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 NEONvmovImm : SDNode<"ARMISD::VMOVIMM",
69 SDTypeProfile<1, 1, [SDTCisVec<0>,
71 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
73 // VDUPLANE can produce a quad-register result from a double-register source,
74 // so the result is not constrained to match the source.
75 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
76 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
79 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
80 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
81 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
83 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
84 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
85 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
86 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
88 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
91 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
92 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
93 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
95 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
97 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
98 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
100 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
101 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
103 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
104 return (EltBits == 32 && EltVal == 0);
107 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
108 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
110 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
111 return (EltBits == 8 && EltVal == 0xff);
114 //===----------------------------------------------------------------------===//
115 // NEON operand definitions
116 //===----------------------------------------------------------------------===//
118 def nModImm : Operand<i32> {
119 let PrintMethod = "printNEONModImmOperand";
122 //===----------------------------------------------------------------------===//
123 // NEON load / store instructions
124 //===----------------------------------------------------------------------===//
126 let mayLoad = 1, neverHasSideEffects = 1 in {
127 // Use vldmia to load a Q register as a D register pair.
128 // This is equivalent to VLDMD except that it has a Q register operand
129 // instead of a pair of D registers.
131 : AXDI5<(outs QPR:$dst), (ins addrmode5:$addr, pred:$p),
132 IndexModeNone, IIC_fpLoadm,
133 "vldm${addr:submode}${p}\t${addr:base}, ${dst:dregpair}", "", []>;
135 // Use vld1 to load a Q register as a D register pair.
136 // This alternative to VLDMQ allows an alignment to be specified.
137 // This is equivalent to VLD1q64 except that it has a Q register operand.
139 : NLdSt<0,0b10,0b1010,0b1100, (outs QPR:$dst), (ins addrmode6:$addr),
140 IIC_VLD1, "vld1", "64", "${dst:dregpair}, $addr", "", []>;
141 } // mayLoad = 1, neverHasSideEffects = 1
143 let mayStore = 1, neverHasSideEffects = 1 in {
144 // Use vstmia to store a Q register as a D register pair.
145 // This is equivalent to VSTMD except that it has a Q register operand
146 // instead of a pair of D registers.
148 : AXDI5<(outs), (ins QPR:$src, addrmode5:$addr, pred:$p),
149 IndexModeNone, IIC_fpStorem,
150 "vstm${addr:submode}${p}\t${addr:base}, ${src:dregpair}", "", []>;
152 // Use vst1 to store a Q register as a D register pair.
153 // This alternative to VSTMQ allows an alignment to be specified.
154 // This is equivalent to VST1q64 except that it has a Q register operand.
156 : NLdSt<0,0b00,0b1010,0b1100, (outs), (ins addrmode6:$addr, QPR:$src),
157 IIC_VST, "vst1", "64", "${src:dregpair}, $addr", "", []>;
158 } // mayStore = 1, neverHasSideEffects = 1
160 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
162 // VLD1 : Vector Load (multiple single elements)
163 class VLD1D<bits<4> op7_4, string Dt>
164 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst),
165 (ins addrmode6:$addr), IIC_VLD1,
166 "vld1", Dt, "\\{$dst\\}, $addr", "", []>;
167 class VLD1Q<bits<4> op7_4, string Dt>
168 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$dst1, DPR:$dst2),
169 (ins addrmode6:$addr), IIC_VLD1,
170 "vld1", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
172 def VLD1d8 : VLD1D<0b0000, "8">;
173 def VLD1d16 : VLD1D<0b0100, "16">;
174 def VLD1d32 : VLD1D<0b1000, "32">;
175 def VLD1d64 : VLD1D<0b1100, "64">;
177 def VLD1q8 : VLD1Q<0b0000, "8">;
178 def VLD1q16 : VLD1Q<0b0100, "16">;
179 def VLD1q32 : VLD1Q<0b1000, "32">;
180 def VLD1q64 : VLD1Q<0b1100, "64">;
182 // ...with address register writeback:
183 class VLD1DWB<bits<4> op7_4, string Dt>
184 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$dst, GPR:$wb),
185 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
186 "vld1", Dt, "\\{$dst\\}, $addr$offset",
187 "$addr.addr = $wb", []>;
188 class VLD1QWB<bits<4> op7_4, string Dt>
189 : NLdSt<0,0b10,0b1010,op7_4, (outs QPR:$dst, GPR:$wb),
190 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1,
191 "vld1", Dt, "${dst:dregpair}, $addr$offset",
192 "$addr.addr = $wb", []>;
194 def VLD1d8_UPD : VLD1DWB<0b0000, "8">;
195 def VLD1d16_UPD : VLD1DWB<0b0100, "16">;
196 def VLD1d32_UPD : VLD1DWB<0b1000, "32">;
197 def VLD1d64_UPD : VLD1DWB<0b1100, "64">;
199 def VLD1q8_UPD : VLD1QWB<0b0000, "8">;
200 def VLD1q16_UPD : VLD1QWB<0b0100, "16">;
201 def VLD1q32_UPD : VLD1QWB<0b1000, "32">;
202 def VLD1q64_UPD : VLD1QWB<0b1100, "64">;
204 // ...with 3 registers (some of these are only for the disassembler):
205 class VLD1D3<bits<4> op7_4, string Dt>
206 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
207 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
208 "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
209 class VLD1D3WB<bits<4> op7_4, string Dt>
210 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
211 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
212 "\\{$dst1, $dst2, $dst3\\}, $addr$offset", "$addr.addr = $wb", []>;
214 def VLD1d8T : VLD1D3<0b0000, "8">;
215 def VLD1d16T : VLD1D3<0b0100, "16">;
216 def VLD1d32T : VLD1D3<0b1000, "32">;
217 def VLD1d64T : VLD1D3<0b1100, "64">;
219 def VLD1d8T_UPD : VLD1D3WB<0b0000, "8">;
220 def VLD1d16T_UPD : VLD1D3WB<0b0100, "16">;
221 def VLD1d32T_UPD : VLD1D3WB<0b1000, "32">;
222 def VLD1d64T_UPD : VLD1D3WB<0b1100, "64">;
224 // ...with 4 registers (some of these are only for the disassembler):
225 class VLD1D4<bits<4> op7_4, string Dt>
226 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
227 (ins addrmode6:$addr), IIC_VLD1, "vld1", Dt,
228 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
229 class VLD1D4WB<bits<4> op7_4, string Dt>
230 : NLdSt<0,0b10,0b0010,op7_4,
231 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
232 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD1, "vld1", Dt,
233 "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset", "$addr.addr = $wb",
236 def VLD1d8Q : VLD1D4<0b0000, "8">;
237 def VLD1d16Q : VLD1D4<0b0100, "16">;
238 def VLD1d32Q : VLD1D4<0b1000, "32">;
239 def VLD1d64Q : VLD1D4<0b1100, "64">;
241 def VLD1d8Q_UPD : VLD1D4WB<0b0000, "8">;
242 def VLD1d16Q_UPD : VLD1D4WB<0b0100, "16">;
243 def VLD1d32Q_UPD : VLD1D4WB<0b1000, "32">;
244 def VLD1d64Q_UPD : VLD1D4WB<0b1100, "64">;
246 // VLD2 : Vector Load (multiple 2-element structures)
247 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
248 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
249 (ins addrmode6:$addr), IIC_VLD2,
250 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr", "", []>;
251 class VLD2Q<bits<4> op7_4, string Dt>
252 : NLdSt<0, 0b10, 0b0011, op7_4,
253 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
254 (ins addrmode6:$addr), IIC_VLD2,
255 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
257 def VLD2d8 : VLD2D<0b1000, 0b0000, "8">;
258 def VLD2d16 : VLD2D<0b1000, 0b0100, "16">;
259 def VLD2d32 : VLD2D<0b1000, 0b1000, "32">;
261 def VLD2q8 : VLD2Q<0b0000, "8">;
262 def VLD2q16 : VLD2Q<0b0100, "16">;
263 def VLD2q32 : VLD2Q<0b1000, "32">;
265 // ...with address register writeback:
266 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
267 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
268 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
269 "vld2", Dt, "\\{$dst1, $dst2\\}, $addr$offset",
270 "$addr.addr = $wb", []>;
271 class VLD2QWB<bits<4> op7_4, string Dt>
272 : NLdSt<0, 0b10, 0b0011, op7_4,
273 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
274 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD2,
275 "vld2", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
276 "$addr.addr = $wb", []>;
278 def VLD2d8_UPD : VLD2DWB<0b1000, 0b0000, "8">;
279 def VLD2d16_UPD : VLD2DWB<0b1000, 0b0100, "16">;
280 def VLD2d32_UPD : VLD2DWB<0b1000, 0b1000, "32">;
282 def VLD2q8_UPD : VLD2QWB<0b0000, "8">;
283 def VLD2q16_UPD : VLD2QWB<0b0100, "16">;
284 def VLD2q32_UPD : VLD2QWB<0b1000, "32">;
286 // ...with double-spaced registers (for disassembly only):
287 def VLD2b8 : VLD2D<0b1001, 0b0000, "8">;
288 def VLD2b16 : VLD2D<0b1001, 0b0100, "16">;
289 def VLD2b32 : VLD2D<0b1001, 0b1000, "32">;
290 def VLD2b8_UPD : VLD2DWB<0b1001, 0b0000, "8">;
291 def VLD2b16_UPD : VLD2DWB<0b1001, 0b0100, "16">;
292 def VLD2b32_UPD : VLD2DWB<0b1001, 0b1000, "32">;
294 // VLD3 : Vector Load (multiple 3-element structures)
295 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
296 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
297 (ins addrmode6:$addr), IIC_VLD3,
298 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr", "", []>;
300 def VLD3d8 : VLD3D<0b0100, 0b0000, "8">;
301 def VLD3d16 : VLD3D<0b0100, 0b0100, "16">;
302 def VLD3d32 : VLD3D<0b0100, 0b1000, "32">;
304 // ...with address register writeback:
305 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
306 : NLdSt<0, 0b10, op11_8, op7_4,
307 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
308 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
309 "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
310 "$addr.addr = $wb", []>;
312 def VLD3d8_UPD : VLD3DWB<0b0100, 0b0000, "8">;
313 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
314 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
316 // ...with double-spaced registers (non-updating versions for disassembly only):
317 def VLD3q8 : VLD3D<0b0101, 0b0000, "8">;
318 def VLD3q16 : VLD3D<0b0101, 0b0100, "16">;
319 def VLD3q32 : VLD3D<0b0101, 0b1000, "32">;
320 def VLD3q8_UPD : VLD3DWB<0b0101, 0b0000, "8">;
321 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
322 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
324 // ...alternate versions to be allocated odd register numbers:
325 def VLD3q8odd_UPD : VLD3DWB<0b0101, 0b0000, "8">;
326 def VLD3q16odd_UPD : VLD3DWB<0b0101, 0b0100, "16">;
327 def VLD3q32odd_UPD : VLD3DWB<0b0101, 0b1000, "32">;
329 // VLD4 : Vector Load (multiple 4-element structures)
330 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
331 : NLdSt<0, 0b10, op11_8, op7_4,
332 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
333 (ins addrmode6:$addr), IIC_VLD4,
334 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr", "", []>;
336 def VLD4d8 : VLD4D<0b0000, 0b0000, "8">;
337 def VLD4d16 : VLD4D<0b0000, 0b0100, "16">;
338 def VLD4d32 : VLD4D<0b0000, 0b1000, "32">;
340 // ...with address register writeback:
341 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
342 : NLdSt<0, 0b10, op11_8, op7_4,
343 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
344 (ins addrmode6:$addr, am6offset:$offset), IIC_VLD4,
345 "vld4", Dt, "\\{$dst1, $dst2, $dst3, $dst4\\}, $addr$offset",
346 "$addr.addr = $wb", []>;
348 def VLD4d8_UPD : VLD4DWB<0b0000, 0b0000, "8">;
349 def VLD4d16_UPD : VLD4DWB<0b0000, 0b0100, "16">;
350 def VLD4d32_UPD : VLD4DWB<0b0000, 0b1000, "32">;
352 // ...with double-spaced registers (non-updating versions for disassembly only):
353 def VLD4q8 : VLD4D<0b0001, 0b0000, "8">;
354 def VLD4q16 : VLD4D<0b0001, 0b0100, "16">;
355 def VLD4q32 : VLD4D<0b0001, 0b1000, "32">;
356 def VLD4q8_UPD : VLD4DWB<0b0001, 0b0000, "8">;
357 def VLD4q16_UPD : VLD4DWB<0b0001, 0b0100, "16">;
358 def VLD4q32_UPD : VLD4DWB<0b0001, 0b1000, "32">;
360 // ...alternate versions to be allocated odd register numbers:
361 def VLD4q8odd_UPD : VLD4DWB<0b0001, 0b0000, "8">;
362 def VLD4q16odd_UPD : VLD4DWB<0b0001, 0b0100, "16">;
363 def VLD4q32odd_UPD : VLD4DWB<0b0001, 0b1000, "32">;
365 // VLD1LN : Vector Load (single element to one lane)
366 // FIXME: Not yet implemented.
368 // VLD2LN : Vector Load (single 2-element structure to one lane)
369 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
370 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2),
371 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
372 IIC_VLD2, "vld2", Dt, "\\{$dst1[$lane], $dst2[$lane]\\}, $addr",
373 "$src1 = $dst1, $src2 = $dst2", []>;
375 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8">;
376 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16">;
377 def VLD2LNd32 : VLD2LN<0b1001, {?,0,?,?}, "32">;
379 // ...with double-spaced registers:
380 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16">;
381 def VLD2LNq32 : VLD2LN<0b1001, {?,1,?,?}, "32">;
383 // ...alternate versions to be allocated odd register numbers:
384 def VLD2LNq16odd : VLD2LN<0b0101, {?,?,1,?}, "16">;
385 def VLD2LNq32odd : VLD2LN<0b1001, {?,1,?,?}, "32">;
387 // ...with address register writeback:
388 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
389 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, GPR:$wb),
390 (ins addrmode6:$addr, am6offset:$offset,
391 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2, "vld2", Dt,
392 "\\{$dst1[$lane], $dst2[$lane]\\}, $addr$offset",
393 "$src1 = $dst1, $src2 = $dst2, $addr.addr = $wb", []>;
395 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8">;
396 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16">;
397 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,?,?}, "32">;
399 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16">;
400 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,?,?}, "32">;
402 // VLD3LN : Vector Load (single 3-element structure to one lane)
403 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
404 : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
405 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
406 nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
407 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
408 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
410 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8">;
411 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
412 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
414 // ...with double-spaced registers:
415 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
416 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
418 // ...alternate versions to be allocated odd register numbers:
419 def VLD3LNq16odd : VLD3LN<0b0110, {?,?,1,0}, "16">;
420 def VLD3LNq32odd : VLD3LN<0b1010, {?,1,0,0}, "32">;
422 // ...with address register writeback:
423 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
424 : NLdSt<1, 0b10, op11_8, op7_4,
425 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
426 (ins addrmode6:$addr, am6offset:$offset,
427 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
428 IIC_VLD3, "vld3", Dt,
429 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
430 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
433 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
434 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
435 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
437 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
438 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
440 // VLD4LN : Vector Load (single 4-element structure to one lane)
441 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
442 : NLdSt<1, 0b10, op11_8, op7_4,
443 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4),
444 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
445 nohash_imm:$lane), IIC_VLD4, "vld4", Dt,
446 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr",
447 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []>;
449 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8">;
450 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16">;
451 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32">;
453 // ...with double-spaced registers:
454 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16">;
455 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32">;
457 // ...alternate versions to be allocated odd register numbers:
458 def VLD4LNq16odd : VLD4LN<0b0111, {?,?,1,?}, "16">;
459 def VLD4LNq32odd : VLD4LN<0b1011, {?,1,?,?}, "32">;
461 // ...with address register writeback:
462 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
463 : NLdSt<1, 0b10, op11_8, op7_4,
464 (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
465 (ins addrmode6:$addr, am6offset:$offset,
466 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
467 IIC_VLD4, "vld4", Dt,
468 "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $addr$offset",
469 "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $addr.addr = $wb",
472 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8">;
473 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16">;
474 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32">;
476 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16">;
477 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32">;
479 // VLD1DUP : Vector Load (single element to all lanes)
480 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
481 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
482 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
483 // FIXME: Not yet implemented.
484 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
486 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
488 // VST1 : Vector Store (multiple single elements)
489 class VST1D<bits<4> op7_4, string Dt>
490 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$addr, DPR:$src), IIC_VST,
491 "vst1", Dt, "\\{$src\\}, $addr", "", []>;
492 class VST1Q<bits<4> op7_4, string Dt>
493 : NLdSt<0,0b00,0b1010,op7_4, (outs),
494 (ins addrmode6:$addr, DPR:$src1, DPR:$src2), IIC_VST,
495 "vst1", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
497 def VST1d8 : VST1D<0b0000, "8">;
498 def VST1d16 : VST1D<0b0100, "16">;
499 def VST1d32 : VST1D<0b1000, "32">;
500 def VST1d64 : VST1D<0b1100, "64">;
502 def VST1q8 : VST1Q<0b0000, "8">;
503 def VST1q16 : VST1Q<0b0100, "16">;
504 def VST1q32 : VST1Q<0b1000, "32">;
505 def VST1q64 : VST1Q<0b1100, "64">;
507 // ...with address register writeback:
508 class VST1DWB<bits<4> op7_4, string Dt>
509 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
510 (ins addrmode6:$addr, am6offset:$offset, DPR:$src), IIC_VST,
511 "vst1", Dt, "\\{$src\\}, $addr$offset", "$addr.addr = $wb", []>;
512 class VST1QWB<bits<4> op7_4, string Dt>
513 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
514 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), IIC_VST,
515 "vst1", Dt, "${src:dregpair}, $addr$offset", "$addr.addr = $wb", []>;
517 def VST1d8_UPD : VST1DWB<0b0000, "8">;
518 def VST1d16_UPD : VST1DWB<0b0100, "16">;
519 def VST1d32_UPD : VST1DWB<0b1000, "32">;
520 def VST1d64_UPD : VST1DWB<0b1100, "64">;
522 def VST1q8_UPD : VST1QWB<0b0000, "8">;
523 def VST1q16_UPD : VST1QWB<0b0100, "16">;
524 def VST1q32_UPD : VST1QWB<0b1000, "32">;
525 def VST1q64_UPD : VST1QWB<0b1100, "64">;
527 // ...with 3 registers (some of these are only for the disassembler):
528 class VST1D3<bits<4> op7_4, string Dt>
529 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
530 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3),
531 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
532 class VST1D3WB<bits<4> op7_4, string Dt>
533 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
534 (ins addrmode6:$addr, am6offset:$offset,
535 DPR:$src1, DPR:$src2, DPR:$src3),
536 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
537 "$addr.addr = $wb", []>;
539 def VST1d8T : VST1D3<0b0000, "8">;
540 def VST1d16T : VST1D3<0b0100, "16">;
541 def VST1d32T : VST1D3<0b1000, "32">;
542 def VST1d64T : VST1D3<0b1100, "64">;
544 def VST1d8T_UPD : VST1D3WB<0b0000, "8">;
545 def VST1d16T_UPD : VST1D3WB<0b0100, "16">;
546 def VST1d32T_UPD : VST1D3WB<0b1000, "32">;
547 def VST1d64T_UPD : VST1D3WB<0b1100, "64">;
549 // ...with 4 registers (some of these are only for the disassembler):
550 class VST1D4<bits<4> op7_4, string Dt>
551 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
552 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
553 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr", "",
555 class VST1D4WB<bits<4> op7_4, string Dt>
556 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
557 (ins addrmode6:$addr, am6offset:$offset,
558 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
559 IIC_VST, "vst1", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
560 "$addr.addr = $wb", []>;
562 def VST1d8Q : VST1D4<0b0000, "8">;
563 def VST1d16Q : VST1D4<0b0100, "16">;
564 def VST1d32Q : VST1D4<0b1000, "32">;
565 def VST1d64Q : VST1D4<0b1100, "64">;
567 def VST1d8Q_UPD : VST1D4WB<0b0000, "8">;
568 def VST1d16Q_UPD : VST1D4WB<0b0100, "16">;
569 def VST1d32Q_UPD : VST1D4WB<0b1000, "32">;
570 def VST1d64Q_UPD : VST1D4WB<0b1100, "64">;
572 // VST2 : Vector Store (multiple 2-element structures)
573 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
574 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
575 (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
576 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr", "", []>;
577 class VST2Q<bits<4> op7_4, string Dt>
578 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
579 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
580 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
583 def VST2d8 : VST2D<0b1000, 0b0000, "8">;
584 def VST2d16 : VST2D<0b1000, 0b0100, "16">;
585 def VST2d32 : VST2D<0b1000, 0b1000, "32">;
587 def VST2q8 : VST2Q<0b0000, "8">;
588 def VST2q16 : VST2Q<0b0100, "16">;
589 def VST2q32 : VST2Q<0b1000, "32">;
591 // ...with address register writeback:
592 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
593 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
594 (ins addrmode6:$addr, am6offset:$offset, DPR:$src1, DPR:$src2),
595 IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr$offset",
596 "$addr.addr = $wb", []>;
597 class VST2QWB<bits<4> op7_4, string Dt>
598 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
599 (ins addrmode6:$addr, am6offset:$offset,
600 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
601 IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
602 "$addr.addr = $wb", []>;
604 def VST2d8_UPD : VST2DWB<0b1000, 0b0000, "8">;
605 def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
606 def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
608 def VST2q8_UPD : VST2QWB<0b0000, "8">;
609 def VST2q16_UPD : VST2QWB<0b0100, "16">;
610 def VST2q32_UPD : VST2QWB<0b1000, "32">;
612 // ...with double-spaced registers (for disassembly only):
613 def VST2b8 : VST2D<0b1001, 0b0000, "8">;
614 def VST2b16 : VST2D<0b1001, 0b0100, "16">;
615 def VST2b32 : VST2D<0b1001, 0b1000, "32">;
616 def VST2b8_UPD : VST2DWB<0b1001, 0b0000, "8">;
617 def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
618 def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
620 // VST3 : Vector Store (multiple 3-element structures)
621 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
622 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
623 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
624 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
626 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
627 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
628 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
630 // ...with address register writeback:
631 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
632 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
633 (ins addrmode6:$addr, am6offset:$offset,
634 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
635 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
636 "$addr.addr = $wb", []>;
638 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
639 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
640 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
642 // ...with double-spaced registers (non-updating versions for disassembly only):
643 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
644 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
645 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
646 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
647 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
648 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
650 // ...alternate versions to be allocated odd register numbers:
651 def VST3q8odd_UPD : VST3DWB<0b0101, 0b0000, "8">;
652 def VST3q16odd_UPD : VST3DWB<0b0101, 0b0100, "16">;
653 def VST3q32odd_UPD : VST3DWB<0b0101, 0b1000, "32">;
655 // VST4 : Vector Store (multiple 4-element structures)
656 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
657 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
658 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
659 IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
662 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
663 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
664 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
666 // ...with address register writeback:
667 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
668 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
669 (ins addrmode6:$addr, am6offset:$offset,
670 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
671 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
672 "$addr.addr = $wb", []>;
674 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
675 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
676 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
678 // ...with double-spaced registers (non-updating versions for disassembly only):
679 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
680 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
681 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
682 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
683 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
684 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
686 // ...alternate versions to be allocated odd register numbers:
687 def VST4q8odd_UPD : VST4DWB<0b0001, 0b0000, "8">;
688 def VST4q16odd_UPD : VST4DWB<0b0001, 0b0100, "16">;
689 def VST4q32odd_UPD : VST4DWB<0b0001, 0b1000, "32">;
691 // VST1LN : Vector Store (single element from one lane)
692 // FIXME: Not yet implemented.
694 // VST2LN : Vector Store (single 2-element structure from one lane)
695 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
696 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
697 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
698 IIC_VST, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
701 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
702 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
703 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
705 // ...with double-spaced registers:
706 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
707 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
709 // ...alternate versions to be allocated odd register numbers:
710 def VST2LNq16odd : VST2LN<0b0101, {?,?,1,?}, "16">;
711 def VST2LNq32odd : VST2LN<0b1001, {?,1,?,?}, "32">;
713 // ...with address register writeback:
714 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
715 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
716 (ins addrmode6:$addr, am6offset:$offset,
717 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST, "vst2", Dt,
718 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
719 "$addr.addr = $wb", []>;
721 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
722 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
723 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
725 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
726 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
728 // VST3LN : Vector Store (single 3-element structure from one lane)
729 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
730 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
731 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
732 nohash_imm:$lane), IIC_VST, "vst3", Dt,
733 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
735 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
736 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
737 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
739 // ...with double-spaced registers:
740 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
741 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
743 // ...alternate versions to be allocated odd register numbers:
744 def VST3LNq16odd : VST3LN<0b0110, {?,?,1,0}, "16">;
745 def VST3LNq32odd : VST3LN<0b1010, {?,1,0,0}, "32">;
747 // ...with address register writeback:
748 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
749 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
750 (ins addrmode6:$addr, am6offset:$offset,
751 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
753 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
754 "$addr.addr = $wb", []>;
756 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
757 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
758 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
760 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
761 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
763 // VST4LN : Vector Store (single 4-element structure from one lane)
764 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
765 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
766 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
767 nohash_imm:$lane), IIC_VST, "vst4", Dt,
768 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
771 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
772 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
773 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
775 // ...with double-spaced registers:
776 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
777 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
779 // ...alternate versions to be allocated odd register numbers:
780 def VST4LNq16odd : VST4LN<0b0111, {?,?,1,?}, "16">;
781 def VST4LNq32odd : VST4LN<0b1011, {?,1,?,?}, "32">;
783 // ...with address register writeback:
784 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
785 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
786 (ins addrmode6:$addr, am6offset:$offset,
787 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
789 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
790 "$addr.addr = $wb", []>;
792 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
793 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
794 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
796 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
797 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
799 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
802 //===----------------------------------------------------------------------===//
803 // NEON pattern fragments
804 //===----------------------------------------------------------------------===//
806 // Extract D sub-registers of Q registers.
807 def DSubReg_i8_reg : SDNodeXForm<imm, [{
808 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
809 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
811 def DSubReg_i16_reg : SDNodeXForm<imm, [{
812 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
813 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
815 def DSubReg_i32_reg : SDNodeXForm<imm, [{
816 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
817 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
819 def DSubReg_f64_reg : SDNodeXForm<imm, [{
820 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
821 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
824 // Extract S sub-registers of Q/D registers.
825 def SSubReg_f32_reg : SDNodeXForm<imm, [{
826 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
827 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
830 // Translate lane numbers from Q registers to D subregs.
831 def SubReg_i8_lane : SDNodeXForm<imm, [{
832 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
834 def SubReg_i16_lane : SDNodeXForm<imm, [{
835 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
837 def SubReg_i32_lane : SDNodeXForm<imm, [{
838 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
841 //===----------------------------------------------------------------------===//
842 // Instruction Classes
843 //===----------------------------------------------------------------------===//
845 // Basic 2-register operations: single-, double- and quad-register.
846 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
847 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
848 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
849 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
850 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
851 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
852 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
853 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
854 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
855 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
856 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
857 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
858 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
859 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
860 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
861 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
862 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
863 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
865 // Basic 2-register intrinsics, both double- and quad-register.
866 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
867 bits<2> op17_16, bits<5> op11_7, bit op4,
868 InstrItinClass itin, string OpcodeStr, string Dt,
869 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
870 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
871 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
872 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
873 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
874 bits<2> op17_16, bits<5> op11_7, bit op4,
875 InstrItinClass itin, string OpcodeStr, string Dt,
876 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
877 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
878 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
879 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
881 // Narrow 2-register intrinsics.
882 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
883 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
884 InstrItinClass itin, string OpcodeStr, string Dt,
885 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
886 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
887 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
888 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
890 // Long 2-register intrinsics (currently only used for VMOVL).
891 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
892 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
893 InstrItinClass itin, string OpcodeStr, string Dt,
894 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
895 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
896 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
897 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src))))]>;
899 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
900 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
901 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
902 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
903 OpcodeStr, Dt, "$dst1, $dst2",
904 "$src1 = $dst1, $src2 = $dst2", []>;
905 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
906 InstrItinClass itin, string OpcodeStr, string Dt>
907 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
908 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
909 "$src1 = $dst1, $src2 = $dst2", []>;
911 // Basic 3-register operations: single-, double- and quad-register.
912 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
913 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
914 SDNode OpNode, bit Commutable>
915 : N3V<op24, op23, op21_20, op11_8, 0, op4,
916 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
917 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
918 let isCommutable = Commutable;
921 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
922 InstrItinClass itin, string OpcodeStr, string Dt,
923 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
924 : N3V<op24, op23, op21_20, op11_8, 0, op4,
925 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
926 OpcodeStr, Dt, "$dst, $src1, $src2", "",
927 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
928 let isCommutable = Commutable;
930 // Same as N3VD but no data type.
931 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
932 InstrItinClass itin, string OpcodeStr,
933 ValueType ResTy, ValueType OpTy,
934 SDNode OpNode, bit Commutable>
935 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
936 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
937 OpcodeStr, "$dst, $src1, $src2", "",
938 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
939 let isCommutable = Commutable;
942 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
943 InstrItinClass itin, string OpcodeStr, string Dt,
944 ValueType Ty, SDNode ShOp>
945 : N3V<0, 1, op21_20, op11_8, 1, 0,
946 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
947 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
949 (Ty (ShOp (Ty DPR:$src1),
950 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
951 let isCommutable = 0;
953 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
954 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
955 : N3V<0, 1, op21_20, op11_8, 1, 0,
956 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
957 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
959 (Ty (ShOp (Ty DPR:$src1),
960 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
961 let isCommutable = 0;
964 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
965 InstrItinClass itin, string OpcodeStr, string Dt,
966 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
967 : N3V<op24, op23, op21_20, op11_8, 1, op4,
968 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
969 OpcodeStr, Dt, "$dst, $src1, $src2", "",
970 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
971 let isCommutable = Commutable;
973 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
974 InstrItinClass itin, string OpcodeStr,
975 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
976 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
977 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
978 OpcodeStr, "$dst, $src1, $src2", "",
979 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
980 let isCommutable = Commutable;
982 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
983 InstrItinClass itin, string OpcodeStr, string Dt,
984 ValueType ResTy, ValueType OpTy, SDNode ShOp>
985 : N3V<1, 1, op21_20, op11_8, 1, 0,
986 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
987 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
988 [(set (ResTy QPR:$dst),
989 (ResTy (ShOp (ResTy QPR:$src1),
990 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
992 let isCommutable = 0;
994 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
995 ValueType ResTy, ValueType OpTy, SDNode ShOp>
996 : N3V<1, 1, op21_20, op11_8, 1, 0,
997 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
998 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
999 [(set (ResTy QPR:$dst),
1000 (ResTy (ShOp (ResTy QPR:$src1),
1001 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1003 let isCommutable = 0;
1006 // Basic 3-register intrinsics, both double- and quad-register.
1007 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1008 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1009 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1010 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1011 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), f, itin,
1012 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1013 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1), (OpTy DPR:$src2))))]> {
1014 let isCommutable = Commutable;
1016 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1017 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1018 : N3V<0, 1, op21_20, op11_8, 1, 0,
1019 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1020 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1021 [(set (Ty DPR:$dst),
1022 (Ty (IntOp (Ty DPR:$src1),
1023 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1025 let isCommutable = 0;
1027 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1028 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1029 : N3V<0, 1, op21_20, op11_8, 1, 0,
1030 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1031 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1032 [(set (Ty DPR:$dst),
1033 (Ty (IntOp (Ty DPR:$src1),
1034 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1035 let isCommutable = 0;
1038 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1039 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1040 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1041 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1042 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), f, itin,
1043 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1044 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1), (OpTy QPR:$src2))))]> {
1045 let isCommutable = Commutable;
1047 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1048 string OpcodeStr, string Dt,
1049 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1050 : N3V<1, 1, op21_20, op11_8, 1, 0,
1051 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1052 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1053 [(set (ResTy QPR:$dst),
1054 (ResTy (IntOp (ResTy QPR:$src1),
1055 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1057 let isCommutable = 0;
1059 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1060 string OpcodeStr, string Dt,
1061 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1062 : N3V<1, 1, op21_20, op11_8, 1, 0,
1063 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1064 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1065 [(set (ResTy QPR:$dst),
1066 (ResTy (IntOp (ResTy QPR:$src1),
1067 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1069 let isCommutable = 0;
1072 // Multiply-Add/Sub operations: single-, double- and quad-register.
1073 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1074 InstrItinClass itin, string OpcodeStr, string Dt,
1075 ValueType Ty, SDNode MulOp, SDNode OpNode>
1076 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1077 (outs DPR_VFP2:$dst),
1078 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1079 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1081 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1082 InstrItinClass itin, string OpcodeStr, string Dt,
1083 ValueType Ty, SDNode MulOp, SDNode OpNode>
1084 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1085 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1086 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1087 [(set DPR:$dst, (Ty (OpNode DPR:$src1,
1088 (Ty (MulOp DPR:$src2, DPR:$src3)))))]>;
1089 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1090 string OpcodeStr, string Dt,
1091 ValueType Ty, SDNode MulOp, SDNode ShOp>
1092 : N3V<0, 1, op21_20, op11_8, 1, 0,
1094 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1096 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1097 [(set (Ty DPR:$dst),
1098 (Ty (ShOp (Ty DPR:$src1),
1099 (Ty (MulOp DPR:$src2,
1100 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1102 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1103 string OpcodeStr, string Dt,
1104 ValueType Ty, SDNode MulOp, SDNode ShOp>
1105 : N3V<0, 1, op21_20, op11_8, 1, 0,
1107 (ins DPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1109 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1110 [(set (Ty DPR:$dst),
1111 (Ty (ShOp (Ty DPR:$src1),
1112 (Ty (MulOp DPR:$src2,
1113 (Ty (NEONvduplane (Ty DPR_8:$src3),
1116 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1117 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1118 SDNode MulOp, SDNode OpNode>
1119 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1120 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1121 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1122 [(set QPR:$dst, (Ty (OpNode QPR:$src1,
1123 (Ty (MulOp QPR:$src2, QPR:$src3)))))]>;
1124 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1125 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1126 SDNode MulOp, SDNode ShOp>
1127 : N3V<1, 1, op21_20, op11_8, 1, 0,
1129 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1131 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1132 [(set (ResTy QPR:$dst),
1133 (ResTy (ShOp (ResTy QPR:$src1),
1134 (ResTy (MulOp QPR:$src2,
1135 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1137 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1138 string OpcodeStr, string Dt,
1139 ValueType ResTy, ValueType OpTy,
1140 SDNode MulOp, SDNode ShOp>
1141 : N3V<1, 1, op21_20, op11_8, 1, 0,
1143 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1145 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1146 [(set (ResTy QPR:$dst),
1147 (ResTy (ShOp (ResTy QPR:$src1),
1148 (ResTy (MulOp QPR:$src2,
1149 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1152 // Neon 3-argument intrinsics, both double- and quad-register.
1153 // The destination register is also used as the first source operand register.
1154 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1155 InstrItinClass itin, string OpcodeStr, string Dt,
1156 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1157 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1158 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1159 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1160 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1161 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1162 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1163 InstrItinClass itin, string OpcodeStr, string Dt,
1164 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1165 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1166 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1167 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1168 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1169 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1171 // Neon Long 3-argument intrinsic. The destination register is
1172 // a quad-register and is also used as the first source operand register.
1173 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1174 InstrItinClass itin, string OpcodeStr, string Dt,
1175 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1176 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1177 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1178 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1180 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2), (TyD DPR:$src3))))]>;
1181 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1182 string OpcodeStr, string Dt,
1183 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1184 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1186 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1188 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1189 [(set (ResTy QPR:$dst),
1190 (ResTy (IntOp (ResTy QPR:$src1),
1192 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1194 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1195 InstrItinClass itin, string OpcodeStr, string Dt,
1196 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1197 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1199 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1201 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1202 [(set (ResTy QPR:$dst),
1203 (ResTy (IntOp (ResTy QPR:$src1),
1205 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1208 // Narrowing 3-register intrinsics.
1209 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1210 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1211 Intrinsic IntOp, bit Commutable>
1212 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1213 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1214 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1215 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1216 let isCommutable = Commutable;
1219 // Long 3-register intrinsics.
1220 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1221 InstrItinClass itin, string OpcodeStr, string Dt,
1222 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1223 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1224 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1225 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1226 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1227 let isCommutable = Commutable;
1229 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1230 string OpcodeStr, string Dt,
1231 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1232 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1233 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1234 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1235 [(set (ResTy QPR:$dst),
1236 (ResTy (IntOp (OpTy DPR:$src1),
1237 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1239 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1240 InstrItinClass itin, string OpcodeStr, string Dt,
1241 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1242 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1243 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1244 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1245 [(set (ResTy QPR:$dst),
1246 (ResTy (IntOp (OpTy DPR:$src1),
1247 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1250 // Wide 3-register intrinsics.
1251 class N3VWInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1252 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1253 Intrinsic IntOp, bit Commutable>
1254 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1255 (outs QPR:$dst), (ins QPR:$src1, DPR:$src2), N3RegFrm, IIC_VSUBiD,
1256 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1257 [(set QPR:$dst, (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$src2))))]> {
1258 let isCommutable = Commutable;
1261 // Pairwise long 2-register intrinsics, both double- and quad-register.
1262 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1263 bits<2> op17_16, bits<5> op11_7, bit op4,
1264 string OpcodeStr, string Dt,
1265 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1266 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1267 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1268 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1269 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1270 bits<2> op17_16, bits<5> op11_7, bit op4,
1271 string OpcodeStr, string Dt,
1272 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1273 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1274 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1275 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1277 // Pairwise long 2-register accumulate intrinsics,
1278 // both double- and quad-register.
1279 // The destination register is also used as the first source operand register.
1280 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1281 bits<2> op17_16, bits<5> op11_7, bit op4,
1282 string OpcodeStr, string Dt,
1283 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1284 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1285 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), IIC_VPALiD,
1286 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1287 [(set DPR:$dst, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$src2))))]>;
1288 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1289 bits<2> op17_16, bits<5> op11_7, bit op4,
1290 string OpcodeStr, string Dt,
1291 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1292 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1293 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), IIC_VPALiQ,
1294 OpcodeStr, Dt, "$dst, $src2", "$src1 = $dst",
1295 [(set QPR:$dst, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$src2))))]>;
1297 // Shift by immediate,
1298 // both double- and quad-register.
1299 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1300 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1301 ValueType Ty, SDNode OpNode>
1302 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1303 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1304 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1305 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1306 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1307 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1308 ValueType Ty, SDNode OpNode>
1309 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1310 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1311 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1312 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1314 // Long shift by immediate.
1315 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1316 string OpcodeStr, string Dt,
1317 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1318 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1319 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1320 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1321 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1322 (i32 imm:$SIMM))))]>;
1324 // Narrow shift by immediate.
1325 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1326 InstrItinClass itin, string OpcodeStr, string Dt,
1327 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1328 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1329 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1330 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1331 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1332 (i32 imm:$SIMM))))]>;
1334 // Shift right by immediate and accumulate,
1335 // both double- and quad-register.
1336 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1337 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1338 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1339 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1340 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1341 [(set DPR:$dst, (Ty (add DPR:$src1,
1342 (Ty (ShOp DPR:$src2, (i32 imm:$SIMM))))))]>;
1343 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1344 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1345 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1346 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1347 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1348 [(set QPR:$dst, (Ty (add QPR:$src1,
1349 (Ty (ShOp QPR:$src2, (i32 imm:$SIMM))))))]>;
1351 // Shift by immediate and insert,
1352 // both double- and quad-register.
1353 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1354 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1355 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$dst),
1356 (ins DPR:$src1, DPR:$src2, i32imm:$SIMM), f, IIC_VSHLiD,
1357 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1358 [(set DPR:$dst, (Ty (ShOp DPR:$src1, DPR:$src2, (i32 imm:$SIMM))))]>;
1359 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1360 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1361 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$dst),
1362 (ins QPR:$src1, QPR:$src2, i32imm:$SIMM), f, IIC_VSHLiQ,
1363 OpcodeStr, Dt, "$dst, $src2, $SIMM", "$src1 = $dst",
1364 [(set QPR:$dst, (Ty (ShOp QPR:$src1, QPR:$src2, (i32 imm:$SIMM))))]>;
1366 // Convert, with fractional bits immediate,
1367 // both double- and quad-register.
1368 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1369 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1371 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1372 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), NVCVTFrm,
1373 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1374 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src), (i32 imm:$SIMM))))]>;
1375 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1376 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1378 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1379 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), NVCVTFrm,
1380 IIC_VUNAQ, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1381 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src), (i32 imm:$SIMM))))]>;
1383 //===----------------------------------------------------------------------===//
1385 //===----------------------------------------------------------------------===//
1387 // Abbreviations used in multiclass suffixes:
1388 // Q = quarter int (8 bit) elements
1389 // H = half int (16 bit) elements
1390 // S = single int (32 bit) elements
1391 // D = double int (64 bit) elements
1393 // Neon 2-register vector operations -- for disassembly only.
1395 // First with only element sizes of 8, 16 and 32 bits:
1396 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1397 bits<5> op11_7, bit op4, string opc, string Dt,
1399 // 64-bit vector types.
1400 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
1401 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1402 opc, !strconcat(Dt, "8"), asm, "", []>;
1403 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
1404 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1405 opc, !strconcat(Dt, "16"), asm, "", []>;
1406 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1407 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1408 opc, !strconcat(Dt, "32"), asm, "", []>;
1409 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
1410 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
1411 opc, "f32", asm, "", []> {
1412 let Inst{10} = 1; // overwrite F = 1
1415 // 128-bit vector types.
1416 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
1417 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1418 opc, !strconcat(Dt, "8"), asm, "", []>;
1419 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
1420 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1421 opc, !strconcat(Dt, "16"), asm, "", []>;
1422 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1423 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1424 opc, !strconcat(Dt, "32"), asm, "", []>;
1425 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
1426 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
1427 opc, "f32", asm, "", []> {
1428 let Inst{10} = 1; // overwrite F = 1
1432 // Neon 3-register vector operations.
1434 // First with only element sizes of 8, 16 and 32 bits:
1435 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1436 InstrItinClass itinD16, InstrItinClass itinD32,
1437 InstrItinClass itinQ16, InstrItinClass itinQ32,
1438 string OpcodeStr, string Dt,
1439 SDNode OpNode, bit Commutable = 0> {
1440 // 64-bit vector types.
1441 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
1442 OpcodeStr, !strconcat(Dt, "8"),
1443 v8i8, v8i8, OpNode, Commutable>;
1444 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
1445 OpcodeStr, !strconcat(Dt, "16"),
1446 v4i16, v4i16, OpNode, Commutable>;
1447 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
1448 OpcodeStr, !strconcat(Dt, "32"),
1449 v2i32, v2i32, OpNode, Commutable>;
1451 // 128-bit vector types.
1452 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
1453 OpcodeStr, !strconcat(Dt, "8"),
1454 v16i8, v16i8, OpNode, Commutable>;
1455 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
1456 OpcodeStr, !strconcat(Dt, "16"),
1457 v8i16, v8i16, OpNode, Commutable>;
1458 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
1459 OpcodeStr, !strconcat(Dt, "32"),
1460 v4i32, v4i32, OpNode, Commutable>;
1463 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
1464 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1466 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
1468 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
1469 v8i16, v4i16, ShOp>;
1470 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
1471 v4i32, v2i32, ShOp>;
1474 // ....then also with element size 64 bits:
1475 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1476 InstrItinClass itinD, InstrItinClass itinQ,
1477 string OpcodeStr, string Dt,
1478 SDNode OpNode, bit Commutable = 0>
1479 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
1480 OpcodeStr, Dt, OpNode, Commutable> {
1481 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
1482 OpcodeStr, !strconcat(Dt, "64"),
1483 v1i64, v1i64, OpNode, Commutable>;
1484 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
1485 OpcodeStr, !strconcat(Dt, "64"),
1486 v2i64, v2i64, OpNode, Commutable>;
1490 // Neon Narrowing 2-register vector intrinsics,
1491 // source operand element sizes of 16, 32 and 64 bits:
1492 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1493 bits<5> op11_7, bit op6, bit op4,
1494 InstrItinClass itin, string OpcodeStr, string Dt,
1496 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
1497 itin, OpcodeStr, !strconcat(Dt, "16"),
1498 v8i8, v8i16, IntOp>;
1499 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
1500 itin, OpcodeStr, !strconcat(Dt, "32"),
1501 v4i16, v4i32, IntOp>;
1502 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
1503 itin, OpcodeStr, !strconcat(Dt, "64"),
1504 v2i32, v2i64, IntOp>;
1508 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
1509 // source operand element sizes of 16, 32 and 64 bits:
1510 multiclass N2VLInt_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
1511 string OpcodeStr, string Dt, Intrinsic IntOp> {
1512 def v8i16 : N2VLInt<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1513 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1514 def v4i32 : N2VLInt<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1515 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1516 def v2i64 : N2VLInt<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
1517 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1521 // Neon 3-register vector intrinsics.
1523 // First with only element sizes of 16 and 32 bits:
1524 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1525 InstrItinClass itinD16, InstrItinClass itinD32,
1526 InstrItinClass itinQ16, InstrItinClass itinQ32,
1527 string OpcodeStr, string Dt,
1528 Intrinsic IntOp, bit Commutable = 0> {
1529 // 64-bit vector types.
1530 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
1531 OpcodeStr, !strconcat(Dt, "16"),
1532 v4i16, v4i16, IntOp, Commutable>;
1533 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
1534 OpcodeStr, !strconcat(Dt, "32"),
1535 v2i32, v2i32, IntOp, Commutable>;
1537 // 128-bit vector types.
1538 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
1539 OpcodeStr, !strconcat(Dt, "16"),
1540 v8i16, v8i16, IntOp, Commutable>;
1541 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
1542 OpcodeStr, !strconcat(Dt, "32"),
1543 v4i32, v4i32, IntOp, Commutable>;
1546 multiclass N3VIntSL_HS<bits<4> op11_8,
1547 InstrItinClass itinD16, InstrItinClass itinD32,
1548 InstrItinClass itinQ16, InstrItinClass itinQ32,
1549 string OpcodeStr, string Dt, Intrinsic IntOp> {
1550 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
1551 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
1552 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
1553 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
1554 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
1555 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
1556 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
1557 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
1560 // ....then also with element size of 8 bits:
1561 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1562 InstrItinClass itinD16, InstrItinClass itinD32,
1563 InstrItinClass itinQ16, InstrItinClass itinQ32,
1564 string OpcodeStr, string Dt,
1565 Intrinsic IntOp, bit Commutable = 0>
1566 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1567 OpcodeStr, Dt, IntOp, Commutable> {
1568 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
1569 OpcodeStr, !strconcat(Dt, "8"),
1570 v8i8, v8i8, IntOp, Commutable>;
1571 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
1572 OpcodeStr, !strconcat(Dt, "8"),
1573 v16i8, v16i8, IntOp, Commutable>;
1576 // ....then also with element size of 64 bits:
1577 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
1578 InstrItinClass itinD16, InstrItinClass itinD32,
1579 InstrItinClass itinQ16, InstrItinClass itinQ32,
1580 string OpcodeStr, string Dt,
1581 Intrinsic IntOp, bit Commutable = 0>
1582 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
1583 OpcodeStr, Dt, IntOp, Commutable> {
1584 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
1585 OpcodeStr, !strconcat(Dt, "64"),
1586 v1i64, v1i64, IntOp, Commutable>;
1587 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
1588 OpcodeStr, !strconcat(Dt, "64"),
1589 v2i64, v2i64, IntOp, Commutable>;
1592 // Neon Narrowing 3-register vector intrinsics,
1593 // source operand element sizes of 16, 32 and 64 bits:
1594 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1595 string OpcodeStr, string Dt,
1596 Intrinsic IntOp, bit Commutable = 0> {
1597 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
1598 OpcodeStr, !strconcat(Dt, "16"),
1599 v8i8, v8i16, IntOp, Commutable>;
1600 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
1601 OpcodeStr, !strconcat(Dt, "32"),
1602 v4i16, v4i32, IntOp, Commutable>;
1603 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
1604 OpcodeStr, !strconcat(Dt, "64"),
1605 v2i32, v2i64, IntOp, Commutable>;
1609 // Neon Long 3-register vector intrinsics.
1611 // First with only element sizes of 16 and 32 bits:
1612 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1613 InstrItinClass itin16, InstrItinClass itin32,
1614 string OpcodeStr, string Dt,
1615 Intrinsic IntOp, bit Commutable = 0> {
1616 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
1617 OpcodeStr, !strconcat(Dt, "16"),
1618 v4i32, v4i16, IntOp, Commutable>;
1619 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
1620 OpcodeStr, !strconcat(Dt, "32"),
1621 v2i64, v2i32, IntOp, Commutable>;
1624 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
1625 InstrItinClass itin, string OpcodeStr, string Dt,
1627 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
1628 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1629 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
1630 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1633 // ....then also with element size of 8 bits:
1634 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1635 InstrItinClass itin16, InstrItinClass itin32,
1636 string OpcodeStr, string Dt,
1637 Intrinsic IntOp, bit Commutable = 0>
1638 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
1639 IntOp, Commutable> {
1640 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
1641 OpcodeStr, !strconcat(Dt, "8"),
1642 v8i16, v8i8, IntOp, Commutable>;
1646 // Neon Wide 3-register vector intrinsics,
1647 // source operand element sizes of 8, 16 and 32 bits:
1648 multiclass N3VWInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1649 string OpcodeStr, string Dt,
1650 Intrinsic IntOp, bit Commutable = 0> {
1651 def v8i16 : N3VWInt<op24, op23, 0b00, op11_8, op4,
1652 OpcodeStr, !strconcat(Dt, "8"),
1653 v8i16, v8i8, IntOp, Commutable>;
1654 def v4i32 : N3VWInt<op24, op23, 0b01, op11_8, op4,
1655 OpcodeStr, !strconcat(Dt, "16"),
1656 v4i32, v4i16, IntOp, Commutable>;
1657 def v2i64 : N3VWInt<op24, op23, 0b10, op11_8, op4,
1658 OpcodeStr, !strconcat(Dt, "32"),
1659 v2i64, v2i32, IntOp, Commutable>;
1663 // Neon Multiply-Op vector operations,
1664 // element sizes of 8, 16 and 32 bits:
1665 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1666 InstrItinClass itinD16, InstrItinClass itinD32,
1667 InstrItinClass itinQ16, InstrItinClass itinQ32,
1668 string OpcodeStr, string Dt, SDNode OpNode> {
1669 // 64-bit vector types.
1670 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
1671 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
1672 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
1673 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
1674 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
1675 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
1677 // 128-bit vector types.
1678 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
1679 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
1680 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
1681 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
1682 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
1683 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
1686 multiclass N3VMulOpSL_HS<bits<4> op11_8,
1687 InstrItinClass itinD16, InstrItinClass itinD32,
1688 InstrItinClass itinQ16, InstrItinClass itinQ32,
1689 string OpcodeStr, string Dt, SDNode ShOp> {
1690 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
1691 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
1692 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
1693 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
1694 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
1695 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
1697 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
1698 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
1702 // Neon 3-argument intrinsics,
1703 // element sizes of 8, 16 and 32 bits:
1704 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1705 InstrItinClass itinD, InstrItinClass itinQ,
1706 string OpcodeStr, string Dt, Intrinsic IntOp> {
1707 // 64-bit vector types.
1708 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
1709 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1710 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
1711 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
1712 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
1713 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
1715 // 128-bit vector types.
1716 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
1717 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
1718 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
1719 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
1720 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
1721 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
1725 // Neon Long 3-argument intrinsics.
1727 // First with only element sizes of 16 and 32 bits:
1728 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
1729 InstrItinClass itin16, InstrItinClass itin32,
1730 string OpcodeStr, string Dt, Intrinsic IntOp> {
1731 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
1732 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
1733 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
1734 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1737 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
1738 string OpcodeStr, string Dt, Intrinsic IntOp> {
1739 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
1740 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
1741 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
1742 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
1745 // ....then also with element size of 8 bits:
1746 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
1747 InstrItinClass itin16, InstrItinClass itin32,
1748 string OpcodeStr, string Dt, Intrinsic IntOp>
1749 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
1750 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
1751 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
1755 // Neon 2-register vector intrinsics,
1756 // element sizes of 8, 16 and 32 bits:
1757 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1758 bits<5> op11_7, bit op4,
1759 InstrItinClass itinD, InstrItinClass itinQ,
1760 string OpcodeStr, string Dt, Intrinsic IntOp> {
1761 // 64-bit vector types.
1762 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1763 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
1764 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1765 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
1766 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1767 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
1769 // 128-bit vector types.
1770 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1771 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
1772 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1773 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
1774 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1775 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
1779 // Neon Pairwise long 2-register intrinsics,
1780 // element sizes of 8, 16 and 32 bits:
1781 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1782 bits<5> op11_7, bit op4,
1783 string OpcodeStr, string Dt, Intrinsic IntOp> {
1784 // 64-bit vector types.
1785 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1786 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1787 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1788 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1789 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1790 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1792 // 128-bit vector types.
1793 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1794 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1795 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1796 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1797 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1798 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1802 // Neon Pairwise long 2-register accumulate intrinsics,
1803 // element sizes of 8, 16 and 32 bits:
1804 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1805 bits<5> op11_7, bit op4,
1806 string OpcodeStr, string Dt, Intrinsic IntOp> {
1807 // 64-bit vector types.
1808 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1809 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
1810 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1811 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
1812 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1813 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
1815 // 128-bit vector types.
1816 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
1817 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
1818 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
1819 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
1820 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
1821 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
1825 // Neon 2-register vector shift by immediate,
1826 // with f of either N2RegVShLFrm or N2RegVShRFrm
1827 // element sizes of 8, 16, 32 and 64 bits:
1828 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1829 InstrItinClass itin, string OpcodeStr, string Dt,
1830 SDNode OpNode, Format f> {
1831 // 64-bit vector types.
1832 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1833 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
1834 let Inst{21-19} = 0b001; // imm6 = 001xxx
1836 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1837 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
1838 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1840 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
1841 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
1842 let Inst{21} = 0b1; // imm6 = 1xxxxx
1844 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
1845 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
1848 // 128-bit vector types.
1849 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1850 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
1851 let Inst{21-19} = 0b001; // imm6 = 001xxx
1853 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1854 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
1855 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1857 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
1858 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
1859 let Inst{21} = 0b1; // imm6 = 1xxxxx
1861 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
1862 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
1866 // Neon Shift-Accumulate vector operations,
1867 // element sizes of 8, 16, 32 and 64 bits:
1868 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1869 string OpcodeStr, string Dt, SDNode ShOp> {
1870 // 64-bit vector types.
1871 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1872 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
1873 let Inst{21-19} = 0b001; // imm6 = 001xxx
1875 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1876 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
1877 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1879 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
1880 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
1881 let Inst{21} = 0b1; // imm6 = 1xxxxx
1883 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
1884 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
1887 // 128-bit vector types.
1888 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1889 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
1890 let Inst{21-19} = 0b001; // imm6 = 001xxx
1892 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1893 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
1894 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1896 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
1897 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
1898 let Inst{21} = 0b1; // imm6 = 1xxxxx
1900 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
1901 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
1906 // Neon Shift-Insert vector operations,
1907 // with f of either N2RegVShLFrm or N2RegVShRFrm
1908 // element sizes of 8, 16, 32 and 64 bits:
1909 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
1910 string OpcodeStr, SDNode ShOp,
1912 // 64-bit vector types.
1913 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
1914 f, OpcodeStr, "8", v8i8, ShOp> {
1915 let Inst{21-19} = 0b001; // imm6 = 001xxx
1917 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
1918 f, OpcodeStr, "16", v4i16, ShOp> {
1919 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1921 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
1922 f, OpcodeStr, "32", v2i32, ShOp> {
1923 let Inst{21} = 0b1; // imm6 = 1xxxxx
1925 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
1926 f, OpcodeStr, "64", v1i64, ShOp>;
1929 // 128-bit vector types.
1930 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
1931 f, OpcodeStr, "8", v16i8, ShOp> {
1932 let Inst{21-19} = 0b001; // imm6 = 001xxx
1934 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
1935 f, OpcodeStr, "16", v8i16, ShOp> {
1936 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1938 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
1939 f, OpcodeStr, "32", v4i32, ShOp> {
1940 let Inst{21} = 0b1; // imm6 = 1xxxxx
1942 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
1943 f, OpcodeStr, "64", v2i64, ShOp>;
1947 // Neon Shift Long operations,
1948 // element sizes of 8, 16, 32 bits:
1949 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1950 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
1951 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1952 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
1953 let Inst{21-19} = 0b001; // imm6 = 001xxx
1955 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1956 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
1957 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1959 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
1960 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
1961 let Inst{21} = 0b1; // imm6 = 1xxxxx
1965 // Neon Shift Narrow operations,
1966 // element sizes of 16, 32, 64 bits:
1967 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
1968 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
1970 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1971 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
1972 let Inst{21-19} = 0b001; // imm6 = 001xxx
1974 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1975 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
1976 let Inst{21-20} = 0b01; // imm6 = 01xxxx
1978 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
1979 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
1980 let Inst{21} = 0b1; // imm6 = 1xxxxx
1984 //===----------------------------------------------------------------------===//
1985 // Instruction Definitions.
1986 //===----------------------------------------------------------------------===//
1988 // Vector Add Operations.
1990 // VADD : Vector Add (integer and floating-point)
1991 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
1993 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
1994 v2f32, v2f32, fadd, 1>;
1995 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
1996 v4f32, v4f32, fadd, 1>;
1997 // VADDL : Vector Add Long (Q = D + D)
1998 defm VADDLs : N3VLInt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
1999 "vaddl", "s", int_arm_neon_vaddls, 1>;
2000 defm VADDLu : N3VLInt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2001 "vaddl", "u", int_arm_neon_vaddlu, 1>;
2002 // VADDW : Vector Add Wide (Q = Q + D)
2003 defm VADDWs : N3VWInt_QHS<0,1,0b0001,0, "vaddw", "s", int_arm_neon_vaddws, 0>;
2004 defm VADDWu : N3VWInt_QHS<1,1,0b0001,0, "vaddw", "u", int_arm_neon_vaddwu, 0>;
2005 // VHADD : Vector Halving Add
2006 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2007 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2008 "vhadd", "s", int_arm_neon_vhadds, 1>;
2009 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2010 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2011 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2012 // VRHADD : Vector Rounding Halving Add
2013 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2014 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2015 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2016 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2017 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2018 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2019 // VQADD : Vector Saturating Add
2020 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2021 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2022 "vqadd", "s", int_arm_neon_vqadds, 1>;
2023 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2024 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2025 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2026 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2027 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2028 int_arm_neon_vaddhn, 1>;
2029 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2030 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2031 int_arm_neon_vraddhn, 1>;
2033 // Vector Multiply Operations.
2035 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2036 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2037 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2038 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2039 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2040 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2041 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2042 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VBIND, "vmul", "f32",
2043 v2f32, v2f32, fmul, 1>;
2044 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VBINQ, "vmul", "f32",
2045 v4f32, v4f32, fmul, 1>;
2046 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2047 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2048 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2051 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2052 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2053 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2054 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2055 (DSubReg_i16_reg imm:$lane))),
2056 (SubReg_i16_lane imm:$lane)))>;
2057 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2058 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2059 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2060 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2061 (DSubReg_i32_reg imm:$lane))),
2062 (SubReg_i32_lane imm:$lane)))>;
2063 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2064 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2065 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2066 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2067 (DSubReg_i32_reg imm:$lane))),
2068 (SubReg_i32_lane imm:$lane)))>;
2070 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2071 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2072 IIC_VMULi16Q, IIC_VMULi32Q,
2073 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2074 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2075 IIC_VMULi16Q, IIC_VMULi32Q,
2076 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2077 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2078 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2080 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2081 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2082 (DSubReg_i16_reg imm:$lane))),
2083 (SubReg_i16_lane imm:$lane)))>;
2084 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2085 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2087 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2088 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2089 (DSubReg_i32_reg imm:$lane))),
2090 (SubReg_i32_lane imm:$lane)))>;
2092 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2093 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2094 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2095 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2096 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2097 IIC_VMULi16Q, IIC_VMULi32Q,
2098 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2099 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2100 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2102 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2103 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2104 (DSubReg_i16_reg imm:$lane))),
2105 (SubReg_i16_lane imm:$lane)))>;
2106 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2107 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2109 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2110 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2111 (DSubReg_i32_reg imm:$lane))),
2112 (SubReg_i32_lane imm:$lane)))>;
2114 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2115 defm VMULLs : N3VLInt_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2116 "vmull", "s", int_arm_neon_vmulls, 1>;
2117 defm VMULLu : N3VLInt_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2118 "vmull", "u", int_arm_neon_vmullu, 1>;
2119 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2120 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2121 defm VMULLsls : N3VLIntSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s",
2122 int_arm_neon_vmulls>;
2123 defm VMULLslu : N3VLIntSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u",
2124 int_arm_neon_vmullu>;
2126 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2127 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2128 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2129 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2130 "vqdmull", "s", int_arm_neon_vqdmull>;
2132 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2134 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2135 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2136 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2137 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2139 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2141 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2142 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2143 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2145 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2146 v4f32, v2f32, fmul, fadd>;
2148 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2149 (mul (v8i16 QPR:$src2),
2150 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2151 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2152 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2153 (DSubReg_i16_reg imm:$lane))),
2154 (SubReg_i16_lane imm:$lane)))>;
2156 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2157 (mul (v4i32 QPR:$src2),
2158 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2159 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2160 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2161 (DSubReg_i32_reg imm:$lane))),
2162 (SubReg_i32_lane imm:$lane)))>;
2164 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2165 (fmul (v4f32 QPR:$src2),
2166 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2167 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2169 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2170 (DSubReg_i32_reg imm:$lane))),
2171 (SubReg_i32_lane imm:$lane)))>;
2173 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2174 defm VMLALs : N3VLInt3_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2175 "vmlal", "s", int_arm_neon_vmlals>;
2176 defm VMLALu : N3VLInt3_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2177 "vmlal", "u", int_arm_neon_vmlalu>;
2179 defm VMLALsls : N3VLInt3SL_HS<0, 0b0010, "vmlal", "s", int_arm_neon_vmlals>;
2180 defm VMLALslu : N3VLInt3SL_HS<1, 0b0010, "vmlal", "u", int_arm_neon_vmlalu>;
2182 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2183 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2184 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2185 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2187 // VMLS : Vector Multiply Subtract (integer and floating-point)
2188 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2189 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2190 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2192 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2194 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2195 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2196 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2198 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2199 v4f32, v2f32, fmul, fsub>;
2201 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2202 (mul (v8i16 QPR:$src2),
2203 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2204 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2205 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2206 (DSubReg_i16_reg imm:$lane))),
2207 (SubReg_i16_lane imm:$lane)))>;
2209 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2210 (mul (v4i32 QPR:$src2),
2211 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2212 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2213 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2214 (DSubReg_i32_reg imm:$lane))),
2215 (SubReg_i32_lane imm:$lane)))>;
2217 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
2218 (fmul (v4f32 QPR:$src2),
2219 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2220 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
2221 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2222 (DSubReg_i32_reg imm:$lane))),
2223 (SubReg_i32_lane imm:$lane)))>;
2225 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
2226 defm VMLSLs : N3VLInt3_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2227 "vmlsl", "s", int_arm_neon_vmlsls>;
2228 defm VMLSLu : N3VLInt3_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
2229 "vmlsl", "u", int_arm_neon_vmlslu>;
2231 defm VMLSLsls : N3VLInt3SL_HS<0, 0b0110, "vmlsl", "s", int_arm_neon_vmlsls>;
2232 defm VMLSLslu : N3VLInt3SL_HS<1, 0b0110, "vmlsl", "u", int_arm_neon_vmlslu>;
2234 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
2235 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
2236 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2237 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
2239 // Vector Subtract Operations.
2241 // VSUB : Vector Subtract (integer and floating-point)
2242 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
2243 "vsub", "i", sub, 0>;
2244 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
2245 v2f32, v2f32, fsub, 0>;
2246 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
2247 v4f32, v4f32, fsub, 0>;
2248 // VSUBL : Vector Subtract Long (Q = D - D)
2249 defm VSUBLs : N3VLInt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2250 "vsubl", "s", int_arm_neon_vsubls, 1>;
2251 defm VSUBLu : N3VLInt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
2252 "vsubl", "u", int_arm_neon_vsublu, 1>;
2253 // VSUBW : Vector Subtract Wide (Q = Q - D)
2254 defm VSUBWs : N3VWInt_QHS<0,1,0b0011,0, "vsubw", "s", int_arm_neon_vsubws, 0>;
2255 defm VSUBWu : N3VWInt_QHS<1,1,0b0011,0, "vsubw", "u", int_arm_neon_vsubwu, 0>;
2256 // VHSUB : Vector Halving Subtract
2257 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
2258 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2259 "vhsub", "s", int_arm_neon_vhsubs, 0>;
2260 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
2261 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2262 "vhsub", "u", int_arm_neon_vhsubu, 0>;
2263 // VQSUB : Vector Saturing Subtract
2264 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
2265 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2266 "vqsub", "s", int_arm_neon_vqsubs, 0>;
2267 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
2268 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2269 "vqsub", "u", int_arm_neon_vqsubu, 0>;
2270 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
2271 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
2272 int_arm_neon_vsubhn, 0>;
2273 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
2274 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
2275 int_arm_neon_vrsubhn, 0>;
2277 // Vector Comparisons.
2279 // VCEQ : Vector Compare Equal
2280 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2281 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
2282 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
2284 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
2286 // For disassembly only.
2287 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
2290 // VCGE : Vector Compare Greater Than or Equal
2291 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2292 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
2293 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2294 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
2295 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
2297 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
2299 // For disassembly only.
2300 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
2302 // For disassembly only.
2303 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
2306 // VCGT : Vector Compare Greater Than
2307 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2308 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
2309 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
2310 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
2311 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
2313 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
2315 // For disassembly only.
2316 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
2318 // For disassembly only.
2319 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
2322 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
2323 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
2324 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
2325 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
2326 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
2327 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
2328 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
2329 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
2330 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
2331 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
2332 // VTST : Vector Test Bits
2333 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
2334 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
2336 // Vector Bitwise Operations.
2338 def vnotd : PatFrag<(ops node:$in),
2339 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
2340 def vnotq : PatFrag<(ops node:$in),
2341 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
2344 // VAND : Vector Bitwise AND
2345 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
2346 v2i32, v2i32, and, 1>;
2347 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
2348 v4i32, v4i32, and, 1>;
2350 // VEOR : Vector Bitwise Exclusive OR
2351 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
2352 v2i32, v2i32, xor, 1>;
2353 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
2354 v4i32, v4i32, xor, 1>;
2356 // VORR : Vector Bitwise OR
2357 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
2358 v2i32, v2i32, or, 1>;
2359 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
2360 v4i32, v4i32, or, 1>;
2362 // VBIC : Vector Bitwise Bit Clear (AND NOT)
2363 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2364 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2365 "vbic", "$dst, $src1, $src2", "",
2366 [(set DPR:$dst, (v2i32 (and DPR:$src1,
2367 (vnotd DPR:$src2))))]>;
2368 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2369 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2370 "vbic", "$dst, $src1, $src2", "",
2371 [(set QPR:$dst, (v4i32 (and QPR:$src1,
2372 (vnotq QPR:$src2))))]>;
2374 // VORN : Vector Bitwise OR NOT
2375 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
2376 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
2377 "vorn", "$dst, $src1, $src2", "",
2378 [(set DPR:$dst, (v2i32 (or DPR:$src1,
2379 (vnotd DPR:$src2))))]>;
2380 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
2381 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
2382 "vorn", "$dst, $src1, $src2", "",
2383 [(set QPR:$dst, (v4i32 (or QPR:$src1,
2384 (vnotq QPR:$src2))))]>;
2386 // VMVN : Vector Bitwise NOT
2387 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
2388 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
2389 "vmvn", "$dst, $src", "",
2390 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
2391 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
2392 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
2393 "vmvn", "$dst, $src", "",
2394 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
2395 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
2396 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
2398 // VBSL : Vector Bitwise Select
2399 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
2400 (ins DPR:$src1, DPR:$src2, DPR:$src3),
2401 N3RegFrm, IIC_VCNTiD,
2402 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2404 (v2i32 (or (and DPR:$src2, DPR:$src1),
2405 (and DPR:$src3, (vnotd DPR:$src1)))))]>;
2406 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
2407 (ins QPR:$src1, QPR:$src2, QPR:$src3),
2408 N3RegFrm, IIC_VCNTiQ,
2409 "vbsl", "$dst, $src2, $src3", "$src1 = $dst",
2411 (v4i32 (or (and QPR:$src2, QPR:$src1),
2412 (and QPR:$src3, (vnotq QPR:$src1)))))]>;
2414 // VBIF : Vector Bitwise Insert if False
2415 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
2416 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
2417 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2418 N3RegFrm, IIC_VBINiD,
2419 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2420 [/* For disassembly only; pattern left blank */]>;
2421 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
2422 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2423 N3RegFrm, IIC_VBINiQ,
2424 "vbif", "$dst, $src2, $src3", "$src1 = $dst",
2425 [/* For disassembly only; pattern left blank */]>;
2427 // VBIT : Vector Bitwise Insert if True
2428 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
2429 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
2430 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3),
2431 N3RegFrm, IIC_VBINiD,
2432 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2433 [/* For disassembly only; pattern left blank */]>;
2434 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
2435 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3),
2436 N3RegFrm, IIC_VBINiQ,
2437 "vbit", "$dst, $src2, $src3", "$src1 = $dst",
2438 [/* For disassembly only; pattern left blank */]>;
2440 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
2441 // for equivalent operations with different register constraints; it just
2444 // Vector Absolute Differences.
2446 // VABD : Vector Absolute Difference
2447 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
2448 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2449 "vabd", "s", int_arm_neon_vabds, 0>;
2450 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
2451 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2452 "vabd", "u", int_arm_neon_vabdu, 0>;
2453 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
2454 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 0>;
2455 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
2456 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 0>;
2458 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
2459 defm VABDLs : N3VLInt_QHS<0,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2460 "vabdl", "s", int_arm_neon_vabdls, 0>;
2461 defm VABDLu : N3VLInt_QHS<1,1,0b0111,0, IIC_VSUBi4Q, IIC_VSUBi4Q,
2462 "vabdl", "u", int_arm_neon_vabdlu, 0>;
2464 // VABA : Vector Absolute Difference and Accumulate
2465 defm VABAs : N3VInt3_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2466 "vaba", "s", int_arm_neon_vabas>;
2467 defm VABAu : N3VInt3_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
2468 "vaba", "u", int_arm_neon_vabau>;
2470 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
2471 defm VABALs : N3VLInt3_QHS<0,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2472 "vabal", "s", int_arm_neon_vabals>;
2473 defm VABALu : N3VLInt3_QHS<1,1,0b0101,0, IIC_VABAD, IIC_VABAD,
2474 "vabal", "u", int_arm_neon_vabalu>;
2476 // Vector Maximum and Minimum.
2478 // VMAX : Vector Maximum
2479 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
2480 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2481 "vmax", "s", int_arm_neon_vmaxs, 1>;
2482 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
2483 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2484 "vmax", "u", int_arm_neon_vmaxu, 1>;
2485 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
2487 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
2488 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2490 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
2492 // VMIN : Vector Minimum
2493 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
2494 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2495 "vmin", "s", int_arm_neon_vmins, 1>;
2496 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
2497 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
2498 "vmin", "u", int_arm_neon_vminu, 1>;
2499 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
2501 v2f32, v2f32, int_arm_neon_vmins, 1>;
2502 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
2504 v4f32, v4f32, int_arm_neon_vmins, 1>;
2506 // Vector Pairwise Operations.
2508 // VPADD : Vector Pairwise Add
2509 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2511 v8i8, v8i8, int_arm_neon_vpadd, 0>;
2512 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2514 v4i16, v4i16, int_arm_neon_vpadd, 0>;
2515 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
2517 v2i32, v2i32, int_arm_neon_vpadd, 0>;
2518 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
2519 IIC_VBIND, "vpadd", "f32",
2520 v2f32, v2f32, int_arm_neon_vpadd, 0>;
2522 // VPADDL : Vector Pairwise Add Long
2523 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
2524 int_arm_neon_vpaddls>;
2525 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
2526 int_arm_neon_vpaddlu>;
2528 // VPADAL : Vector Pairwise Add and Accumulate Long
2529 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
2530 int_arm_neon_vpadals>;
2531 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
2532 int_arm_neon_vpadalu>;
2534 // VPMAX : Vector Pairwise Maximum
2535 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2536 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
2537 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2538 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
2539 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2540 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
2541 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2542 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
2543 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2544 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
2545 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2546 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
2547 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
2548 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
2550 // VPMIN : Vector Pairwise Minimum
2551 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2552 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
2553 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2554 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
2555 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2556 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
2557 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2558 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
2559 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2560 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
2561 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
2562 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
2563 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VSUBi4D, "vpmin",
2564 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
2566 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
2568 // VRECPE : Vector Reciprocal Estimate
2569 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2570 IIC_VUNAD, "vrecpe", "u32",
2571 v2i32, v2i32, int_arm_neon_vrecpe>;
2572 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
2573 IIC_VUNAQ, "vrecpe", "u32",
2574 v4i32, v4i32, int_arm_neon_vrecpe>;
2575 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2576 IIC_VUNAD, "vrecpe", "f32",
2577 v2f32, v2f32, int_arm_neon_vrecpe>;
2578 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
2579 IIC_VUNAQ, "vrecpe", "f32",
2580 v4f32, v4f32, int_arm_neon_vrecpe>;
2582 // VRECPS : Vector Reciprocal Step
2583 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2584 IIC_VRECSD, "vrecps", "f32",
2585 v2f32, v2f32, int_arm_neon_vrecps, 1>;
2586 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
2587 IIC_VRECSQ, "vrecps", "f32",
2588 v4f32, v4f32, int_arm_neon_vrecps, 1>;
2590 // VRSQRTE : Vector Reciprocal Square Root Estimate
2591 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2592 IIC_VUNAD, "vrsqrte", "u32",
2593 v2i32, v2i32, int_arm_neon_vrsqrte>;
2594 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
2595 IIC_VUNAQ, "vrsqrte", "u32",
2596 v4i32, v4i32, int_arm_neon_vrsqrte>;
2597 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2598 IIC_VUNAD, "vrsqrte", "f32",
2599 v2f32, v2f32, int_arm_neon_vrsqrte>;
2600 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
2601 IIC_VUNAQ, "vrsqrte", "f32",
2602 v4f32, v4f32, int_arm_neon_vrsqrte>;
2604 // VRSQRTS : Vector Reciprocal Square Root Step
2605 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2606 IIC_VRECSD, "vrsqrts", "f32",
2607 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
2608 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
2609 IIC_VRECSQ, "vrsqrts", "f32",
2610 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
2614 // VSHL : Vector Shift
2615 defm VSHLs : N3VInt_QHSD<0, 0, 0b0100, 0, N3RegVShFrm,
2616 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2617 "vshl", "s", int_arm_neon_vshifts, 0>;
2618 defm VSHLu : N3VInt_QHSD<1, 0, 0b0100, 0, N3RegVShFrm,
2619 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
2620 "vshl", "u", int_arm_neon_vshiftu, 0>;
2621 // VSHL : Vector Shift Left (Immediate)
2622 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
2624 // VSHR : Vector Shift Right (Immediate)
2625 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
2627 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
2630 // VSHLL : Vector Shift Left Long
2631 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
2632 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
2634 // VSHLL : Vector Shift Left Long (with maximum shift count)
2635 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
2636 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
2637 ValueType OpTy, SDNode OpNode>
2638 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
2639 ResTy, OpTy, OpNode> {
2640 let Inst{21-16} = op21_16;
2642 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
2643 v8i16, v8i8, NEONvshlli>;
2644 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
2645 v4i32, v4i16, NEONvshlli>;
2646 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
2647 v2i64, v2i32, NEONvshlli>;
2649 // VSHRN : Vector Shift Right and Narrow
2650 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
2653 // VRSHL : Vector Rounding Shift
2654 defm VRSHLs : N3VInt_QHSD<0, 0, 0b0101, 0, N3RegVShFrm,
2655 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2656 "vrshl", "s", int_arm_neon_vrshifts, 0>;
2657 defm VRSHLu : N3VInt_QHSD<1, 0, 0b0101, 0, N3RegVShFrm,
2658 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2659 "vrshl", "u", int_arm_neon_vrshiftu, 0>;
2660 // VRSHR : Vector Rounding Shift Right
2661 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
2663 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
2666 // VRSHRN : Vector Rounding Shift Right and Narrow
2667 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
2670 // VQSHL : Vector Saturating Shift
2671 defm VQSHLs : N3VInt_QHSD<0, 0, 0b0100, 1, N3RegVShFrm,
2672 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2673 "vqshl", "s", int_arm_neon_vqshifts, 0>;
2674 defm VQSHLu : N3VInt_QHSD<1, 0, 0b0100, 1, N3RegVShFrm,
2675 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2676 "vqshl", "u", int_arm_neon_vqshiftu, 0>;
2677 // VQSHL : Vector Saturating Shift Left (Immediate)
2678 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
2680 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
2682 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
2683 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
2686 // VQSHRN : Vector Saturating Shift Right and Narrow
2687 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
2689 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
2692 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
2693 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
2696 // VQRSHL : Vector Saturating Rounding Shift
2697 defm VQRSHLs : N3VInt_QHSD<0, 0, 0b0101, 1, N3RegVShFrm,
2698 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2699 "vqrshl", "s", int_arm_neon_vqrshifts, 0>;
2700 defm VQRSHLu : N3VInt_QHSD<1, 0, 0b0101, 1, N3RegVShFrm,
2701 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
2702 "vqrshl", "u", int_arm_neon_vqrshiftu, 0>;
2704 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
2705 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
2707 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
2710 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
2711 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
2714 // VSRA : Vector Shift Right and Accumulate
2715 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
2716 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
2717 // VRSRA : Vector Rounding Shift Right and Accumulate
2718 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
2719 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
2721 // VSLI : Vector Shift Left and Insert
2722 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
2723 // VSRI : Vector Shift Right and Insert
2724 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
2726 // Vector Absolute and Saturating Absolute.
2728 // VABS : Vector Absolute Value
2729 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
2730 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
2732 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2733 IIC_VUNAD, "vabs", "f32",
2734 v2f32, v2f32, int_arm_neon_vabs>;
2735 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
2736 IIC_VUNAQ, "vabs", "f32",
2737 v4f32, v4f32, int_arm_neon_vabs>;
2739 // VQABS : Vector Saturating Absolute Value
2740 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
2741 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
2742 int_arm_neon_vqabs>;
2746 def vnegd : PatFrag<(ops node:$in),
2747 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
2748 def vnegq : PatFrag<(ops node:$in),
2749 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
2751 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2752 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
2753 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2754 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
2755 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
2756 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
2757 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2758 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
2760 // VNEG : Vector Negate (integer)
2761 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
2762 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
2763 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
2764 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
2765 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
2766 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
2768 // VNEG : Vector Negate (floating-point)
2769 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
2770 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
2771 "vneg", "f32", "$dst, $src", "",
2772 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
2773 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
2774 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
2775 "vneg", "f32", "$dst, $src", "",
2776 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
2778 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
2779 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
2780 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
2781 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
2782 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
2783 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
2785 // VQNEG : Vector Saturating Negate
2786 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
2787 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
2788 int_arm_neon_vqneg>;
2790 // Vector Bit Counting Operations.
2792 // VCLS : Vector Count Leading Sign Bits
2793 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
2794 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
2796 // VCLZ : Vector Count Leading Zeros
2797 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
2798 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
2800 // VCNT : Vector Count One Bits
2801 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2802 IIC_VCNTiD, "vcnt", "8",
2803 v8i8, v8i8, int_arm_neon_vcnt>;
2804 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
2805 IIC_VCNTiQ, "vcnt", "8",
2806 v16i8, v16i8, int_arm_neon_vcnt>;
2808 // Vector Swap -- for disassembly only.
2809 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
2810 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2811 "vswp", "$dst, $src", "", []>;
2812 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
2813 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2814 "vswp", "$dst, $src", "", []>;
2816 // Vector Move Operations.
2818 // VMOV : Vector Move (Register)
2820 let neverHasSideEffects = 1 in {
2821 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
2822 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2823 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
2824 N3RegFrm, IIC_VMOVD, "vmov", "$dst, $src", "", []>;
2826 // Pseudo vector move instructions for QQ and QQQQ registers. This should
2827 // be expanded after register allocation is completed.
2828 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
2829 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2831 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
2832 NoItinerary, "${:comment} vmov\t$dst, $src", []>;
2833 } // neverHasSideEffects
2835 // VMOV : Vector Move (Immediate)
2837 let isReMaterializable = 1 in {
2838 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
2839 (ins nModImm:$SIMM), IIC_VMOVImm,
2840 "vmov", "i8", "$dst, $SIMM", "",
2841 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
2842 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
2843 (ins nModImm:$SIMM), IIC_VMOVImm,
2844 "vmov", "i8", "$dst, $SIMM", "",
2845 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
2847 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
2848 (ins nModImm:$SIMM), IIC_VMOVImm,
2849 "vmov", "i16", "$dst, $SIMM", "",
2850 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]>;
2851 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
2852 (ins nModImm:$SIMM), IIC_VMOVImm,
2853 "vmov", "i16", "$dst, $SIMM", "",
2854 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]>;
2856 def VMOVv2i32 : N1ModImm<1, 0b000, {0,?,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
2857 (ins nModImm:$SIMM), IIC_VMOVImm,
2858 "vmov", "i32", "$dst, $SIMM", "",
2859 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]>;
2860 def VMOVv4i32 : N1ModImm<1, 0b000, {0,?,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
2861 (ins nModImm:$SIMM), IIC_VMOVImm,
2862 "vmov", "i32", "$dst, $SIMM", "",
2863 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]>;
2865 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
2866 (ins nModImm:$SIMM), IIC_VMOVImm,
2867 "vmov", "i64", "$dst, $SIMM", "",
2868 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
2869 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
2870 (ins nModImm:$SIMM), IIC_VMOVImm,
2871 "vmov", "i64", "$dst, $SIMM", "",
2872 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
2873 } // isReMaterializable
2875 // VMOV : Vector Get Lane (move scalar to ARM core register)
2877 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
2878 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2879 IIC_VMOVSI, "vmov", "s8", "$dst, $src[$lane]",
2880 [(set GPR:$dst, (NEONvgetlanes (v8i8 DPR:$src),
2882 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
2883 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2884 IIC_VMOVSI, "vmov", "s16", "$dst, $src[$lane]",
2885 [(set GPR:$dst, (NEONvgetlanes (v4i16 DPR:$src),
2887 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
2888 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2889 IIC_VMOVSI, "vmov", "u8", "$dst, $src[$lane]",
2890 [(set GPR:$dst, (NEONvgetlaneu (v8i8 DPR:$src),
2892 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
2893 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2894 IIC_VMOVSI, "vmov", "u16", "$dst, $src[$lane]",
2895 [(set GPR:$dst, (NEONvgetlaneu (v4i16 DPR:$src),
2897 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
2898 (outs GPR:$dst), (ins DPR:$src, nohash_imm:$lane),
2899 IIC_VMOVSI, "vmov", "32", "$dst, $src[$lane]",
2900 [(set GPR:$dst, (extractelt (v2i32 DPR:$src),
2902 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
2903 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
2904 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2905 (DSubReg_i8_reg imm:$lane))),
2906 (SubReg_i8_lane imm:$lane))>;
2907 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
2908 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2909 (DSubReg_i16_reg imm:$lane))),
2910 (SubReg_i16_lane imm:$lane))>;
2911 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
2912 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
2913 (DSubReg_i8_reg imm:$lane))),
2914 (SubReg_i8_lane imm:$lane))>;
2915 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
2916 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
2917 (DSubReg_i16_reg imm:$lane))),
2918 (SubReg_i16_lane imm:$lane))>;
2919 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
2920 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
2921 (DSubReg_i32_reg imm:$lane))),
2922 (SubReg_i32_lane imm:$lane))>;
2923 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
2924 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
2925 (SSubReg_f32_reg imm:$src2))>;
2926 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
2927 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
2928 (SSubReg_f32_reg imm:$src2))>;
2929 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
2930 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2931 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
2932 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
2935 // VMOV : Vector Set Lane (move ARM core register to scalar)
2937 let Constraints = "$src1 = $dst" in {
2938 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$dst),
2939 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2940 IIC_VMOVISL, "vmov", "8", "$dst[$lane], $src2",
2941 [(set DPR:$dst, (vector_insert (v8i8 DPR:$src1),
2942 GPR:$src2, imm:$lane))]>;
2943 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$dst),
2944 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2945 IIC_VMOVISL, "vmov", "16", "$dst[$lane], $src2",
2946 [(set DPR:$dst, (vector_insert (v4i16 DPR:$src1),
2947 GPR:$src2, imm:$lane))]>;
2948 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$dst),
2949 (ins DPR:$src1, GPR:$src2, nohash_imm:$lane),
2950 IIC_VMOVISL, "vmov", "32", "$dst[$lane], $src2",
2951 [(set DPR:$dst, (insertelt (v2i32 DPR:$src1),
2952 GPR:$src2, imm:$lane))]>;
2954 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
2955 (v16i8 (INSERT_SUBREG QPR:$src1,
2956 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
2957 (DSubReg_i8_reg imm:$lane))),
2958 GPR:$src2, (SubReg_i8_lane imm:$lane))),
2959 (DSubReg_i8_reg imm:$lane)))>;
2960 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
2961 (v8i16 (INSERT_SUBREG QPR:$src1,
2962 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
2963 (DSubReg_i16_reg imm:$lane))),
2964 GPR:$src2, (SubReg_i16_lane imm:$lane))),
2965 (DSubReg_i16_reg imm:$lane)))>;
2966 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
2967 (v4i32 (INSERT_SUBREG QPR:$src1,
2968 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
2969 (DSubReg_i32_reg imm:$lane))),
2970 GPR:$src2, (SubReg_i32_lane imm:$lane))),
2971 (DSubReg_i32_reg imm:$lane)))>;
2973 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
2974 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
2975 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2976 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
2977 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
2978 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
2980 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2981 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2982 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
2983 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
2985 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
2986 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
2987 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
2988 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
2989 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
2990 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
2992 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
2993 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2994 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
2995 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2996 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
2997 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
2999 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3000 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3001 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3003 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3004 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3005 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3007 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3008 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3009 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3012 // VDUP : Vector Duplicate (from ARM core register to all elements)
3014 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3015 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3016 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3017 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3018 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3019 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3020 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3021 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3023 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3024 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3025 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3026 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3027 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3028 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3030 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3031 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3032 [(set DPR:$dst, (v2f32 (NEONvdup
3033 (f32 (bitconvert GPR:$src)))))]>;
3034 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3035 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3036 [(set QPR:$dst, (v4f32 (NEONvdup
3037 (f32 (bitconvert GPR:$src)))))]>;
3039 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3041 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3043 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3044 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3045 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3047 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3048 ValueType ResTy, ValueType OpTy>
3049 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3050 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3051 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3054 // Inst{19-16} is partially specified depending on the element size.
3056 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8>;
3057 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16>;
3058 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32>;
3059 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32>;
3060 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8>;
3061 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16>;
3062 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32>;
3063 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32>;
3065 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3066 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3067 (DSubReg_i8_reg imm:$lane))),
3068 (SubReg_i8_lane imm:$lane)))>;
3069 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3070 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3071 (DSubReg_i16_reg imm:$lane))),
3072 (SubReg_i16_lane imm:$lane)))>;
3073 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3074 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3075 (DSubReg_i32_reg imm:$lane))),
3076 (SubReg_i32_lane imm:$lane)))>;
3077 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3078 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3079 (DSubReg_i32_reg imm:$lane))),
3080 (SubReg_i32_lane imm:$lane)))>;
3082 def VDUPfdf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 0, 0,
3083 (outs DPR:$dst), (ins SPR:$src),
3084 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3085 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3087 def VDUPfqf : N2V<0b11, 0b11, {?,1}, {0,0}, 0b11000, 1, 0,
3088 (outs QPR:$dst), (ins SPR:$src),
3089 IIC_VMOVD, "vdup", "32", "$dst, ${src:lane}", "",
3090 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3092 // VMOVN : Vector Narrowing Move
3093 defm VMOVN : N2VNInt_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVD,
3094 "vmovn", "i", int_arm_neon_vmovn>;
3095 // VQMOVN : Vector Saturating Narrowing Move
3096 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3097 "vqmovn", "s", int_arm_neon_vqmovns>;
3098 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3099 "vqmovn", "u", int_arm_neon_vqmovnu>;
3100 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3101 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3102 // VMOVL : Vector Lengthening Move
3103 defm VMOVLs : N2VLInt_QHS<0b01,0b10100,0,1, "vmovl", "s",
3104 int_arm_neon_vmovls>;
3105 defm VMOVLu : N2VLInt_QHS<0b11,0b10100,0,1, "vmovl", "u",
3106 int_arm_neon_vmovlu>;
3108 // Vector Conversions.
3110 // VCVT : Vector Convert Between Floating-Point and Integers
3111 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3112 v2i32, v2f32, fp_to_sint>;
3113 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3114 v2i32, v2f32, fp_to_uint>;
3115 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3116 v2f32, v2i32, sint_to_fp>;
3117 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3118 v2f32, v2i32, uint_to_fp>;
3120 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3121 v4i32, v4f32, fp_to_sint>;
3122 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3123 v4i32, v4f32, fp_to_uint>;
3124 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3125 v4f32, v4i32, sint_to_fp>;
3126 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3127 v4f32, v4i32, uint_to_fp>;
3129 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3130 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3131 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3132 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3133 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3134 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3135 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
3136 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3137 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
3139 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3140 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
3141 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3142 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
3143 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3144 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
3145 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
3146 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
3150 // VREV64 : Vector Reverse elements within 64-bit doublewords
3152 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3153 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
3154 (ins DPR:$src), IIC_VMOVD,
3155 OpcodeStr, Dt, "$dst, $src", "",
3156 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
3157 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3158 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
3159 (ins QPR:$src), IIC_VMOVD,
3160 OpcodeStr, Dt, "$dst, $src", "",
3161 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
3163 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
3164 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
3165 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
3166 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
3168 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
3169 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
3170 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
3171 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
3173 // VREV32 : Vector Reverse elements within 32-bit words
3175 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3176 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
3177 (ins DPR:$src), IIC_VMOVD,
3178 OpcodeStr, Dt, "$dst, $src", "",
3179 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
3180 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3181 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
3182 (ins QPR:$src), IIC_VMOVD,
3183 OpcodeStr, Dt, "$dst, $src", "",
3184 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
3186 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
3187 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
3189 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
3190 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
3192 // VREV16 : Vector Reverse elements within 16-bit halfwords
3194 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3195 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
3196 (ins DPR:$src), IIC_VMOVD,
3197 OpcodeStr, Dt, "$dst, $src", "",
3198 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
3199 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
3200 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
3201 (ins QPR:$src), IIC_VMOVD,
3202 OpcodeStr, Dt, "$dst, $src", "",
3203 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
3205 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
3206 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
3208 // Other Vector Shuffles.
3210 // VEXT : Vector Extract
3212 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
3213 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
3214 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
3215 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3216 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
3217 (Ty DPR:$rhs), imm:$index)))]>;
3219 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
3220 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
3221 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
3222 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
3223 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
3224 (Ty QPR:$rhs), imm:$index)))]>;
3226 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
3227 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
3228 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
3229 def VEXTdf : VEXTd<"vext", "32", v2f32>;
3231 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
3232 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
3233 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
3234 def VEXTqf : VEXTq<"vext", "32", v4f32>;
3236 // VTRN : Vector Transpose
3238 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
3239 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
3240 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
3242 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
3243 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
3244 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
3246 // VUZP : Vector Unzip (Deinterleave)
3248 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
3249 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
3250 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
3252 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
3253 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
3254 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
3256 // VZIP : Vector Zip (Interleave)
3258 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
3259 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
3260 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
3262 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
3263 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
3264 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
3266 // Vector Table Lookup and Table Extension.
3268 // VTBL : Vector Table Lookup
3270 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$dst),
3271 (ins DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTB1,
3272 "vtbl", "8", "$dst, \\{$tbl1\\}, $src", "",
3273 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbl1 DPR:$tbl1, DPR:$src)))]>;
3274 let hasExtraSrcRegAllocReq = 1 in {
3276 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$dst),
3277 (ins DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTB2,
3278 "vtbl", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "", []>;
3280 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$dst),
3281 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src), NVTBLFrm, IIC_VTB3,
3282 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src", "", []>;
3284 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$dst),
3285 (ins DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src),
3287 "vtbl", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src", "", []>;
3288 } // hasExtraSrcRegAllocReq = 1
3290 // VTBX : Vector Table Extension
3292 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$dst),
3293 (ins DPR:$orig, DPR:$tbl1, DPR:$src), NVTBLFrm, IIC_VTBX1,
3294 "vtbx", "8", "$dst, \\{$tbl1\\}, $src", "$orig = $dst",
3295 [(set DPR:$dst, (v8i8 (int_arm_neon_vtbx1
3296 DPR:$orig, DPR:$tbl1, DPR:$src)))]>;
3297 let hasExtraSrcRegAllocReq = 1 in {
3299 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$dst),
3300 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$src), NVTBLFrm, IIC_VTBX2,
3301 "vtbx", "8", "$dst, \\{$tbl1, $tbl2\\}, $src", "$orig = $dst", []>;
3303 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$dst),
3304 (ins DPR:$orig, DPR:$tbl1, DPR:$tbl2, DPR:$tbl3, DPR:$src),
3305 NVTBLFrm, IIC_VTBX3,
3306 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3\\}, $src",
3307 "$orig = $dst", []>;
3309 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$dst), (ins DPR:$orig, DPR:$tbl1,
3310 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$src), NVTBLFrm, IIC_VTBX4,
3311 "vtbx", "8", "$dst, \\{$tbl1, $tbl2, $tbl3, $tbl4\\}, $src",
3312 "$orig = $dst", []>;
3313 } // hasExtraSrcRegAllocReq = 1
3315 //===----------------------------------------------------------------------===//
3316 // NEON instructions for single-precision FP math
3317 //===----------------------------------------------------------------------===//
3319 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
3320 : NEONFPPat<(ResTy (OpNode SPR:$a)),
3321 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
3325 class N3VSPat<SDNode OpNode, NeonI Inst>
3326 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
3327 (EXTRACT_SUBREG (v2f32
3328 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3330 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3334 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
3335 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
3336 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3338 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3340 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
3344 // These need separate instructions because they must use DPR_VFP2 register
3345 // class which have SPR sub-registers.
3347 // Vector Add Operations used for single-precision FP
3348 let neverHasSideEffects = 1 in
3349 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
3350 def : N3VSPat<fadd, VADDfd_sfp>;
3352 // Vector Sub Operations used for single-precision FP
3353 let neverHasSideEffects = 1 in
3354 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
3355 def : N3VSPat<fsub, VSUBfd_sfp>;
3357 // Vector Multiply Operations used for single-precision FP
3358 let neverHasSideEffects = 1 in
3359 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
3360 def : N3VSPat<fmul, VMULfd_sfp>;
3362 // Vector Multiply-Accumulate/Subtract used for single-precision FP
3363 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
3364 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
3366 //let neverHasSideEffects = 1 in
3367 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
3368 // v2f32, fmul, fadd>;
3369 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
3371 //let neverHasSideEffects = 1 in
3372 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
3373 // v2f32, fmul, fsub>;
3374 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
3376 // Vector Absolute used for single-precision FP
3377 let neverHasSideEffects = 1 in
3378 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
3379 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3380 "vabs", "f32", "$dst, $src", "", []>;
3381 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
3383 // Vector Negate used for single-precision FP
3384 let neverHasSideEffects = 1 in
3385 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3386 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
3387 "vneg", "f32", "$dst, $src", "", []>;
3388 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
3390 // Vector Maximum used for single-precision FP
3391 let neverHasSideEffects = 1 in
3392 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3393 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3394 "vmax", "f32", "$dst, $src1, $src2", "", []>;
3395 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
3397 // Vector Minimum used for single-precision FP
3398 let neverHasSideEffects = 1 in
3399 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
3400 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
3401 "vmin", "f32", "$dst, $src1, $src2", "", []>;
3402 def : N3VSPat<NEONfmin, VMINfd_sfp>;
3404 // Vector Convert between single-precision FP and integer
3405 let neverHasSideEffects = 1 in
3406 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3407 v2i32, v2f32, fp_to_sint>;
3408 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
3410 let neverHasSideEffects = 1 in
3411 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3412 v2i32, v2f32, fp_to_uint>;
3413 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
3415 let neverHasSideEffects = 1 in
3416 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3417 v2f32, v2i32, sint_to_fp>;
3418 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
3420 let neverHasSideEffects = 1 in
3421 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3422 v2f32, v2i32, uint_to_fp>;
3423 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
3425 //===----------------------------------------------------------------------===//
3426 // Non-Instruction Patterns
3427 //===----------------------------------------------------------------------===//
3430 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
3431 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
3432 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
3433 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
3434 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
3435 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
3436 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
3437 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
3438 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
3439 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
3440 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
3441 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
3442 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
3443 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
3444 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
3445 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
3446 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
3447 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
3448 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
3449 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
3450 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
3451 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
3452 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
3453 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
3454 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
3455 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
3456 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
3457 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
3458 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
3459 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
3461 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
3462 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
3463 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
3464 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
3465 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
3466 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
3467 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
3468 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
3469 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
3470 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
3471 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
3472 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
3473 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
3474 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
3475 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
3476 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
3477 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
3478 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
3479 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
3480 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
3481 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
3482 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
3483 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
3484 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
3485 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
3486 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
3487 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
3488 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
3489 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
3490 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;