1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
14 //===----------------------------------------------------------------------===//
15 // NEON-specific DAG Nodes.
16 //===----------------------------------------------------------------------===//
18 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
20 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
21 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
22 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
23 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
24 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
25 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
27 // Types for vector shift by immediates. The "SHX" version is for long and
28 // narrow operations where the source and destination vectors have different
29 // types. The "SHINS" version is for shift and insert operations.
30 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
32 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
34 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
35 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
37 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
38 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
39 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
40 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
41 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
42 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
43 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
45 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
46 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
47 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
49 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
50 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
51 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
52 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
53 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
54 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
56 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
57 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
58 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
60 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
61 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
63 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
65 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
66 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
68 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
69 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
70 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
72 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
74 // VDUPLANE can produce a quad-register result from a double-register source,
75 // so the result is not constrained to match the source.
76 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
77 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
80 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
81 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
82 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
84 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
85 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
86 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
87 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
89 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
92 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
93 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
94 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
96 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
98 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
99 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
101 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
102 SDTCisSameAs<0, 2>]>;
103 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
104 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
106 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
107 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
108 unsigned EltBits = 0;
109 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
110 return (EltBits == 32 && EltVal == 0);
113 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
114 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
115 unsigned EltBits = 0;
116 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
117 return (EltBits == 8 && EltVal == 0xff);
120 //===----------------------------------------------------------------------===//
121 // NEON operand definitions
122 //===----------------------------------------------------------------------===//
124 def nModImm : Operand<i32> {
125 let PrintMethod = "printNEONModImmOperand";
128 //===----------------------------------------------------------------------===//
129 // NEON load / store instructions
130 //===----------------------------------------------------------------------===//
132 // Use VLDM to load a Q register as a D register pair.
133 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
135 : PseudoVFPLdStM<(outs QPR:$dst), (ins addrmode4:$addr), IIC_fpLoad_m, "",
136 [(set QPR:$dst, (v2f64 (load addrmode4:$addr)))]>;
138 // Use VSTM to store a Q register as a D register pair.
139 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
141 : PseudoVFPLdStM<(outs), (ins QPR:$src, addrmode4:$addr), IIC_fpStore_m, "",
142 [(store (v2f64 QPR:$src), addrmode4:$addr)]>;
144 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
146 // Classes for VLD* pseudo-instructions with multi-register operands.
147 // These are expanded to real instructions after register allocation.
148 class VLDQPseudo<InstrItinClass itin>
149 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
150 class VLDQWBPseudo<InstrItinClass itin>
151 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
152 (ins addrmode6:$addr, am6offset:$offset), itin,
154 class VLDQQPseudo<InstrItinClass itin>
155 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
156 class VLDQQWBPseudo<InstrItinClass itin>
157 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
158 (ins addrmode6:$addr, am6offset:$offset), itin,
160 class VLDQQQQWBPseudo<InstrItinClass itin>
161 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
162 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
163 "$addr.addr = $wb, $src = $dst">;
165 // VLD1 : Vector Load (multiple single elements)
166 class VLD1D<bits<4> op7_4, string Dt>
167 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd),
168 (ins addrmode6:$Rn), IIC_VLD1,
169 "vld1", Dt, "\\{$Vd\\}, $Rn", "", []> {
173 class VLD1Q<bits<4> op7_4, string Dt>
174 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2),
175 (ins addrmode6:$Rn), IIC_VLD1x2,
176 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
178 let Inst{5-4} = Rn{5-4};
181 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
182 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
183 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
184 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
186 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
187 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
188 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
189 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
191 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
192 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
193 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
194 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
196 // ...with address register writeback:
197 class VLD1DWB<bits<4> op7_4, string Dt>
198 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
199 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
200 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
201 "$Rn.addr = $wb", []> {
204 class VLD1QWB<bits<4> op7_4, string Dt>
205 : NLdSt<0,0b10,0b1010,op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
206 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
207 "vld1", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
208 "$Rn.addr = $wb", []> {
209 let Inst{5-4} = Rn{5-4};
212 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
213 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
214 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
215 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
217 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
218 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
219 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
220 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
222 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
223 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
224 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
225 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
227 // ...with 3 registers (some of these are only for the disassembler):
228 class VLD1D3<bits<4> op7_4, string Dt>
229 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
230 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
231 "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
235 class VLD1D3WB<bits<4> op7_4, string Dt>
236 : NLdSt<0,0b10,0b0110,op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
237 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
238 "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
242 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
243 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
244 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
245 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
247 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
248 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
249 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
250 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
252 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
253 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
255 // ...with 4 registers (some of these are only for the disassembler):
256 class VLD1D4<bits<4> op7_4, string Dt>
257 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
258 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
259 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
261 let Inst{5-4} = Rn{5-4};
263 class VLD1D4WB<bits<4> op7_4, string Dt>
264 : NLdSt<0,0b10,0b0010,op7_4,
265 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
266 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4, "vld1", Dt,
267 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
269 let Inst{5-4} = Rn{5-4};
272 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
273 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
274 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
275 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
277 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
278 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
279 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
280 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
282 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
283 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
285 // VLD2 : Vector Load (multiple 2-element structures)
286 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
287 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
288 (ins addrmode6:$Rn), IIC_VLD2,
289 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
291 let Inst{5-4} = Rn{5-4};
293 class VLD2Q<bits<4> op7_4, string Dt>
294 : NLdSt<0, 0b10, 0b0011, op7_4,
295 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
296 (ins addrmode6:$Rn), IIC_VLD2x2,
297 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
299 let Inst{5-4} = Rn{5-4};
302 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
303 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
304 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
306 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
307 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
308 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
310 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
311 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
312 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
314 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
315 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
316 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
318 // ...with address register writeback:
319 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
320 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
321 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
322 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
323 "$Rn.addr = $wb", []> {
324 let Inst{5-4} = Rn{5-4};
326 class VLD2QWB<bits<4> op7_4, string Dt>
327 : NLdSt<0, 0b10, 0b0011, op7_4,
328 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
329 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
330 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
331 "$Rn.addr = $wb", []> {
332 let Inst{5-4} = Rn{5-4};
335 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
336 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
337 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
339 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
340 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
341 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
343 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
344 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
345 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
347 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
348 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
349 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
351 // ...with double-spaced registers (for disassembly only):
352 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
353 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
354 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
355 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
356 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
357 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
359 // VLD3 : Vector Load (multiple 3-element structures)
360 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
361 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
362 (ins addrmode6:$Rn), IIC_VLD3,
363 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
368 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
369 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
370 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
372 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
373 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
374 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
376 // ...with address register writeback:
377 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
378 : NLdSt<0, 0b10, op11_8, op7_4,
379 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
380 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
381 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
382 "$Rn.addr = $wb", []> {
386 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
387 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
388 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
390 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
391 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
392 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
394 // ...with double-spaced registers (non-updating versions for disassembly only):
395 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
396 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
397 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
398 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
399 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
400 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
402 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
403 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
404 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
406 // ...alternate versions to be allocated odd register numbers:
407 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
408 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
409 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
411 // VLD4 : Vector Load (multiple 4-element structures)
412 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
413 : NLdSt<0, 0b10, op11_8, op7_4,
414 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
415 (ins addrmode6:$Rn), IIC_VLD4,
416 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
418 let Inst{5-4} = Rn{5-4};
421 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
422 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
423 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
425 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
426 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
427 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
429 // ...with address register writeback:
430 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
431 : NLdSt<0, 0b10, op11_8, op7_4,
432 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
433 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4,
434 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
435 "$Rn.addr = $wb", []> {
436 let Inst{5-4} = Rn{5-4};
439 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
440 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
441 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
443 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
444 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
445 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4>;
447 // ...with double-spaced registers (non-updating versions for disassembly only):
448 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
449 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
450 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
451 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
452 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
453 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
455 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
456 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
457 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
459 // ...alternate versions to be allocated odd register numbers:
460 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
461 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
462 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4>;
464 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
466 // Classes for VLD*LN pseudo-instructions with multi-register operands.
467 // These are expanded to real instructions after register allocation.
468 class VLDQLNPseudo<InstrItinClass itin>
469 : PseudoNLdSt<(outs QPR:$dst),
470 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
471 itin, "$src = $dst">;
472 class VLDQLNWBPseudo<InstrItinClass itin>
473 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
474 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
475 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
476 class VLDQQLNPseudo<InstrItinClass itin>
477 : PseudoNLdSt<(outs QQPR:$dst),
478 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
479 itin, "$src = $dst">;
480 class VLDQQLNWBPseudo<InstrItinClass itin>
481 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
482 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
483 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
484 class VLDQQQQLNPseudo<InstrItinClass itin>
485 : PseudoNLdSt<(outs QQQQPR:$dst),
486 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
487 itin, "$src = $dst">;
488 class VLDQQQQLNWBPseudo<InstrItinClass itin>
489 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
490 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
491 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
493 // VLD1LN : Vector Load (single element to one lane)
494 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
496 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
497 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
498 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
500 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
501 (i32 (LoadOp addrmode6:$Rn)),
505 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
506 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
507 (i32 (LoadOp addrmode6:$addr)),
511 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
512 let Inst{7-5} = lane{2-0};
514 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
515 let Inst{7-6} = lane{1-0};
518 def VLD1LNd32 : VLD1LN<0b1000, {?,0,?,?}, "32", v2i32, load> {
519 let Inst{7} = lane{0};
524 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
525 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
526 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
528 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
530 // ...with address register writeback:
531 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
532 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
533 (ins addrmode6:$Rn, am6offset:$Rm,
534 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
535 "\\{$Vd[$lane]\\}, $Rn$Rm",
536 "$src = $Vd, $Rn.addr = $wb", []>;
538 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
539 let Inst{7-5} = lane{2-0};
541 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
542 let Inst{7-6} = lane{1-0};
545 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
546 let Inst{7} = lane{0};
551 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
552 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
553 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
555 // VLD2LN : Vector Load (single 2-element structure to one lane)
556 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
557 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
558 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
559 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
560 "$src1 = $Vd, $src2 = $dst2", []> {
565 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
566 let Inst{7-5} = lane{2-0};
568 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
569 let Inst{7-6} = lane{1-0};
571 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
572 let Inst{7} = lane{0};
575 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
576 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
577 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
579 // ...with double-spaced registers:
580 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
581 let Inst{7-6} = lane{1-0};
583 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
584 let Inst{7} = lane{0};
587 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
588 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
590 // ...with address register writeback:
591 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
592 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
593 (ins addrmode6:$Rn, am6offset:$Rm,
594 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
595 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
596 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
600 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
601 let Inst{7-5} = lane{2-0};
603 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
604 let Inst{7-6} = lane{1-0};
606 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
607 let Inst{7} = lane{0};
610 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
611 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
612 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
614 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
615 let Inst{7-6} = lane{1-0};
617 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
618 let Inst{7} = lane{0};
621 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
622 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
624 // VLD3LN : Vector Load (single 3-element structure to one lane)
625 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
626 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
627 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
628 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
629 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
630 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
634 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
635 let Inst{7-5} = lane{2-0};
637 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
638 let Inst{7-6} = lane{1-0};
640 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
641 let Inst{7} = lane{0};
644 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
645 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
646 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
648 // ...with double-spaced registers:
649 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
650 let Inst{7-6} = lane{1-0};
652 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
653 let Inst{7} = lane{0};
656 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
657 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
659 // ...with address register writeback:
660 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
661 : NLdStLn<1, 0b10, op11_8, op7_4,
662 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
663 (ins addrmode6:$Rn, am6offset:$Rm,
664 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
665 IIC_VLD3lnu, "vld3", Dt,
666 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
667 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
670 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
671 let Inst{7-5} = lane{2-0};
673 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
674 let Inst{7-6} = lane{1-0};
676 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
677 let Inst{7} = lane{0};
680 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
681 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
682 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
684 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
685 let Inst{7-6} = lane{1-0};
687 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
688 let Inst{7} = lane{0};
691 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
692 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
694 // VLD4LN : Vector Load (single 4-element structure to one lane)
695 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
696 : NLdStLn<1, 0b10, op11_8, op7_4,
697 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
698 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
699 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
700 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
701 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
706 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
707 let Inst{7-5} = lane{2-0};
709 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
710 let Inst{7-6} = lane{1-0};
712 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
713 let Inst{7} = lane{0};
717 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
718 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
719 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
721 // ...with double-spaced registers:
722 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
723 let Inst{7-6} = lane{1-0};
725 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
726 let Inst{7} = lane{0};
730 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
731 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
733 // ...with address register writeback:
734 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
735 : NLdStLn<1, 0b10, op11_8, op7_4,
736 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
737 (ins addrmode6:$Rn, am6offset:$Rm,
738 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
739 IIC_VLD4ln, "vld4", Dt,
740 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
741 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
746 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
747 let Inst{7-5} = lane{2-0};
749 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
750 let Inst{7-6} = lane{1-0};
752 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
753 let Inst{7} = lane{0};
757 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
758 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
759 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
761 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
762 let Inst{7-6} = lane{1-0};
764 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
765 let Inst{7} = lane{0};
769 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
770 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
772 // VLD1DUP : Vector Load (single element to all lanes)
773 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
774 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
775 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
776 // FIXME: Not yet implemented.
777 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
779 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
781 // Classes for VST* pseudo-instructions with multi-register operands.
782 // These are expanded to real instructions after register allocation.
783 class VSTQPseudo<InstrItinClass itin>
784 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
785 class VSTQWBPseudo<InstrItinClass itin>
786 : PseudoNLdSt<(outs GPR:$wb),
787 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
789 class VSTQQPseudo<InstrItinClass itin>
790 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
791 class VSTQQWBPseudo<InstrItinClass itin>
792 : PseudoNLdSt<(outs GPR:$wb),
793 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
795 class VSTQQQQWBPseudo<InstrItinClass itin>
796 : PseudoNLdSt<(outs GPR:$wb),
797 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
800 // VST1 : Vector Store (multiple single elements)
801 class VST1D<bits<4> op7_4, string Dt>
802 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, DPR:$Vd),
803 IIC_VST1, "vst1", Dt, "\\{$Vd\\}, $Rn", "", []> {
807 class VST1Q<bits<4> op7_4, string Dt>
808 : NLdSt<0,0b00,0b1010,op7_4, (outs),
809 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
810 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
812 let Inst{5-4} = Rn{5-4};
815 def VST1d8 : VST1D<{0,0,0,?}, "8">;
816 def VST1d16 : VST1D<{0,1,0,?}, "16">;
817 def VST1d32 : VST1D<{1,0,0,?}, "32">;
818 def VST1d64 : VST1D<{1,1,0,?}, "64">;
820 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
821 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
822 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
823 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
825 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
826 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
827 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
828 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
830 // ...with address register writeback:
831 class VST1DWB<bits<4> op7_4, string Dt>
832 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
833 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
834 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
837 class VST1QWB<bits<4> op7_4, string Dt>
838 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
839 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
840 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
841 "$Rn.addr = $wb", []> {
842 let Inst{5-4} = Rn{5-4};
845 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
846 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
847 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
848 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
850 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
851 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
852 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
853 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
855 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
856 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
857 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
858 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
860 // ...with 3 registers (some of these are only for the disassembler):
861 class VST1D3<bits<4> op7_4, string Dt>
862 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
863 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
864 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
868 class VST1D3WB<bits<4> op7_4, string Dt>
869 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
870 (ins addrmode6:$Rn, am6offset:$Rm,
871 DPR:$Vd, DPR:$src2, DPR:$src3),
872 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
873 "$Rn.addr = $wb", []> {
877 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
878 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
879 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
880 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
882 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
883 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
884 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
885 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
887 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
888 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
890 // ...with 4 registers (some of these are only for the disassembler):
891 class VST1D4<bits<4> op7_4, string Dt>
892 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
893 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
894 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
897 let Inst{5-4} = Rn{5-4};
899 class VST1D4WB<bits<4> op7_4, string Dt>
900 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
901 (ins addrmode6:$Rn, am6offset:$Rm,
902 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
903 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
904 "$Rn.addr = $wb", []> {
905 let Inst{5-4} = Rn{5-4};
908 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
909 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
910 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
911 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
913 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
914 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
915 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
916 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
918 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
919 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
921 // VST2 : Vector Store (multiple 2-element structures)
922 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
923 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
924 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
925 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
927 let Inst{5-4} = Rn{5-4};
929 class VST2Q<bits<4> op7_4, string Dt>
930 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
931 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
932 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
935 let Inst{5-4} = Rn{5-4};
938 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
939 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
940 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
942 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
943 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
944 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
946 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
947 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
948 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
950 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
951 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
952 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
954 // ...with address register writeback:
955 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
956 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
957 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
958 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
959 "$Rn.addr = $wb", []> {
960 let Inst{5-4} = Rn{5-4};
962 class VST2QWB<bits<4> op7_4, string Dt>
963 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
964 (ins addrmode6:$Rn, am6offset:$Rm,
965 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
966 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
967 "$Rn.addr = $wb", []> {
968 let Inst{5-4} = Rn{5-4};
971 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
972 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
973 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
975 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
976 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
977 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
979 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
980 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
981 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
983 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
984 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
985 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
987 // ...with double-spaced registers (for disassembly only):
988 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
989 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
990 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
991 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
992 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
993 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
995 // VST3 : Vector Store (multiple 3-element structures)
996 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
997 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
998 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
999 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1001 let Inst{4} = Rn{4};
1004 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1005 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1006 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1008 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1009 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1010 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1012 // ...with address register writeback:
1013 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1014 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1015 (ins addrmode6:$Rn, am6offset:$Rm,
1016 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1017 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1018 "$Rn.addr = $wb", []> {
1019 let Inst{4} = Rn{4};
1022 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1023 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1024 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1026 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1027 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1028 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1030 // ...with double-spaced registers (non-updating versions for disassembly only):
1031 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1032 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1033 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1034 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1035 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1036 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1038 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1039 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1040 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1042 // ...alternate versions to be allocated odd register numbers:
1043 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1044 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1045 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1047 // VST4 : Vector Store (multiple 4-element structures)
1048 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1049 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1050 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1051 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1054 let Inst{5-4} = Rn{5-4};
1057 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1058 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1059 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1061 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1062 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1063 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1065 // ...with address register writeback:
1066 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1067 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1068 (ins addrmode6:$Rn, am6offset:$Rm,
1069 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1070 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1071 "$Rn.addr = $wb", []> {
1072 let Inst{5-4} = Rn{5-4};
1075 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1076 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1077 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1079 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1080 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1081 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1083 // ...with double-spaced registers (non-updating versions for disassembly only):
1084 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1085 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1086 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1087 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1088 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1089 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1091 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1092 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1093 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1095 // ...alternate versions to be allocated odd register numbers:
1096 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1097 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1098 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1100 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1102 // Classes for VST*LN pseudo-instructions with multi-register operands.
1103 // These are expanded to real instructions after register allocation.
1104 class VSTQLNPseudo<InstrItinClass itin>
1105 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1107 class VSTQLNWBPseudo<InstrItinClass itin>
1108 : PseudoNLdSt<(outs GPR:$wb),
1109 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1110 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1111 class VSTQQLNPseudo<InstrItinClass itin>
1112 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1114 class VSTQQLNWBPseudo<InstrItinClass itin>
1115 : PseudoNLdSt<(outs GPR:$wb),
1116 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1117 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1118 class VSTQQQQLNPseudo<InstrItinClass itin>
1119 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1121 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1122 : PseudoNLdSt<(outs GPR:$wb),
1123 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1124 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1126 // VST1LN : Vector Store (single element from one lane)
1127 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1128 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1129 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1130 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "", []> {
1134 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8"> {
1135 let Inst{7-5} = lane{2-0};
1137 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16"> {
1138 let Inst{7-6} = lane{1-0};
1139 let Inst{4} = Rn{5};
1141 def VST1LNd32 : VST1LN<0b1000, {?,0,?,?}, "32"> {
1142 let Inst{7} = lane{0};
1143 let Inst{5-4} = Rn{5-4};
1146 def VST1LNq8Pseudo : VSTQLNPseudo<IIC_VST1ln>;
1147 def VST1LNq16Pseudo : VSTQLNPseudo<IIC_VST1ln>;
1148 def VST1LNq32Pseudo : VSTQLNPseudo<IIC_VST1ln>;
1150 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1152 // ...with address register writeback:
1153 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1154 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1155 (ins addrmode6:$Rn, am6offset:$Rm,
1156 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1157 "\\{$Vd[$lane]\\}, $Rn$Rm",
1158 "$Rn.addr = $wb", []>;
1160 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8"> {
1161 let Inst{7-5} = lane{2-0};
1163 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16"> {
1164 let Inst{7-6} = lane{1-0};
1165 let Inst{4} = Rn{5};
1167 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32"> {
1168 let Inst{7} = lane{0};
1169 let Inst{5-4} = Rn{5-4};
1172 def VST1LNq8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1173 def VST1LNq16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1174 def VST1LNq32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST1lnu>;
1176 // VST2LN : Vector Store (single 2-element structure from one lane)
1177 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1178 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1179 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1180 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1183 let Inst{4} = Rn{4};
1186 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1187 let Inst{7-5} = lane{2-0};
1189 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1190 let Inst{7-6} = lane{1-0};
1192 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1193 let Inst{7} = lane{0};
1196 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1197 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1198 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1200 // ...with double-spaced registers:
1201 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1202 let Inst{7-6} = lane{1-0};
1203 let Inst{4} = Rn{4};
1205 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1206 let Inst{7} = lane{0};
1207 let Inst{4} = Rn{4};
1210 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1211 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1213 // ...with address register writeback:
1214 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1215 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1216 (ins addrmode6:$addr, am6offset:$offset,
1217 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1218 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1219 "$addr.addr = $wb", []> {
1220 let Inst{4} = Rn{4};
1223 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1224 let Inst{7-5} = lane{2-0};
1226 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1227 let Inst{7-6} = lane{1-0};
1229 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1230 let Inst{7} = lane{0};
1233 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1234 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1235 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1237 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1238 let Inst{7-6} = lane{1-0};
1240 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1241 let Inst{7} = lane{0};
1244 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1245 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1247 // VST3LN : Vector Store (single 3-element structure from one lane)
1248 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1249 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1250 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1251 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1252 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1256 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1257 let Inst{7-5} = lane{2-0};
1259 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1260 let Inst{7-6} = lane{1-0};
1262 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1263 let Inst{7} = lane{0};
1266 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1267 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1268 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1270 // ...with double-spaced registers:
1271 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1272 let Inst{7-6} = lane{1-0};
1274 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1275 let Inst{7} = lane{0};
1278 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1279 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1281 // ...with address register writeback:
1282 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1283 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1284 (ins addrmode6:$Rn, am6offset:$Rm,
1285 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1286 IIC_VST3lnu, "vst3", Dt,
1287 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1288 "$Rn.addr = $wb", []>;
1290 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1291 let Inst{7-5} = lane{2-0};
1293 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1294 let Inst{7-6} = lane{1-0};
1296 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1297 let Inst{7} = lane{0};
1300 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1301 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1302 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1304 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1305 let Inst{7-6} = lane{1-0};
1307 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1308 let Inst{7} = lane{0};
1311 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1312 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1314 // VST4LN : Vector Store (single 4-element structure from one lane)
1315 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1316 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1317 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1318 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1319 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1322 let Inst{4} = Rn{4};
1325 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1326 let Inst{7-5} = lane{2-0};
1328 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1329 let Inst{7-6} = lane{1-0};
1331 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1332 let Inst{7} = lane{0};
1333 let Inst{5} = Rn{5};
1336 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1337 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1338 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1340 // ...with double-spaced registers:
1341 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1342 let Inst{7-6} = lane{1-0};
1344 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1345 let Inst{7} = lane{0};
1346 let Inst{5} = Rn{5};
1349 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1350 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1352 // ...with address register writeback:
1353 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1354 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1355 (ins addrmode6:$Rn, am6offset:$Rm,
1356 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1357 IIC_VST4lnu, "vst4", Dt,
1358 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1359 "$Rn.addr = $wb", []> {
1360 let Inst{4} = Rn{4};
1363 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1364 let Inst{7-5} = lane{2-0};
1366 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1367 let Inst{7-6} = lane{1-0};
1369 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1370 let Inst{7} = lane{0};
1371 let Inst{5} = Rn{5};
1374 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1375 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1376 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1378 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1379 let Inst{7-6} = lane{1-0};
1381 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1382 let Inst{7} = lane{0};
1383 let Inst{5} = Rn{5};
1386 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1387 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1389 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1392 //===----------------------------------------------------------------------===//
1393 // NEON pattern fragments
1394 //===----------------------------------------------------------------------===//
1396 // Extract D sub-registers of Q registers.
1397 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1398 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1399 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1401 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1402 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1403 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1405 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1406 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1407 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1409 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1410 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1411 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1414 // Extract S sub-registers of Q/D registers.
1415 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1416 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1417 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1420 // Translate lane numbers from Q registers to D subregs.
1421 def SubReg_i8_lane : SDNodeXForm<imm, [{
1422 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1424 def SubReg_i16_lane : SDNodeXForm<imm, [{
1425 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1427 def SubReg_i32_lane : SDNodeXForm<imm, [{
1428 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1431 //===----------------------------------------------------------------------===//
1432 // Instruction Classes
1433 //===----------------------------------------------------------------------===//
1435 // Basic 2-register operations: single-, double- and quad-register.
1436 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1437 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1438 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1439 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1440 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1441 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1442 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1443 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1444 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1445 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1446 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1447 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1448 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1449 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1450 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1451 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1452 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1453 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1455 // Basic 2-register intrinsics, both double- and quad-register.
1456 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1457 bits<2> op17_16, bits<5> op11_7, bit op4,
1458 InstrItinClass itin, string OpcodeStr, string Dt,
1459 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1460 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1461 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1462 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1463 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1464 bits<2> op17_16, bits<5> op11_7, bit op4,
1465 InstrItinClass itin, string OpcodeStr, string Dt,
1466 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1467 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1468 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1469 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1471 // Narrow 2-register operations.
1472 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1473 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1474 InstrItinClass itin, string OpcodeStr, string Dt,
1475 ValueType TyD, ValueType TyQ, SDNode OpNode>
1476 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1477 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1478 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1480 // Narrow 2-register intrinsics.
1481 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1482 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1483 InstrItinClass itin, string OpcodeStr, string Dt,
1484 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1485 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1486 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1487 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1489 // Long 2-register operations (currently only used for VMOVL).
1490 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1491 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1492 InstrItinClass itin, string OpcodeStr, string Dt,
1493 ValueType TyQ, ValueType TyD, SDNode OpNode>
1494 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1495 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1496 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1498 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1499 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1500 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1501 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1502 OpcodeStr, Dt, "$dst1, $dst2",
1503 "$src1 = $dst1, $src2 = $dst2", []>;
1504 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1505 InstrItinClass itin, string OpcodeStr, string Dt>
1506 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1507 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1508 "$src1 = $dst1, $src2 = $dst2", []>;
1510 // Basic 3-register operations: single-, double- and quad-register.
1511 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1512 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1513 SDNode OpNode, bit Commutable>
1514 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1515 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1516 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1517 let isCommutable = Commutable;
1520 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1521 InstrItinClass itin, string OpcodeStr, string Dt,
1522 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1523 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1524 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1525 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1526 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1527 let isCommutable = Commutable;
1529 // Same as N3VD but no data type.
1530 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1531 InstrItinClass itin, string OpcodeStr,
1532 ValueType ResTy, ValueType OpTy,
1533 SDNode OpNode, bit Commutable>
1534 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1535 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1536 OpcodeStr, "$dst, $src1, $src2", "",
1537 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1538 let isCommutable = Commutable;
1541 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1542 InstrItinClass itin, string OpcodeStr, string Dt,
1543 ValueType Ty, SDNode ShOp>
1544 : N3V<0, 1, op21_20, op11_8, 1, 0,
1545 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1546 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1547 [(set (Ty DPR:$dst),
1548 (Ty (ShOp (Ty DPR:$src1),
1549 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1550 let isCommutable = 0;
1552 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1553 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1554 : N3V<0, 1, op21_20, op11_8, 1, 0,
1555 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1556 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1557 [(set (Ty DPR:$dst),
1558 (Ty (ShOp (Ty DPR:$src1),
1559 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1560 let isCommutable = 0;
1563 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1564 InstrItinClass itin, string OpcodeStr, string Dt,
1565 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1566 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1567 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1568 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1569 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1570 let isCommutable = Commutable;
1572 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1573 InstrItinClass itin, string OpcodeStr,
1574 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1575 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1576 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1577 OpcodeStr, "$dst, $src1, $src2", "",
1578 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1579 let isCommutable = Commutable;
1581 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1582 InstrItinClass itin, string OpcodeStr, string Dt,
1583 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1584 : N3V<1, 1, op21_20, op11_8, 1, 0,
1585 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1586 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1587 [(set (ResTy QPR:$dst),
1588 (ResTy (ShOp (ResTy QPR:$src1),
1589 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1591 let isCommutable = 0;
1593 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1594 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1595 : N3V<1, 1, op21_20, op11_8, 1, 0,
1596 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1597 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1598 [(set (ResTy QPR:$dst),
1599 (ResTy (ShOp (ResTy QPR:$src1),
1600 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1602 let isCommutable = 0;
1605 // Basic 3-register intrinsics, both double- and quad-register.
1606 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1607 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1608 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1609 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1610 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1611 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1612 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1613 let isCommutable = Commutable;
1615 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1616 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1617 : N3V<0, 1, op21_20, op11_8, 1, 0,
1618 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1619 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1620 [(set (Ty DPR:$dst),
1621 (Ty (IntOp (Ty DPR:$src1),
1622 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1624 let isCommutable = 0;
1626 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1627 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1628 : N3V<0, 1, op21_20, op11_8, 1, 0,
1629 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1630 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1631 [(set (Ty DPR:$dst),
1632 (Ty (IntOp (Ty DPR:$src1),
1633 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1634 let isCommutable = 0;
1636 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1637 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1638 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1639 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1640 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1641 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1642 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1643 let isCommutable = 0;
1646 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1647 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1648 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1649 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1650 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1651 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1652 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1653 let isCommutable = Commutable;
1655 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1656 string OpcodeStr, string Dt,
1657 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1658 : N3V<1, 1, op21_20, op11_8, 1, 0,
1659 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1660 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1661 [(set (ResTy QPR:$dst),
1662 (ResTy (IntOp (ResTy QPR:$src1),
1663 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1665 let isCommutable = 0;
1667 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1668 string OpcodeStr, string Dt,
1669 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1670 : N3V<1, 1, op21_20, op11_8, 1, 0,
1671 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1672 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1673 [(set (ResTy QPR:$dst),
1674 (ResTy (IntOp (ResTy QPR:$src1),
1675 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1677 let isCommutable = 0;
1679 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1680 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1681 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1682 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1683 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1684 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1685 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1686 let isCommutable = 0;
1689 // Multiply-Add/Sub operations: single-, double- and quad-register.
1690 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1691 InstrItinClass itin, string OpcodeStr, string Dt,
1692 ValueType Ty, SDNode MulOp, SDNode OpNode>
1693 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1694 (outs DPR_VFP2:$dst),
1695 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1696 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1698 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1699 InstrItinClass itin, string OpcodeStr, string Dt,
1700 ValueType Ty, SDNode MulOp, SDNode OpNode>
1701 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1702 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1703 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1704 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1705 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1707 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1708 string OpcodeStr, string Dt,
1709 ValueType Ty, SDNode MulOp, SDNode ShOp>
1710 : N3V<0, 1, op21_20, op11_8, 1, 0,
1712 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1714 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1715 [(set (Ty DPR:$dst),
1716 (Ty (ShOp (Ty DPR:$src1),
1717 (Ty (MulOp DPR:$src2,
1718 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1720 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1721 string OpcodeStr, string Dt,
1722 ValueType Ty, SDNode MulOp, SDNode ShOp>
1723 : N3V<0, 1, op21_20, op11_8, 1, 0,
1725 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1727 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1729 (Ty (ShOp (Ty DPR:$src1),
1731 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1734 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1735 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1736 SDNode MulOp, SDNode OpNode>
1737 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1738 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1739 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1740 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1741 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1742 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1743 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1744 SDNode MulOp, SDNode ShOp>
1745 : N3V<1, 1, op21_20, op11_8, 1, 0,
1747 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1749 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1750 [(set (ResTy QPR:$dst),
1751 (ResTy (ShOp (ResTy QPR:$src1),
1752 (ResTy (MulOp QPR:$src2,
1753 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1755 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1756 string OpcodeStr, string Dt,
1757 ValueType ResTy, ValueType OpTy,
1758 SDNode MulOp, SDNode ShOp>
1759 : N3V<1, 1, op21_20, op11_8, 1, 0,
1761 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1763 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1764 [(set (ResTy QPR:$dst),
1765 (ResTy (ShOp (ResTy QPR:$src1),
1766 (ResTy (MulOp QPR:$src2,
1767 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1770 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1771 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1772 InstrItinClass itin, string OpcodeStr, string Dt,
1773 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1774 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1775 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1776 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1777 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1778 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1779 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1780 InstrItinClass itin, string OpcodeStr, string Dt,
1781 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1782 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1783 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1784 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1785 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1786 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1788 // Neon 3-argument intrinsics, both double- and quad-register.
1789 // The destination register is also used as the first source operand register.
1790 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1791 InstrItinClass itin, string OpcodeStr, string Dt,
1792 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1793 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1794 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1795 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1796 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1797 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1798 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1799 InstrItinClass itin, string OpcodeStr, string Dt,
1800 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1801 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1802 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1803 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1804 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1805 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1807 // Long Multiply-Add/Sub operations.
1808 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1809 InstrItinClass itin, string OpcodeStr, string Dt,
1810 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1811 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1812 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1813 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1814 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1815 (TyQ (MulOp (TyD DPR:$Vn),
1816 (TyD DPR:$Vm)))))]>;
1817 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1818 InstrItinClass itin, string OpcodeStr, string Dt,
1819 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1820 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1821 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1823 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1825 (OpNode (TyQ QPR:$src1),
1826 (TyQ (MulOp (TyD DPR:$src2),
1827 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1829 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1830 InstrItinClass itin, string OpcodeStr, string Dt,
1831 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1832 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1833 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1835 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1837 (OpNode (TyQ QPR:$src1),
1838 (TyQ (MulOp (TyD DPR:$src2),
1839 (TyD (NEONvduplane (TyD DPR_8:$src3),
1842 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1843 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1844 InstrItinClass itin, string OpcodeStr, string Dt,
1845 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1847 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1848 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1849 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1850 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1851 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1852 (TyD DPR:$Vm)))))))]>;
1854 // Neon Long 3-argument intrinsic. The destination register is
1855 // a quad-register and is also used as the first source operand register.
1856 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1857 InstrItinClass itin, string OpcodeStr, string Dt,
1858 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1859 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1860 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1861 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1863 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1864 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1865 string OpcodeStr, string Dt,
1866 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1867 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1869 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1871 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1872 [(set (ResTy QPR:$dst),
1873 (ResTy (IntOp (ResTy QPR:$src1),
1875 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1877 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1878 InstrItinClass itin, string OpcodeStr, string Dt,
1879 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1880 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1882 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1884 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1885 [(set (ResTy QPR:$dst),
1886 (ResTy (IntOp (ResTy QPR:$src1),
1888 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1891 // Narrowing 3-register intrinsics.
1892 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1893 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1894 Intrinsic IntOp, bit Commutable>
1895 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1896 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1897 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1898 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1899 let isCommutable = Commutable;
1902 // Long 3-register operations.
1903 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1904 InstrItinClass itin, string OpcodeStr, string Dt,
1905 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1906 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1907 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1908 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1909 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1910 let isCommutable = Commutable;
1912 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1913 InstrItinClass itin, string OpcodeStr, string Dt,
1914 ValueType TyQ, ValueType TyD, SDNode OpNode>
1915 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1916 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1917 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1919 (TyQ (OpNode (TyD DPR:$src1),
1920 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1921 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1922 InstrItinClass itin, string OpcodeStr, string Dt,
1923 ValueType TyQ, ValueType TyD, SDNode OpNode>
1924 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1925 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1926 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1928 (TyQ (OpNode (TyD DPR:$src1),
1929 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1931 // Long 3-register operations with explicitly extended operands.
1932 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1933 InstrItinClass itin, string OpcodeStr, string Dt,
1934 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1936 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1937 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
1938 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
1939 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
1940 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1941 let isCommutable = Commutable;
1944 // Long 3-register intrinsics with explicit extend (VABDL).
1945 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1946 InstrItinClass itin, string OpcodeStr, string Dt,
1947 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1949 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1950 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1951 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1952 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1953 (TyD DPR:$src2))))))]> {
1954 let isCommutable = Commutable;
1957 // Long 3-register intrinsics.
1958 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1959 InstrItinClass itin, string OpcodeStr, string Dt,
1960 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1961 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1962 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1963 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1964 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1965 let isCommutable = Commutable;
1967 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1968 string OpcodeStr, string Dt,
1969 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1970 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1971 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1972 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1973 [(set (ResTy QPR:$dst),
1974 (ResTy (IntOp (OpTy DPR:$src1),
1975 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1977 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1978 InstrItinClass itin, string OpcodeStr, string Dt,
1979 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1980 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1981 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1982 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1983 [(set (ResTy QPR:$dst),
1984 (ResTy (IntOp (OpTy DPR:$src1),
1985 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1988 // Wide 3-register operations.
1989 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1990 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1991 SDNode OpNode, SDNode ExtOp, bit Commutable>
1992 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1993 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
1994 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
1995 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
1996 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1997 let isCommutable = Commutable;
2000 // Pairwise long 2-register intrinsics, both double- and quad-register.
2001 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2002 bits<2> op17_16, bits<5> op11_7, bit op4,
2003 string OpcodeStr, string Dt,
2004 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2005 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
2006 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2007 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
2008 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2009 bits<2> op17_16, bits<5> op11_7, bit op4,
2010 string OpcodeStr, string Dt,
2011 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2012 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
2013 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
2014 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
2016 // Pairwise long 2-register accumulate intrinsics,
2017 // both double- and quad-register.
2018 // The destination register is also used as the first source operand register.
2019 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2020 bits<2> op17_16, bits<5> op11_7, bit op4,
2021 string OpcodeStr, string Dt,
2022 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2023 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2024 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2025 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2026 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2027 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2028 bits<2> op17_16, bits<5> op11_7, bit op4,
2029 string OpcodeStr, string Dt,
2030 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2031 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2032 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2033 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2034 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2036 // Shift by immediate,
2037 // both double- and quad-register.
2038 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2039 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2040 ValueType Ty, SDNode OpNode>
2041 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2042 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
2043 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2044 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
2045 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2046 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2047 ValueType Ty, SDNode OpNode>
2048 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2049 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
2050 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2051 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
2053 // Long shift by immediate.
2054 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2055 string OpcodeStr, string Dt,
2056 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2057 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2058 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
2059 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2060 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
2061 (i32 imm:$SIMM))))]>;
2063 // Narrow shift by immediate.
2064 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2065 InstrItinClass itin, string OpcodeStr, string Dt,
2066 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2067 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2068 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
2069 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
2070 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
2071 (i32 imm:$SIMM))))]>;
2073 // Shift right by immediate and accumulate,
2074 // both double- and quad-register.
2075 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2076 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2077 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2078 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2079 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2080 [(set DPR:$Vd, (Ty (add DPR:$src1,
2081 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2082 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2083 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2084 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2085 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2086 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2087 [(set QPR:$Vd, (Ty (add QPR:$src1,
2088 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2090 // Shift by immediate and insert,
2091 // both double- and quad-register.
2092 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2093 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2094 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2095 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
2096 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2097 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2098 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2099 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
2100 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2101 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
2102 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2103 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2105 // Convert, with fractional bits immediate,
2106 // both double- and quad-register.
2107 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2108 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2110 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2111 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2112 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2113 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2114 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2115 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2117 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2118 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2119 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2120 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2122 //===----------------------------------------------------------------------===//
2124 //===----------------------------------------------------------------------===//
2126 // Abbreviations used in multiclass suffixes:
2127 // Q = quarter int (8 bit) elements
2128 // H = half int (16 bit) elements
2129 // S = single int (32 bit) elements
2130 // D = double int (64 bit) elements
2132 // Neon 2-register vector operations -- for disassembly only.
2134 // First with only element sizes of 8, 16 and 32 bits:
2135 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2136 bits<5> op11_7, bit op4, string opc, string Dt,
2138 // 64-bit vector types.
2139 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2140 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2141 opc, !strconcat(Dt, "8"), asm, "", []>;
2142 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2143 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2144 opc, !strconcat(Dt, "16"), asm, "", []>;
2145 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2146 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2147 opc, !strconcat(Dt, "32"), asm, "", []>;
2148 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2149 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2150 opc, "f32", asm, "", []> {
2151 let Inst{10} = 1; // overwrite F = 1
2154 // 128-bit vector types.
2155 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2156 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2157 opc, !strconcat(Dt, "8"), asm, "", []>;
2158 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2159 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2160 opc, !strconcat(Dt, "16"), asm, "", []>;
2161 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2162 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2163 opc, !strconcat(Dt, "32"), asm, "", []>;
2164 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2165 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2166 opc, "f32", asm, "", []> {
2167 let Inst{10} = 1; // overwrite F = 1
2171 // Neon 3-register vector operations.
2173 // First with only element sizes of 8, 16 and 32 bits:
2174 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2175 InstrItinClass itinD16, InstrItinClass itinD32,
2176 InstrItinClass itinQ16, InstrItinClass itinQ32,
2177 string OpcodeStr, string Dt,
2178 SDNode OpNode, bit Commutable = 0> {
2179 // 64-bit vector types.
2180 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2181 OpcodeStr, !strconcat(Dt, "8"),
2182 v8i8, v8i8, OpNode, Commutable>;
2183 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2184 OpcodeStr, !strconcat(Dt, "16"),
2185 v4i16, v4i16, OpNode, Commutable>;
2186 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2187 OpcodeStr, !strconcat(Dt, "32"),
2188 v2i32, v2i32, OpNode, Commutable>;
2190 // 128-bit vector types.
2191 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2192 OpcodeStr, !strconcat(Dt, "8"),
2193 v16i8, v16i8, OpNode, Commutable>;
2194 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2195 OpcodeStr, !strconcat(Dt, "16"),
2196 v8i16, v8i16, OpNode, Commutable>;
2197 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2198 OpcodeStr, !strconcat(Dt, "32"),
2199 v4i32, v4i32, OpNode, Commutable>;
2202 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2203 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2205 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2207 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2208 v8i16, v4i16, ShOp>;
2209 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2210 v4i32, v2i32, ShOp>;
2213 // ....then also with element size 64 bits:
2214 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2215 InstrItinClass itinD, InstrItinClass itinQ,
2216 string OpcodeStr, string Dt,
2217 SDNode OpNode, bit Commutable = 0>
2218 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2219 OpcodeStr, Dt, OpNode, Commutable> {
2220 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2221 OpcodeStr, !strconcat(Dt, "64"),
2222 v1i64, v1i64, OpNode, Commutable>;
2223 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2224 OpcodeStr, !strconcat(Dt, "64"),
2225 v2i64, v2i64, OpNode, Commutable>;
2229 // Neon Narrowing 2-register vector operations,
2230 // source operand element sizes of 16, 32 and 64 bits:
2231 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2232 bits<5> op11_7, bit op6, bit op4,
2233 InstrItinClass itin, string OpcodeStr, string Dt,
2235 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2236 itin, OpcodeStr, !strconcat(Dt, "16"),
2237 v8i8, v8i16, OpNode>;
2238 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2239 itin, OpcodeStr, !strconcat(Dt, "32"),
2240 v4i16, v4i32, OpNode>;
2241 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2242 itin, OpcodeStr, !strconcat(Dt, "64"),
2243 v2i32, v2i64, OpNode>;
2246 // Neon Narrowing 2-register vector intrinsics,
2247 // source operand element sizes of 16, 32 and 64 bits:
2248 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2249 bits<5> op11_7, bit op6, bit op4,
2250 InstrItinClass itin, string OpcodeStr, string Dt,
2252 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2253 itin, OpcodeStr, !strconcat(Dt, "16"),
2254 v8i8, v8i16, IntOp>;
2255 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2256 itin, OpcodeStr, !strconcat(Dt, "32"),
2257 v4i16, v4i32, IntOp>;
2258 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2259 itin, OpcodeStr, !strconcat(Dt, "64"),
2260 v2i32, v2i64, IntOp>;
2264 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2265 // source operand element sizes of 16, 32 and 64 bits:
2266 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2267 string OpcodeStr, string Dt, SDNode OpNode> {
2268 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2269 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2270 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2271 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2272 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2273 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2277 // Neon 3-register vector intrinsics.
2279 // First with only element sizes of 16 and 32 bits:
2280 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2281 InstrItinClass itinD16, InstrItinClass itinD32,
2282 InstrItinClass itinQ16, InstrItinClass itinQ32,
2283 string OpcodeStr, string Dt,
2284 Intrinsic IntOp, bit Commutable = 0> {
2285 // 64-bit vector types.
2286 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2287 OpcodeStr, !strconcat(Dt, "16"),
2288 v4i16, v4i16, IntOp, Commutable>;
2289 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2290 OpcodeStr, !strconcat(Dt, "32"),
2291 v2i32, v2i32, IntOp, Commutable>;
2293 // 128-bit vector types.
2294 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2295 OpcodeStr, !strconcat(Dt, "16"),
2296 v8i16, v8i16, IntOp, Commutable>;
2297 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2298 OpcodeStr, !strconcat(Dt, "32"),
2299 v4i32, v4i32, IntOp, Commutable>;
2301 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2302 InstrItinClass itinD16, InstrItinClass itinD32,
2303 InstrItinClass itinQ16, InstrItinClass itinQ32,
2304 string OpcodeStr, string Dt,
2306 // 64-bit vector types.
2307 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2308 OpcodeStr, !strconcat(Dt, "16"),
2309 v4i16, v4i16, IntOp>;
2310 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2311 OpcodeStr, !strconcat(Dt, "32"),
2312 v2i32, v2i32, IntOp>;
2314 // 128-bit vector types.
2315 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2316 OpcodeStr, !strconcat(Dt, "16"),
2317 v8i16, v8i16, IntOp>;
2318 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2319 OpcodeStr, !strconcat(Dt, "32"),
2320 v4i32, v4i32, IntOp>;
2323 multiclass N3VIntSL_HS<bits<4> op11_8,
2324 InstrItinClass itinD16, InstrItinClass itinD32,
2325 InstrItinClass itinQ16, InstrItinClass itinQ32,
2326 string OpcodeStr, string Dt, Intrinsic IntOp> {
2327 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2328 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2329 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2330 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2331 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2332 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2333 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2334 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2337 // ....then also with element size of 8 bits:
2338 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2339 InstrItinClass itinD16, InstrItinClass itinD32,
2340 InstrItinClass itinQ16, InstrItinClass itinQ32,
2341 string OpcodeStr, string Dt,
2342 Intrinsic IntOp, bit Commutable = 0>
2343 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2344 OpcodeStr, Dt, IntOp, Commutable> {
2345 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2346 OpcodeStr, !strconcat(Dt, "8"),
2347 v8i8, v8i8, IntOp, Commutable>;
2348 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2349 OpcodeStr, !strconcat(Dt, "8"),
2350 v16i8, v16i8, IntOp, Commutable>;
2352 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2353 InstrItinClass itinD16, InstrItinClass itinD32,
2354 InstrItinClass itinQ16, InstrItinClass itinQ32,
2355 string OpcodeStr, string Dt,
2357 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2358 OpcodeStr, Dt, IntOp> {
2359 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2360 OpcodeStr, !strconcat(Dt, "8"),
2362 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2363 OpcodeStr, !strconcat(Dt, "8"),
2364 v16i8, v16i8, IntOp>;
2368 // ....then also with element size of 64 bits:
2369 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2370 InstrItinClass itinD16, InstrItinClass itinD32,
2371 InstrItinClass itinQ16, InstrItinClass itinQ32,
2372 string OpcodeStr, string Dt,
2373 Intrinsic IntOp, bit Commutable = 0>
2374 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2375 OpcodeStr, Dt, IntOp, Commutable> {
2376 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2377 OpcodeStr, !strconcat(Dt, "64"),
2378 v1i64, v1i64, IntOp, Commutable>;
2379 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2380 OpcodeStr, !strconcat(Dt, "64"),
2381 v2i64, v2i64, IntOp, Commutable>;
2383 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2384 InstrItinClass itinD16, InstrItinClass itinD32,
2385 InstrItinClass itinQ16, InstrItinClass itinQ32,
2386 string OpcodeStr, string Dt,
2388 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2389 OpcodeStr, Dt, IntOp> {
2390 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2391 OpcodeStr, !strconcat(Dt, "64"),
2392 v1i64, v1i64, IntOp>;
2393 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2394 OpcodeStr, !strconcat(Dt, "64"),
2395 v2i64, v2i64, IntOp>;
2398 // Neon Narrowing 3-register vector intrinsics,
2399 // source operand element sizes of 16, 32 and 64 bits:
2400 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2401 string OpcodeStr, string Dt,
2402 Intrinsic IntOp, bit Commutable = 0> {
2403 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2404 OpcodeStr, !strconcat(Dt, "16"),
2405 v8i8, v8i16, IntOp, Commutable>;
2406 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2407 OpcodeStr, !strconcat(Dt, "32"),
2408 v4i16, v4i32, IntOp, Commutable>;
2409 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2410 OpcodeStr, !strconcat(Dt, "64"),
2411 v2i32, v2i64, IntOp, Commutable>;
2415 // Neon Long 3-register vector operations.
2417 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2418 InstrItinClass itin16, InstrItinClass itin32,
2419 string OpcodeStr, string Dt,
2420 SDNode OpNode, bit Commutable = 0> {
2421 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2422 OpcodeStr, !strconcat(Dt, "8"),
2423 v8i16, v8i8, OpNode, Commutable>;
2424 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2425 OpcodeStr, !strconcat(Dt, "16"),
2426 v4i32, v4i16, OpNode, Commutable>;
2427 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2428 OpcodeStr, !strconcat(Dt, "32"),
2429 v2i64, v2i32, OpNode, Commutable>;
2432 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2433 InstrItinClass itin, string OpcodeStr, string Dt,
2435 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2436 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2437 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2438 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2441 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2442 InstrItinClass itin16, InstrItinClass itin32,
2443 string OpcodeStr, string Dt,
2444 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2445 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2446 OpcodeStr, !strconcat(Dt, "8"),
2447 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2448 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2449 OpcodeStr, !strconcat(Dt, "16"),
2450 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2451 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2452 OpcodeStr, !strconcat(Dt, "32"),
2453 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2456 // Neon Long 3-register vector intrinsics.
2458 // First with only element sizes of 16 and 32 bits:
2459 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2460 InstrItinClass itin16, InstrItinClass itin32,
2461 string OpcodeStr, string Dt,
2462 Intrinsic IntOp, bit Commutable = 0> {
2463 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2464 OpcodeStr, !strconcat(Dt, "16"),
2465 v4i32, v4i16, IntOp, Commutable>;
2466 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2467 OpcodeStr, !strconcat(Dt, "32"),
2468 v2i64, v2i32, IntOp, Commutable>;
2471 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2472 InstrItinClass itin, string OpcodeStr, string Dt,
2474 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2475 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2476 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2477 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2480 // ....then also with element size of 8 bits:
2481 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2482 InstrItinClass itin16, InstrItinClass itin32,
2483 string OpcodeStr, string Dt,
2484 Intrinsic IntOp, bit Commutable = 0>
2485 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2486 IntOp, Commutable> {
2487 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2488 OpcodeStr, !strconcat(Dt, "8"),
2489 v8i16, v8i8, IntOp, Commutable>;
2492 // ....with explicit extend (VABDL).
2493 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2494 InstrItinClass itin, string OpcodeStr, string Dt,
2495 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2496 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2497 OpcodeStr, !strconcat(Dt, "8"),
2498 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2499 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2500 OpcodeStr, !strconcat(Dt, "16"),
2501 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2502 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2503 OpcodeStr, !strconcat(Dt, "32"),
2504 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2508 // Neon Wide 3-register vector intrinsics,
2509 // source operand element sizes of 8, 16 and 32 bits:
2510 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2511 string OpcodeStr, string Dt,
2512 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2513 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2514 OpcodeStr, !strconcat(Dt, "8"),
2515 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2516 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2517 OpcodeStr, !strconcat(Dt, "16"),
2518 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2519 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2520 OpcodeStr, !strconcat(Dt, "32"),
2521 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2525 // Neon Multiply-Op vector operations,
2526 // element sizes of 8, 16 and 32 bits:
2527 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2528 InstrItinClass itinD16, InstrItinClass itinD32,
2529 InstrItinClass itinQ16, InstrItinClass itinQ32,
2530 string OpcodeStr, string Dt, SDNode OpNode> {
2531 // 64-bit vector types.
2532 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2533 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2534 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2535 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2536 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2537 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2539 // 128-bit vector types.
2540 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2541 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2542 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2543 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2544 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2545 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2548 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2549 InstrItinClass itinD16, InstrItinClass itinD32,
2550 InstrItinClass itinQ16, InstrItinClass itinQ32,
2551 string OpcodeStr, string Dt, SDNode ShOp> {
2552 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2553 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2554 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2555 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2556 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2557 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2559 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2560 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2564 // Neon Intrinsic-Op vector operations,
2565 // element sizes of 8, 16 and 32 bits:
2566 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2567 InstrItinClass itinD, InstrItinClass itinQ,
2568 string OpcodeStr, string Dt, Intrinsic IntOp,
2570 // 64-bit vector types.
2571 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2572 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2573 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2574 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2575 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2576 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2578 // 128-bit vector types.
2579 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2580 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2581 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2582 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2583 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2584 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2587 // Neon 3-argument intrinsics,
2588 // element sizes of 8, 16 and 32 bits:
2589 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2590 InstrItinClass itinD, InstrItinClass itinQ,
2591 string OpcodeStr, string Dt, Intrinsic IntOp> {
2592 // 64-bit vector types.
2593 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2594 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2595 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2596 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2597 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2598 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2600 // 128-bit vector types.
2601 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2602 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2603 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2604 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2605 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2606 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2610 // Neon Long Multiply-Op vector operations,
2611 // element sizes of 8, 16 and 32 bits:
2612 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2613 InstrItinClass itin16, InstrItinClass itin32,
2614 string OpcodeStr, string Dt, SDNode MulOp,
2616 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2617 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2618 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2619 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2620 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2621 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2624 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2625 string Dt, SDNode MulOp, SDNode OpNode> {
2626 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2627 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2628 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2629 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2633 // Neon Long 3-argument intrinsics.
2635 // First with only element sizes of 16 and 32 bits:
2636 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2637 InstrItinClass itin16, InstrItinClass itin32,
2638 string OpcodeStr, string Dt, Intrinsic IntOp> {
2639 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2640 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2641 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2642 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2645 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2646 string OpcodeStr, string Dt, Intrinsic IntOp> {
2647 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2648 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2649 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2650 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2653 // ....then also with element size of 8 bits:
2654 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2655 InstrItinClass itin16, InstrItinClass itin32,
2656 string OpcodeStr, string Dt, Intrinsic IntOp>
2657 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2658 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2659 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2662 // ....with explicit extend (VABAL).
2663 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2664 InstrItinClass itin, string OpcodeStr, string Dt,
2665 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2666 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2667 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2668 IntOp, ExtOp, OpNode>;
2669 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2670 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2671 IntOp, ExtOp, OpNode>;
2672 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2673 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2674 IntOp, ExtOp, OpNode>;
2678 // Neon 2-register vector intrinsics,
2679 // element sizes of 8, 16 and 32 bits:
2680 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2681 bits<5> op11_7, bit op4,
2682 InstrItinClass itinD, InstrItinClass itinQ,
2683 string OpcodeStr, string Dt, Intrinsic IntOp> {
2684 // 64-bit vector types.
2685 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2686 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2687 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2688 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2689 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2690 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2692 // 128-bit vector types.
2693 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2694 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2695 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2696 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2697 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2698 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2702 // Neon Pairwise long 2-register intrinsics,
2703 // element sizes of 8, 16 and 32 bits:
2704 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2705 bits<5> op11_7, bit op4,
2706 string OpcodeStr, string Dt, Intrinsic IntOp> {
2707 // 64-bit vector types.
2708 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2709 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2710 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2711 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2712 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2713 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2715 // 128-bit vector types.
2716 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2717 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2718 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2719 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2720 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2721 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2725 // Neon Pairwise long 2-register accumulate intrinsics,
2726 // element sizes of 8, 16 and 32 bits:
2727 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2728 bits<5> op11_7, bit op4,
2729 string OpcodeStr, string Dt, Intrinsic IntOp> {
2730 // 64-bit vector types.
2731 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2732 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2733 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2734 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2735 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2736 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2738 // 128-bit vector types.
2739 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2740 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2741 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2742 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2743 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2744 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2748 // Neon 2-register vector shift by immediate,
2749 // with f of either N2RegVShLFrm or N2RegVShRFrm
2750 // element sizes of 8, 16, 32 and 64 bits:
2751 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2752 InstrItinClass itin, string OpcodeStr, string Dt,
2753 SDNode OpNode, Format f> {
2754 // 64-bit vector types.
2755 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2756 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2757 let Inst{21-19} = 0b001; // imm6 = 001xxx
2759 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2760 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2761 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2763 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2764 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2765 let Inst{21} = 0b1; // imm6 = 1xxxxx
2767 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2768 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2771 // 128-bit vector types.
2772 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2773 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2774 let Inst{21-19} = 0b001; // imm6 = 001xxx
2776 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2777 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2778 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2780 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2781 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2782 let Inst{21} = 0b1; // imm6 = 1xxxxx
2784 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2785 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2789 // Neon Shift-Accumulate vector operations,
2790 // element sizes of 8, 16, 32 and 64 bits:
2791 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2792 string OpcodeStr, string Dt, SDNode ShOp> {
2793 // 64-bit vector types.
2794 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2795 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2796 let Inst{21-19} = 0b001; // imm6 = 001xxx
2798 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2799 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2800 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2802 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2803 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2804 let Inst{21} = 0b1; // imm6 = 1xxxxx
2806 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2807 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2810 // 128-bit vector types.
2811 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2812 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2813 let Inst{21-19} = 0b001; // imm6 = 001xxx
2815 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2816 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2817 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2819 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2820 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2821 let Inst{21} = 0b1; // imm6 = 1xxxxx
2823 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2824 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2829 // Neon Shift-Insert vector operations,
2830 // with f of either N2RegVShLFrm or N2RegVShRFrm
2831 // element sizes of 8, 16, 32 and 64 bits:
2832 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2833 string OpcodeStr, SDNode ShOp,
2835 // 64-bit vector types.
2836 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2837 f, OpcodeStr, "8", v8i8, ShOp> {
2838 let Inst{21-19} = 0b001; // imm6 = 001xxx
2840 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2841 f, OpcodeStr, "16", v4i16, ShOp> {
2842 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2844 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2845 f, OpcodeStr, "32", v2i32, ShOp> {
2846 let Inst{21} = 0b1; // imm6 = 1xxxxx
2848 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2849 f, OpcodeStr, "64", v1i64, ShOp>;
2852 // 128-bit vector types.
2853 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2854 f, OpcodeStr, "8", v16i8, ShOp> {
2855 let Inst{21-19} = 0b001; // imm6 = 001xxx
2857 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2858 f, OpcodeStr, "16", v8i16, ShOp> {
2859 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2861 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2862 f, OpcodeStr, "32", v4i32, ShOp> {
2863 let Inst{21} = 0b1; // imm6 = 1xxxxx
2865 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2866 f, OpcodeStr, "64", v2i64, ShOp>;
2870 // Neon Shift Long operations,
2871 // element sizes of 8, 16, 32 bits:
2872 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2873 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2874 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2875 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2876 let Inst{21-19} = 0b001; // imm6 = 001xxx
2878 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2879 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2880 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2882 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2883 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2884 let Inst{21} = 0b1; // imm6 = 1xxxxx
2888 // Neon Shift Narrow operations,
2889 // element sizes of 16, 32, 64 bits:
2890 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2891 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2893 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2894 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2895 let Inst{21-19} = 0b001; // imm6 = 001xxx
2897 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2898 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2899 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2901 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2902 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2903 let Inst{21} = 0b1; // imm6 = 1xxxxx
2907 //===----------------------------------------------------------------------===//
2908 // Instruction Definitions.
2909 //===----------------------------------------------------------------------===//
2911 // Vector Add Operations.
2913 // VADD : Vector Add (integer and floating-point)
2914 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2916 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2917 v2f32, v2f32, fadd, 1>;
2918 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2919 v4f32, v4f32, fadd, 1>;
2920 // VADDL : Vector Add Long (Q = D + D)
2921 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2922 "vaddl", "s", add, sext, 1>;
2923 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2924 "vaddl", "u", add, zext, 1>;
2925 // VADDW : Vector Add Wide (Q = Q + D)
2926 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2927 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2928 // VHADD : Vector Halving Add
2929 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2930 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2931 "vhadd", "s", int_arm_neon_vhadds, 1>;
2932 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2933 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2934 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2935 // VRHADD : Vector Rounding Halving Add
2936 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2937 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2938 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2939 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2940 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2941 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2942 // VQADD : Vector Saturating Add
2943 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2944 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2945 "vqadd", "s", int_arm_neon_vqadds, 1>;
2946 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2947 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2948 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2949 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2950 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2951 int_arm_neon_vaddhn, 1>;
2952 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2953 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2954 int_arm_neon_vraddhn, 1>;
2956 // Vector Multiply Operations.
2958 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2959 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2960 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2961 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2962 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2963 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2964 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2965 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
2966 v2f32, v2f32, fmul, 1>;
2967 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
2968 v4f32, v4f32, fmul, 1>;
2969 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2970 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2971 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2974 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2975 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2976 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2977 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2978 (DSubReg_i16_reg imm:$lane))),
2979 (SubReg_i16_lane imm:$lane)))>;
2980 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2981 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2982 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2983 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2984 (DSubReg_i32_reg imm:$lane))),
2985 (SubReg_i32_lane imm:$lane)))>;
2986 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2987 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2988 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2989 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2990 (DSubReg_i32_reg imm:$lane))),
2991 (SubReg_i32_lane imm:$lane)))>;
2993 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2994 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2995 IIC_VMULi16Q, IIC_VMULi32Q,
2996 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2997 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2998 IIC_VMULi16Q, IIC_VMULi32Q,
2999 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3000 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3001 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3003 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3004 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3005 (DSubReg_i16_reg imm:$lane))),
3006 (SubReg_i16_lane imm:$lane)))>;
3007 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3008 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3010 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3011 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3012 (DSubReg_i32_reg imm:$lane))),
3013 (SubReg_i32_lane imm:$lane)))>;
3015 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3016 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3017 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3018 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3019 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3020 IIC_VMULi16Q, IIC_VMULi32Q,
3021 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3022 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3023 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3025 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3026 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3027 (DSubReg_i16_reg imm:$lane))),
3028 (SubReg_i16_lane imm:$lane)))>;
3029 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3030 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3032 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3033 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3034 (DSubReg_i32_reg imm:$lane))),
3035 (SubReg_i32_lane imm:$lane)))>;
3037 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3038 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3039 "vmull", "s", NEONvmulls, 1>;
3040 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3041 "vmull", "u", NEONvmullu, 1>;
3042 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3043 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3044 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3045 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3047 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3048 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3049 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3050 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3051 "vqdmull", "s", int_arm_neon_vqdmull>;
3053 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3055 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3056 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3057 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3058 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3060 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3062 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3063 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3064 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3066 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3067 v4f32, v2f32, fmul, fadd>;
3069 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3070 (mul (v8i16 QPR:$src2),
3071 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3072 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3073 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3074 (DSubReg_i16_reg imm:$lane))),
3075 (SubReg_i16_lane imm:$lane)))>;
3077 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3078 (mul (v4i32 QPR:$src2),
3079 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3080 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3081 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3082 (DSubReg_i32_reg imm:$lane))),
3083 (SubReg_i32_lane imm:$lane)))>;
3085 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
3086 (fmul (v4f32 QPR:$src2),
3087 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3088 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3090 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3091 (DSubReg_i32_reg imm:$lane))),
3092 (SubReg_i32_lane imm:$lane)))>;
3094 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3095 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3096 "vmlal", "s", NEONvmulls, add>;
3097 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3098 "vmlal", "u", NEONvmullu, add>;
3100 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3101 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3103 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3104 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3105 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3106 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3108 // VMLS : Vector Multiply Subtract (integer and floating-point)
3109 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3110 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3111 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3113 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3115 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3116 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3117 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3119 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3120 v4f32, v2f32, fmul, fsub>;
3122 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3123 (mul (v8i16 QPR:$src2),
3124 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3125 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3126 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3127 (DSubReg_i16_reg imm:$lane))),
3128 (SubReg_i16_lane imm:$lane)))>;
3130 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3131 (mul (v4i32 QPR:$src2),
3132 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3133 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3134 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3135 (DSubReg_i32_reg imm:$lane))),
3136 (SubReg_i32_lane imm:$lane)))>;
3138 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3139 (fmul (v4f32 QPR:$src2),
3140 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3141 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3142 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3143 (DSubReg_i32_reg imm:$lane))),
3144 (SubReg_i32_lane imm:$lane)))>;
3146 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3147 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3148 "vmlsl", "s", NEONvmulls, sub>;
3149 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3150 "vmlsl", "u", NEONvmullu, sub>;
3152 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3153 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3155 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3156 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3157 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3158 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3160 // Vector Subtract Operations.
3162 // VSUB : Vector Subtract (integer and floating-point)
3163 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3164 "vsub", "i", sub, 0>;
3165 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3166 v2f32, v2f32, fsub, 0>;
3167 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3168 v4f32, v4f32, fsub, 0>;
3169 // VSUBL : Vector Subtract Long (Q = D - D)
3170 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3171 "vsubl", "s", sub, sext, 0>;
3172 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3173 "vsubl", "u", sub, zext, 0>;
3174 // VSUBW : Vector Subtract Wide (Q = Q - D)
3175 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3176 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3177 // VHSUB : Vector Halving Subtract
3178 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3179 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3180 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3181 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3182 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3183 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3184 // VQSUB : Vector Saturing Subtract
3185 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3186 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3187 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3188 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3189 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3190 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3191 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3192 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3193 int_arm_neon_vsubhn, 0>;
3194 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3195 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3196 int_arm_neon_vrsubhn, 0>;
3198 // Vector Comparisons.
3200 // VCEQ : Vector Compare Equal
3201 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3202 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3203 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3205 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3207 // For disassembly only.
3208 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3211 // VCGE : Vector Compare Greater Than or Equal
3212 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3213 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3214 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3215 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3216 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3218 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3220 // For disassembly only.
3221 // FIXME: This instruction's encoding MAY NOT BE correct.
3222 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3224 // For disassembly only.
3225 // FIXME: This instruction's encoding MAY NOT BE correct.
3226 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3229 // VCGT : Vector Compare Greater Than
3230 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3231 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3232 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3233 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3234 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3236 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3238 // For disassembly only.
3239 // FIXME: This instruction's encoding MAY NOT BE correct.
3240 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3242 // For disassembly only.
3243 // FIXME: This instruction's encoding MAY NOT BE correct.
3244 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3247 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3248 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3249 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3250 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3251 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3252 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3253 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3254 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3255 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3256 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3257 // VTST : Vector Test Bits
3258 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3259 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3261 // Vector Bitwise Operations.
3263 def vnotd : PatFrag<(ops node:$in),
3264 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3265 def vnotq : PatFrag<(ops node:$in),
3266 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3269 // VAND : Vector Bitwise AND
3270 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3271 v2i32, v2i32, and, 1>;
3272 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3273 v4i32, v4i32, and, 1>;
3275 // VEOR : Vector Bitwise Exclusive OR
3276 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3277 v2i32, v2i32, xor, 1>;
3278 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3279 v4i32, v4i32, xor, 1>;
3281 // VORR : Vector Bitwise OR
3282 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3283 v2i32, v2i32, or, 1>;
3284 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3285 v4i32, v4i32, or, 1>;
3287 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3288 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3289 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3290 "vbic", "$dst, $src1, $src2", "",
3291 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3292 (vnotd DPR:$src2))))]>;
3293 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3294 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3295 "vbic", "$dst, $src1, $src2", "",
3296 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3297 (vnotq QPR:$src2))))]>;
3299 // VORN : Vector Bitwise OR NOT
3300 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3301 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3302 "vorn", "$dst, $src1, $src2", "",
3303 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3304 (vnotd DPR:$src2))))]>;
3305 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3306 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3307 "vorn", "$dst, $src1, $src2", "",
3308 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3309 (vnotq QPR:$src2))))]>;
3311 // VMVN : Vector Bitwise NOT (Immediate)
3313 let isReMaterializable = 1 in {
3315 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3316 (ins nModImm:$SIMM), IIC_VMOVImm,
3317 "vmvn", "i16", "$dst, $SIMM", "",
3318 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3319 let Inst{9} = SIMM{9};
3322 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3323 (ins nModImm:$SIMM), IIC_VMOVImm,
3324 "vmvn", "i16", "$dst, $SIMM", "",
3325 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3326 let Inst{9} = SIMM{9};
3329 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3330 (ins nModImm:$SIMM), IIC_VMOVImm,
3331 "vmvn", "i32", "$dst, $SIMM", "",
3332 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3333 let Inst{11-8} = SIMM{11-8};
3336 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3337 (ins nModImm:$SIMM), IIC_VMOVImm,
3338 "vmvn", "i32", "$dst, $SIMM", "",
3339 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3340 let Inst{11-8} = SIMM{11-8};
3344 // VMVN : Vector Bitwise NOT
3345 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3346 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3347 "vmvn", "$dst, $src", "",
3348 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3349 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3350 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3351 "vmvn", "$dst, $src", "",
3352 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3353 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3354 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3356 // VBSL : Vector Bitwise Select
3357 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3358 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3359 N3RegFrm, IIC_VCNTiD,
3360 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3362 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3363 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3364 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3365 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3366 N3RegFrm, IIC_VCNTiQ,
3367 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3369 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3370 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3372 // VBIF : Vector Bitwise Insert if False
3373 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3374 // FIXME: This instruction's encoding MAY NOT BE correct.
3375 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3376 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3377 N3RegFrm, IIC_VBINiD,
3378 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3379 [/* For disassembly only; pattern left blank */]>;
3380 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3381 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3382 N3RegFrm, IIC_VBINiQ,
3383 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3384 [/* For disassembly only; pattern left blank */]>;
3386 // VBIT : Vector Bitwise Insert if True
3387 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3388 // FIXME: This instruction's encoding MAY NOT BE correct.
3389 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3390 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3391 N3RegFrm, IIC_VBINiD,
3392 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3393 [/* For disassembly only; pattern left blank */]>;
3394 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3395 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3396 N3RegFrm, IIC_VBINiQ,
3397 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3398 [/* For disassembly only; pattern left blank */]>;
3400 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3401 // for equivalent operations with different register constraints; it just
3404 // Vector Absolute Differences.
3406 // VABD : Vector Absolute Difference
3407 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3408 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3409 "vabd", "s", int_arm_neon_vabds, 1>;
3410 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3411 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3412 "vabd", "u", int_arm_neon_vabdu, 1>;
3413 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3414 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3415 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3416 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3418 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3419 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3420 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3421 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3422 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3424 // VABA : Vector Absolute Difference and Accumulate
3425 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3426 "vaba", "s", int_arm_neon_vabds, add>;
3427 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3428 "vaba", "u", int_arm_neon_vabdu, add>;
3430 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3431 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3432 "vabal", "s", int_arm_neon_vabds, zext, add>;
3433 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3434 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3436 // Vector Maximum and Minimum.
3438 // VMAX : Vector Maximum
3439 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3440 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3441 "vmax", "s", int_arm_neon_vmaxs, 1>;
3442 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3443 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3444 "vmax", "u", int_arm_neon_vmaxu, 1>;
3445 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3447 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3448 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3450 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3452 // VMIN : Vector Minimum
3453 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3454 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3455 "vmin", "s", int_arm_neon_vmins, 1>;
3456 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3457 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3458 "vmin", "u", int_arm_neon_vminu, 1>;
3459 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3461 v2f32, v2f32, int_arm_neon_vmins, 1>;
3462 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3464 v4f32, v4f32, int_arm_neon_vmins, 1>;
3466 // Vector Pairwise Operations.
3468 // VPADD : Vector Pairwise Add
3469 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3471 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3472 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3474 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3475 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3477 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3478 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3479 IIC_VPBIND, "vpadd", "f32",
3480 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3482 // VPADDL : Vector Pairwise Add Long
3483 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3484 int_arm_neon_vpaddls>;
3485 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3486 int_arm_neon_vpaddlu>;
3488 // VPADAL : Vector Pairwise Add and Accumulate Long
3489 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3490 int_arm_neon_vpadals>;
3491 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3492 int_arm_neon_vpadalu>;
3494 // VPMAX : Vector Pairwise Maximum
3495 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3496 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3497 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3498 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3499 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3500 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3501 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3502 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3503 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3504 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3505 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3506 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3507 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3508 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3510 // VPMIN : Vector Pairwise Minimum
3511 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3512 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3513 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3514 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3515 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3516 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3517 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3518 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3519 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3520 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3521 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3522 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3523 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3524 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3526 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3528 // VRECPE : Vector Reciprocal Estimate
3529 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3530 IIC_VUNAD, "vrecpe", "u32",
3531 v2i32, v2i32, int_arm_neon_vrecpe>;
3532 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3533 IIC_VUNAQ, "vrecpe", "u32",
3534 v4i32, v4i32, int_arm_neon_vrecpe>;
3535 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3536 IIC_VUNAD, "vrecpe", "f32",
3537 v2f32, v2f32, int_arm_neon_vrecpe>;
3538 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3539 IIC_VUNAQ, "vrecpe", "f32",
3540 v4f32, v4f32, int_arm_neon_vrecpe>;
3542 // VRECPS : Vector Reciprocal Step
3543 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3544 IIC_VRECSD, "vrecps", "f32",
3545 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3546 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3547 IIC_VRECSQ, "vrecps", "f32",
3548 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3550 // VRSQRTE : Vector Reciprocal Square Root Estimate
3551 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3552 IIC_VUNAD, "vrsqrte", "u32",
3553 v2i32, v2i32, int_arm_neon_vrsqrte>;
3554 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3555 IIC_VUNAQ, "vrsqrte", "u32",
3556 v4i32, v4i32, int_arm_neon_vrsqrte>;
3557 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3558 IIC_VUNAD, "vrsqrte", "f32",
3559 v2f32, v2f32, int_arm_neon_vrsqrte>;
3560 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3561 IIC_VUNAQ, "vrsqrte", "f32",
3562 v4f32, v4f32, int_arm_neon_vrsqrte>;
3564 // VRSQRTS : Vector Reciprocal Square Root Step
3565 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3566 IIC_VRECSD, "vrsqrts", "f32",
3567 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3568 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3569 IIC_VRECSQ, "vrsqrts", "f32",
3570 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3574 // VSHL : Vector Shift
3575 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3576 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3577 "vshl", "s", int_arm_neon_vshifts>;
3578 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3579 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3580 "vshl", "u", int_arm_neon_vshiftu>;
3581 // VSHL : Vector Shift Left (Immediate)
3582 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3584 // VSHR : Vector Shift Right (Immediate)
3585 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3587 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3590 // VSHLL : Vector Shift Left Long
3591 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3592 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3594 // VSHLL : Vector Shift Left Long (with maximum shift count)
3595 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3596 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3597 ValueType OpTy, SDNode OpNode>
3598 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3599 ResTy, OpTy, OpNode> {
3600 let Inst{21-16} = op21_16;
3602 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3603 v8i16, v8i8, NEONvshlli>;
3604 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3605 v4i32, v4i16, NEONvshlli>;
3606 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3607 v2i64, v2i32, NEONvshlli>;
3609 // VSHRN : Vector Shift Right and Narrow
3610 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3613 // VRSHL : Vector Rounding Shift
3614 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3615 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3616 "vrshl", "s", int_arm_neon_vrshifts>;
3617 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3618 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3619 "vrshl", "u", int_arm_neon_vrshiftu>;
3620 // VRSHR : Vector Rounding Shift Right
3621 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3623 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3626 // VRSHRN : Vector Rounding Shift Right and Narrow
3627 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3630 // VQSHL : Vector Saturating Shift
3631 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3632 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3633 "vqshl", "s", int_arm_neon_vqshifts>;
3634 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3635 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3636 "vqshl", "u", int_arm_neon_vqshiftu>;
3637 // VQSHL : Vector Saturating Shift Left (Immediate)
3638 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3640 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3642 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3643 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3646 // VQSHRN : Vector Saturating Shift Right and Narrow
3647 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3649 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3652 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3653 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3656 // VQRSHL : Vector Saturating Rounding Shift
3657 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3658 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3659 "vqrshl", "s", int_arm_neon_vqrshifts>;
3660 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3661 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3662 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3664 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3665 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3667 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3670 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3671 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3674 // VSRA : Vector Shift Right and Accumulate
3675 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3676 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3677 // VRSRA : Vector Rounding Shift Right and Accumulate
3678 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3679 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3681 // VSLI : Vector Shift Left and Insert
3682 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3683 // VSRI : Vector Shift Right and Insert
3684 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3686 // Vector Absolute and Saturating Absolute.
3688 // VABS : Vector Absolute Value
3689 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3690 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3692 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3693 IIC_VUNAD, "vabs", "f32",
3694 v2f32, v2f32, int_arm_neon_vabs>;
3695 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3696 IIC_VUNAQ, "vabs", "f32",
3697 v4f32, v4f32, int_arm_neon_vabs>;
3699 // VQABS : Vector Saturating Absolute Value
3700 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3701 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3702 int_arm_neon_vqabs>;
3706 def vnegd : PatFrag<(ops node:$in),
3707 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3708 def vnegq : PatFrag<(ops node:$in),
3709 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3711 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3712 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3713 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3714 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3715 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3716 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3717 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3718 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3720 // VNEG : Vector Negate (integer)
3721 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3722 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3723 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3724 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3725 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3726 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3728 // VNEG : Vector Negate (floating-point)
3729 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3730 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3731 "vneg", "f32", "$dst, $src", "",
3732 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3733 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3734 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3735 "vneg", "f32", "$dst, $src", "",
3736 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3738 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3739 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3740 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3741 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3742 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3743 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3745 // VQNEG : Vector Saturating Negate
3746 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3747 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3748 int_arm_neon_vqneg>;
3750 // Vector Bit Counting Operations.
3752 // VCLS : Vector Count Leading Sign Bits
3753 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3754 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3756 // VCLZ : Vector Count Leading Zeros
3757 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3758 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3760 // VCNT : Vector Count One Bits
3761 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3762 IIC_VCNTiD, "vcnt", "8",
3763 v8i8, v8i8, int_arm_neon_vcnt>;
3764 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3765 IIC_VCNTiQ, "vcnt", "8",
3766 v16i8, v16i8, int_arm_neon_vcnt>;
3768 // Vector Swap -- for disassembly only.
3769 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3770 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3771 "vswp", "$dst, $src", "", []>;
3772 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3773 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3774 "vswp", "$dst, $src", "", []>;
3776 // Vector Move Operations.
3778 // VMOV : Vector Move (Register)
3780 let neverHasSideEffects = 1 in {
3781 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3782 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3783 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3784 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3786 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3787 // be expanded after register allocation is completed.
3788 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3789 NoItinerary, "", []>;
3791 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3792 NoItinerary, "", []>;
3793 } // neverHasSideEffects
3795 // VMOV : Vector Move (Immediate)
3797 let isReMaterializable = 1 in {
3798 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3799 (ins nModImm:$SIMM), IIC_VMOVImm,
3800 "vmov", "i8", "$dst, $SIMM", "",
3801 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3802 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3803 (ins nModImm:$SIMM), IIC_VMOVImm,
3804 "vmov", "i8", "$dst, $SIMM", "",
3805 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3807 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3808 (ins nModImm:$SIMM), IIC_VMOVImm,
3809 "vmov", "i16", "$dst, $SIMM", "",
3810 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3811 let Inst{9} = SIMM{9};
3814 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3815 (ins nModImm:$SIMM), IIC_VMOVImm,
3816 "vmov", "i16", "$dst, $SIMM", "",
3817 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3818 let Inst{9} = SIMM{9};
3821 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3822 (ins nModImm:$SIMM), IIC_VMOVImm,
3823 "vmov", "i32", "$dst, $SIMM", "",
3824 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3825 let Inst{11-8} = SIMM{11-8};
3828 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3829 (ins nModImm:$SIMM), IIC_VMOVImm,
3830 "vmov", "i32", "$dst, $SIMM", "",
3831 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
3832 let Inst{11-8} = SIMM{11-8};
3835 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3836 (ins nModImm:$SIMM), IIC_VMOVImm,
3837 "vmov", "i64", "$dst, $SIMM", "",
3838 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3839 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3840 (ins nModImm:$SIMM), IIC_VMOVImm,
3841 "vmov", "i64", "$dst, $SIMM", "",
3842 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3843 } // isReMaterializable
3845 // VMOV : Vector Get Lane (move scalar to ARM core register)
3847 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3848 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3849 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
3850 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
3852 let Inst{21} = lane{2};
3853 let Inst{6-5} = lane{1-0};
3855 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3856 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3857 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
3858 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
3860 let Inst{21} = lane{1};
3861 let Inst{6} = lane{0};
3863 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3864 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3865 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
3866 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
3868 let Inst{21} = lane{2};
3869 let Inst{6-5} = lane{1-0};
3871 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3872 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3873 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
3874 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
3876 let Inst{21} = lane{1};
3877 let Inst{6} = lane{0};
3879 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3880 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3881 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
3882 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
3884 let Inst{21} = lane{0};
3886 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3887 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3888 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3889 (DSubReg_i8_reg imm:$lane))),
3890 (SubReg_i8_lane imm:$lane))>;
3891 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3892 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3893 (DSubReg_i16_reg imm:$lane))),
3894 (SubReg_i16_lane imm:$lane))>;
3895 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3896 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3897 (DSubReg_i8_reg imm:$lane))),
3898 (SubReg_i8_lane imm:$lane))>;
3899 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3900 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3901 (DSubReg_i16_reg imm:$lane))),
3902 (SubReg_i16_lane imm:$lane))>;
3903 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3904 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3905 (DSubReg_i32_reg imm:$lane))),
3906 (SubReg_i32_lane imm:$lane))>;
3907 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3908 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3909 (SSubReg_f32_reg imm:$src2))>;
3910 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3911 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3912 (SSubReg_f32_reg imm:$src2))>;
3913 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3914 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3915 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3916 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3919 // VMOV : Vector Set Lane (move ARM core register to scalar)
3921 let Constraints = "$src1 = $V" in {
3922 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
3923 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3924 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
3925 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
3926 GPR:$R, imm:$lane))]> {
3927 let Inst{21} = lane{2};
3928 let Inst{6-5} = lane{1-0};
3930 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
3931 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3932 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
3933 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
3934 GPR:$R, imm:$lane))]> {
3935 let Inst{21} = lane{1};
3936 let Inst{6} = lane{0};
3938 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
3939 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3940 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
3941 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
3942 GPR:$R, imm:$lane))]> {
3943 let Inst{21} = lane{0};
3946 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3947 (v16i8 (INSERT_SUBREG QPR:$src1,
3948 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3949 (DSubReg_i8_reg imm:$lane))),
3950 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3951 (DSubReg_i8_reg imm:$lane)))>;
3952 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3953 (v8i16 (INSERT_SUBREG QPR:$src1,
3954 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3955 (DSubReg_i16_reg imm:$lane))),
3956 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3957 (DSubReg_i16_reg imm:$lane)))>;
3958 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3959 (v4i32 (INSERT_SUBREG QPR:$src1,
3960 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3961 (DSubReg_i32_reg imm:$lane))),
3962 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3963 (DSubReg_i32_reg imm:$lane)))>;
3965 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3966 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3967 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3968 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3969 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3970 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3972 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3973 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3974 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3975 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3977 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3978 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3979 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3980 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3981 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3982 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3984 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3985 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3986 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3987 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3988 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3989 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3991 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3992 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3993 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3995 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3996 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3997 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3999 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4000 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4001 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4004 // VDUP : Vector Duplicate (from ARM core register to all elements)
4006 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4007 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
4008 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4009 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4010 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4011 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
4012 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
4013 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
4015 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4016 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4017 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4018 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4019 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4020 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4022 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
4023 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4024 [(set DPR:$dst, (v2f32 (NEONvdup
4025 (f32 (bitconvert GPR:$src)))))]>;
4026 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
4027 IIC_VMOVIS, "vdup", "32", "$dst, $src",
4028 [(set QPR:$dst, (v4f32 (NEONvdup
4029 (f32 (bitconvert GPR:$src)))))]>;
4031 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4033 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4035 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4036 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
4037 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
4039 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4040 ValueType ResTy, ValueType OpTy>
4041 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
4042 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
4043 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
4046 // Inst{19-16} is partially specified depending on the element size.
4048 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
4049 let Inst{19-17} = lane{2-0};
4051 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
4052 let Inst{19-18} = lane{1-0};
4054 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
4055 let Inst{19} = lane{0};
4057 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
4058 let Inst{19} = lane{0};
4060 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
4061 let Inst{19-17} = lane{2-0};
4063 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
4064 let Inst{19-18} = lane{1-0};
4066 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
4067 let Inst{19} = lane{0};
4069 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
4070 let Inst{19} = lane{0};
4073 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4074 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4075 (DSubReg_i8_reg imm:$lane))),
4076 (SubReg_i8_lane imm:$lane)))>;
4077 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4078 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4079 (DSubReg_i16_reg imm:$lane))),
4080 (SubReg_i16_lane imm:$lane)))>;
4081 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4082 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4083 (DSubReg_i32_reg imm:$lane))),
4084 (SubReg_i32_lane imm:$lane)))>;
4085 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4086 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
4087 (DSubReg_i32_reg imm:$lane))),
4088 (SubReg_i32_lane imm:$lane)))>;
4090 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4091 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4092 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4093 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4095 // VMOVN : Vector Narrowing Move
4096 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4097 "vmovn", "i", trunc>;
4098 // VQMOVN : Vector Saturating Narrowing Move
4099 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4100 "vqmovn", "s", int_arm_neon_vqmovns>;
4101 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4102 "vqmovn", "u", int_arm_neon_vqmovnu>;
4103 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4104 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4105 // VMOVL : Vector Lengthening Move
4106 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4107 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4109 // Vector Conversions.
4111 // VCVT : Vector Convert Between Floating-Point and Integers
4112 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4113 v2i32, v2f32, fp_to_sint>;
4114 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4115 v2i32, v2f32, fp_to_uint>;
4116 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4117 v2f32, v2i32, sint_to_fp>;
4118 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4119 v2f32, v2i32, uint_to_fp>;
4121 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4122 v4i32, v4f32, fp_to_sint>;
4123 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4124 v4i32, v4f32, fp_to_uint>;
4125 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4126 v4f32, v4i32, sint_to_fp>;
4127 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4128 v4f32, v4i32, uint_to_fp>;
4130 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4131 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4132 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4133 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4134 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4135 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4136 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4137 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4138 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4140 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4141 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4142 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4143 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4144 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4145 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4146 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4147 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4151 // VREV64 : Vector Reverse elements within 64-bit doublewords
4153 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4154 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
4155 (ins DPR:$src), IIC_VMOVD,
4156 OpcodeStr, Dt, "$dst, $src", "",
4157 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
4158 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4159 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
4160 (ins QPR:$src), IIC_VMOVQ,
4161 OpcodeStr, Dt, "$dst, $src", "",
4162 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
4164 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4165 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4166 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4167 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4169 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4170 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4171 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4172 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4174 // VREV32 : Vector Reverse elements within 32-bit words
4176 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4177 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
4178 (ins DPR:$src), IIC_VMOVD,
4179 OpcodeStr, Dt, "$dst, $src", "",
4180 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
4181 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4182 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
4183 (ins QPR:$src), IIC_VMOVQ,
4184 OpcodeStr, Dt, "$dst, $src", "",
4185 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
4187 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4188 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4190 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4191 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4193 // VREV16 : Vector Reverse elements within 16-bit halfwords
4195 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4196 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
4197 (ins DPR:$src), IIC_VMOVD,
4198 OpcodeStr, Dt, "$dst, $src", "",
4199 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
4200 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4201 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
4202 (ins QPR:$src), IIC_VMOVQ,
4203 OpcodeStr, Dt, "$dst, $src", "",
4204 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
4206 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4207 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4209 // Other Vector Shuffles.
4211 // VEXT : Vector Extract
4213 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4214 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
4215 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
4216 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4217 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
4218 (Ty DPR:$rhs), imm:$index)))]> {
4220 let Inst{11-8} = index{3-0};
4223 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4224 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
4225 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
4226 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4227 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
4228 (Ty QPR:$rhs), imm:$index)))]> {
4230 let Inst{11-8} = index{3-0};
4233 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
4234 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
4235 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
4236 def VEXTdf : VEXTd<"vext", "32", v2f32>;
4238 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
4239 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
4240 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
4241 def VEXTqf : VEXTq<"vext", "32", v4f32>;
4243 // VTRN : Vector Transpose
4245 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4246 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4247 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4249 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4250 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4251 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4253 // VUZP : Vector Unzip (Deinterleave)
4255 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4256 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4257 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4259 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4260 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4261 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4263 // VZIP : Vector Zip (Interleave)
4265 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4266 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4267 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4269 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4270 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4271 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4273 // Vector Table Lookup and Table Extension.
4275 // VTBL : Vector Table Lookup
4277 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4278 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4279 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4280 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4281 let hasExtraSrcRegAllocReq = 1 in {
4283 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4284 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4285 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4287 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4288 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4289 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4291 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4292 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4294 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4295 } // hasExtraSrcRegAllocReq = 1
4298 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4300 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4302 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4304 // VTBX : Vector Table Extension
4306 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4307 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4308 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4309 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4310 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4311 let hasExtraSrcRegAllocReq = 1 in {
4313 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4314 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4315 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4317 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4318 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4319 NVTBLFrm, IIC_VTBX3,
4320 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4323 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4324 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4325 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4327 } // hasExtraSrcRegAllocReq = 1
4330 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4331 IIC_VTBX2, "$orig = $dst", []>;
4333 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4334 IIC_VTBX3, "$orig = $dst", []>;
4336 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4337 IIC_VTBX4, "$orig = $dst", []>;
4339 //===----------------------------------------------------------------------===//
4340 // NEON instructions for single-precision FP math
4341 //===----------------------------------------------------------------------===//
4343 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4344 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4345 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4349 class N3VSPat<SDNode OpNode, NeonI Inst>
4350 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4351 (EXTRACT_SUBREG (v2f32
4352 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4354 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4358 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4359 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4360 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4362 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4364 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4368 // These need separate instructions because they must use DPR_VFP2 register
4369 // class which have SPR sub-registers.
4371 // Vector Add Operations used for single-precision FP
4372 let neverHasSideEffects = 1 in
4373 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4374 def : N3VSPat<fadd, VADDfd_sfp>;
4376 // Vector Sub Operations used for single-precision FP
4377 let neverHasSideEffects = 1 in
4378 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4379 def : N3VSPat<fsub, VSUBfd_sfp>;
4381 // Vector Multiply Operations used for single-precision FP
4382 let neverHasSideEffects = 1 in
4383 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4384 def : N3VSPat<fmul, VMULfd_sfp>;
4386 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4387 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4388 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4390 //let neverHasSideEffects = 1 in
4391 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4392 // v2f32, fmul, fadd>;
4393 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4395 //let neverHasSideEffects = 1 in
4396 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4397 // v2f32, fmul, fsub>;
4398 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4400 // Vector Absolute used for single-precision FP
4401 let neverHasSideEffects = 1 in
4402 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4403 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4404 "vabs", "f32", "$dst, $src", "", []>;
4405 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4407 // Vector Negate used for single-precision FP
4408 let neverHasSideEffects = 1 in
4409 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4410 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4411 "vneg", "f32", "$dst, $src", "", []>;
4412 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4414 // Vector Maximum used for single-precision FP
4415 let neverHasSideEffects = 1 in
4416 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4417 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4418 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4419 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4421 // Vector Minimum used for single-precision FP
4422 let neverHasSideEffects = 1 in
4423 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4424 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4425 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4426 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4428 // Vector Convert between single-precision FP and integer
4429 let neverHasSideEffects = 1 in
4430 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4431 v2i32, v2f32, fp_to_sint>;
4432 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4434 let neverHasSideEffects = 1 in
4435 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4436 v2i32, v2f32, fp_to_uint>;
4437 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4439 let neverHasSideEffects = 1 in
4440 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4441 v2f32, v2i32, sint_to_fp>;
4442 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4444 let neverHasSideEffects = 1 in
4445 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4446 v2f32, v2i32, uint_to_fp>;
4447 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4449 //===----------------------------------------------------------------------===//
4450 // Non-Instruction Patterns
4451 //===----------------------------------------------------------------------===//
4454 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4455 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4456 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4457 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4458 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4459 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4460 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4461 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4462 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4463 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4464 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4465 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4466 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4467 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4468 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4469 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4470 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4471 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4472 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4473 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4474 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4475 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4476 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4477 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4478 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4479 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4480 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4481 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4482 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4483 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4485 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4486 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4487 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4488 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4489 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4490 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4491 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4492 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4493 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4494 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4495 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4496 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4497 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4498 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4499 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4500 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4501 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4502 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4503 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4504 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4505 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4506 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4507 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4508 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4509 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4510 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4511 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4512 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4513 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4514 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;