1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
69 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
70 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
72 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
74 // VDUPLANE can produce a quad-register result from a double-register source,
75 // so the result is not constrained to match the source.
76 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
77 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
80 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
81 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
82 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
84 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
85 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
86 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
87 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
89 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
93 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
94 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
96 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
98 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
99 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
101 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 2>]>;
103 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
104 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
106 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
107 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
108 unsigned EltBits = 0;
109 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
110 return (EltBits == 32 && EltVal == 0);
113 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
114 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
115 unsigned EltBits = 0;
116 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
117 return (EltBits == 8 && EltVal == 0xff);
120 //===----------------------------------------------------------------------===//
121 // NEON operand definitions
122 //===----------------------------------------------------------------------===//
124 def nModImm : Operand<i32> {
125 let PrintMethod = "printNEONModImmOperand";
128 //===----------------------------------------------------------------------===//
129 // NEON load / store instructions
130 //===----------------------------------------------------------------------===//
132 // Use vldmia to load a Q register as a D register pair.
133 // This is equivalent to VLDMD except that it has a Q register operand
134 // instead of a pair of D registers.
136 : AXDI5<(outs QPR:$dst), (ins addrmode4:$addr, pred:$p),
137 IndexModeNone, IIC_fpLoadm,
138 "vldm${addr:submode}${p}\t$addr, ${dst:dregpair}", "",
139 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]>;
141 let mayLoad = 1, neverHasSideEffects = 1 in {
142 // Use vld1 to load a Q register as a D register pair.
143 // This alternative to VLDMQ allows an alignment to be specified.
144 // This is equivalent to VLD1q64 except that it has a Q register operand.
146 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
147 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
148 } // mayLoad = 1, neverHasSideEffects = 1
150 // Use vstmia to store a Q register as a D register pair.
151 // This is equivalent to VSTMD except that it has a Q register operand
152 // instead of a pair of D registers.
154 : AXDI5<(outs), (ins QPR:$src, addrmode4:$addr, pred:$p),
155 IndexModeNone, IIC_fpStorem,
156 "vstm${addr:submode}${p}\t$addr, ${src:dregpair}", "",
157 [(store (v2f64 QPR:$src), addrmode4:$addr)]>;
159 let mayStore = 1, neverHasSideEffects = 1 in {
160 // Use vst1 to store a Q register as a D register pair.
161 // This alternative to VSTMQ allows an alignment to be specified.
162 // This is equivalent to VST1q64 except that it has a Q register operand.
164 : NLdSt<0,0b00,0b1010,0b1100, (outs), (ins addrmode6:$addr, QPR:$src),
165 IIC_VST, "vst1", "64", "${src:dregpair}, $addr", "", []>;
166 } // mayStore = 1, neverHasSideEffects = 1
168 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
170 // VLD1 : Vector Load (multiple single elements)
171 class VLD1D<bits<4> op7_4, string Dt>
172 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
173 (ins addrmode6:$addr), IIC_VLD1,
174 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
175 class VLD1Q<bits<4> op7_4, string Dt>
176 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
177 (ins addrmode6:$addr), IIC_VLD1,
178 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
180 def VLD1d8 : VLD1D<0b0000, "8">;
181 def VLD1d16 : VLD1D<0b0100, "16">;
182 def VLD1d32 : VLD1D<0b1000, "32">;
183 def VLD1d64 : VLD1D<0b1100, "64">;
185 def VLD1q8 : VLD1Q<0b0000, "8">;
186 def VLD1q16 : VLD1Q<0b0100, "16">;
187 def VLD1q32 : VLD1Q<0b1000, "32">;
188 def VLD1q64 : VLD1Q<0b1100, "64">;
190 // ...with address register writeback:
191 class VLD1DWB<bits<4> op7_4, string Dt>
192 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
193 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
194 "vld1", Dt, "\\{$dst\\}, $addr$offset",
195 "$addr.addr = $wb", []>;
196 class VLD1QWB<bits<4> op7_4, string Dt>
197 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
198 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
199 "vld1", Dt, "${dst:dregpair}, $addr$offset",
200 "$addr.addr = $wb", []>;
202 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
203 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
204 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
205 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
207 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
208 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
209 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
210 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
212 // ...with 3 registers (some of these are only for the disassembler):
213 class VLD1D3<bits<4> op7_4, string Dt>
214 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
215 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
216 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
217 class VLD1D3WB<bits<4> op7_4, string Dt>
218 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
219 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
220 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
222 def VLD1d8T : VLD1D3<0b0000, "8">;
223 def VLD1d16T : VLD1D3<0b0100, "16">;
224 def VLD1d32T : VLD1D3<0b1000, "32">;
225 def VLD1d64T : VLD1D3<0b1100, "64">;
227 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
228 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
229 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
230 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
232 // ...with 4 registers (some of these are only for the disassembler):
233 class VLD1D4<bits<4> op7_4, string Dt>
234 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
235 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
236 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
237 class VLD1D4WB<bits<4> op7_4, string Dt>
238 : NLdSt<0,0b10,0b0010,op7_4,
239 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
240 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
241 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
244 def VLD1d8Q : VLD1D4<0b0000, "8">;
245 def VLD1d16Q : VLD1D4<0b0100, "16">;
246 def VLD1d32Q : VLD1D4<0b1000, "32">;
247 def VLD1d64Q : VLD1D4<0b1100, "64">;
249 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
250 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
251 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
252 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
254 // VLD2 : Vector Load (multiple 2-element structures)
255 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
256 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
257 (ins addrmode6:$addr), IIC_VLD2,
258 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
259 class VLD2Q<bits<4> op7_4, string Dt>
260 : NLdSt<0, 0b10, 0b0011, op7_4,
261 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
262 (ins addrmode6:$addr), IIC_VLD2,
263 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
265 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
266 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
267 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
269 def VLD2q8 : VLD2Q<0b0000, "8">;
270 def VLD2q16 : VLD2Q<0b0100, "16">;
271 def VLD2q32 : VLD2Q<0b1000, "32">;
273 // ...with address register writeback:
274 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
275 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
276 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
277 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
278 "$addr.addr = $wb", []>;
279 class VLD2QWB<bits<4> op7_4, string Dt>
280 : NLdSt<0, 0b10, 0b0011, op7_4,
281 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
282 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
283 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
284 "$addr.addr = $wb", []>;
286 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
287 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
288 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
290 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
291 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
292 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
294 // ...with double-spaced registers (for disassembly only):
295 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
296 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
297 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
298 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
299 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
300 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
302 // VLD3 : Vector Load (multiple 3-element structures)
303 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
304 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
305 (ins addrmode6:$addr), IIC_VLD3,
306 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
308 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
309 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
310 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
312 // ...with address register writeback:
313 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
314 : NLdSt<0, 0b10, op11_8, op7_4,
315 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
316 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
317 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
318 "$addr.addr = $wb", []>;
320 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
321 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
322 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
324 // ...with double-spaced registers (non-updating versions for disassembly only):
325 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
326 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
327 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
328 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
329 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
330 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
332 // ...alternate versions to be allocated odd register numbers:
333 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
334 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
335 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
337 // VLD4 : Vector Load (multiple 4-element structures)
338 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
339 : NLdSt<0, 0b10, op11_8, op7_4,
340 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
341 (ins addrmode6:$addr), IIC_VLD4,
342 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
344 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
345 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
346 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
348 // ...with address register writeback:
349 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
350 : NLdSt<0, 0b10, op11_8, op7_4,
351 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
352 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
353 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
354 "$addr.addr = $wb", []>;
356 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
357 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
358 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
360 // ...with double-spaced registers (non-updating versions for disassembly only):
361 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
362 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
363 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
364 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
365 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
366 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
368 // ...alternate versions to be allocated odd register numbers:
369 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
370 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
371 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
373 // VLD1LN : Vector Load (single element to one lane)
374 // FIXME: Not yet implemented.
376 // VLD2LN : Vector Load (single 2-element structure to one lane)
377 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
378 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
379 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
380 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
381 "$src1 = $dst1, $src2 = $dst2", []>;
383 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
384 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
385 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
387 // ...with double-spaced registers:
388 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
389 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
391 // ...alternate versions to be allocated odd register numbers:
392 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
393 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
395 // ...with address register writeback:
396 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
397 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
398 (ins addrmode6:$addr, am6offset:$offset,
399 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
400 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
401 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
403 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
404 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
405 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
407 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
408 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
410 // VLD3LN : Vector Load (single 3-element structure to one lane)
411 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
412 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
413 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
414 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
415 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
416 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
418 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
419 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
420 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
422 // ...with double-spaced registers:
423 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
424 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
426 // ...alternate versions to be allocated odd register numbers:
427 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
428 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
430 // ...with address register writeback:
431 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
432 : NLdSt<1, 0b10, op11_8, op7_4,
433 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
434 (ins addrmode6:$addr, am6offset:$offset,
435 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
436 IIC_VLD3, "vld3", Dt,
437 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
438 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
441 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
442 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
443 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
445 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
446 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
448 // VLD4LN : Vector Load (single 4-element structure to one lane)
449 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
450 : NLdSt<1, 0b10, op11_8, op7_4,
451 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
452 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
453 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
454 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
455 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
457 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
458 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
459 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
461 // ...with double-spaced registers:
462 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
463 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
465 // ...alternate versions to be allocated odd register numbers:
466 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
467 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
469 // ...with address register writeback:
470 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
471 : NLdSt<1, 0b10, op11_8, op7_4,
472 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
473 (ins addrmode6:$addr, am6offset:$offset,
474 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
475 IIC_VLD4, "vld4", Dt,
476 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
477 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
480 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
481 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
482 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
484 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
485 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
487 // VLD1DUP : Vector Load (single element to all lanes)
488 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
489 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
490 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
491 // FIXME: Not yet implemented.
492 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
494 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
496 // Classes for VST* pseudo-instructions with multi-register operands.
497 // These are expanded to real instructions after register allocation.
499 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), IIC_VST, "">;
501 : PseudoNLdSt<(outs GPR:$wb),
502 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
505 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), IIC_VST, "">;
507 : PseudoNLdSt<(outs GPR:$wb),
508 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), IIC_VST,
510 class VSTQQQQWBPseudo
511 : PseudoNLdSt<(outs GPR:$wb),
512 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), IIC_VST,
515 // VST1 : Vector Store (multiple single elements)
516 class VST1D<bits<4> op7_4, string Dt>
517 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
518 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
519 class VST1Q<bits<4> op7_4, string Dt>
520 : NLdSt<0,0b00,0b1010,op7_4, (outs),
521 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
522 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
524 def VST1d8 : VST1D<0b0000, "8">;
525 def VST1d16 : VST1D<0b0100, "16">;
526 def VST1d32 : VST1D<0b1000, "32">;
527 def VST1d64 : VST1D<0b1100, "64">;
529 def VST1q8 : VST1Q<0b0000, "8">;
530 def VST1q16 : VST1Q<0b0100, "16">;
531 def VST1q32 : VST1Q<0b1000, "32">;
532 def VST1q64 : VST1Q<0b1100, "64">;
534 def VST1q8Pseudo : VSTQPseudo;
535 def VST1q16Pseudo : VSTQPseudo;
536 def VST1q32Pseudo : VSTQPseudo;
537 def VST1q64Pseudo : VSTQPseudo;
539 // ...with address register writeback:
540 class VST1DWB<bits<4> op7_4, string Dt>
541 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
542 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
543 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
544 class VST1QWB<bits<4> op7_4, string Dt>
545 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
546 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
547 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
549 def VST1d8_UPD : VST1DWB<0b0000, "8">;
550 def VST1d16_UPD : VST1DWB<0b0100, "16">;
551 def VST1d32_UPD : VST1DWB<0b1000, "32">;
552 def VST1d64_UPD : VST1DWB<0b1100, "64">;
554 def VST1q8_UPD : VST1QWB<0b0000, "8">;
555 def VST1q16_UPD : VST1QWB<0b0100, "16">;
556 def VST1q32_UPD : VST1QWB<0b1000, "32">;
557 def VST1q64_UPD : VST1QWB<0b1100, "64">;
559 def VST1q8Pseudo_UPD : VSTQWBPseudo;
560 def VST1q16Pseudo_UPD : VSTQWBPseudo;
561 def VST1q32Pseudo_UPD : VSTQWBPseudo;
562 def VST1q64Pseudo_UPD : VSTQWBPseudo;
564 // ...with 3 registers (some of these are only for the disassembler):
565 class VST1D3<bits<4> op7_4, string Dt>
566 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
567 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
568 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
569 class VST1D3WB<bits<4> op7_4, string Dt>
570 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
571 (ins addrmode6:$addr, am6offset:$offset,
572 DPR:$src1, DPR:$src2, DPR:$src3),
573 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
574 "$addr.addr = $wb", []>;
576 def VST1d8T : VST1D3<0b0000, "8">;
577 def VST1d16T : VST1D3<0b0100, "16">;
578 def VST1d32T : VST1D3<0b1000, "32">;
579 def VST1d64T : VST1D3<0b1100, "64">;
581 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
582 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
583 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
584 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
586 def VST1d64TPseudo : VSTQQPseudo;
587 def VST1d64TPseudo_UPD : VSTQQWBPseudo;
589 // ...with 4 registers (some of these are only for the disassembler):
590 class VST1D4<bits<4> op7_4, string Dt>
591 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
592 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
593 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
595 class VST1D4WB<bits<4> op7_4, string Dt>
596 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
597 (ins addrmode6:$addr, am6offset:$offset,
598 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
599 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
600 "$addr.addr = $wb", []>;
602 def VST1d8Q : VST1D4<0b0000, "8">;
603 def VST1d16Q : VST1D4<0b0100, "16">;
604 def VST1d32Q : VST1D4<0b1000, "32">;
605 def VST1d64Q : VST1D4<0b1100, "64">;
607 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
608 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
609 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
610 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
612 def VST1d64QPseudo : VSTQQPseudo;
613 def VST1d64QPseudo_UPD : VSTQQWBPseudo;
615 // VST2 : Vector Store (multiple 2-element structures)
616 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
617 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
618 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
619 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
620 class VST2Q<bits<4> op7_4, string Dt>
621 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
622 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
623 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
626 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
627 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
628 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
630 def VST2q8 : VST2Q<0b0000, "8">;
631 def VST2q16 : VST2Q<0b0100, "16">;
632 def VST2q32 : VST2Q<0b1000, "32">;
634 def VST2d8Pseudo : VSTQPseudo;
635 def VST2d16Pseudo : VSTQPseudo;
636 def VST2d32Pseudo : VSTQPseudo;
638 def VST2q8Pseudo : VSTQQPseudo;
639 def VST2q16Pseudo : VSTQQPseudo;
640 def VST2q32Pseudo : VSTQQPseudo;
642 // ...with address register writeback:
643 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
644 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
645 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
646 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
647 "$addr.addr = $wb", []>;
648 class VST2QWB<bits<4> op7_4, string Dt>
649 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
650 (ins addrmode6:$addr, am6offset:$offset,
651 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
652 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
653 "$addr.addr = $wb", []>;
655 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
656 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
657 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
659 def VST2q8_UPD : VST2QWB<0b0000, "8">;
660 def VST2q16_UPD : VST2QWB<0b0100, "16">;
661 def VST2q32_UPD : VST2QWB<0b1000, "32">;
663 def VST2d8Pseudo_UPD : VSTQWBPseudo;
664 def VST2d16Pseudo_UPD : VSTQWBPseudo;
665 def VST2d32Pseudo_UPD : VSTQWBPseudo;
667 def VST2q8Pseudo_UPD : VSTQQWBPseudo;
668 def VST2q16Pseudo_UPD : VSTQQWBPseudo;
669 def VST2q32Pseudo_UPD : VSTQQWBPseudo;
671 // ...with double-spaced registers (for disassembly only):
672 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
673 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
674 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
675 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
676 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
677 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
679 // VST3 : Vector Store (multiple 3-element structures)
680 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
681 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
682 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
683 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
685 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
686 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
687 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
689 def VST3d8Pseudo : VSTQQPseudo;
690 def VST3d16Pseudo : VSTQQPseudo;
691 def VST3d32Pseudo : VSTQQPseudo;
693 // ...with address register writeback:
694 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
695 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
696 (ins addrmode6:$addr, am6offset:$offset,
697 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
698 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
699 "$addr.addr = $wb", []>;
701 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
702 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
703 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
705 def VST3d8Pseudo_UPD : VSTQQWBPseudo;
706 def VST3d16Pseudo_UPD : VSTQQWBPseudo;
707 def VST3d32Pseudo_UPD : VSTQQWBPseudo;
709 // ...with double-spaced registers (non-updating versions for disassembly only):
710 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
711 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
712 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
713 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
714 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
715 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
717 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo;
718 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo;
719 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo;
721 // ...alternate versions to be allocated odd register numbers:
722 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo;
723 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo;
724 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo;
726 // VST4 : Vector Store (multiple 4-element structures)
727 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
728 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
729 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
730 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
733 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
734 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
735 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
737 def VST4d8Pseudo : VSTQQPseudo;
738 def VST4d16Pseudo : VSTQQPseudo;
739 def VST4d32Pseudo : VSTQQPseudo;
741 // ...with address register writeback:
742 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
743 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
744 (ins addrmode6:$addr, am6offset:$offset,
745 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
746 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
747 "$addr.addr = $wb", []>;
749 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
750 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
751 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
753 def VST4d8Pseudo_UPD : VSTQQWBPseudo;
754 def VST4d16Pseudo_UPD : VSTQQWBPseudo;
755 def VST4d32Pseudo_UPD : VSTQQWBPseudo;
757 // ...with double-spaced registers (non-updating versions for disassembly only):
758 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
759 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
760 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
761 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
762 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
763 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
765 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo;
766 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo;
767 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo;
769 // ...alternate versions to be allocated odd register numbers:
770 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo;
771 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo;
772 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo;
774 // VST1LN : Vector Store (single element from one lane)
775 // FIXME: Not yet implemented.
777 // VST2LN : Vector Store (single 2-element structure from one lane)
778 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
779 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
780 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
781 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
784 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
785 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
786 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
788 // ...with double-spaced registers:
789 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
790 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
792 // ...alternate versions to be allocated odd register numbers:
793 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
794 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
796 // ...with address register writeback:
797 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
798 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
799 (ins addrmode6:$addr, am6offset:$offset,
800 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
801 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
802 "$addr.addr = $wb", []>;
804 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
805 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
806 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
808 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
809 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
811 // VST3LN : Vector Store (single 3-element structure from one lane)
812 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
813 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
814 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
815 nohash_imm:$lane), IIC_VST, "vst3", Dt,
816 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
818 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
819 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
820 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
822 // ...with double-spaced registers:
823 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
824 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
826 // ...alternate versions to be allocated odd register numbers:
827 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
828 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
830 // ...with address register writeback:
831 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
832 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
833 (ins addrmode6:$addr, am6offset:$offset,
834 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
836 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
837 "$addr.addr = $wb", []>;
839 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
840 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
841 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
843 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
844 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
846 // VST4LN : Vector Store (single 4-element structure from one lane)
847 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
848 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
849 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
850 nohash_imm:$lane), IIC_VST, "vst4", Dt,
851 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
854 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
855 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
856 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
858 // ...with double-spaced registers:
859 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
860 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
862 // ...alternate versions to be allocated odd register numbers:
863 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
864 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
866 // ...with address register writeback:
867 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
868 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
869 (ins addrmode6:$addr, am6offset:$offset,
870 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
872 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
873 "$addr.addr = $wb", []>;
875 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
876 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
877 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
879 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
880 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
882 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
885 //===----------------------------------------------------------------------===//
886 // NEON pattern fragments
887 //===----------------------------------------------------------------------===//
889 // Extract D sub-registers of Q registers.
890 def DSubReg_i8_reg : SDNodeXForm<imm, [{
891 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
892 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
894 def DSubReg_i16_reg : SDNodeXForm<imm, [{
895 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
896 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
898 def DSubReg_i32_reg : SDNodeXForm<imm, [{
899 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
900 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
902 def DSubReg_f64_reg : SDNodeXForm<imm, [{
903 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
904 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
907 // Extract S sub-registers of Q/D registers.
908 def SSubReg_f32_reg : SDNodeXForm<imm, [{
909 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
910 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
913 // Translate lane numbers from Q registers to D subregs.
914 def SubReg_i8_lane : SDNodeXForm<imm, [{
915 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
917 def SubReg_i16_lane : SDNodeXForm<imm, [{
918 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
920 def SubReg_i32_lane : SDNodeXForm<imm, [{
921 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
924 //===----------------------------------------------------------------------===//
925 // Instruction Classes
926 //===----------------------------------------------------------------------===//
928 // Basic 2-register operations: single-, double- and quad-register.
929 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
930 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
931 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
932 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
933 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
934 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
935 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
936 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
937 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
938 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
939 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
940 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
941 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
942 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
943 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
944 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
945 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
946 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
948 // Basic 2-register intrinsics, both double- and quad-register.
949 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
950 bits<2> op17_16, bits<5> op11_7, bit op4,
951 InstrItinClass itin, string OpcodeStr, string Dt,
952 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
953 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
954 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
955 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
956 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
957 bits<2> op17_16, bits<5> op11_7, bit op4,
958 InstrItinClass itin, string OpcodeStr, string Dt,
959 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
960 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
961 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
962 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
964 // Narrow 2-register operations.
965 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
966 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
967 InstrItinClass itin, string OpcodeStr, string Dt,
968 ValueType TyD, ValueType TyQ, SDNode OpNode>
969 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
970 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
971 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
973 // Narrow 2-register intrinsics.
974 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
975 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
976 InstrItinClass itin, string OpcodeStr, string Dt,
977 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
978 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
979 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
980 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
982 // Long 2-register operations (currently only used for VMOVL).
983 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
984 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
985 InstrItinClass itin, string OpcodeStr, string Dt,
986 ValueType TyQ, ValueType TyD, SDNode OpNode>
987 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
988 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
989 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
991 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
992 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
993 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
994 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
995 OpcodeStr, Dt, "$dst1, $dst2",
996 "$src1 = $dst1, $src2 = $dst2", []>;
997 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
998 InstrItinClass itin, string OpcodeStr, string Dt>
999 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1000 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1001 "$src1 = $dst1, $src2 = $dst2", []>;
1003 // Basic 3-register operations: single-, double- and quad-register.
1004 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1005 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1006 SDNode OpNode, bit Commutable>
1007 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1008 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1009 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1010 let isCommutable = Commutable;
1013 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1014 InstrItinClass itin, string OpcodeStr, string Dt,
1015 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1016 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1017 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1018 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1019 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1020 let isCommutable = Commutable;
1022 // Same as N3VD but no data type.
1023 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1024 InstrItinClass itin, string OpcodeStr,
1025 ValueType ResTy, ValueType OpTy,
1026 SDNode OpNode, bit Commutable>
1027 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1028 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1029 OpcodeStr, "$dst, $src1, $src2", "",
1030 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1031 let isCommutable = Commutable;
1034 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1035 InstrItinClass itin, string OpcodeStr, string Dt,
1036 ValueType Ty, SDNode ShOp>
1037 : N3V<0, 1, op21_20, op11_8, 1, 0,
1038 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1039 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1040 [(set (Ty DPR:$dst),
1041 (Ty (ShOp (Ty DPR:$src1),
1042 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1043 let isCommutable = 0;
1045 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1046 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1047 : N3V<0, 1, op21_20, op11_8, 1, 0,
1048 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1049 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1050 [(set (Ty DPR:$dst),
1051 (Ty (ShOp (Ty DPR:$src1),
1052 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1053 let isCommutable = 0;
1056 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1057 InstrItinClass itin, string OpcodeStr, string Dt,
1058 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1059 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1060 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1061 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1062 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1063 let isCommutable = Commutable;
1065 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1066 InstrItinClass itin, string OpcodeStr,
1067 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1068 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1069 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1070 OpcodeStr, "$dst, $src1, $src2", "",
1071 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1072 let isCommutable = Commutable;
1074 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1075 InstrItinClass itin, string OpcodeStr, string Dt,
1076 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1077 : N3V<1, 1, op21_20, op11_8, 1, 0,
1078 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1079 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1080 [(set (ResTy QPR:$dst),
1081 (ResTy (ShOp (ResTy QPR:$src1),
1082 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1084 let isCommutable = 0;
1086 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1087 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1088 : N3V<1, 1, op21_20, op11_8, 1, 0,
1089 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1090 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1091 [(set (ResTy QPR:$dst),
1092 (ResTy (ShOp (ResTy QPR:$src1),
1093 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1095 let isCommutable = 0;
1098 // Basic 3-register intrinsics, both double- and quad-register.
1099 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1100 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1101 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1102 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1103 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1104 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1105 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1106 let isCommutable = Commutable;
1108 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1109 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1110 : N3V<0, 1, op21_20, op11_8, 1, 0,
1111 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1112 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1113 [(set (Ty DPR:$dst),
1114 (Ty (IntOp (Ty DPR:$src1),
1115 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1117 let isCommutable = 0;
1119 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1120 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1121 : N3V<0, 1, op21_20, op11_8, 1, 0,
1122 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1123 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1124 [(set (Ty DPR:$dst),
1125 (Ty (IntOp (Ty DPR:$src1),
1126 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1127 let isCommutable = 0;
1130 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1131 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1132 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1133 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1134 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1135 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1136 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1137 let isCommutable = Commutable;
1139 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1140 string OpcodeStr, string Dt,
1141 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1142 : N3V<1, 1, op21_20, op11_8, 1, 0,
1143 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1144 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1145 [(set (ResTy QPR:$dst),
1146 (ResTy (IntOp (ResTy QPR:$src1),
1147 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1149 let isCommutable = 0;
1151 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1152 string OpcodeStr, string Dt,
1153 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1154 : N3V<1, 1, op21_20, op11_8, 1, 0,
1155 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1156 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1157 [(set (ResTy QPR:$dst),
1158 (ResTy (IntOp (ResTy QPR:$src1),
1159 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1161 let isCommutable = 0;
1164 // Multiply-Add/Sub operations: single-, double- and quad-register.
1165 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1166 InstrItinClass itin, string OpcodeStr, string Dt,
1167 ValueType Ty, SDNode MulOp, SDNode OpNode>
1168 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1169 (outs DPR_VFP2:$dst),
1170 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1171 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1173 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1174 InstrItinClass itin, string OpcodeStr, string Dt,
1175 ValueType Ty, SDNode MulOp, SDNode OpNode>
1176 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1177 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1178 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1179 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1180 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1181 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1182 string OpcodeStr, string Dt,
1183 ValueType Ty, SDNode MulOp, SDNode ShOp>
1184 : N3V<0, 1, op21_20, op11_8, 1, 0,
1186 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1188 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1189 [(set (Ty DPR:$dst),
1190 (Ty (ShOp (Ty DPR:$src1),
1191 (Ty (MulOp DPR:$src2,
1192 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1194 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1195 string OpcodeStr, string Dt,
1196 ValueType Ty, SDNode MulOp, SDNode ShOp>
1197 : N3V<0, 1, op21_20, op11_8, 1, 0,
1199 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1201 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1202 [(set (Ty DPR:$dst),
1203 (Ty (ShOp (Ty DPR:$src1),
1204 (Ty (MulOp DPR:$src2,
1205 (Ty (NEONvduplane (Ty DPR_8:$src3),
1208 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1209 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1210 SDNode MulOp, SDNode OpNode>
1211 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1212 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1213 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1214 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1215 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1216 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1217 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1218 SDNode MulOp, SDNode ShOp>
1219 : N3V<1, 1, op21_20, op11_8, 1, 0,
1221 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1223 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1224 [(set (ResTy QPR:$dst),
1225 (ResTy (ShOp (ResTy QPR:$src1),
1226 (ResTy (MulOp QPR:$src2,
1227 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1229 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1230 string OpcodeStr, string Dt,
1231 ValueType ResTy, ValueType OpTy,
1232 SDNode MulOp, SDNode ShOp>
1233 : N3V<1, 1, op21_20, op11_8, 1, 0,
1235 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1237 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1238 [(set (ResTy QPR:$dst),
1239 (ResTy (ShOp (ResTy QPR:$src1),
1240 (ResTy (MulOp QPR:$src2,
1241 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1244 // Neon 3-argument intrinsics, both double- and quad-register.
1245 // The destination register is also used as the first source operand register.
1246 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1247 InstrItinClass itin, string OpcodeStr, string Dt,
1248 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1249 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1250 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1251 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1252 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1253 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1254 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1255 InstrItinClass itin, string OpcodeStr, string Dt,
1256 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1257 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1258 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1259 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1260 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1261 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1263 // Long Multiply-Add/Sub operations.
1264 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1265 InstrItinClass itin, string OpcodeStr, string Dt,
1266 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1267 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1268 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1269 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1270 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1271 (TyQ (MulOp (TyD DPR:$src2),
1272 (TyD DPR:$src3)))))]>;
1273 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1274 InstrItinClass itin, string OpcodeStr, string Dt,
1275 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1276 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1277 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1279 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1281 (OpNode (TyQ QPR:$src1),
1282 (TyQ (MulOp (TyD DPR:$src2),
1283 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1285 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1286 InstrItinClass itin, string OpcodeStr, string Dt,
1287 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1288 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1289 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1291 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1293 (OpNode (TyQ QPR:$src1),
1294 (TyQ (MulOp (TyD DPR:$src2),
1295 (TyD (NEONvduplane (TyD DPR_8:$src3),
1299 // Neon Long 3-argument intrinsic. The destination register is
1300 // a quad-register and is also used as the first source operand register.
1301 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1302 InstrItinClass itin, string OpcodeStr, string Dt,
1303 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1304 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1305 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1306 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1308 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1309 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1310 string OpcodeStr, string Dt,
1311 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1312 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1314 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1316 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1317 [(set (ResTy QPR:$dst),
1318 (ResTy (IntOp (ResTy QPR:$src1),
1320 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1322 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1323 InstrItinClass itin, string OpcodeStr, string Dt,
1324 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1325 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1327 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1329 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1330 [(set (ResTy QPR:$dst),
1331 (ResTy (IntOp (ResTy QPR:$src1),
1333 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1336 // Narrowing 3-register intrinsics.
1337 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1338 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1339 Intrinsic IntOp, bit Commutable>
1340 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1341 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1342 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1343 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1344 let isCommutable = Commutable;
1347 // Long 3-register operations.
1348 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1349 InstrItinClass itin, string OpcodeStr, string Dt,
1350 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1351 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1352 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1353 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1354 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1355 let isCommutable = Commutable;
1357 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1358 InstrItinClass itin, string OpcodeStr, string Dt,
1359 ValueType TyQ, ValueType TyD, SDNode OpNode>
1360 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1361 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1362 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1364 (TyQ (OpNode (TyD DPR:$src1),
1365 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1366 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1367 InstrItinClass itin, string OpcodeStr, string Dt,
1368 ValueType TyQ, ValueType TyD, SDNode OpNode>
1369 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1370 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1371 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1373 (TyQ (OpNode (TyD DPR:$src1),
1374 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1376 // Long 3-register operations with explicitly extended operands.
1377 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1378 InstrItinClass itin, string OpcodeStr, string Dt,
1379 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1381 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1382 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1383 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1384 [(set QPR:$dst, (OpNode (TyQ (ExtOp (TyD DPR:$src1))),
1385 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1386 let isCommutable = Commutable;
1389 // Long 3-register intrinsics.
1390 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1391 InstrItinClass itin, string OpcodeStr, string Dt,
1392 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1393 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1394 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1395 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1396 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1397 let isCommutable = Commutable;
1399 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1400 string OpcodeStr, string Dt,
1401 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1402 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1403 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1404 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1405 [(set (ResTy QPR:$dst),
1406 (ResTy (IntOp (OpTy DPR:$src1),
1407 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1409 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1410 InstrItinClass itin, string OpcodeStr, string Dt,
1411 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1412 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1413 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1414 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1415 [(set (ResTy QPR:$dst),
1416 (ResTy (IntOp (OpTy DPR:$src1),
1417 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1420 // Wide 3-register operations.
1421 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1422 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1423 SDNode OpNode, SDNode ExtOp, bit Commutable>
1424 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1425 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1426 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1427 [(set QPR:$dst, (OpNode (TyQ QPR:$src1),
1428 (TyQ (ExtOp (TyD DPR:$src2)))))]> {
1429 let isCommutable = Commutable;
1432 // Pairwise long 2-register intrinsics, both double- and quad-register.
1433 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1434 bits<2> op17_16, bits<5> op11_7, bit op4,
1435 string OpcodeStr, string Dt,
1436 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1437 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1438 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1439 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1440 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1441 bits<2> op17_16, bits<5> op11_7, bit op4,
1442 string OpcodeStr, string Dt,
1443 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1444 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1445 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1446 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1448 // Pairwise long 2-register accumulate intrinsics,
1449 // both double- and quad-register.
1450 // The destination register is also used as the first source operand register.
1451 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1452 bits<2> op17_16, bits<5> op11_7, bit op4,
1453 string OpcodeStr, string Dt,
1454 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1455 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1456 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1457 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1458 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1459 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1460 bits<2> op17_16, bits<5> op11_7, bit op4,
1461 string OpcodeStr, string Dt,
1462 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1463 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1464 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1465 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1466 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1468 // Shift by immediate,
1469 // both double- and quad-register.
1470 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1471 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1472 ValueType Ty, SDNode OpNode>
1473 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1474 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1475 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1476 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1477 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1478 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1479 ValueType Ty, SDNode OpNode>
1480 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1481 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1482 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1483 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1485 // Long shift by immediate.
1486 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1487 string OpcodeStr, string Dt,
1488 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1489 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1490 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1491 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1492 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1493 (i32 imm:$SIMM))))]>;
1495 // Narrow shift by immediate.
1496 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1497 InstrItinClass itin, string OpcodeStr, string Dt,
1498 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1499 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1500 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1501 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1502 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1503 (i32 imm:$SIMM))))]>;
1505 // Shift right by immediate and accumulate,
1506 // both double- and quad-register.
1507 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1508 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1509 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1510 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1511 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1512 [(set DPR:$dst, (Ty (add DPR:$src1,
1513 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1514 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1515 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1516 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1517 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1518 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1519 [(set QPR:$dst, (Ty (add QPR:$src1,
1520 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1522 // Shift by immediate and insert,
1523 // both double- and quad-register.
1524 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1525 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1526 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1527 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1528 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1529 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1530 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1531 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1532 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1533 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1534 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1535 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1537 // Convert, with fractional bits immediate,
1538 // both double- and quad-register.
1539 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1540 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1542 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1543 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1544 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1545 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1546 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1547 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1549 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1550 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1551 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1552 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1554 //===----------------------------------------------------------------------===//
1556 //===----------------------------------------------------------------------===//
1558 // Abbreviations used in multiclass suffixes:
1559 // Q = quarter int (8 bit) elements
1560 // H = half int (16 bit) elements
1561 // S = single int (32 bit) elements
1562 // D = double int (64 bit) elements
1564 // Neon 2-register vector operations -- for disassembly only.
1566 // First with only element sizes of 8, 16 and 32 bits:
1567 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1568 bits<5> op11_7, bit op4, string opc, string Dt,
1570 // 64-bit vector types.
1571 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1572 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1573 opc, !strconcat(Dt, "8"), asm, "", []>;
1574 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1575 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1576 opc, !strconcat(Dt, "16"), asm, "", []>;
1577 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1578 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1579 opc, !strconcat(Dt, "32"), asm, "", []>;
1580 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1581 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1582 opc, "f32", asm, "", []> {
1583 let Inst{10} = 1; // overwrite F = 1
1586 // 128-bit vector types.
1587 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1588 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1589 opc, !strconcat(Dt, "8"), asm, "", []>;
1590 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1591 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1592 opc, !strconcat(Dt, "16"), asm, "", []>;
1593 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1594 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1595 opc, !strconcat(Dt, "32"), asm, "", []>;
1596 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1597 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1598 opc, "f32", asm, "", []> {
1599 let Inst{10} = 1; // overwrite F = 1
1603 // Neon 3-register vector operations.
1605 // First with only element sizes of 8, 16 and 32 bits:
1606 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1607 InstrItinClass itinD16, InstrItinClass itinD32,
1608 InstrItinClass itinQ16, InstrItinClass itinQ32,
1609 string OpcodeStr, string Dt,
1610 SDNode OpNode, bit Commutable = 0> {
1611 // 64-bit vector types.
1612 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1613 OpcodeStr, !strconcat(Dt, "8"),
1614 v8i8, v8i8, OpNode, Commutable>;
1615 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1616 OpcodeStr, !strconcat(Dt, "16"),
1617 v4i16, v4i16, OpNode, Commutable>;
1618 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1619 OpcodeStr, !strconcat(Dt, "32"),
1620 v2i32, v2i32, OpNode, Commutable>;
1622 // 128-bit vector types.
1623 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1624 OpcodeStr, !strconcat(Dt, "8"),
1625 v16i8, v16i8, OpNode, Commutable>;
1626 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1627 OpcodeStr, !strconcat(Dt, "16"),
1628 v8i16, v8i16, OpNode, Commutable>;
1629 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1630 OpcodeStr, !strconcat(Dt, "32"),
1631 v4i32, v4i32, OpNode, Commutable>;
1634 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1635 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1637 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1639 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1640 v8i16, v4i16, ShOp>;
1641 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1642 v4i32, v2i32, ShOp>;
1645 // ....then also with element size 64 bits:
1646 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1647 InstrItinClass itinD, InstrItinClass itinQ,
1648 string OpcodeStr, string Dt,
1649 SDNode OpNode, bit Commutable = 0>
1650 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1651 OpcodeStr, Dt, OpNode, Commutable> {
1652 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1653 OpcodeStr, !strconcat(Dt, "64"),
1654 v1i64, v1i64, OpNode, Commutable>;
1655 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1656 OpcodeStr, !strconcat(Dt, "64"),
1657 v2i64, v2i64, OpNode, Commutable>;
1661 // Neon Narrowing 2-register vector operations,
1662 // source operand element sizes of 16, 32 and 64 bits:
1663 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1664 bits<5> op11_7, bit op6, bit op4,
1665 InstrItinClass itin, string OpcodeStr, string Dt,
1667 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1668 itin, OpcodeStr, !strconcat(Dt, "16"),
1669 v8i8, v8i16, OpNode>;
1670 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1671 itin, OpcodeStr, !strconcat(Dt, "32"),
1672 v4i16, v4i32, OpNode>;
1673 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1674 itin, OpcodeStr, !strconcat(Dt, "64"),
1675 v2i32, v2i64, OpNode>;
1678 // Neon Narrowing 2-register vector intrinsics,
1679 // source operand element sizes of 16, 32 and 64 bits:
1680 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1681 bits<5> op11_7, bit op6, bit op4,
1682 InstrItinClass itin, string OpcodeStr, string Dt,
1684 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1685 itin, OpcodeStr, !strconcat(Dt, "16"),
1686 v8i8, v8i16, IntOp>;
1687 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1688 itin, OpcodeStr, !strconcat(Dt, "32"),
1689 v4i16, v4i32, IntOp>;
1690 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1691 itin, OpcodeStr, !strconcat(Dt, "64"),
1692 v2i32, v2i64, IntOp>;
1696 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1697 // source operand element sizes of 16, 32 and 64 bits:
1698 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1699 string OpcodeStr, string Dt, SDNode OpNode> {
1700 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1701 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
1702 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1703 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1704 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1705 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1709 // Neon 3-register vector intrinsics.
1711 // First with only element sizes of 16 and 32 bits:
1712 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1713 InstrItinClass itinD16, InstrItinClass itinD32,
1714 InstrItinClass itinQ16, InstrItinClass itinQ32,
1715 string OpcodeStr, string Dt,
1716 Intrinsic IntOp, bit Commutable = 0> {
1717 // 64-bit vector types.
1718 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1719 OpcodeStr, !strconcat(Dt, "16"),
1720 v4i16, v4i16, IntOp, Commutable>;
1721 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1722 OpcodeStr, !strconcat(Dt, "32"),
1723 v2i32, v2i32, IntOp, Commutable>;
1725 // 128-bit vector types.
1726 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1727 OpcodeStr, !strconcat(Dt, "16"),
1728 v8i16, v8i16, IntOp, Commutable>;
1729 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1730 OpcodeStr, !strconcat(Dt, "32"),
1731 v4i32, v4i32, IntOp, Commutable>;
1734 multiclass N3VIntSL_HS<bits<4> op11_8,
1735 InstrItinClass itinD16, InstrItinClass itinD32,
1736 InstrItinClass itinQ16, InstrItinClass itinQ32,
1737 string OpcodeStr, string Dt, Intrinsic IntOp> {
1738 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1739 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1740 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1741 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1742 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1743 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1744 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1745 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1748 // ....then also with element size of 8 bits:
1749 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1750 InstrItinClass itinD16, InstrItinClass itinD32,
1751 InstrItinClass itinQ16, InstrItinClass itinQ32,
1752 string OpcodeStr, string Dt,
1753 Intrinsic IntOp, bit Commutable = 0>
1754 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1755 OpcodeStr, Dt, IntOp, Commutable> {
1756 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1757 OpcodeStr, !strconcat(Dt, "8"),
1758 v8i8, v8i8, IntOp, Commutable>;
1759 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1760 OpcodeStr, !strconcat(Dt, "8"),
1761 v16i8, v16i8, IntOp, Commutable>;
1764 // ....then also with element size of 64 bits:
1765 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1766 InstrItinClass itinD16, InstrItinClass itinD32,
1767 InstrItinClass itinQ16, InstrItinClass itinQ32,
1768 string OpcodeStr, string Dt,
1769 Intrinsic IntOp, bit Commutable = 0>
1770 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1771 OpcodeStr, Dt, IntOp, Commutable> {
1772 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1773 OpcodeStr, !strconcat(Dt, "64"),
1774 v1i64, v1i64, IntOp, Commutable>;
1775 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1776 OpcodeStr, !strconcat(Dt, "64"),
1777 v2i64, v2i64, IntOp, Commutable>;
1780 // Neon Narrowing 3-register vector intrinsics,
1781 // source operand element sizes of 16, 32 and 64 bits:
1782 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1783 string OpcodeStr, string Dt,
1784 Intrinsic IntOp, bit Commutable = 0> {
1785 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1786 OpcodeStr, !strconcat(Dt, "16"),
1787 v8i8, v8i16, IntOp, Commutable>;
1788 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1789 OpcodeStr, !strconcat(Dt, "32"),
1790 v4i16, v4i32, IntOp, Commutable>;
1791 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1792 OpcodeStr, !strconcat(Dt, "64"),
1793 v2i32, v2i64, IntOp, Commutable>;
1797 // Neon Long 3-register vector operations.
1799 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1800 InstrItinClass itin16, InstrItinClass itin32,
1801 string OpcodeStr, string Dt,
1802 SDNode OpNode, bit Commutable = 0> {
1803 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
1804 OpcodeStr, !strconcat(Dt, "8"),
1805 v8i16, v8i8, OpNode, Commutable>;
1806 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
1807 OpcodeStr, !strconcat(Dt, "16"),
1808 v4i32, v4i16, OpNode, Commutable>;
1809 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
1810 OpcodeStr, !strconcat(Dt, "32"),
1811 v2i64, v2i32, OpNode, Commutable>;
1814 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
1815 InstrItinClass itin, string OpcodeStr, string Dt,
1817 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
1818 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
1819 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
1820 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
1823 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1824 InstrItinClass itin16, InstrItinClass itin32,
1825 string OpcodeStr, string Dt,
1826 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
1827 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
1828 OpcodeStr, !strconcat(Dt, "8"),
1829 v8i16, v8i8, OpNode, ExtOp, Commutable>;
1830 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
1831 OpcodeStr, !strconcat(Dt, "16"),
1832 v4i32, v4i16, OpNode, ExtOp, Commutable>;
1833 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
1834 OpcodeStr, !strconcat(Dt, "32"),
1835 v2i64, v2i32, OpNode, ExtOp, Commutable>;
1838 // Neon Long 3-register vector intrinsics.
1840 // First with only element sizes of 16 and 32 bits:
1841 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1842 InstrItinClass itin16, InstrItinClass itin32,
1843 string OpcodeStr, string Dt,
1844 Intrinsic IntOp, bit Commutable = 0> {
1845 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1846 OpcodeStr, !strconcat(Dt, "16"),
1847 v4i32, v4i16, IntOp, Commutable>;
1848 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1849 OpcodeStr, !strconcat(Dt, "32"),
1850 v2i64, v2i32, IntOp, Commutable>;
1853 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1854 InstrItinClass itin, string OpcodeStr, string Dt,
1856 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1857 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1858 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1859 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1862 // ....then also with element size of 8 bits:
1863 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1864 InstrItinClass itin16, InstrItinClass itin32,
1865 string OpcodeStr, string Dt,
1866 Intrinsic IntOp, bit Commutable = 0>
1867 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1868 IntOp, Commutable> {
1869 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1870 OpcodeStr, !strconcat(Dt, "8"),
1871 v8i16, v8i8, IntOp, Commutable>;
1875 // Neon Wide 3-register vector intrinsics,
1876 // source operand element sizes of 8, 16 and 32 bits:
1877 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1878 string OpcodeStr, string Dt,
1879 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
1880 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
1881 OpcodeStr, !strconcat(Dt, "8"),
1882 v8i16, v8i8, OpNode, ExtOp, Commutable>;
1883 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
1884 OpcodeStr, !strconcat(Dt, "16"),
1885 v4i32, v4i16, OpNode, ExtOp, Commutable>;
1886 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
1887 OpcodeStr, !strconcat(Dt, "32"),
1888 v2i64, v2i32, OpNode, ExtOp, Commutable>;
1892 // Neon Multiply-Op vector operations,
1893 // element sizes of 8, 16 and 32 bits:
1894 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1895 InstrItinClass itinD16, InstrItinClass itinD32,
1896 InstrItinClass itinQ16, InstrItinClass itinQ32,
1897 string OpcodeStr, string Dt, SDNode OpNode> {
1898 // 64-bit vector types.
1899 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1900 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1901 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1902 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1903 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1904 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1906 // 128-bit vector types.
1907 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1908 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1909 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1910 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1911 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1912 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1915 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1916 InstrItinClass itinD16, InstrItinClass itinD32,
1917 InstrItinClass itinQ16, InstrItinClass itinQ32,
1918 string OpcodeStr, string Dt, SDNode ShOp> {
1919 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1920 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1921 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1922 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1923 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1924 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1926 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1927 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1931 // Neon 3-argument intrinsics,
1932 // element sizes of 8, 16 and 32 bits:
1933 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1934 InstrItinClass itinD, InstrItinClass itinQ,
1935 string OpcodeStr, string Dt, Intrinsic IntOp> {
1936 // 64-bit vector types.
1937 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1938 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1939 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1940 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1941 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1942 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1944 // 128-bit vector types.
1945 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1946 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1947 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1948 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1949 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1950 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1954 // Neon Long Multiply-Op vector operations,
1955 // element sizes of 8, 16 and 32 bits:
1956 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1957 InstrItinClass itin16, InstrItinClass itin32,
1958 string OpcodeStr, string Dt, SDNode MulOp,
1960 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
1961 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
1962 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
1963 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
1964 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
1965 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
1968 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
1969 string Dt, SDNode MulOp, SDNode OpNode> {
1970 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
1971 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
1972 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
1973 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
1977 // Neon Long 3-argument intrinsics.
1979 // First with only element sizes of 16 and 32 bits:
1980 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1981 InstrItinClass itin16, InstrItinClass itin32,
1982 string OpcodeStr, string Dt, Intrinsic IntOp> {
1983 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
1984 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1985 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
1986 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1989 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1990 string OpcodeStr, string Dt, Intrinsic IntOp> {
1991 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1992 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1993 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1994 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1997 // ....then also with element size of 8 bits:
1998 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1999 InstrItinClass itin16, InstrItinClass itin32,
2000 string OpcodeStr, string Dt, Intrinsic IntOp>
2001 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2002 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2003 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2007 // Neon 2-register vector intrinsics,
2008 // element sizes of 8, 16 and 32 bits:
2009 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2010 bits<5> op11_7, bit op4,
2011 InstrItinClass itinD, InstrItinClass itinQ,
2012 string OpcodeStr, string Dt, Intrinsic IntOp> {
2013 // 64-bit vector types.
2014 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2015 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2016 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2017 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2018 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2019 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2021 // 128-bit vector types.
2022 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2023 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2024 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2025 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2026 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2027 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2031 // Neon Pairwise long 2-register intrinsics,
2032 // element sizes of 8, 16 and 32 bits:
2033 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2034 bits<5> op11_7, bit op4,
2035 string OpcodeStr, string Dt, Intrinsic IntOp> {
2036 // 64-bit vector types.
2037 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2038 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2039 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2040 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2041 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2042 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2044 // 128-bit vector types.
2045 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2046 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2047 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2048 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2049 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2050 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2054 // Neon Pairwise long 2-register accumulate intrinsics,
2055 // element sizes of 8, 16 and 32 bits:
2056 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2057 bits<5> op11_7, bit op4,
2058 string OpcodeStr, string Dt, Intrinsic IntOp> {
2059 // 64-bit vector types.
2060 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2061 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2062 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2063 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2064 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2065 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2067 // 128-bit vector types.
2068 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2069 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2070 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2071 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2072 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2073 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2077 // Neon 2-register vector shift by immediate,
2078 // with f of either N2RegVShLFrm or N2RegVShRFrm
2079 // element sizes of 8, 16, 32 and 64 bits:
2080 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2081 InstrItinClass itin, string OpcodeStr, string Dt,
2082 SDNode OpNode, Format f> {
2083 // 64-bit vector types.
2084 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2085 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2086 let Inst{21-19} = 0b001; // imm6 = 001xxx
2088 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2089 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2090 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2092 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2093 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2094 let Inst{21} = 0b1; // imm6 = 1xxxxx
2096 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2097 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2100 // 128-bit vector types.
2101 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2102 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2103 let Inst{21-19} = 0b001; // imm6 = 001xxx
2105 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2106 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2107 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2109 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2110 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2111 let Inst{21} = 0b1; // imm6 = 1xxxxx
2113 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2114 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2118 // Neon Shift-Accumulate vector operations,
2119 // element sizes of 8, 16, 32 and 64 bits:
2120 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2121 string OpcodeStr, string Dt, SDNode ShOp> {
2122 // 64-bit vector types.
2123 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2124 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2125 let Inst{21-19} = 0b001; // imm6 = 001xxx
2127 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2128 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2129 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2131 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2132 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2133 let Inst{21} = 0b1; // imm6 = 1xxxxx
2135 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2136 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2139 // 128-bit vector types.
2140 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2141 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2142 let Inst{21-19} = 0b001; // imm6 = 001xxx
2144 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2145 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2146 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2148 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2149 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2150 let Inst{21} = 0b1; // imm6 = 1xxxxx
2152 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2153 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2158 // Neon Shift-Insert vector operations,
2159 // with f of either N2RegVShLFrm or N2RegVShRFrm
2160 // element sizes of 8, 16, 32 and 64 bits:
2161 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2162 string OpcodeStr, SDNode ShOp,
2164 // 64-bit vector types.
2165 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2166 f, OpcodeStr, "8", v8i8, ShOp> {
2167 let Inst{21-19} = 0b001; // imm6 = 001xxx
2169 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2170 f, OpcodeStr, "16", v4i16, ShOp> {
2171 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2173 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2174 f, OpcodeStr, "32", v2i32, ShOp> {
2175 let Inst{21} = 0b1; // imm6 = 1xxxxx
2177 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2178 f, OpcodeStr, "64", v1i64, ShOp>;
2181 // 128-bit vector types.
2182 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2183 f, OpcodeStr, "8", v16i8, ShOp> {
2184 let Inst{21-19} = 0b001; // imm6 = 001xxx
2186 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2187 f, OpcodeStr, "16", v8i16, ShOp> {
2188 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2190 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2191 f, OpcodeStr, "32", v4i32, ShOp> {
2192 let Inst{21} = 0b1; // imm6 = 1xxxxx
2194 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2195 f, OpcodeStr, "64", v2i64, ShOp>;
2199 // Neon Shift Long operations,
2200 // element sizes of 8, 16, 32 bits:
2201 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2202 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2203 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2204 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2205 let Inst{21-19} = 0b001; // imm6 = 001xxx
2207 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2208 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2209 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2211 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2212 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2213 let Inst{21} = 0b1; // imm6 = 1xxxxx
2217 // Neon Shift Narrow operations,
2218 // element sizes of 16, 32, 64 bits:
2219 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2220 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2222 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2223 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2224 let Inst{21-19} = 0b001; // imm6 = 001xxx
2226 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2227 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2228 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2230 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2231 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2232 let Inst{21} = 0b1; // imm6 = 1xxxxx
2236 //===----------------------------------------------------------------------===//
2237 // Instruction Definitions.
2238 //===----------------------------------------------------------------------===//
2240 // Vector Add Operations.
2242 // VADD : Vector Add (integer and floating-point)
2243 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2245 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2246 v2f32, v2f32, fadd, 1>;
2247 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2248 v4f32, v4f32, fadd, 1>;
2249 // VADDL : Vector Add Long (Q = D + D)
2250 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2251 "vaddl", "s", add, sext, 1>;
2252 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2253 "vaddl", "u", add, zext, 1>;
2254 // VADDW : Vector Add Wide (Q = Q + D)
2255 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2256 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2257 // VHADD : Vector Halving Add
2258 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2259 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2260 "vhadd", "s", int_arm_neon_vhadds, 1>;
2261 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2262 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2263 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2264 // VRHADD : Vector Rounding Halving Add
2265 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2266 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2267 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2268 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2269 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2270 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2271 // VQADD : Vector Saturating Add
2272 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2273 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2274 "vqadd", "s", int_arm_neon_vqadds, 1>;
2275 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2276 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2277 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2278 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2279 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2280 int_arm_neon_vaddhn, 1>;
2281 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2282 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2283 int_arm_neon_vraddhn, 1>;
2285 // Vector Multiply Operations.
2287 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2288 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2289 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2290 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2291 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2292 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2293 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2294 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2295 v2f32, v2f32, fmul, 1>;
2296 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2297 v4f32, v4f32, fmul, 1>;
2298 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2299 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2300 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2303 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2304 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2305 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2306 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2307 (DSubReg_i16_reg imm:$lane))),
2308 (SubReg_i16_lane imm:$lane)))>;
2309 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2310 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2311 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2312 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2313 (DSubReg_i32_reg imm:$lane))),
2314 (SubReg_i32_lane imm:$lane)))>;
2315 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2316 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2317 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2318 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2319 (DSubReg_i32_reg imm:$lane))),
2320 (SubReg_i32_lane imm:$lane)))>;
2322 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2323 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2324 IIC_VMULi16Q, IIC_VMULi32Q,
2325 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2326 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2327 IIC_VMULi16Q, IIC_VMULi32Q,
2328 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2329 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2330 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2332 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2333 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2334 (DSubReg_i16_reg imm:$lane))),
2335 (SubReg_i16_lane imm:$lane)))>;
2336 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2337 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2339 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2340 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2341 (DSubReg_i32_reg imm:$lane))),
2342 (SubReg_i32_lane imm:$lane)))>;
2344 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2345 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2346 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2347 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2348 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2349 IIC_VMULi16Q, IIC_VMULi32Q,
2350 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2351 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2352 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2354 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2355 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2356 (DSubReg_i16_reg imm:$lane))),
2357 (SubReg_i16_lane imm:$lane)))>;
2358 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2359 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2361 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2362 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2363 (DSubReg_i32_reg imm:$lane))),
2364 (SubReg_i32_lane imm:$lane)))>;
2366 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2367 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2368 "vmull", "s", NEONvmulls, 1>;
2369 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2370 "vmull", "u", NEONvmullu, 1>;
2371 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2372 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2373 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
2374 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
2376 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2377 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2378 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2379 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2380 "vqdmull", "s", int_arm_neon_vqdmull>;
2382 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2384 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2385 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2386 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2387 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2389 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2391 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2392 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2393 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2395 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2396 v4f32, v2f32, fmul, fadd>;
2398 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2399 (mul (v8i16 QPR:$src2),
2400 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2401 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2402 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2403 (DSubReg_i16_reg imm:$lane))),
2404 (SubReg_i16_lane imm:$lane)))>;
2406 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2407 (mul (v4i32 QPR:$src2),
2408 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2409 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2410 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2411 (DSubReg_i32_reg imm:$lane))),
2412 (SubReg_i32_lane imm:$lane)))>;
2414 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2415 (fmul (v4f32 QPR:$src2),
2416 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2417 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2419 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2420 (DSubReg_i32_reg imm:$lane))),
2421 (SubReg_i32_lane imm:$lane)))>;
2423 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2424 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2425 "vmlal", "s", NEONvmulls, add>;
2426 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2427 "vmlal", "u", NEONvmullu, add>;
2429 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
2430 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
2432 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2433 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2434 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2435 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2437 // VMLS : Vector Multiply Subtract (integer and floating-point)
2438 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2439 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2440 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2442 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2444 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2445 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2446 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2448 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2449 v4f32, v2f32, fmul, fsub>;
2451 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2452 (mul (v8i16 QPR:$src2),
2453 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2454 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2455 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2456 (DSubReg_i16_reg imm:$lane))),
2457 (SubReg_i16_lane imm:$lane)))>;
2459 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2460 (mul (v4i32 QPR:$src2),
2461 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2462 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2463 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2464 (DSubReg_i32_reg imm:$lane))),
2465 (SubReg_i32_lane imm:$lane)))>;
2467 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2468 (fmul (v4f32 QPR:$src2),
2469 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2470 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2471 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2472 (DSubReg_i32_reg imm:$lane))),
2473 (SubReg_i32_lane imm:$lane)))>;
2475 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2476 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2477 "vmlsl", "s", NEONvmulls, sub>;
2478 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2479 "vmlsl", "u", NEONvmullu, sub>;
2481 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
2482 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
2484 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2485 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2486 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2487 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2489 // Vector Subtract Operations.
2491 // VSUB : Vector Subtract (integer and floating-point)
2492 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2493 "vsub", "i", sub, 0>;
2494 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2495 v2f32, v2f32, fsub, 0>;
2496 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2497 v4f32, v4f32, fsub, 0>;
2498 // VSUBL : Vector Subtract Long (Q = D - D)
2499 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2500 "vsubl", "s", sub, sext, 0>;
2501 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2502 "vsubl", "u", sub, zext, 0>;
2503 // VSUBW : Vector Subtract Wide (Q = Q - D)
2504 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
2505 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
2506 // VHSUB : Vector Halving Subtract
2507 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2508 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2509 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2510 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2511 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2512 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2513 // VQSUB : Vector Saturing Subtract
2514 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2515 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2516 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2517 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2518 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2519 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2520 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2521 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2522 int_arm_neon_vsubhn, 0>;
2523 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2524 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2525 int_arm_neon_vrsubhn, 0>;
2527 // Vector Comparisons.
2529 // VCEQ : Vector Compare Equal
2530 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2531 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2532 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2534 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2536 // For disassembly only.
2537 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2540 // VCGE : Vector Compare Greater Than or Equal
2541 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2542 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2543 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2544 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2545 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2547 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2549 // For disassembly only.
2550 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2552 // For disassembly only.
2553 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2556 // VCGT : Vector Compare Greater Than
2557 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2558 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2559 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2560 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2561 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2563 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2565 // For disassembly only.
2566 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2568 // For disassembly only.
2569 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2572 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2573 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2574 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2575 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2576 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2577 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2578 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2579 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2580 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2581 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2582 // VTST : Vector Test Bits
2583 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2584 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2586 // Vector Bitwise Operations.
2588 def vnotd : PatFrag<(ops node:$in),
2589 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
2590 def vnotq : PatFrag<(ops node:$in),
2591 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
2594 // VAND : Vector Bitwise AND
2595 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2596 v2i32, v2i32, and, 1>;
2597 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2598 v4i32, v4i32, and, 1>;
2600 // VEOR : Vector Bitwise Exclusive OR
2601 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2602 v2i32, v2i32, xor, 1>;
2603 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2604 v4i32, v4i32, xor, 1>;
2606 // VORR : Vector Bitwise OR
2607 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2608 v2i32, v2i32, or, 1>;
2609 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2610 v4i32, v4i32, or, 1>;
2612 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2613 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2614 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2615 "vbic", "$dst, $src1, $src2", "",
2616 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2617 (vnotd DPR:$src2))))]>;
2618 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2619 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2620 "vbic", "$dst, $src1, $src2", "",
2621 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2622 (vnotq QPR:$src2))))]>;
2624 // VORN : Vector Bitwise OR NOT
2625 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2626 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2627 "vorn", "$dst, $src1, $src2", "",
2628 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2629 (vnotd DPR:$src2))))]>;
2630 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2631 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2632 "vorn", "$dst, $src1, $src2", "",
2633 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2634 (vnotq QPR:$src2))))]>;
2636 // VMVN : Vector Bitwise NOT (Immediate)
2638 let isReMaterializable = 1 in {
2639 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
2640 (ins nModImm:$SIMM), IIC_VMOVImm,
2641 "vmvn", "i16", "$dst, $SIMM", "",
2642 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]>;
2643 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
2644 (ins nModImm:$SIMM), IIC_VMOVImm,
2645 "vmvn", "i16", "$dst, $SIMM", "",
2646 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]>;
2648 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
2649 (ins nModImm:$SIMM), IIC_VMOVImm,
2650 "vmvn", "i32", "$dst, $SIMM", "",
2651 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]>;
2652 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
2653 (ins nModImm:$SIMM), IIC_VMOVImm,
2654 "vmvn", "i32", "$dst, $SIMM", "",
2655 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]>;
2658 // VMVN : Vector Bitwise NOT
2659 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2660 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2661 "vmvn", "$dst, $src", "",
2662 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
2663 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2664 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2665 "vmvn", "$dst, $src", "",
2666 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
2667 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
2668 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
2670 // VBSL : Vector Bitwise Select
2671 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2672 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2673 N3RegFrm, IIC_VCNTiD,
2674 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2676 (v2i32 (or (and DPR:$src2, DPR:$src1),
2677 (and DPR:$src3, (vnotd DPR:$src1)))))]>;
2678 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2679 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2680 N3RegFrm, IIC_VCNTiQ,
2681 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2683 (v4i32 (or (and QPR:$src2, QPR:$src1),
2684 (and QPR:$src3, (vnotq QPR:$src1)))))]>;
2686 // VBIF : Vector Bitwise Insert if False
2687 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2688 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2689 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2690 N3RegFrm, IIC_VBINiD,
2691 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2692 [/* For disassembly only; pattern left blank */]>;
2693 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2694 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2695 N3RegFrm, IIC_VBINiQ,
2696 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2697 [/* For disassembly only; pattern left blank */]>;
2699 // VBIT : Vector Bitwise Insert if True
2700 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2701 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2702 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2703 N3RegFrm, IIC_VBINiD,
2704 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2705 [/* For disassembly only; pattern left blank */]>;
2706 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2707 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2708 N3RegFrm, IIC_VBINiQ,
2709 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2710 [/* For disassembly only; pattern left blank */]>;
2712 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2713 // for equivalent operations with different register constraints; it just
2716 // Vector Absolute Differences.
2718 // VABD : Vector Absolute Difference
2719 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2720 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2721 "vabd", "s", int_arm_neon_vabds, 0>;
2722 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2723 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2724 "vabd", "u", int_arm_neon_vabdu, 0>;
2725 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2726 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2727 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2728 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2730 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2731 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2732 "vabdl", "s", int_arm_neon_vabdls, 0>;
2733 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2734 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2736 // VABA : Vector Absolute Difference and Accumulate
2737 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2738 "vaba", "s", int_arm_neon_vabas>;
2739 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2740 "vaba", "u", int_arm_neon_vabau>;
2742 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2743 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2744 "vabal", "s", int_arm_neon_vabals>;
2745 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2746 "vabal", "u", int_arm_neon_vabalu>;
2748 // Vector Maximum and Minimum.
2750 // VMAX : Vector Maximum
2751 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2752 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2753 "vmax", "s", int_arm_neon_vmaxs, 1>;
2754 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2755 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2756 "vmax", "u", int_arm_neon_vmaxu, 1>;
2757 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2759 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2760 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2762 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2764 // VMIN : Vector Minimum
2765 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2766 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2767 "vmin", "s", int_arm_neon_vmins, 1>;
2768 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2769 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2770 "vmin", "u", int_arm_neon_vminu, 1>;
2771 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2773 v2f32, v2f32, int_arm_neon_vmins, 1>;
2774 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2776 v4f32, v4f32, int_arm_neon_vmins, 1>;
2778 // Vector Pairwise Operations.
2780 // VPADD : Vector Pairwise Add
2781 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2783 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2784 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2786 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2787 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2789 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2790 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2791 IIC_VBIND, "vpadd", "f32",
2792 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2794 // VPADDL : Vector Pairwise Add Long
2795 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2796 int_arm_neon_vpaddls>;
2797 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2798 int_arm_neon_vpaddlu>;
2800 // VPADAL : Vector Pairwise Add and Accumulate Long
2801 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2802 int_arm_neon_vpadals>;
2803 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2804 int_arm_neon_vpadalu>;
2806 // VPMAX : Vector Pairwise Maximum
2807 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2808 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2809 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2810 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2811 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2812 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2813 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2814 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2815 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2816 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2817 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2818 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2819 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2820 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2822 // VPMIN : Vector Pairwise Minimum
2823 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2824 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2825 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2826 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2827 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2828 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2829 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2830 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2831 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2832 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2833 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2834 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2835 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2836 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2838 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2840 // VRECPE : Vector Reciprocal Estimate
2841 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2842 IIC_VUNAD, "vrecpe", "u32",
2843 v2i32, v2i32, int_arm_neon_vrecpe>;
2844 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2845 IIC_VUNAQ, "vrecpe", "u32",
2846 v4i32, v4i32, int_arm_neon_vrecpe>;
2847 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2848 IIC_VUNAD, "vrecpe", "f32",
2849 v2f32, v2f32, int_arm_neon_vrecpe>;
2850 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2851 IIC_VUNAQ, "vrecpe", "f32",
2852 v4f32, v4f32, int_arm_neon_vrecpe>;
2854 // VRECPS : Vector Reciprocal Step
2855 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2856 IIC_VRECSD, "vrecps", "f32",
2857 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2858 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2859 IIC_VRECSQ, "vrecps", "f32",
2860 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2862 // VRSQRTE : Vector Reciprocal Square Root Estimate
2863 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2864 IIC_VUNAD, "vrsqrte", "u32",
2865 v2i32, v2i32, int_arm_neon_vrsqrte>;
2866 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2867 IIC_VUNAQ, "vrsqrte", "u32",
2868 v4i32, v4i32, int_arm_neon_vrsqrte>;
2869 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2870 IIC_VUNAD, "vrsqrte", "f32",
2871 v2f32, v2f32, int_arm_neon_vrsqrte>;
2872 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2873 IIC_VUNAQ, "vrsqrte", "f32",
2874 v4f32, v4f32, int_arm_neon_vrsqrte>;
2876 // VRSQRTS : Vector Reciprocal Square Root Step
2877 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2878 IIC_VRECSD, "vrsqrts", "f32",
2879 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2880 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2881 IIC_VRECSQ, "vrsqrts", "f32",
2882 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2886 // VSHL : Vector Shift
2887 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2888 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2889 "vshl", "s", int_arm_neon_vshifts, 0>;
2890 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2891 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2892 "vshl", "u", int_arm_neon_vshiftu, 0>;
2893 // VSHL : Vector Shift Left (Immediate)
2894 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2896 // VSHR : Vector Shift Right (Immediate)
2897 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2899 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2902 // VSHLL : Vector Shift Left Long
2903 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2904 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2906 // VSHLL : Vector Shift Left Long (with maximum shift count)
2907 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2908 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2909 ValueType OpTy, SDNode OpNode>
2910 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2911 ResTy, OpTy, OpNode> {
2912 let Inst{21-16} = op21_16;
2914 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2915 v8i16, v8i8, NEONvshlli>;
2916 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2917 v4i32, v4i16, NEONvshlli>;
2918 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2919 v2i64, v2i32, NEONvshlli>;
2921 // VSHRN : Vector Shift Right and Narrow
2922 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2925 // VRSHL : Vector Rounding Shift
2926 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2927 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2928 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2929 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2930 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2931 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2932 // VRSHR : Vector Rounding Shift Right
2933 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2935 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2938 // VRSHRN : Vector Rounding Shift Right and Narrow
2939 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2942 // VQSHL : Vector Saturating Shift
2943 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2944 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2945 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2946 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2947 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2948 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2949 // VQSHL : Vector Saturating Shift Left (Immediate)
2950 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2952 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2954 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2955 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2958 // VQSHRN : Vector Saturating Shift Right and Narrow
2959 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2961 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2964 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2965 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2968 // VQRSHL : Vector Saturating Rounding Shift
2969 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
2970 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2971 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
2972 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
2973 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2974 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
2976 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2977 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2979 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2982 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2983 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2986 // VSRA : Vector Shift Right and Accumulate
2987 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2988 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2989 // VRSRA : Vector Rounding Shift Right and Accumulate
2990 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2991 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2993 // VSLI : Vector Shift Left and Insert
2994 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2995 // VSRI : Vector Shift Right and Insert
2996 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2998 // Vector Absolute and Saturating Absolute.
3000 // VABS : Vector Absolute Value
3001 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3002 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3004 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3005 IIC_VUNAD, "vabs", "f32",
3006 v2f32, v2f32, int_arm_neon_vabs>;
3007 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3008 IIC_VUNAQ, "vabs", "f32",
3009 v4f32, v4f32, int_arm_neon_vabs>;
3011 // VQABS : Vector Saturating Absolute Value
3012 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3013 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3014 int_arm_neon_vqabs>;
3018 def vnegd : PatFrag<(ops node:$in),
3019 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3020 def vnegq : PatFrag<(ops node:$in),
3021 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3023 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3024 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3025 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3026 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3027 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3028 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3029 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3030 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3032 // VNEG : Vector Negate (integer)
3033 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3034 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3035 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3036 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3037 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3038 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3040 // VNEG : Vector Negate (floating-point)
3041 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3042 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3043 "vneg", "f32", "$dst, $src", "",
3044 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3045 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3046 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3047 "vneg", "f32", "$dst, $src", "",
3048 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3050 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3051 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3052 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3053 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3054 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3055 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3057 // VQNEG : Vector Saturating Negate
3058 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3059 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3060 int_arm_neon_vqneg>;
3062 // Vector Bit Counting Operations.
3064 // VCLS : Vector Count Leading Sign Bits
3065 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3066 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3068 // VCLZ : Vector Count Leading Zeros
3069 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3070 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3072 // VCNT : Vector Count One Bits
3073 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3074 IIC_VCNTiD, "vcnt", "8",
3075 v8i8, v8i8, int_arm_neon_vcnt>;
3076 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3077 IIC_VCNTiQ, "vcnt", "8",
3078 v16i8, v16i8, int_arm_neon_vcnt>;
3080 // Vector Swap -- for disassembly only.
3081 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3082 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3083 "vswp", "$dst, $src", "", []>;
3084 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3085 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3086 "vswp", "$dst, $src", "", []>;
3088 // Vector Move Operations.
3090 // VMOV : Vector Move (Register)
3092 let neverHasSideEffects = 1 in {
3093 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3094 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3095 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3096 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
3098 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3099 // be expanded after register allocation is completed.
3100 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3101 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3103 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3104 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
3105 } // neverHasSideEffects
3107 // VMOV : Vector Move (Immediate)
3109 let isReMaterializable = 1 in {
3110 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3111 (ins nModImm:$SIMM), IIC_VMOVImm,
3112 "vmov", "i8", "$dst, $SIMM", "",
3113 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3114 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3115 (ins nModImm:$SIMM), IIC_VMOVImm,
3116 "vmov", "i8", "$dst, $SIMM", "",
3117 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3119 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3120 (ins nModImm:$SIMM), IIC_VMOVImm,
3121 "vmov", "i16", "$dst, $SIMM", "",
3122 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]>;
3123 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3124 (ins nModImm:$SIMM), IIC_VMOVImm,
3125 "vmov", "i16", "$dst, $SIMM", "",
3126 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]>;
3128 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3129 (ins nModImm:$SIMM), IIC_VMOVImm,
3130 "vmov", "i32", "$dst, $SIMM", "",
3131 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]>;
3132 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3133 (ins nModImm:$SIMM), IIC_VMOVImm,
3134 "vmov", "i32", "$dst, $SIMM", "",
3135 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]>;
3137 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3138 (ins nModImm:$SIMM), IIC_VMOVImm,
3139 "vmov", "i64", "$dst, $SIMM", "",
3140 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3141 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3142 (ins nModImm:$SIMM), IIC_VMOVImm,
3143 "vmov", "i64", "$dst, $SIMM", "",
3144 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3145 } // isReMaterializable
3147 // VMOV : Vector Get Lane (move scalar to ARM core register)
3149 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3150 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3151 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
3152 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
3154 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3155 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3156 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
3157 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
3159 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3160 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3161 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
3162 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
3164 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3165 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3166 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
3167 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
3169 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3170 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3171 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
3172 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
3174 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3175 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3176 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3177 (DSubReg_i8_reg imm:$lane))),
3178 (SubReg_i8_lane imm:$lane))>;
3179 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3180 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3181 (DSubReg_i16_reg imm:$lane))),
3182 (SubReg_i16_lane imm:$lane))>;
3183 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3184 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3185 (DSubReg_i8_reg imm:$lane))),
3186 (SubReg_i8_lane imm:$lane))>;
3187 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3188 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3189 (DSubReg_i16_reg imm:$lane))),
3190 (SubReg_i16_lane imm:$lane))>;
3191 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3192 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3193 (DSubReg_i32_reg imm:$lane))),
3194 (SubReg_i32_lane imm:$lane))>;
3195 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3196 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3197 (SSubReg_f32_reg imm:$src2))>;
3198 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3199 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3200 (SSubReg_f32_reg imm:$src2))>;
3201 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3202 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3203 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3204 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3207 // VMOV : Vector Set Lane (move ARM core register to scalar)
3209 let Constraints = "$src1 = $dst" in {
3210 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
3211 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3212 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
3213 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
3214 GPR:$src2, imm:$lane))]>;
3215 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
3216 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3217 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
3218 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
3219 GPR:$src2, imm:$lane))]>;
3220 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
3221 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
3222 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
3223 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
3224 GPR:$src2, imm:$lane))]>;
3226 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3227 (v16i8 (INSERT_SUBREG QPR:$src1,
3228 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3229 (DSubReg_i8_reg imm:$lane))),
3230 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3231 (DSubReg_i8_reg imm:$lane)))>;
3232 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3233 (v8i16 (INSERT_SUBREG QPR:$src1,
3234 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3235 (DSubReg_i16_reg imm:$lane))),
3236 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3237 (DSubReg_i16_reg imm:$lane)))>;
3238 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3239 (v4i32 (INSERT_SUBREG QPR:$src1,
3240 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3241 (DSubReg_i32_reg imm:$lane))),
3242 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3243 (DSubReg_i32_reg imm:$lane)))>;
3245 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3246 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3247 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3248 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3249 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3250 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3252 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3253 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3254 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3255 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3257 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3258 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3259 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3260 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3261 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3262 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3264 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3265 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3266 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3267 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3268 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3269 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3271 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3272 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3273 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3275 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3276 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3277 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3279 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3280 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3281 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3284 // VDUP : Vector Duplicate (from ARM core register to all elements)
3286 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3287 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3288 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3289 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3290 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3291 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3292 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3293 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3295 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3296 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3297 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3298 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3299 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3300 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3302 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3303 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3304 [(set DPR:$dst, (v2f32 (NEONvdup
3305 (f32 (bitconvert GPR:$src)))))]>;
3306 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3307 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3308 [(set QPR:$dst, (v4f32 (NEONvdup
3309 (f32 (bitconvert GPR:$src)))))]>;
3311 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3313 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3315 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3316 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3317 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3319 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3320 ValueType ResTy, ValueType OpTy>
3321 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3322 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3323 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3326 // Inst{19-16} is partially specified depending on the element size.
3328 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3329 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3330 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3331 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3332 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3333 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3334 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3335 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3337 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3338 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3339 (DSubReg_i8_reg imm:$lane))),
3340 (SubReg_i8_lane imm:$lane)))>;
3341 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3342 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3343 (DSubReg_i16_reg imm:$lane))),
3344 (SubReg_i16_lane imm:$lane)))>;
3345 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3346 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3347 (DSubReg_i32_reg imm:$lane))),
3348 (SubReg_i32_lane imm:$lane)))>;
3349 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3350 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3351 (DSubReg_i32_reg imm:$lane))),
3352 (SubReg_i32_lane imm:$lane)))>;
3354 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3355 (outs DPR:$dst), (ins SPR:$src),
3356 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3357 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3359 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3360 (outs QPR:$dst), (ins SPR:$src),
3361 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3362 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3364 // VMOVN : Vector Narrowing Move
3365 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3366 "vmovn", "i", trunc>;
3367 // VQMOVN : Vector Saturating Narrowing Move
3368 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3369 "vqmovn", "s", int_arm_neon_vqmovns>;
3370 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3371 "vqmovn", "u", int_arm_neon_vqmovnu>;
3372 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3373 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3374 // VMOVL : Vector Lengthening Move
3375 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3376 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3378 // Vector Conversions.
3380 // VCVT : Vector Convert Between Floating-Point and Integers
3381 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3382 v2i32, v2f32, fp_to_sint>;
3383 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3384 v2i32, v2f32, fp_to_uint>;
3385 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3386 v2f32, v2i32, sint_to_fp>;
3387 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3388 v2f32, v2i32, uint_to_fp>;
3390 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3391 v4i32, v4f32, fp_to_sint>;
3392 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3393 v4i32, v4f32, fp_to_uint>;
3394 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3395 v4f32, v4i32, sint_to_fp>;
3396 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3397 v4f32, v4i32, uint_to_fp>;
3399 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3400 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3401 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3402 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3403 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3404 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3405 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3406 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3407 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3409 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3410 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3411 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3412 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3413 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3414 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3415 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3416 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3420 // VREV64 : Vector Reverse elements within 64-bit doublewords
3422 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3423 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3424 (ins DPR:$src), IIC_VMOVD,
3425 OpcodeStr, Dt, "$dst, $src", "",
3426 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3427 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3428 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3429 (ins QPR:$src), IIC_VMOVD,
3430 OpcodeStr, Dt, "$dst, $src", "",
3431 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3433 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3434 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3435 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3436 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3438 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3439 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3440 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3441 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3443 // VREV32 : Vector Reverse elements within 32-bit words
3445 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3446 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3447 (ins DPR:$src), IIC_VMOVD,
3448 OpcodeStr, Dt, "$dst, $src", "",
3449 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3450 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3451 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3452 (ins QPR:$src), IIC_VMOVD,
3453 OpcodeStr, Dt, "$dst, $src", "",
3454 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3456 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3457 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3459 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3460 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3462 // VREV16 : Vector Reverse elements within 16-bit halfwords
3464 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3465 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3466 (ins DPR:$src), IIC_VMOVD,
3467 OpcodeStr, Dt, "$dst, $src", "",
3468 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3469 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3470 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3471 (ins QPR:$src), IIC_VMOVD,
3472 OpcodeStr, Dt, "$dst, $src", "",
3473 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3475 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3476 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3478 // Other Vector Shuffles.
3480 // VEXT : Vector Extract
3482 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3483 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3484 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3485 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3486 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3487 (Ty DPR:$rhs), imm:$index)))]>;
3489 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3490 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3491 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3492 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3493 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3494 (Ty QPR:$rhs), imm:$index)))]>;
3496 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3497 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3498 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3499 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3501 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3502 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3503 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3504 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3506 // VTRN : Vector Transpose
3508 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3509 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3510 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3512 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3513 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3514 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3516 // VUZP : Vector Unzip (Deinterleave)
3518 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3519 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3520 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3522 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3523 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3524 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3526 // VZIP : Vector Zip (Interleave)
3528 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3529 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3530 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3532 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3533 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3534 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3536 // Vector Table Lookup and Table Extension.
3538 // VTBL : Vector Table Lookup
3540 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3541 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3542 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3543 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3544 let hasExtraSrcRegAllocReq = 1 in {
3546 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3547 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3548 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "", []>;
3550 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3551 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3552 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "", []>;
3554 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3555 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3557 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "", []>;
3558 } // hasExtraSrcRegAllocReq = 1
3560 // VTBX : Vector Table Extension
3562 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3563 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3564 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3565 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3566 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3567 let hasExtraSrcRegAllocReq = 1 in {
3569 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3570 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3571 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst", []>;
3573 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3574 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3575 NVTBLFrm, IIC_VTBX3,
3576 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src",
3577 "$orig = $dst", []>;
3579 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3580 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3581 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3582 "$orig = $dst", []>;
3583 } // hasExtraSrcRegAllocReq = 1
3585 //===----------------------------------------------------------------------===//
3586 // NEON instructions for single-precision FP math
3587 //===----------------------------------------------------------------------===//
3589 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3590 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3591 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3595 class N3VSPat<SDNode OpNode, NeonI Inst>
3596 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3597 (EXTRACT_SUBREG (v2f32
3598 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3600 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3604 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3605 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3606 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3608 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3610 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3614 // These need separate instructions because they must use DPR_VFP2 register
3615 // class which have SPR sub-registers.
3617 // Vector Add Operations used for single-precision FP
3618 let neverHasSideEffects = 1 in
3619 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3620 def : N3VSPat<fadd, VADDfd_sfp>;
3622 // Vector Sub Operations used for single-precision FP
3623 let neverHasSideEffects = 1 in
3624 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3625 def : N3VSPat<fsub, VSUBfd_sfp>;
3627 // Vector Multiply Operations used for single-precision FP
3628 let neverHasSideEffects = 1 in
3629 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3630 def : N3VSPat<fmul, VMULfd_sfp>;
3632 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3633 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3634 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3636 //let neverHasSideEffects = 1 in
3637 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3638 // v2f32, fmul, fadd>;
3639 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3641 //let neverHasSideEffects = 1 in
3642 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3643 // v2f32, fmul, fsub>;
3644 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3646 // Vector Absolute used for single-precision FP
3647 let neverHasSideEffects = 1 in
3648 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3649 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3650 "vabs", "f32", "$dst, $src", "", []>;
3651 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3653 // Vector Negate used for single-precision FP
3654 let neverHasSideEffects = 1 in
3655 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3656 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3657 "vneg", "f32", "$dst, $src", "", []>;
3658 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3660 // Vector Maximum used for single-precision FP
3661 let neverHasSideEffects = 1 in
3662 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3663 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3664 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3665 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3667 // Vector Minimum used for single-precision FP
3668 let neverHasSideEffects = 1 in
3669 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3670 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3671 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3672 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3674 // Vector Convert between single-precision FP and integer
3675 let neverHasSideEffects = 1 in
3676 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3677 v2i32, v2f32, fp_to_sint>;
3678 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3680 let neverHasSideEffects = 1 in
3681 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3682 v2i32, v2f32, fp_to_uint>;
3683 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3685 let neverHasSideEffects = 1 in
3686 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3687 v2f32, v2i32, sint_to_fp>;
3688 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3690 let neverHasSideEffects = 1 in
3691 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3692 v2f32, v2i32, uint_to_fp>;
3693 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3695 //===----------------------------------------------------------------------===//
3696 // Non-Instruction Patterns
3697 //===----------------------------------------------------------------------===//
3700 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3701 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3702 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3703 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3704 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3705 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3706 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3707 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3708 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3709 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3710 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3711 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3712 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3713 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3714 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3715 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3716 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3717 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3718 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3719 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3720 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3721 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3722 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3723 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3724 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3725 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3726 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3727 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3728 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3729 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3731 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3732 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3733 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3734 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3735 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3736 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3737 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3738 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3739 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3740 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3741 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3742 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3743 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3744 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3745 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3746 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3747 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3748 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3749 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3750 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3751 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3752 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3753 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3754 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3755 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3756 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3757 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3758 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3759 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3760 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;