 def VLD2q8Pseudo_UPD : VLDQQWBPseudo;
342 def VLD2q16Pseudo_UPD : VLDQQWBPseudo;
343 def VLD2q32Pseudo_UPD : VLDQQWBPseudo;
345 // ...with double-spaced registers (for disassembly only):
346 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
347 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
348 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
349 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
350 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
351 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
353 // VLD3 : Vector Load (multiple 3-element structures)
354 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
355 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
356 (ins addrmode6:$addr), IIC_VLD3,
357 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
359 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
360 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
361 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
363 def VLD3d8Pseudo : VLDQQPseudo;
364 def VLD3d16Pseudo : VLDQQPseudo;
365 def VLD3d32Pseudo : VLDQQPseudo;
367 // ...with address register writeback:
368 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
369 : NLdSt<0, 0b10, op11_8, op7_4,
370 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
371 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
372 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
373 "$addr.addr = $wb", []>;
375 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
376 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
377 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
379 def VLD3d8Pseudo_UPD : VLDQQWBPseudo;
380 def VLD3d16Pseudo_UPD : VLDQQWBPseudo;
381 def VLD3d32Pseudo_UPD : VLDQQWBPseudo;
383 // ...with double-spaced registers (non-updating versions for disassembly only):
384 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
385 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
386 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
387 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
388 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
389 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
391 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo;
392 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo;
393 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo;
395 // ...alternate versions to be allocated odd register numbers:
396 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo;
397 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo;
398 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo;
400 // VLD4 : Vector Load (multiple 4-element structures)
401 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
402 : NLdSt<0, 0b10, op11_8, op7_4,
403 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
404 (ins addrmode6:$addr), IIC_VLD4,
405 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
407 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
408 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
409 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
411 def VLD4d8Pseudo : VLDQQPseudo;
412 def VLD4d16Pseudo : VLDQQPseudo;
413 def VLD4d32Pseudo : VLDQQPseudo;
415 // ...with address register writeback:
416 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
417 : NLdSt<0, 0b10, op11_8, op7_4,
418 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
419 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
420 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
421 "$addr.addr = $wb", []>;
423 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
424 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
425 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
427 def VLD4d8Pseudo_UPD : VLDQQWBPseudo;
428 def VLD4d16Pseudo_UPD : VLDQQWBPseudo;
429 def VLD4d32Pseudo_UPD : VLDQQWBPseudo;
431 // ...with double-spaced registers (non-updating versions for disassembly only):
432 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
433 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
434 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
435 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
436 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
437 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
439 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo;
440 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo;
441 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo;
443 // ...alternate versions to be allocated odd register numbers:
444 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo;
445 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo;
446 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo;
448 // Classes for VLD*LN pseudo-instructions with multi-register operands.
449 // These are expanded to real instructions after register allocation.
450 class VLDQLNPseudo<InstrItinClass itin>
451 : PseudoNLdSt<(outs QPR:$dst),
452 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
453 itin, "$src = $dst">;
454 class VLDQLNWBPseudo<InstrItinClass itin>
455 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
456 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
457 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
458 class VLDQQLNPseudo<InstrItinClass itin>
459 : PseudoNLdSt<(outs QQPR:$dst),
460 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
461 itin, "$src = $dst">;
462 class VLDQQLNWBPseudo<InstrItinClass itin>
463 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
464 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
465 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
466 class VLDQQQQLNPseudo<InstrItinClass itin>
467 : PseudoNLdSt<(outs QQQQPR:$dst),
468 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
469 itin, "$src = $dst">;
470 class VLDQQQQLNWBPseudo<InstrItinClass itin>
471 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
472 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
473 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
475 // VLD1LN : Vector Load (single element to one lane)
476 // FIXME: Not yet implemented.
478 // VLD2LN : Vector Load (single 2-element structure to one lane)
479 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
480 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
481 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
482 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
483 "$src1 = $dst1, $src2 = $dst2", []>;
485 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
486 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
487 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
489 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2>;
490 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2>;
491 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2>;
493 // ...with double-spaced registers:
494 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
495 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
497 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2>;
498 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2>;
500 // ...with address register writeback:
501 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
502 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
503 (ins addrmode6:$addr, am6offset:$offset,
504 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
505 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
506 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
508 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
509 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
510 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
512 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2>;
513 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2>;
514 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2>;
516 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
517 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
519 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2>;
520 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2>;
522 // VLD3LN : Vector Load (single 3-element structure to one lane)
523 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
524 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
525 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
526 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
527 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
528 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
530 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
531 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
532 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
534 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3>;
535 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3>;
536 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3>;
538 // ...with double-spaced registers:
539 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
540 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
542 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3>;
543 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3>;
545 // ...with address register writeback:
546 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
547 : NLdSt<1, 0b10, op11_8, op7_4,
548 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
549 (ins addrmode6:$addr, am6offset:$offset,
550 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
551 IIC_VLD3, "vld3", Dt,
552 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
553 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
556 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
557 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
558 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
560 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
561 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
562 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
564 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
565 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
567 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3>;
568 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3>;
570 // VLD4LN : Vector Load (single 4-element structure to one lane)
571 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
572 : NLdSt<1, 0b10, op11_8, op7_4,
573 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
574 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
575 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
576 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
577 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
579 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
580 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
581 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
583 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4>;
584 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4>;
585 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4>;
587 // ...with double-spaced registers:
588 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
589 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
591 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4>;
592 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4>;
594 // ...with address register writeback:
595 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
596 : NLdSt<1, 0b10, op11_8, op7_4,
597 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
598 (ins addrmode6:$addr, am6offset:$offset,
599 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
600 IIC_VLD4, "vld4", Dt,
601 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
602 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
605 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
606 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
607 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
609 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4>;
610 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4>;
611 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4>;
613 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
614 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
616 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4>;
617 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4>;
619 // VLD1DUP : Vector Load (single element to all lanes)
620 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
621 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
622 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
623 // FIXME: Not yet implemented.
624 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
626 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
628 // Classes for VST* pseudo-instructions with multi-register operands.
629 // These are expanded to real instructions after register allocation.
631 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), IIC_VST, "">;
633 : PseudoNLdSt<(outs GPR:$wb),
634 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
637 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), IIC_VST, "">;
639 : PseudoNLdSt<(outs GPR:$wb),
640 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), IIC_VST,
642 class VSTQQQQWBPseudo
643 : PseudoNLdSt<(outs GPR:$wb),
644 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), IIC_VST,
647 // VST1 : Vector Store (multiple single elements)
648 class VST1D<bits<4> op7_4, string Dt>
649 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
650 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
651 class VST1Q<bits<4> op7_4, string Dt>
652 : NLdSt<0,0b00,0b1010,op7_4, (outs),
653 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
654 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
656 def VST1d8 : VST1D<0b0000, "8">;
657 def VST1d16 : VST1D<0b0100, "16">;
658 def VST1d32 : VST1D<0b1000, "32">;
659 def VST1d64 : VST1D<0b1100, "64">;
661 def VST1q8 : VST1Q<0b0000, "8">;
662 def VST1q16 : VST1Q<0b0100, "16">;
663 def VST1q32 : VST1Q<0b1000, "32">;
664 def VST1q64 : VST1Q<0b1100, "64">;
666 def VST1q8Pseudo : VSTQPseudo;
667 def VST1q16Pseudo : VSTQPseudo;
668 def VST1q32Pseudo : VSTQPseudo;
669 def VST1q64Pseudo : VSTQPseudo;
671 // ...with address register writeback:
672 class VST1DWB<bits<4> op7_4, string Dt>
673 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
674 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
675 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
676 class VST1QWB<bits<4> op7_4, string Dt>
677 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
678 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
679 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
681 def VST1d8_UPD : VST1DWB<0b0000, "8">;
682 def VST1d16_UPD : VST1DWB<0b0100, "16">;
683 def VST1d32_UPD : VST1DWB<0b1000, "32">;
684 def VST1d64_UPD : VST1DWB<0b1100, "64">;
686 def VST1q8_UPD : VST1QWB<0b0000, "8">;
687 def VST1q16_UPD : VST1QWB<0b0100, "16">;
688 def VST1q32_UPD : VST1QWB<0b1000, "32">;
689 def VST1q64_UPD : VST1QWB<0b1100, "64">;
691 def VST1q8Pseudo_UPD : VSTQWBPseudo;
692 def VST1q16Pseudo_UPD : VSTQWBPseudo;
693 def VST1q32Pseudo_UPD : VSTQWBPseudo;
694 def VST1q64Pseudo_UPD : VSTQWBPseudo;
696 // ...with 3 registers (some of these are only for the disassembler):
697 class VST1D3<bits<4> op7_4, string Dt>
698 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
699 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
700 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
701 class VST1D3WB<bits<4> op7_4, string Dt>
702 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
703 (ins addrmode6:$addr, am6offset:$offset,
704 DPR:$src1, DPR:$src2, DPR:$src3),
705 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
706 "$addr.addr = $wb", []>;
708 def VST1d8T : VST1D3<0b0000, "8">;
709 def VST1d16T : VST1D3<0b0100, "16">;
710 def VST1d32T : VST1D3<0b1000, "32">;
711 def VST1d64T : VST1D3<0b1100, "64">;
713 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
714 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
715 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
716 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
718 def VST1d64TPseudo : VSTQQPseudo;
719 def VST1d64TPseudo_UPD : VSTQQWBPseudo;
721 // ...with 4 registers (some of these are only for the disassembler):
722 class VST1D4<bits<4> op7_4, string Dt>
723 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
724 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
725 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
727 class VST1D4WB<bits<4> op7_4, string Dt>
728 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
729 (ins addrmode6:$addr, am6offset:$offset,
730 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
731 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
732 "$addr.addr = $wb", []>;
734 def VST1d8Q : VST1D4<0b0000, "8">;
735 def VST1d16Q : VST1D4<0b0100, "16">;
736 def VST1d32Q : VST1D4<0b1000, "32">;
737 def VST1d64Q : VST1D4<0b1100, "64">;
739 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
740 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
741 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
742 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
744 def VST1d64QPseudo : VSTQQPseudo;
745 def VST1d64QPseudo_UPD : VSTQQWBPseudo;
747 // VST2 : Vector Store (multiple 2-element structures)
748 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
749 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
750 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
751 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
752 class VST2Q<bits<4> op7_4, string Dt>
753 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
754 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
755 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
758 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
759 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
760 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
762 def VST2q8 : VST2Q<0b0000, "8">;
763 def VST2q16 : VST2Q<0b0100, "16">;
764 def VST2q32 : VST2Q<0b1000, "32">;
766 def VST2d8Pseudo : VSTQPseudo;
767 def VST2d16Pseudo : VSTQPseudo;
768 def VST2d32Pseudo : VSTQPseudo;
770 def VST2q8Pseudo : VSTQQPseudo;
771 def VST2q16Pseudo : VSTQQPseudo;
772 def VST2q32Pseudo : VSTQQPseudo;
774 // ...with address register writeback:
775 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
776 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
777 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
778 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
779 "$addr.addr = $wb", []>;
780 class VST2QWB<bits<4> op7_4, string Dt>
781 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
782 (ins addrmode6:$addr, am6offset:$offset,
783 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
784 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
785 "$addr.addr = $wb", []>;
787 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
788 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
789 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
791 def VST2q8_UPD : VST2QWB<0b0000, "8">;
792 def VST2q16_UPD : VST2QWB<0b0100, "16">;
793 def VST2q32_UPD : VST2QWB<0b1000, "32">;
795 def VST2d8Pseudo_UPD : VSTQWBPseudo;
796 def VST2d16Pseudo_UPD : VSTQWBPseudo;
797 def VST2d32Pseudo_UPD : VSTQWBPseudo;
799 def VST2q8Pseudo_UPD : VSTQQWBPseudo;
800 def VST2q16Pseudo_UPD : VSTQQWBPseudo;
801 def VST2q32Pseudo_UPD : VSTQQWBPseudo;
803 // ...with double-spaced registers (for disassembly only):
804 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
805 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
806 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
807 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
808 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
809 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
811 // VST3 : Vector Store (multiple 3-element structures)
812 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
813 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
814 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
815 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
817 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
818 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
819 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
821 def VST3d8Pseudo : VSTQQPseudo;
822 def VST3d16Pseudo : VSTQQPseudo;
823 def VST3d32Pseudo : VSTQQPseudo;
825 // ...with address register writeback:
826 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
827 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
828 (ins addrmode6:$addr, am6offset:$offset,
829 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
830 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
831 "$addr.addr = $wb", []>;
833 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
834 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
835 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
837 def VST3d8Pseudo_UPD : VSTQQWBPseudo;
838 def VST3d16Pseudo_UPD : VSTQQWBPseudo;
839 def VST3d32Pseudo_UPD : VSTQQWBPseudo;
841 // ...with double-spaced registers (non-updating versions for disassembly only):
842 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
843 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
844 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
845 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
846 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
847 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
849 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo;
850 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo;
851 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo;
853 // ...alternate versions to be allocated odd register numbers:
854 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo;
855 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo;
856 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo;
858 // VST4 : Vector Store (multiple 4-element structures)
859 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
860 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
861 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
862 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
865 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
866 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
867 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
869 def VST4d8Pseudo : VSTQQPseudo;
870 def VST4d16Pseudo : VSTQQPseudo;
871 def VST4d32Pseudo : VSTQQPseudo;
873 // ...with address register writeback:
874 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
875 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
876 (ins addrmode6:$addr, am6offset:$offset,
877 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
878 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
879 "$addr.addr = $wb", []>;
881 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
882 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
883 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
885 def VST4d8Pseudo_UPD : VSTQQWBPseudo;
886 def VST4d16Pseudo_UPD : VSTQQWBPseudo;
887 def VST4d32Pseudo_UPD : VSTQQWBPseudo;
889 // ...with double-spaced registers (non-updating versions for disassembly only):
890 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
891 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
892 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
893 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
894 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
895 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
897 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo;
898 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo;
899 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo;
901 // ...alternate versions to be allocated odd register numbers:
902 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo;
903 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo;
904 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo;
906 // Classes for VST*LN pseudo-instructions with multi-register operands.
907 // These are expanded to real instructions after register allocation.
908 class VSTQLNPseudo<InstrItinClass itin>
909 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
911 class VSTQLNWBPseudo<InstrItinClass itin>
912 : PseudoNLdSt<(outs GPR:$wb),
913 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
914 nohash_imm:$lane), itin, "$addr.addr = $wb">;
915 class VSTQQLNPseudo<InstrItinClass itin>
916 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
918 class VSTQQLNWBPseudo<InstrItinClass itin>
919 : PseudoNLdSt<(outs GPR:$wb),
920 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
921 nohash_imm:$lane), itin, "$addr.addr = $wb">;
922 class VSTQQQQLNPseudo<InstrItinClass itin>
923 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
925 class VSTQQQQLNWBPseudo<InstrItinClass itin>
926 : PseudoNLdSt<(outs GPR:$wb),
927 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
928 nohash_imm:$lane), itin, "$addr.addr = $wb">;
930 // VST1LN : Vector Store (single element from one lane)
931 // FIXME: Not yet implemented.
933 // VST2LN : Vector Store (single 2-element structure from one lane)
934 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
935 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
936 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
937 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
940 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
941 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
942 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
944 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST>;
945 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST>;
946 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST>;
948 // ...with double-spaced registers:
949 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
950 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
952 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST>;
953 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST>;
955 // ...with address register writeback:
956 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
957 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
958 (ins addrmode6:$addr, am6offset:$offset,
959 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
960 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
961 "$addr.addr = $wb", []>;
963 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
964 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
965 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
967 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST>;
968 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST>;
969 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST>;
971 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
972 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
974 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
975 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
977 // VST3LN : Vector Store (single 3-element structure from one lane)
978 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
979 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
980 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
981 nohash_imm:$lane), IIC_VST, "vst3", Dt,
982 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
984 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
985 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
986 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
988 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST>;
989 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST>;
990 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST>;
992 // ...with double-spaced registers:
993 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
994 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
996 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST>;
997 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST>;
999 // ...with address register writeback:
1000 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1001 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1002 (ins addrmode6:$addr, am6offset:$offset,
1003 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1004 IIC_VST, "vst3", Dt,
1005 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
1006 "$addr.addr = $wb", []>;
1008 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
1009 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
1010 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
1012 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1013 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1014 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1016 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
1017 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
1019 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1020 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1022 // VST4LN : Vector Store (single 4-element structure from one lane)
1023 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1024 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1025 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1026 nohash_imm:$lane), IIC_VST, "vst4", Dt,
1027 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
1030 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
1031 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
1032 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
1034 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST>;
1035 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST>;
1036 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST>;
1038 // ...with double-spaced registers:
1039 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
1040 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
1042 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST>;
1043 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST>;
1045 // ...with address register writeback:
1046 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1047 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1048 (ins addrmode6:$addr, am6offset:$offset,
1049 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1050 IIC_VST, "vst4", Dt,
1051 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
1052 "$addr.addr = $wb", []>;
1054 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
1055 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
1056 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
1058 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1059 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1060 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST>;
1062 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
1063 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
1065 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1066 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST>;
1068 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1071 //===----------------------------------------------------------------------===//
1072 // NEON pattern fragments
1073 //===----------------------------------------------------------------------===//
1075 // Extract D sub-registers of Q registers.
1076 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1077 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1078 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1080 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1081 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1082 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1084 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1085 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1086 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1088 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1089 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1090 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1093 // Extract S sub-registers of Q/D registers.
1094 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1095 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1096 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1099 // Translate lane numbers from Q registers to D subregs.
1100 def SubReg_i8_lane : SDNodeXForm<imm, [{
1101 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1103 def SubReg_i16_lane : SDNodeXForm<imm, [{
1104 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1106 def SubReg_i32_lane : SDNodeXForm<imm, [{
1107 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1110 //===----------------------------------------------------------------------===//
1111 // Instruction Classes
1112 //===----------------------------------------------------------------------===//
1114 // Basic 2-register operations: single-, double- and quad-register.
1115 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1116 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1117 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1118 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1119 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1120 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1121 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1122 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1123 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1124 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1125 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1126 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1127 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1128 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1129 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1130 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1131 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1132 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1134 // Basic 2-register intrinsics, both double- and quad-register.
1135 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1136 bits<2> op17_16, bits<5> op11_7, bit op4,
1137 InstrItinClass itin, string OpcodeStr, string Dt,
1138 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1139 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1140 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1141 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1142 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1143 bits<2> op17_16, bits<5> op11_7, bit op4,
1144 InstrItinClass itin, string OpcodeStr, string Dt,
1145 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1146 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1147 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1148 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1150 // Narrow 2-register operations.
1151 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1152 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1153 InstrItinClass itin, string OpcodeStr, string Dt,
1154 ValueType TyD, ValueType TyQ, SDNode OpNode>
1155 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1156 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1157 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1159 // Narrow 2-register intrinsics.
1160 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1161 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1162 InstrItinClass itin, string OpcodeStr, string Dt,
1163 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1164 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1165 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1166 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1168 // Long 2-register operations (currently only used for VMOVL).
1169 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1170 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1171 InstrItinClass itin, string OpcodeStr, string Dt,
1172 ValueType TyQ, ValueType TyD, SDNode OpNode>
1173 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1174 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1175 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1177 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1178 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1179 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1180 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1181 OpcodeStr, Dt, "$dst1, $dst2",
1182 "$src1 = $dst1, $src2 = $dst2", []>;
1183 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1184 InstrItinClass itin, string OpcodeStr, string Dt>
1185 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1186 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1187 "$src1 = $dst1, $src2 = $dst2", []>;
1189 // Basic 3-register operations: single-, double- and quad-register.
1190 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1191 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1192 SDNode OpNode, bit Commutable>
1193 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1194 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1195 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1196 let isCommutable = Commutable;
1199 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1200 InstrItinClass itin, string OpcodeStr, string Dt,
1201 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1202 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1203 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1204 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1205 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1206 let isCommutable = Commutable;
1208 // Same as N3VD but no data type.
1209 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1210 InstrItinClass itin, string OpcodeStr,
1211 ValueType ResTy, ValueType OpTy,
1212 SDNode OpNode, bit Commutable>
1213 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1214 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1215 OpcodeStr, "$dst, $src1, $src2", "",
1216 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1217 let isCommutable = Commutable;
1220 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1221 InstrItinClass itin, string OpcodeStr, string Dt,
1222 ValueType Ty, SDNode ShOp>
1223 : N3V<0, 1, op21_20, op11_8, 1, 0,
1224 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1225 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1226 [(set (Ty DPR:$dst),
1227 (Ty (ShOp (Ty DPR:$src1),
1228 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1229 let isCommutable = 0;
1231 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1232 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1233 : N3V<0, 1, op21_20, op11_8, 1, 0,
1234 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1235 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1236 [(set (Ty DPR:$dst),
1237 (Ty (ShOp (Ty DPR:$src1),
1238 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1239 let isCommutable = 0;
1242 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1243 InstrItinClass itin, string OpcodeStr, string Dt,
1244 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1245 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1246 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1247 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1248 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1249 let isCommutable = Commutable;
1251 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1252 InstrItinClass itin, string OpcodeStr,
1253 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1254 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1255 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1256 OpcodeStr, "$dst, $src1, $src2", "",
1257 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1258 let isCommutable = Commutable;
1260 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1261 InstrItinClass itin, string OpcodeStr, string Dt,
1262 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1263 : N3V<1, 1, op21_20, op11_8, 1, 0,
1264 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1265 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1266 [(set (ResTy QPR:$dst),
1267 (ResTy (ShOp (ResTy QPR:$src1),
1268 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1270 let isCommutable = 0;
1272 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1273 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1274 : N3V<1, 1, op21_20, op11_8, 1, 0,
1275 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1276 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1277 [(set (ResTy QPR:$dst),
1278 (ResTy (ShOp (ResTy QPR:$src1),
1279 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1281 let isCommutable = 0;
1284 // Basic 3-register intrinsics, both double- and quad-register.
1285 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1286 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1287 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1288 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1289 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1290 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1291 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1292 let isCommutable = Commutable;
1294 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1295 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1296 : N3V<0, 1, op21_20, op11_8, 1, 0,
1297 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1298 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1299 [(set (Ty DPR:$dst),
1300 (Ty (IntOp (Ty DPR:$src1),
1301 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1303 let isCommutable = 0;
1305 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1306 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1307 : N3V<0, 1, op21_20, op11_8, 1, 0,
1308 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1309 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1310 [(set (Ty DPR:$dst),
1311 (Ty (IntOp (Ty DPR:$src1),
1312 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1313 let isCommutable = 0;
1316 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1317 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1318 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1319 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1320 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1321 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1322 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1323 let isCommutable = Commutable;
1325 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1326 string OpcodeStr, string Dt,
1327 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1328 : N3V<1, 1, op21_20, op11_8, 1, 0,
1329 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1330 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1331 [(set (ResTy QPR:$dst),
1332 (ResTy (IntOp (ResTy QPR:$src1),
1333 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1335 let isCommutable = 0;
1337 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1338 string OpcodeStr, string Dt,
1339 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1340 : N3V<1, 1, op21_20, op11_8, 1, 0,
1341 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1342 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1343 [(set (ResTy QPR:$dst),
1344 (ResTy (IntOp (ResTy QPR:$src1),
1345 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1347 let isCommutable = 0;
1350 // Multiply-Add/Sub operations: single-, double- and quad-register.
1351 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1352 InstrItinClass itin, string OpcodeStr, string Dt,
1353 ValueType Ty, SDNode MulOp, SDNode OpNode>
1354 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1355 (outs DPR_VFP2:$dst),
1356 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1357 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1359 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1360 InstrItinClass itin, string OpcodeStr, string Dt,
1361 ValueType Ty, SDNode MulOp, SDNode OpNode>
1362 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1363 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1364 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1365 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1366 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1367 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1368 string OpcodeStr, string Dt,
1369 ValueType Ty, SDNode MulOp, SDNode ShOp>
1370 : N3V<0, 1, op21_20, op11_8, 1, 0,
1372 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1374 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1375 [(set (Ty DPR:$dst),
1376 (Ty (ShOp (Ty DPR:$src1),
1377 (Ty (MulOp DPR:$src2,
1378 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1380 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1381 string OpcodeStr, string Dt,
1382 ValueType Ty, SDNode MulOp, SDNode ShOp>
1383 : N3V<0, 1, op21_20, op11_8, 1, 0,
1385 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1387 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1388 [(set (Ty DPR:$dst),
1389 (Ty (ShOp (Ty DPR:$src1),
1390 (Ty (MulOp DPR:$src2,
1391 (Ty (NEONvduplane (Ty DPR_8:$src3),
1394 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1395 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1396 SDNode MulOp, SDNode OpNode>
1397 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1398 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1399 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1400 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1401 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1402 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1403 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1404 SDNode MulOp, SDNode ShOp>
1405 : N3V<1, 1, op21_20, op11_8, 1, 0,
1407 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1409 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1410 [(set (ResTy QPR:$dst),
1411 (ResTy (ShOp (ResTy QPR:$src1),
1412 (ResTy (MulOp QPR:$src2,
1413 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1415 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1416 string OpcodeStr, string Dt,
1417 ValueType ResTy, ValueType OpTy,
1418 SDNode MulOp, SDNode ShOp>
1419 : N3V<1, 1, op21_20, op11_8, 1, 0,
1421 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1423 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1424 [(set (ResTy QPR:$dst),
1425 (ResTy (ShOp (ResTy QPR:$src1),
1426 (ResTy (MulOp QPR:$src2,
1427 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1430 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1431 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1432 InstrItinClass itin, string OpcodeStr, string Dt,
1433 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1434 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1435 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1436 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1437 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1438 (Ty (IntOp (Ty DPR:$src2), (Ty DPR:$src3))))))]>;
1439 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1440 InstrItinClass itin, string OpcodeStr, string Dt,
1441 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1442 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1443 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1444 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1445 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1446 (Ty (IntOp (Ty QPR:$src2), (Ty QPR:$src3))))))]>;
1448 // Neon 3-argument intrinsics, both double- and quad-register.
1449 // The destination register is also used as the first source operand register.
1450 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1451 InstrItinClass itin, string OpcodeStr, string Dt,
1452 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1453 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1454 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1455 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1456 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1457 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1458 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1459 InstrItinClass itin, string OpcodeStr, string Dt,
1460 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1461 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1462 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1463 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1464 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1465 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1467 // Long Multiply-Add/Sub operations.
1468 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1469 InstrItinClass itin, string OpcodeStr, string Dt,
1470 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1471 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1472 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1473 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1474 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1475 (TyQ (MulOp (TyD DPR:$src2),
1476 (TyD DPR:$src3)))))]>;
1477 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1478 InstrItinClass itin, string OpcodeStr, string Dt,
1479 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1480 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1481 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1483 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1485 (OpNode (TyQ QPR:$src1),
1486 (TyQ (MulOp (TyD DPR:$src2),
1487 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1489 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1490 InstrItinClass itin, string OpcodeStr, string Dt,
1491 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1492 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1493 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1495 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1497 (OpNode (TyQ QPR:$src1),
1498 (TyQ (MulOp (TyD DPR:$src2),
1499 (TyD (NEONvduplane (TyD DPR_8:$src3),
1502 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1503 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1504 InstrItinClass itin, string OpcodeStr, string Dt,
1505 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1507 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1508 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1509 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1510 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1511 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src2),
1512 (TyD DPR:$src3)))))))]>;
1514 // Neon Long 3-argument intrinsic. The destination register is
1515 // a quad-register and is also used as the first source operand register.
1516 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1517 InstrItinClass itin, string OpcodeStr, string Dt,
1518 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1519 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1520 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1521 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1523 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1524 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1525 string OpcodeStr, string Dt,
1526 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1527 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1529 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1531 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1532 [(set (ResTy QPR:$dst),
1533 (ResTy (IntOp (ResTy QPR:$src1),
1535 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1537 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1538 InstrItinClass itin, string OpcodeStr, string Dt,
1539 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1540 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1542 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1544 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1545 [(set (ResTy QPR:$dst),
1546 (ResTy (IntOp (ResTy QPR:$src1),
1548 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1551 // Narrowing 3-register intrinsics.
1552 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1553 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1554 Intrinsic IntOp, bit Commutable>
1555 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1556 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1557 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1558 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1559 let isCommutable = Commutable;
1562 // Long 3-register operations.
1563 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1564 InstrItinClass itin, string OpcodeStr, string Dt,
1565 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1566 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1567 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1568 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1569 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1570 let isCommutable = Commutable;
1572 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1573 InstrItinClass itin, string OpcodeStr, string Dt,
1574 ValueType TyQ, ValueType TyD, SDNode OpNode>
1575 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1576 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1577 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1579 (TyQ (OpNode (TyD DPR:$src1),
1580 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1581 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1582 InstrItinClass itin, string OpcodeStr, string Dt,
1583 ValueType TyQ, ValueType TyD, SDNode OpNode>
1584 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1585 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1586 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1588 (TyQ (OpNode (TyD DPR:$src1),
1589 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1591 // Long 3-register operations with explicitly extended operands.
1592 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1593 InstrItinClass itin, string OpcodeStr, string Dt,
1594 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1596 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1597 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1598 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1599 [(set QPR:$dst, (OpNode (TyQ (ExtOp (TyD DPR:$src1))),
1600 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1601 let isCommutable = Commutable;
1604 // Long 3-register intrinsics with explicit extend (VABDL).
1605 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1606 InstrItinClass itin, string OpcodeStr, string Dt,
1607 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1609 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1610 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1611 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1612 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1613 (TyD DPR:$src2))))))]> {
1614 let isCommutable = Commutable;
1617 // Long 3-register intrinsics.
1618 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1619 InstrItinClass itin, string OpcodeStr, string Dt,
1620 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1621 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1622 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1623 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1624 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1625 let isCommutable = Commutable;
1627 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1628 string OpcodeStr, string Dt,
1629 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1630 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1631 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1632 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1633 [(set (ResTy QPR:$dst),
1634 (ResTy (IntOp (OpTy DPR:$src1),
1635 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1637 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1638 InstrItinClass itin, string OpcodeStr, string Dt,
1639 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1640 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1641 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1642 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1643 [(set (ResTy QPR:$dst),
1644 (ResTy (IntOp (OpTy DPR:$src1),
1645 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1648 // Wide 3-register operations.
1649 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1650 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1651 SDNode OpNode, SDNode ExtOp, bit Commutable>
1652 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1653 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1654 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1655 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1656 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1657 let isCommutable = Commutable;
1660 // Pairwise long 2-register intrinsics, both double- and quad-register.
1661 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1662 bits<2> op17_16, bits<5> op11_7, bit op4,
1663 string OpcodeStr, string Dt,
1664 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1665 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1666 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1667 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1668 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1669 bits<2> op17_16, bits<5> op11_7, bit op4,
1670 string OpcodeStr, string Dt,
1671 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1672 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1673 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1674 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1676 // Pairwise long 2-register accumulate intrinsics,
1677 // both double- and quad-register.
1678 // The destination register is also used as the first source operand register.
1679 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1680 bits<2> op17_16, bits<5> op11_7, bit op4,
1681 string OpcodeStr, string Dt,
1682 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1683 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1684 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1685 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1686 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1687 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1688 bits<2> op17_16, bits<5> op11_7, bit op4,
1689 string OpcodeStr, string Dt,
1690 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1691 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1692 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1693 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1694 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1696 // Shift by immediate,
1697 // both double- and quad-register.
1698 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1699 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1700 ValueType Ty, SDNode OpNode>
1701 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1702 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1703 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1704 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1705 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1706 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1707 ValueType Ty, SDNode OpNode>
1708 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1709 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1710 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1711 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1713 // Long shift by immediate.
1714 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1715 string OpcodeStr, string Dt,
1716 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1717 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1718 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1719 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1720 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1721 (i32 imm:$SIMM))))]>;
1723 // Narrow shift by immediate.
1724 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1725 InstrItinClass itin, string OpcodeStr, string Dt,
1726 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1727 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1728 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1729 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1730 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1731 (i32 imm:$SIMM))))]>;
1733 // Shift right by immediate and accumulate,
1734 // both double- and quad-register.
1735 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1736 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1737 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1738 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1739 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1740 [(set DPR:$dst, (Ty (add DPR:$src1,
1741 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1742 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1743 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1744 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1745 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1746 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1747 [(set QPR:$dst, (Ty (add QPR:$src1,
1748 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1750 // Shift by immediate and insert,
1751 // both double- and quad-register.
1752 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1753 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1754 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1755 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1756 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1757 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1758 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1759 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1760 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1761 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1762 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1763 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1765 // Convert, with fractional bits immediate,
1766 // both double- and quad-register.
1767 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1768 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1770 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1771 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1772 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1773 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1774 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1775 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1777 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1778 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1779 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1780 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1782 //===----------------------------------------------------------------------===//
1784 //===----------------------------------------------------------------------===//
1786 // Abbreviations used in multiclass suffixes:
1787 // Q = quarter int (8 bit) elements
1788 // H = half int (16 bit) elements
1789 // S = single int (32 bit) elements
1790 // D = double int (64 bit) elements
1792 // Neon 2-register vector operations -- for disassembly only.
1794 // First with only element sizes of 8, 16 and 32 bits:
1795 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1796 bits<5> op11_7, bit op4, string opc, string Dt,
1798 // 64-bit vector types.
1799 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1800 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1801 opc, !strconcat(Dt, "8"), asm, "", []>;
1802 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1803 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1804 opc, !strconcat(Dt, "16"), asm, "", []>;
1805 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1806 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1807 opc, !strconcat(Dt, "32"), asm, "", []>;
1808 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1809 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1810 opc, "f32", asm, "", []> {
1811 let Inst{10} = 1; // overwrite F = 1
1814 // 128-bit vector types.
1815 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1816 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1817 opc, !strconcat(Dt, "8"), asm, "", []>;
1818 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1819 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1820 opc, !strconcat(Dt, "16"), asm, "", []>;
1821 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1822 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1823 opc, !strconcat(Dt, "32"), asm, "", []>;
1824 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1825 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1826 opc, "f32", asm, "", []> {
1827 let Inst{10} = 1; // overwrite F = 1
1831 // Neon 3-register vector operations.
1833 // First with only element sizes of 8, 16 and 32 bits:
1834 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1835 InstrItinClass itinD16, InstrItinClass itinD32,
1836 InstrItinClass itinQ16, InstrItinClass itinQ32,
1837 string OpcodeStr, string Dt,
1838 SDNode OpNode, bit Commutable = 0> {
1839 // 64-bit vector types.
1840 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1841 OpcodeStr, !strconcat(Dt, "8"),
1842 v8i8, v8i8, OpNode, Commutable>;
1843 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1844 OpcodeStr, !strconcat(Dt, "16"),
1845 v4i16, v4i16, OpNode, Commutable>;
1846 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1847 OpcodeStr, !strconcat(Dt, "32"),
1848 v2i32, v2i32, OpNode, Commutable>;
1850 // 128-bit vector types.
1851 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1852 OpcodeStr, !strconcat(Dt, "8"),
1853 v16i8, v16i8, OpNode, Commutable>;
1854 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1855 OpcodeStr, !strconcat(Dt, "16"),
1856 v8i16, v8i16, OpNode, Commutable>;
1857 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1858 OpcodeStr, !strconcat(Dt, "32"),
1859 v4i32, v4i32, OpNode, Commutable>;
1862 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1863 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1865 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1867 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1868 v8i16, v4i16, ShOp>;
1869 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1870 v4i32, v2i32, ShOp>;
1873 // ....then also with element size 64 bits:
1874 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1875 InstrItinClass itinD, InstrItinClass itinQ,
1876 string OpcodeStr, string Dt,
1877 SDNode OpNode, bit Commutable = 0>
1878 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1879 OpcodeStr, Dt, OpNode, Commutable> {
1880 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1881 OpcodeStr, !strconcat(Dt, "64"),
1882 v1i64, v1i64, OpNode, Commutable>;
1883 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1884 OpcodeStr, !strconcat(Dt, "64"),
1885 v2i64, v2i64, OpNode, Commutable>;
1889 // Neon Narrowing 2-register vector operations,
1890 // source operand element sizes of 16, 32 and 64 bits:
1891 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1892 bits<5> op11_7, bit op6, bit op4,
1893 InstrItinClass itin, string OpcodeStr, string Dt,
1895 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1896 itin, OpcodeStr, !strconcat(Dt, "16"),
1897 v8i8, v8i16, OpNode>;
1898 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1899 itin, OpcodeStr, !strconcat(Dt, "32"),
1900 v4i16, v4i32, OpNode>;
1901 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1902 itin, OpcodeStr, !strconcat(Dt, "64"),
1903 v2i32, v2i64, OpNode>;
1906 // Neon Narrowing 2-register vector intrinsics,
1907 // source operand element sizes of 16, 32 and 64 bits:
1908 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1909 bits<5> op11_7, bit op6, bit op4,
1910 InstrItinClass itin, string OpcodeStr, string Dt,
1912 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1913 itin, OpcodeStr, !strconcat(Dt, "16"),
1914 v8i8, v8i16, IntOp>;
1915 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1916 itin, OpcodeStr, !strconcat(Dt, "32"),
1917 v4i16, v4i32, IntOp>;
1918 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1919 itin, OpcodeStr, !strconcat(Dt, "64"),
1920 v2i32, v2i64, IntOp>;
1924 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1925 // source operand element sizes of 16, 32 and 64 bits:
1926 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1927 string OpcodeStr, string Dt, SDNode OpNode> {
1928 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1929 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
1930 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1931 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1932 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1933 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1937 // Neon 3-register vector intrinsics.
1939 // First with only element sizes of 16 and 32 bits:
1940 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1941 InstrItinClass itinD16, InstrItinClass itinD32,
1942 InstrItinClass itinQ16, InstrItinClass itinQ32,
1943 string OpcodeStr, string Dt,
1944 Intrinsic IntOp, bit Commutable = 0> {
1945 // 64-bit vector types.
1946 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1947 OpcodeStr, !strconcat(Dt, "16"),
1948 v4i16, v4i16, IntOp, Commutable>;
1949 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1950 OpcodeStr, !strconcat(Dt, "32"),
1951 v2i32, v2i32, IntOp, Commutable>;
1953 // 128-bit vector types.
1954 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1955 OpcodeStr, !strconcat(Dt, "16"),
1956 v8i16, v8i16, IntOp, Commutable>;
1957 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1958 OpcodeStr, !strconcat(Dt, "32"),
1959 v4i32, v4i32, IntOp, Commutable>;
1962 multiclass N3VIntSL_HS<bits<4> op11_8,
1963 InstrItinClass itinD16, InstrItinClass itinD32,
1964 InstrItinClass itinQ16, InstrItinClass itinQ32,
1965 string OpcodeStr, string Dt, Intrinsic IntOp> {
1966 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1967 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1968 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1969 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1970 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1971 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1972 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1973 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1976 // ....then also with element size of 8 bits:
1977 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1978 InstrItinClass itinD16, InstrItinClass itinD32,
1979 InstrItinClass itinQ16, InstrItinClass itinQ32,
1980 string OpcodeStr, string Dt,
1981 Intrinsic IntOp, bit Commutable = 0>
1982 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1983 OpcodeStr, Dt, IntOp, Commutable> {
1984 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1985 OpcodeStr, !strconcat(Dt, "8"),
1986 v8i8, v8i8, IntOp, Commutable>;
1987 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1988 OpcodeStr, !strconcat(Dt, "8"),
1989 v16i8, v16i8, IntOp, Commutable>;
1992 // ....then also with element size of 64 bits:
1993 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1994 InstrItinClass itinD16, InstrItinClass itinD32,
1995 InstrItinClass itinQ16, InstrItinClass itinQ32,
1996 string OpcodeStr, string Dt,
1997 Intrinsic IntOp, bit Commutable = 0>
1998 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1999 OpcodeStr, Dt, IntOp, Commutable> {
2000 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2001 OpcodeStr, !strconcat(Dt, "64"),
2002 v1i64, v1i64, IntOp, Commutable>;
2003 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2004 OpcodeStr, !strconcat(Dt, "64"),
2005 v2i64, v2i64, IntOp, Commutable>;
2008 // Neon Narrowing 3-register vector intrinsics,
2009 // source operand element sizes of 16, 32 and 64 bits:
2010 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2011 string OpcodeStr, string Dt,
2012 Intrinsic IntOp, bit Commutable = 0> {
2013 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2014 OpcodeStr, !strconcat(Dt, "16"),
2015 v8i8, v8i16, IntOp, Commutable>;
2016 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2017 OpcodeStr, !strconcat(Dt, "32"),
2018 v4i16, v4i32, IntOp, Commutable>;
2019 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2020 OpcodeStr, !strconcat(Dt, "64"),
2021 v2i32, v2i64, IntOp, Commutable>;
2025 // Neon Long 3-register vector operations.
2027 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2028 InstrItinClass itin16, InstrItinClass itin32,
2029 string OpcodeStr, string Dt,
2030 SDNode OpNode, bit Commutable = 0> {
2031 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2032 OpcodeStr, !strconcat(Dt, "8"),
2033 v8i16, v8i8, OpNode, Commutable>;
2034 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2035 OpcodeStr, !strconcat(Dt, "16"),
2036 v4i32, v4i16, OpNode, Commutable>;
2037 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2038 OpcodeStr, !strconcat(Dt, "32"),
2039 v2i64, v2i32, OpNode, Commutable>;
2042 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2043 InstrItinClass itin, string OpcodeStr, string Dt,
2045 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2046 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2047 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2048 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2051 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2052 InstrItinClass itin16, InstrItinClass itin32,
2053 string OpcodeStr, string Dt,
2054 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2055 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2056 OpcodeStr, !strconcat(Dt, "8"),
2057 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2058 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2059 OpcodeStr, !strconcat(Dt, "16"),
2060 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2061 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2062 OpcodeStr, !strconcat(Dt, "32"),
2063 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2066 // Neon Long 3-register vector intrinsics.
2068 // First with only element sizes of 16 and 32 bits:
2069 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2070 InstrItinClass itin16, InstrItinClass itin32,
2071 string OpcodeStr, string Dt,
2072 Intrinsic IntOp, bit Commutable = 0> {
2073 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2074 OpcodeStr, !strconcat(Dt, "16"),
2075 v4i32, v4i16, IntOp, Commutable>;
2076 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2077 OpcodeStr, !strconcat(Dt, "32"),
2078 v2i64, v2i32, IntOp, Commutable>;
2081 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2082 InstrItinClass itin, string OpcodeStr, string Dt,
2084 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2085 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2086 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2087 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2090 // ....then also with element size of 8 bits:
2091 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2092 InstrItinClass itin16, InstrItinClass itin32,
2093 string OpcodeStr, string Dt,
2094 Intrinsic IntOp, bit Commutable = 0>
2095 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2096 IntOp, Commutable> {
2097 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2098 OpcodeStr, !strconcat(Dt, "8"),
2099 v8i16, v8i8, IntOp, Commutable>;
2102 // ....with explicit extend (VABDL).
2103 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2104 InstrItinClass itin, string OpcodeStr, string Dt,
2105 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2106 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2107 OpcodeStr, !strconcat(Dt, "8"),
2108 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2109 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2110 OpcodeStr, !strconcat(Dt, "16"),
2111 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2112 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2113 OpcodeStr, !strconcat(Dt, "32"),
2114 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2118 // Neon Wide 3-register vector intrinsics,
2119 // source operand element sizes of 8, 16 and 32 bits:
2120 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2121 string OpcodeStr, string Dt,
2122 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2123 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2124 OpcodeStr, !strconcat(Dt, "8"),
2125 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2126 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2127 OpcodeStr, !strconcat(Dt, "16"),
2128 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2129 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2130 OpcodeStr, !strconcat(Dt, "32"),
2131 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2135 // Neon Multiply-Op vector operations,
2136 // element sizes of 8, 16 and 32 bits:
2137 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2138 InstrItinClass itinD16, InstrItinClass itinD32,
2139 InstrItinClass itinQ16, InstrItinClass itinQ32,
2140 string OpcodeStr, string Dt, SDNode OpNode> {
2141 // 64-bit vector types.
2142 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2143 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2144 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2145 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2146 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2147 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2149 // 128-bit vector types.
2150 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2151 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2152 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2153 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2154 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2155 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2158 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2159 InstrItinClass itinD16, InstrItinClass itinD32,
2160 InstrItinClass itinQ16, InstrItinClass itinQ32,
2161 string OpcodeStr, string Dt, SDNode ShOp> {
2162 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2163 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2164 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2165 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2166 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2167 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2169 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2170 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2174 // Neon Intrinsic-Op vector operations,
2175 // element sizes of 8, 16 and 32 bits:
2176 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2177 InstrItinClass itinD, InstrItinClass itinQ,
2178 string OpcodeStr, string Dt, Intrinsic IntOp,
2180 // 64-bit vector types.
2181 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2182 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2183 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2184 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2185 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2186 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2188 // 128-bit vector types.
2189 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2190 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2191 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2192 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2193 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2194 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2197 // Neon 3-argument intrinsics,
2198 // element sizes of 8, 16 and 32 bits:
2199 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2200 InstrItinClass itinD, InstrItinClass itinQ,
2201 string OpcodeStr, string Dt, Intrinsic IntOp> {
2202 // 64-bit vector types.
2203 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2204 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2205 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2206 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2207 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2208 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2210 // 128-bit vector types.
2211 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2212 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2213 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2214 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2215 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2216 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2220 // Neon Long Multiply-Op vector operations,
2221 // element sizes of 8, 16 and 32 bits:
2222 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2223 InstrItinClass itin16, InstrItinClass itin32,
2224 string OpcodeStr, string Dt, SDNode MulOp,
2226 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2227 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2228 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2229 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2230 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2231 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2234 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2235 string Dt, SDNode MulOp, SDNode OpNode> {
2236 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2237 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2238 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2239 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2243 // Neon Long 3-argument intrinsics.
2245 // First with only element sizes of 16 and 32 bits:
2246 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2247 InstrItinClass itin16, InstrItinClass itin32,
2248 string OpcodeStr, string Dt, Intrinsic IntOp> {
2249 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2250 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2251 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2252 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2255 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2256 string OpcodeStr, string Dt, Intrinsic IntOp> {
2257 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2258 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2259 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2260 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2263 // ....then also with element size of 8 bits:
2264 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2265 InstrItinClass itin16, InstrItinClass itin32,
2266 string OpcodeStr, string Dt, Intrinsic IntOp>
2267 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2268 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2269 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2272 // ....with explicit extend (VABAL).
2273 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2274 InstrItinClass itin, string OpcodeStr, string Dt,
2275 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2276 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2277 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2278 IntOp, ExtOp, OpNode>;
2279 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2280 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2281 IntOp, ExtOp, OpNode>;
2282 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2283 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2284 IntOp, ExtOp, OpNode>;
2288 // Neon 2-register vector intrinsics,
2289 // element sizes of 8, 16 and 32 bits:
2290 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2291 bits<5> op11_7, bit op4,
2292 InstrItinClass itinD, InstrItinClass itinQ,
2293 string OpcodeStr, string Dt, Intrinsic IntOp> {
2294 // 64-bit vector types.
2295 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2296 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2297 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2298 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2299 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2300 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2302 // 128-bit vector types.
2303 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2304 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2305 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2306 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2307 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2308 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2312 // Neon Pairwise long 2-register intrinsics,
2313 // element sizes of 8, 16 and 32 bits:
2314 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2315 bits<5> op11_7, bit op4,
2316 string OpcodeStr, string Dt, Intrinsic IntOp> {
2317 // 64-bit vector types.
2318 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2319 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2320 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2321 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2322 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2323 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2325 // 128-bit vector types.
2326 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2327 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2328 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2329 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2330 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2331 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2335 // Neon Pairwise long 2-register accumulate intrinsics,
2336 // element sizes of 8, 16 and 32 bits:
2337 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2338 bits<5> op11_7, bit op4,
2339 string OpcodeStr, string Dt, Intrinsic IntOp> {
2340 // 64-bit vector types.
2341 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2342 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2343 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2344 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2345 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2346 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2348 // 128-bit vector types.
2349 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2350 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2351 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2352 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2353 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2354 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2358 // Neon 2-register vector shift by immediate,
2359 // with f of either N2RegVShLFrm or N2RegVShRFrm
2360 // element sizes of 8, 16, 32 and 64 bits:
2361 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2362 InstrItinClass itin, string OpcodeStr, string Dt,
2363 SDNode OpNode, Format f> {
2364 // 64-bit vector types.
2365 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2366 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2367 let Inst{21-19} = 0b001; // imm6 = 001xxx
2369 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2370 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2371 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2373 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2374 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2375 let Inst{21} = 0b1; // imm6 = 1xxxxx
2377 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2378 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2381 // 128-bit vector types.
2382 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2383 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2384 let Inst{21-19} = 0b001; // imm6 = 001xxx
2386 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2387 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2388 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2390 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2391 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2392 let Inst{21} = 0b1; // imm6 = 1xxxxx
2394 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2395 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2399 // Neon Shift-Accumulate vector operations,
2400 // element sizes of 8, 16, 32 and 64 bits:
2401 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2402 string OpcodeStr, string Dt, SDNode ShOp> {
2403 // 64-bit vector types.
2404 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2405 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2406 let Inst{21-19} = 0b001; // imm6 = 001xxx
2408 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2409 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2410 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2412 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2413 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2414 let Inst{21} = 0b1; // imm6 = 1xxxxx
2416 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2417 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2420 // 128-bit vector types.
2421 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2422 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2423 let Inst{21-19} = 0b001; // imm6 = 001xxx
2425 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2426 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2427 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2429 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2430 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2431 let Inst{21} = 0b1; // imm6 = 1xxxxx
2433 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2434 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2439 // Neon Shift-Insert vector operations,
2440 // with f of either N2RegVShLFrm or N2RegVShRFrm
2441 // element sizes of 8, 16, 32 and 64 bits:
2442 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2443 string OpcodeStr, SDNode ShOp,
2445 // 64-bit vector types.
2446 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2447 f, OpcodeStr, "8", v8i8, ShOp> {
2448 let Inst{21-19} = 0b001; // imm6 = 001xxx
2450 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2451 f, OpcodeStr, "16", v4i16, ShOp> {
2452 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2454 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2455 f, OpcodeStr, "32", v2i32, ShOp> {
2456 let Inst{21} = 0b1; // imm6 = 1xxxxx
2458 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2459 f, OpcodeStr, "64", v1i64, ShOp>;
2462 // 128-bit vector types.
2463 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2464 f, OpcodeStr, "8", v16i8, ShOp> {
2465 let Inst{21-19} = 0b001; // imm6 = 001xxx
2467 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2468 f, OpcodeStr, "16", v8i16, ShOp> {
2469 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2471 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2472 f, OpcodeStr, "32", v4i32, ShOp> {
2473 let Inst{21} = 0b1; // imm6 = 1xxxxx
2475 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2476 f, OpcodeStr, "64", v2i64, ShOp>;
2480 // Neon Shift Long operations,
2481 // element sizes of 8, 16, 32 bits:
2482 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2483 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2484 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2485 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2486 let Inst{21-19} = 0b001; // imm6 = 001xxx
2488 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2489 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2490 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2492 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2493 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2494 let Inst{21} = 0b1; // imm6 = 1xxxxx
2498 // Neon Shift Narrow operations,
2499 // element sizes of 16, 32, 64 bits:
2500 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2501 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2503 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2504 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2505 let Inst{21-19} = 0b001; // imm6 = 001xxx
2507 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2508 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2509 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2511 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2512 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2513 let Inst{21} = 0b1; // imm6 = 1xxxxx
2517 //===----------------------------------------------------------------------===//
2518 // Instruction Definitions.
2519 //===----------------------------------------------------------------------===//
2521 // Vector Add Operations.
2523 // VADD : Vector Add (integer and floating-point)
2524 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2526 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2527 v2f32, v2f32, fadd, 1>;
2528 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2529 v4f32, v4f32, fadd, 1>;
2530 // VADDL : Vector Add Long (Q = D + D)
2531 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2532 "vaddl", "s", add, sext, 1>;
2533 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2534 "vaddl", "u", add, zext, 1>;
2535 // VADDW : Vector Add Wide (Q = Q + D)
2536 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2537 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2538 // VHADD : Vector Halving Add
2539 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2540 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2541 "vhadd", "s", int_arm_neon_vhadds, 1>;
2542 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2543 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2544 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2545 // VRHADD : Vector Rounding Halving Add
2546 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2547 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2548 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2549 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2550 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2551 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2552 // VQADD : Vector Saturating Add
2553 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2554 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2555 "vqadd", "s", int_arm_neon_vqadds, 1>;
2556 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2557 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2558 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2559 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2560 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2561 int_arm_neon_vaddhn, 1>;
2562 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2563 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2564 int_arm_neon_vraddhn, 1>;
2566 // Vector Multiply Operations.
2568 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2569 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2570 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2571 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2572 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2573 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2574 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2575 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2576 v2f32, v2f32, fmul, 1>;
2577 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2578 v4f32, v4f32, fmul, 1>;
2579 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2580 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2581 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2584 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2585 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2586 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2587 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2588 (DSubReg_i16_reg imm:$lane))),
2589 (SubReg_i16_lane imm:$lane)))>;
2590 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2591 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2592 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2593 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2594 (DSubReg_i32_reg imm:$lane))),
2595 (SubReg_i32_lane imm:$lane)))>;
2596 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2597 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2598 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2599 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2600 (DSubReg_i32_reg imm:$lane))),
2601 (SubReg_i32_lane imm:$lane)))>;
2603 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2604 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2605 IIC_VMULi16Q, IIC_VMULi32Q,
2606 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2607 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2608 IIC_VMULi16Q, IIC_VMULi32Q,
2609 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2610 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2611 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2613 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2614 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2615 (DSubReg_i16_reg imm:$lane))),
2616 (SubReg_i16_lane imm:$lane)))>;
2617 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2618 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2620 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2621 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2622 (DSubReg_i32_reg imm:$lane))),
2623 (SubReg_i32_lane imm:$lane)))>;
2625 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2626 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2627 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2628 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2629 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2630 IIC_VMULi16Q, IIC_VMULi32Q,
2631 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2632 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2633 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2635 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2636 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2637 (DSubReg_i16_reg imm:$lane))),
2638 (SubReg_i16_lane imm:$lane)))>;
2639 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2640 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2642 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2643 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2644 (DSubReg_i32_reg imm:$lane))),
2645 (SubReg_i32_lane imm:$lane)))>;
2647 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2648 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2649 "vmull", "s", NEONvmulls, 1>;
2650 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2651 "vmull", "u", NEONvmullu, 1>;
2652 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2653 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2654 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
2655 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
2657 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2658 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2659 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2660 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2661 "vqdmull", "s", int_arm_neon_vqdmull>;
2663 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2665 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2666 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2667 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2668 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2670 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2672 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2673 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2674 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2676 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2677 v4f32, v2f32, fmul, fadd>;
2679 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2680 (mul (v8i16 QPR:$src2),
2681 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2682 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2683 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2684 (DSubReg_i16_reg imm:$lane))),
2685 (SubReg_i16_lane imm:$lane)))>;
2687 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2688 (mul (v4i32 QPR:$src2),
2689 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2690 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2691 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2692 (DSubReg_i32_reg imm:$lane))),
2693 (SubReg_i32_lane imm:$lane)))>;
2695 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2696 (fmul (v4f32 QPR:$src2),
2697 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2698 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2700 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2701 (DSubReg_i32_reg imm:$lane))),
2702 (SubReg_i32_lane imm:$lane)))>;
2704 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2705 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2706 "vmlal", "s", NEONvmulls, add>;
2707 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2708 "vmlal", "u", NEONvmullu, add>;
2710 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
2711 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
2713 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2714 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2715 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2716 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2718 // VMLS : Vector Multiply Subtract (integer and floating-point)
2719 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2720 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2721 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2723 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2725 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2726 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2727 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2729 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2730 v4f32, v2f32, fmul, fsub>;
2732 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2733 (mul (v8i16 QPR:$src2),
2734 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2735 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2736 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2737 (DSubReg_i16_reg imm:$lane))),
2738 (SubReg_i16_lane imm:$lane)))>;
2740 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2741 (mul (v4i32 QPR:$src2),
2742 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2743 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2744 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2745 (DSubReg_i32_reg imm:$lane))),
2746 (SubReg_i32_lane imm:$lane)))>;
2748 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2749 (fmul (v4f32 QPR:$src2),
2750 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2751 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2752 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2753 (DSubReg_i32_reg imm:$lane))),
2754 (SubReg_i32_lane imm:$lane)))>;
2756 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2757 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2758 "vmlsl", "s", NEONvmulls, sub>;
2759 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2760 "vmlsl", "u", NEONvmullu, sub>;
2762 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
2763 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
2765 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2766 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2767 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2768 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2770 // Vector Subtract Operations.
2772 // VSUB : Vector Subtract (integer and floating-point)
2773 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2774 "vsub", "i", sub, 0>;
2775 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2776 v2f32, v2f32, fsub, 0>;
2777 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2778 v4f32, v4f32, fsub, 0>;
2779 // VSUBL : Vector Subtract Long (Q = D - D)
2780 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2781 "vsubl", "s", sub, sext, 0>;
2782 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2783 "vsubl", "u", sub, zext, 0>;
2784 // VSUBW : Vector Subtract Wide (Q = Q - D)
2785 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
2786 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
2787 // VHSUB : Vector Halving Subtract
2788 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2789 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2790 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2791 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2792 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2793 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2794 // VQSUB : Vector Saturing Subtract
2795 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2796 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2797 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2798 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2799 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2800 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2801 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2802 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2803 int_arm_neon_vsubhn, 0>;
2804 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2805 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2806 int_arm_neon_vrsubhn, 0>;
2808 // Vector Comparisons.
2810 // VCEQ : Vector Compare Equal
2811 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2812 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2813 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2815 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2817 // For disassembly only.
2818 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2821 // VCGE : Vector Compare Greater Than or Equal
2822 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2823 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2824 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2825 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2826 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2828 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2830 // For disassembly only.
2831 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2833 // For disassembly only.
2834 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2837 // VCGT : Vector Compare Greater Than
2838 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2839 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2840 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2841 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2842 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2844 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2846 // For disassembly only.
2847 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2849 // For disassembly only.
2850 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2853 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2854 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2855 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2856 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2857 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2858 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2859 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2860 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2861 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2862 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2863 // VTST : Vector Test Bits
2864 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2865 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2867 // Vector Bitwise Operations.
2869 def vnotd : PatFrag<(ops node:$in),
2870 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
2871 def vnotq : PatFrag<(ops node:$in),
2872 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
2875 // VAND : Vector Bitwise AND
2876 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2877 v2i32, v2i32, and, 1>;
2878 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2879 v4i32, v4i32, and, 1>;
2881 // VEOR : Vector Bitwise Exclusive OR
2882 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2883 v2i32, v2i32, xor, 1>;
2884 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2885 v4i32, v4i32, xor, 1>;
2887 // VORR : Vector Bitwise OR
2888 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2889 v2i32, v2i32, or, 1>;
2890 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2891 v4i32, v4i32, or, 1>;
2893 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2894 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2895 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2896 "vbic", "$dst, $src1, $src2", "",
2897 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2898 (vnotd DPR:$src2))))]>;
2899 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2900 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2901 "vbic", "$dst, $src1, $src2", "",
2902 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2903 (vnotq QPR:$src2))))]>;
2905 // VORN : Vector Bitwise OR NOT
2906 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2907 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2908 "vorn", "$dst, $src1, $src2", "",
2909 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2910 (vnotd DPR:$src2))))]>;
2911 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2912 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2913 "vorn", "$dst, $src1, $src2", "",
2914 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2915 (vnotq QPR:$src2))))]>;
2917 // VMVN : Vector Bitwise NOT (Immediate)
2919 let isReMaterializable = 1 in {
2920 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
2921 (ins nModImm:$SIMM), IIC_VMOVImm,
2922 "vmvn", "i16", "$dst, $SIMM", "",
2923 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]>;
2924 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
2925 (ins nModImm:$SIMM), IIC_VMOVImm,
2926 "vmvn", "i16", "$dst, $SIMM", "",
2927 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]>;
2929 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
2930 (ins nModImm:$SIMM), IIC_VMOVImm,
2931 "vmvn", "i32", "$dst, $SIMM", "",
2932 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]>;
2933 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
2934 (ins nModImm:$SIMM), IIC_VMOVImm,
2935 "vmvn", "i32", "$dst, $SIMM", "",
2936 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]>;
2939 // VMVN : Vector Bitwise NOT
2940 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2941 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2942 "vmvn", "$dst, $src", "",
2943 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
2944 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2945 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2946 "vmvn", "$dst, $src", "",
2947 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
2948 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
2949 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
2951 // VBSL : Vector Bitwise Select
2952 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2953 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2954 N3RegFrm, IIC_VCNTiD,
2955 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2957 (v2i32 (or (and DPR:$src2, DPR:$src1),
2958 (and DPR:$src3, (vnotd DPR:$src1)))))]>;
2959 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2960 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2961 N3RegFrm, IIC_VCNTiQ,
2962 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2964 (v4i32 (or (and QPR:$src2, QPR:$src1),
2965 (and QPR:$src3, (vnotq QPR:$src1)))))]>;
2967 // VBIF : Vector Bitwise Insert if False
2968 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2969 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2970 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2971 N3RegFrm, IIC_VBINiD,
2972 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2973 [/* For disassembly only; pattern left blank */]>;
2974 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2975 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2976 N3RegFrm, IIC_VBINiQ,
2977 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2978 [/* For disassembly only; pattern left blank */]>;
2980 // VBIT : Vector Bitwise Insert if True
2981 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2982 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2983 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2984 N3RegFrm, IIC_VBINiD,
2985 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2986 [/* For disassembly only; pattern left blank */]>;
2987 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2988 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2989 N3RegFrm, IIC_VBINiQ,
2990 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2991 [/* For disassembly only; pattern left blank */]>;
2993 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2994 // for equivalent operations with different register constraints; it just
2997 // Vector Absolute Differences.
2999 // VABD : Vector Absolute Difference
3000 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3001 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3002 "vabd", "s", int_arm_neon_vabds, 1>;
3003 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3004 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3005 "vabd", "u", int_arm_neon_vabdu, 1>;
3006 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3007 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3008 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3009 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3011 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3012 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3013 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3014 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3015 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3017 // VABA : Vector Absolute Difference and Accumulate
3018 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3019 "vaba", "s", int_arm_neon_vabds, add>;
3020 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3021 "vaba", "u", int_arm_neon_vabdu, add>;
3023 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3024 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3025 "vabal", "s", int_arm_neon_vabds, zext, add>;
3026 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3027 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3029 // Vector Maximum and Minimum.
3031 // VMAX : Vector Maximum
3032 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3033 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3034 "vmax", "s", int_arm_neon_vmaxs, 1>;
3035 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3036 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3037 "vmax", "u", int_arm_neon_vmaxu, 1>;
3038 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3040 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3041 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3043 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3045 // VMIN : Vector Minimum
3046 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3047 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3048 "vmin", "s", int_arm_neon_vmins, 1>;
3049 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3050 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3051 "vmin", "u", int_arm_neon_vminu, 1>;
3052 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3054 v2f32, v2f32, int_arm_neon_vmins, 1>;
3055 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3057 v4f32, v4f32, int_arm_neon_vmins, 1>;
3059 // Vector Pairwise Operations.
3061 // VPADD : Vector Pairwise Add
3062 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3064 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3065 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3067 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3068 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3070 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3071 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3072 IIC_VBIND, "vpadd", "f32",
3073 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3075 // VPADDL : Vector Pairwise Add Long
3076 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3077 int_arm_neon_vpaddls>;
3078 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3079 int_arm_neon_vpaddlu>;
3081 // VPADAL : Vector Pairwise Add and Accumulate Long
3082 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3083 int_arm_neon_vpadals>;
3084 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3085 int_arm_neon_vpadalu>;
3087 // VPMAX : Vector Pairwise Maximum
3088 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3089 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3090 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3091 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3092 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3093 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3094 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3095 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3096 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3097 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3098 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3099 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3100 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3101 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3103 // VPMIN : Vector Pairwise Minimum
3104 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3105 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3106 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3107 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3108 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3109 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3110 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3111 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3112 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3113 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3114 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3115 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3116 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
3117 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3119 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3121 // VRECPE : Vector Reciprocal Estimate
3122 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3123 IIC_VUNAD, "vrecpe", "u32",
3124 v2i32, v2i32, int_arm_neon_vrecpe>;
3125 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3126 IIC_VUNAQ, "vrecpe", "u32",
3127 v4i32, v4i32, int_arm_neon_vrecpe>;
3128 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3129 IIC_VUNAD, "vrecpe", "f32",
3130 v2f32, v2f32, int_arm_neon_vrecpe>;
3131 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3132 IIC_VUNAQ, "vrecpe", "f32",
3133 v4f32, v4f32, int_arm_neon_vrecpe>;
3135 // VRECPS : Vector Reciprocal Step
3136 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3137 IIC_VRECSD, "vrecps", "f32",
3138 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3139 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3140 IIC_VRECSQ, "vrecps", "f32",
3141 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3143 // VRSQRTE : Vector Reciprocal Square Root Estimate
3144 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3145 IIC_VUNAD, "vrsqrte", "u32",
3146 v2i32, v2i32, int_arm_neon_vrsqrte>;
3147 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3148 IIC_VUNAQ, "vrsqrte", "u32",
3149 v4i32, v4i32, int_arm_neon_vrsqrte>;
3150 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3151 IIC_VUNAD, "vrsqrte", "f32",
3152 v2f32, v2f32, int_arm_neon_vrsqrte>;
3153 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3154 IIC_VUNAQ, "vrsqrte", "f32",
3155 v4f32, v4f32, int_arm_neon_vrsqrte>;
3157 // VRSQRTS : Vector Reciprocal Square Root Step
3158 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3159 IIC_VRECSD, "vrsqrts", "f32",
3160 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3161 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3162 IIC_VRECSQ, "vrsqrts", "f32",
3163 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3167 // VSHL : Vector Shift
3168 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
3169 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3170 "vshl", "s", int_arm_neon_vshifts, 0>;
3171 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
3172 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3173 "vshl", "u", int_arm_neon_vshiftu, 0>;
3174 // VSHL : Vector Shift Left (Immediate)
3175 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3177 // VSHR : Vector Shift Right (Immediate)
3178 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3180 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3183 // VSHLL : Vector Shift Left Long
3184 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3185 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3187 // VSHLL : Vector Shift Left Long (with maximum shift count)
3188 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3189 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3190 ValueType OpTy, SDNode OpNode>
3191 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3192 ResTy, OpTy, OpNode> {
3193 let Inst{21-16} = op21_16;
3195 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3196 v8i16, v8i8, NEONvshlli>;
3197 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3198 v4i32, v4i16, NEONvshlli>;
3199 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3200 v2i64, v2i32, NEONvshlli>;
3202 // VSHRN : Vector Shift Right and Narrow
3203 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3206 // VRSHL : Vector Rounding Shift
3207 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
3208 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3209 "vrshl", "s", int_arm_neon_vrshifts, 0>;
3210 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
3211 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3212 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
3213 // VRSHR : Vector Rounding Shift Right
3214 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3216 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3219 // VRSHRN : Vector Rounding Shift Right and Narrow
3220 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3223 // VQSHL : Vector Saturating Shift
3224 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
3225 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3226 "vqshl", "s", int_arm_neon_vqshifts, 0>;
3227 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
3228 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3229 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
3230 // VQSHL : Vector Saturating Shift Left (Immediate)
3231 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3233 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3235 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3236 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3239 // VQSHRN : Vector Saturating Shift Right and Narrow
3240 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3242 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3245 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3246 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3249 // VQRSHL : Vector Saturating Rounding Shift
3250 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
3251 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3252 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
3253 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
3254 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3255 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
3257 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3258 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3260 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3263 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3264 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3267 // VSRA : Vector Shift Right and Accumulate
3268 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3269 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3270 // VRSRA : Vector Rounding Shift Right and Accumulate
3271 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3272 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3274 // VSLI : Vector Shift Left and Insert
3275 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3276 // VSRI : Vector Shift Right and Insert
3277 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3279 // Vector Absolute and Saturating Absolute.
3281 // VABS : Vector Absolute Value
3282 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3283 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3285 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3286 IIC_VUNAD, "vabs", "f32",
3287 v2f32, v2f32, int_arm_neon_vabs>;
3288 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3289 IIC_VUNAQ, "vabs", "f32",
3290 v4f32, v4f32, int_arm_neon_vabs>;
3292 // VQABS : Vector Saturating Absolute Value
3293 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3294 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3295 int_arm_neon_vqabs>;
3299 def vnegd : PatFrag<(ops node:$in),
3300 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3301 def vnegq : PatFrag<(ops node:$in),
3302 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3304 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3305 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3306 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3307 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3308 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3309 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3310 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3311 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3313 // VNEG : Vector Negate (integer)
3314 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3315 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3316 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3317 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3318 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3319 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3321 // VNEG : Vector Negate (floating-point)
3322 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3323 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3324 "vneg", "f32", "$dst, $src", "",
3325 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3326 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3327 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3328 "vneg", "f32", "$dst, $src", "",
3329 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3331 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3332 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3333 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3334 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3335 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3336 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3338 // VQNEG : Vector Saturating Negate
3339 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3340 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3341 int_arm_neon_vqneg>;
3343 // Vector Bit Counting Operations.
3345 // VCLS : Vector Count Leading Sign Bits
3346 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3347 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3349 // VCLZ : Vector Count Leading Zeros
3350 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3351 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3353 // VCNT : Vector Count One Bits
3354 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3355 IIC_VCNTiD, "vcnt", "8",
3356 v8i8, v8i8, int_arm_neon_vcnt>;
3357 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3358 IIC_VCNTiQ, "vcnt", "8",
3359 v16i8, v16i8, int_arm_neon_vcnt>;
3361 // Vector Swap -- for disassembly only.
3362 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3363 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3364 "vswp", "$dst, $src", "", []>;
3365 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3366 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3367 "vswp", "$dst, $src", "", []>;
3369 // Vector Move Operations.
3371 // VMOV : Vector Move (Register)
3373 let neverHasSideEffects = 1 in {
3374 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3375 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3376 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3377 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3379 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3380 // be expanded after register allocation is completed.
3381 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3382 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3384 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3385 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3386 } // neverHasSideEffects
3388 // VMOV : Vector Move (Immediate)
3390 let isReMaterializable = 1 in {
3391 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3392 (ins nModImm:$SIMM), IIC_VMOVImm,
3393 "vmov", "i8", "$dst, $SIMM", "",
3394 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3395 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3396 (ins nModImm:$SIMM), IIC_VMOVImm,
3397 "vmov", "i8", "$dst, $SIMM", "",
3398 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3400 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3401 (ins nModImm:$SIMM), IIC_VMOVImm,
3402 "vmov", "i16", "$dst, $SIMM", "",
3403 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]>;
3404 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3405 (ins nModImm:$SIMM), IIC_VMOVImm,
3406 "vmov", "i16", "$dst, $SIMM", "",
3407 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]>;
3409 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3410 (ins nModImm:$SIMM), IIC_VMOVImm,
3411 "vmov", "i32", "$dst, $SIMM", "",
3412 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]>;
3413 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3414 (ins nModImm:$SIMM), IIC_VMOVImm,
3415 "vmov", "i32", "$dst, $SIMM", "",
3416 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]>;
3418 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3419 (ins nModImm:$SIMM), IIC_VMOVImm,
3420 "vmov", "i64", "$dst, $SIMM", "",
3421 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3422 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3423 (ins nModImm:$SIMM), IIC_VMOVImm,
3424 "vmov", "i64", "$dst, $SIMM", "",
3425 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3426 } // isReMaterializable
3428 // VMOV : Vector Get Lane (move scalar to ARM core register)
3430 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3431 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3432 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
3433 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
3435 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3436 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3437 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
3438 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
3440 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3441 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3442 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
3443 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
3445 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3446 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3447 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
3448 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
3450 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3451 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3452 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
3453 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
3455 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3456 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3457 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3458 (DSubReg_i8_reg imm:$lane))),
3459 (SubReg_i8_lane imm:$lane))>;
3460 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3461 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3462 (DSubReg_i16_reg imm:$lane))),
3463 (SubReg_i16_lane imm:$lane))>;
3464 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3465 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3466 (DSubReg_i8_reg imm:$lane))),
3467 (SubReg_i8_lane imm:$lane))>;
3468 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3469 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3470 (DSubReg_i16_reg imm:$lane))),
3471 (SubReg_i16_lane imm:$lane))>;
3472 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3473 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3474 (DSubReg_i32_reg imm:$lane))),
3475 (SubReg_i32_lane imm:$lane))>;
3476 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3477 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3478 (SSubReg_f32_reg imm:$src2))>;
3479 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3480 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3481 (SSubReg_f32_reg imm:$src2))>;
3482 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3483 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3484 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3485 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3488 // VMOV : Vector Set Lane (move ARM core register to scalar)
3490 let Constraints = "$src1 = $dst" in {
3491 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
3492 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3493 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
3494 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
3495 GPR:$src2, imm:$lane))]>;
3496 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
3497 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3498 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
3499 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
3500 GPR:$src2, imm:$lane))]>;
3501 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
3502 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3503 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
3504 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
3505 GPR:$src2, imm:$lane))]>;
3507 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3508 (v16i8 (INSERT_SUBREG QPR:$src1,
3509 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3510 (DSubReg_i8_reg imm:$lane))),
3511 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3512 (DSubReg_i8_reg imm:$lane)))>;
3513 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3514 (v8i16 (INSERT_SUBREG QPR:$src1,
3515 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3516 (DSubReg_i16_reg imm:$lane))),
3517 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3518 (DSubReg_i16_reg imm:$lane)))>;
3519 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3520 (v4i32 (INSERT_SUBREG QPR:$src1,
3521 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3522 (DSubReg_i32_reg imm:$lane))),
3523 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3524 (DSubReg_i32_reg imm:$lane)))>;
3526 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3527 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3528 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3529 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3530 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3531 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3533 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3534 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3535 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3536 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3538 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3539 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3540 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3541 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3542 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3543 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3545 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3546 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3547 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3548 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3549 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3550 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3552 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3553 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3554 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3556 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3557 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3558 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3560 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3561 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3562 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3565 // VDUP : Vector Duplicate (from ARM core register to all elements)
3567 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3568 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3569 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3570 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3571 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3572 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3573 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3574 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3576 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3577 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3578 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3579 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3580 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3581 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3583 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3584 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3585 [(set DPR:$dst, (v2f32 (NEONvdup
3586 (f32 (bitconvert GPR:$src)))))]>;
3587 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3588 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3589 [(set QPR:$dst, (v4f32 (NEONvdup
3590 (f32 (bitconvert GPR:$src)))))]>;
3592 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3594 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3596 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3597 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3598 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3600 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3601 ValueType ResTy, ValueType OpTy>
3602 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3603 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3604 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3607 // Inst{19-16} is partially specified depending on the element size.
3609 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3610 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3611 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3612 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3613 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3614 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3615 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3616 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3618 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3619 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3620 (DSubReg_i8_reg imm:$lane))),
3621 (SubReg_i8_lane imm:$lane)))>;
3622 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3623 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3624 (DSubReg_i16_reg imm:$lane))),
3625 (SubReg_i16_lane imm:$lane)))>;
3626 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3627 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3628 (DSubReg_i32_reg imm:$lane))),
3629 (SubReg_i32_lane imm:$lane)))>;
3630 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3631 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3632 (DSubReg_i32_reg imm:$lane))),
3633 (SubReg_i32_lane imm:$lane)))>;
3635 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3636 (outs DPR:$dst), (ins SPR:$src),
3637 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3638 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3640 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3641 (outs QPR:$dst), (ins SPR:$src),
3642 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3643 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3645 // VMOVN : Vector Narrowing Move
3646 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3647 "vmovn", "i", trunc>;
3648 // VQMOVN : Vector Saturating Narrowing Move
3649 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3650 "vqmovn", "s", int_arm_neon_vqmovns>;
3651 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3652 "vqmovn", "u", int_arm_neon_vqmovnu>;
3653 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3654 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3655 // VMOVL : Vector Lengthening Move
3656 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3657 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3659 // Vector Conversions.
3661 // VCVT : Vector Convert Between Floating-Point and Integers
3662 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3663 v2i32, v2f32, fp_to_sint>;
3664 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3665 v2i32, v2f32, fp_to_uint>;
3666 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3667 v2f32, v2i32, sint_to_fp>;
3668 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3669 v2f32, v2i32, uint_to_fp>;
3671 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3672 v4i32, v4f32, fp_to_sint>;
3673 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3674 v4i32, v4f32, fp_to_uint>;
3675 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3676 v4f32, v4i32, sint_to_fp>;
3677 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3678 v4f32, v4i32, uint_to_fp>;
3680 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3681 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3682 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3683 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3684 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3685 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3686 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3687 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3688 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3690 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3691 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3692 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3693 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3694 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3695 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3696 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3697 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3701 // VREV64 : Vector Reverse elements within 64-bit doublewords
3703 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3704 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3705 (ins DPR:$src), IIC_VMOVD,
3706 OpcodeStr, Dt, "$dst, $src", "",
3707 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3708 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3709 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3710 (ins QPR:$src), IIC_VMOVD,
3711 OpcodeStr, Dt, "$dst, $src", "",
3712 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3714 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3715 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3716 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3717 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3719 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3720 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3721 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3722 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3724 // VREV32 : Vector Reverse elements within 32-bit words
3726 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3727 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3728 (ins DPR:$src), IIC_VMOVD,
3729 OpcodeStr, Dt, "$dst, $src", "",
3730 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3731 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3732 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3733 (ins QPR:$src), IIC_VMOVD,
3734 OpcodeStr, Dt, "$dst, $src", "",
3735 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3737 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3738 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3740 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3741 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3743 // VREV16 : Vector Reverse elements within 16-bit halfwords
3745 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3746 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3747 (ins DPR:$src), IIC_VMOVD,
3748 OpcodeStr, Dt, "$dst, $src", "",
3749 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3750 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3751 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3752 (ins QPR:$src), IIC_VMOVD,
3753 OpcodeStr, Dt, "$dst, $src", "",
3754 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3756 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3757 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3759 // Other Vector Shuffles.
3761 // VEXT : Vector Extract
3763 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3764 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3765 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3766 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3767 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3768 (Ty DPR:$rhs), imm:$index)))]>;
3770 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3771 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3772 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3773 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3774 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3775 (Ty QPR:$rhs), imm:$index)))]>;
3777 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3778 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3779 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3780 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3782 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3783 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3784 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3785 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3787 // VTRN : Vector Transpose
3789 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3790 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3791 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3793 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3794 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3795 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3797 // VUZP : Vector Unzip (Deinterleave)
3799 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3800 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3801 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3803 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3804 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3805 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3807 // VZIP : Vector Zip (Interleave)
3809 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3810 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3811 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3813 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3814 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3815 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3817 // Vector Table Lookup and Table Extension.
3819 // VTBL : Vector Table Lookup
3821 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3822 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3823 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3824 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3825 let hasExtraSrcRegAllocReq = 1 in {
3827 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3828 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3829 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "", []>;
3831 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3832 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3833 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "", []>;
3835 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3836 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3838 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "", []>;
3839 } // hasExtraSrcRegAllocReq = 1
3841 // VTBX : Vector Table Extension
3843 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3844 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3845 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3846 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3847 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3848 let hasExtraSrcRegAllocReq = 1 in {
3850 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3851 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3852 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst", []>;
3854 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3855 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3856 NVTBLFrm, IIC_VTBX3,
3857 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src",
3858 "$orig = $dst", []>;
3860 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3861 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3862 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3863 "$orig = $dst", []>;
3864 } // hasExtraSrcRegAllocReq = 1
3866 //===----------------------------------------------------------------------===//
3867 // NEON instructions for single-precision FP math
3868 //===----------------------------------------------------------------------===//
3870 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3871 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3872 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3876 class N3VSPat<SDNode OpNode, NeonI Inst>
3877 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3878 (EXTRACT_SUBREG (v2f32
3879 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3881 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3885 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3886 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3887 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3889 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3891 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3895 // These need separate instructions because they must use DPR_VFP2 register
3896 // class which have SPR sub-registers.
3898 // Vector Add Operations used for single-precision FP
3899 let neverHasSideEffects = 1 in
3900 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3901 def : N3VSPat<fadd, VADDfd_sfp>;
3903 // Vector Sub Operations used for single-precision FP
3904 let neverHasSideEffects = 1 in
3905 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3906 def : N3VSPat<fsub, VSUBfd_sfp>;
3908 // Vector Multiply Operations used for single-precision FP
3909 let neverHasSideEffects = 1 in
3910 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3911 def : N3VSPat<fmul, VMULfd_sfp>;
3913 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3914 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3915 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3917 //let neverHasSideEffects = 1 in
3918 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3919 // v2f32, fmul, fadd>;
3920 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3922 //let neverHasSideEffects = 1 in
3923 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3924 // v2f32, fmul, fsub>;
3925 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3927 // Vector Absolute used for single-precision FP
3928 let neverHasSideEffects = 1 in
3929 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3930 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3931 "vabs", "f32", "$dst, $src", "", []>;
3932 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3934 // Vector Negate used for single-precision FP
3935 let neverHasSideEffects = 1 in
3936 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3937 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3938 "vneg", "f32", "$dst, $src", "", []>;
3939 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3941 // Vector Maximum used for single-precision FP
3942 let neverHasSideEffects = 1 in
3943 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3944 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3945 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3946 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3948 // Vector Minimum used for single-precision FP
3949 let neverHasSideEffects = 1 in
3950 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3951 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3952 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3953 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3955 // Vector Convert between single-precision FP and integer
3956 let neverHasSideEffects = 1 in
3957 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3958 v2i32, v2f32, fp_to_sint>;
3959 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3961 let neverHasSideEffects = 1 in
3962 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3963 v2i32, v2f32, fp_to_uint>;
3964 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3966 let neverHasSideEffects = 1 in
3967 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3968 v2f32, v2i32, sint_to_fp>;
3969 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3971 let neverHasSideEffects = 1 in
3972 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3973 v2f32, v2i32, uint_to_fp>;
3974 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3976 //===----------------------------------------------------------------------===//
3977 // Non-Instruction Patterns
3978 //===----------------------------------------------------------------------===//
3981 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3982 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3983 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3984 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3985 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3986 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3987 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3988 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3989 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3990 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3991 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3992 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3993 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3994 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3995 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3996 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3997 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3998 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3999 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4000 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4001 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4002 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4003 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4004 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4005 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4006 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4007 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4008 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4009 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4010 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4012 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4013 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4014 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4015 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4016 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4017 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4018 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4019 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4020 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4021 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4022 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4023 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4024 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4025 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4026 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4027 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4028 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4029 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4030 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4031 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4032 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4033 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4034 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4035 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4036 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4037 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4038 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4039 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4040 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4041 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;