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 = $wb", []> {
968 let Inst{5-4} = Rn{5-4};
971 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
972 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
973 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
975 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
976 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
977 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
979 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
980 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
981 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
983 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
984 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
985 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
987 // ...with double-spaced registers (for disassembly only):
988 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
989 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
990 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
991 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
992 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
993 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
995 // VST3 : Vector Store (multiple 3-element structures)
996 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
997 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
998 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST3,
999 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
1001 def VST3d8 : VST3D<0b0100, 0b0000, "8">;
1002 def VST3d16 : VST3D<0b0100, 0b0100, "16">;
1003 def VST3d32 : VST3D<0b0100, 0b1000, "32">;
1005 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1006 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1007 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1009 // ...with address register writeback:
1010 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1011 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1012 (ins addrmode6:$addr, am6offset:$offset,
1013 DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST3u,
1014 "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr$offset",
1015 "$addr.addr = $wb", []>;
1017 def VST3d8_UPD : VST3DWB<0b0100, 0b0000, "8">;
1018 def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
1019 def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
1021 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1022 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1023 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1025 // ...with double-spaced registers (non-updating versions for disassembly only):
1026 def VST3q8 : VST3D<0b0101, 0b0000, "8">;
1027 def VST3q16 : VST3D<0b0101, 0b0100, "16">;
1028 def VST3q32 : VST3D<0b0101, 0b1000, "32">;
1029 def VST3q8_UPD : VST3DWB<0b0101, 0b0000, "8">;
1030 def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
1031 def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
1033 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1034 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1035 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1037 // ...alternate versions to be allocated odd register numbers:
1038 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1039 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1040 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1042 // VST4 : Vector Store (multiple 4-element structures)
1043 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1044 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1045 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
1046 IIC_VST4, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
1049 def VST4d8 : VST4D<0b0000, 0b0000, "8">;
1050 def VST4d16 : VST4D<0b0000, 0b0100, "16">;
1051 def VST4d32 : VST4D<0b0000, 0b1000, "32">;
1053 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1054 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1055 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1057 // ...with address register writeback:
1058 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1059 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1060 (ins addrmode6:$addr, am6offset:$offset,
1061 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1062 "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr$offset",
1063 "$addr.addr = $wb", []>;
1065 def VST4d8_UPD : VST4DWB<0b0000, 0b0000, "8">;
1066 def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
1067 def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
1069 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1070 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1071 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1073 // ...with double-spaced registers (non-updating versions for disassembly only):
1074 def VST4q8 : VST4D<0b0001, 0b0000, "8">;
1075 def VST4q16 : VST4D<0b0001, 0b0100, "16">;
1076 def VST4q32 : VST4D<0b0001, 0b1000, "32">;
1077 def VST4q8_UPD : VST4DWB<0b0001, 0b0000, "8">;
1078 def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
1079 def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
1081 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1082 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1083 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1085 // ...alternate versions to be allocated odd register numbers:
1086 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1087 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1088 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1090 // Classes for VST*LN pseudo-instructions with multi-register operands.
1091 // These are expanded to real instructions after register allocation.
1092 class VSTQLNPseudo<InstrItinClass itin>
1093 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1095 class VSTQLNWBPseudo<InstrItinClass itin>
1096 : PseudoNLdSt<(outs GPR:$wb),
1097 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1098 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1099 class VSTQQLNPseudo<InstrItinClass itin>
1100 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1102 class VSTQQLNWBPseudo<InstrItinClass itin>
1103 : PseudoNLdSt<(outs GPR:$wb),
1104 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1105 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1106 class VSTQQQQLNPseudo<InstrItinClass itin>
1107 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1109 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1110 : PseudoNLdSt<(outs GPR:$wb),
1111 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1112 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1114 // VST1LN : Vector Store (single element from one lane)
1115 // FIXME: Not yet implemented.
1117 // VST2LN : Vector Store (single 2-element structure from one lane)
1118 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1119 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1120 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, nohash_imm:$lane),
1121 IIC_VST2ln, "vst2", Dt, "\\{$src1[$lane], $src2[$lane]\\}, $addr",
1124 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8">;
1125 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16">;
1126 def VST2LNd32 : VST2LN<0b1001, {?,0,?,?}, "32">;
1128 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1129 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1130 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1132 // ...with double-spaced registers:
1133 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16">;
1134 def VST2LNq32 : VST2LN<0b1001, {?,1,?,?}, "32">;
1136 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1137 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1139 // ...with address register writeback:
1140 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1141 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1142 (ins addrmode6:$addr, am6offset:$offset,
1143 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1144 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1145 "$addr.addr = $wb", []>;
1147 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8">;
1148 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16">;
1149 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,?,?}, "32">;
1151 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1152 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1153 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1155 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16">;
1156 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,?,?}, "32">;
1158 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1159 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1161 // VST3LN : Vector Store (single 3-element structure from one lane)
1162 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1163 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1164 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
1165 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1166 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr", "", []>;
1168 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8">;
1169 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16">;
1170 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32">;
1172 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1173 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1174 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1176 // ...with double-spaced registers:
1177 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16">;
1178 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32">;
1180 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1181 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1183 // ...with address register writeback:
1184 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1185 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1186 (ins addrmode6:$addr, am6offset:$offset,
1187 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1188 IIC_VST3lnu, "vst3", Dt,
1189 "\\{$src1[$lane], $src2[$lane], $src3[$lane]\\}, $addr$offset",
1190 "$addr.addr = $wb", []>;
1192 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8">;
1193 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16">;
1194 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32">;
1196 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1197 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1198 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1200 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16">;
1201 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32">;
1203 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1204 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1206 // VST4LN : Vector Store (single 4-element structure from one lane)
1207 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1208 : NLdSt<1, 0b00, op11_8, op7_4, (outs),
1209 (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
1210 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1211 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr",
1214 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8">;
1215 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16">;
1216 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32">;
1218 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1219 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1220 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1222 // ...with double-spaced registers:
1223 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16">;
1224 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32">;
1226 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1227 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1229 // ...with address register writeback:
1230 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1231 : NLdSt<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1232 (ins addrmode6:$addr, am6offset:$offset,
1233 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1234 IIC_VST4lnu, "vst4", Dt,
1235 "\\{$src1[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $addr$offset",
1236 "$addr.addr = $wb", []>;
1238 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8">;
1239 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16">;
1240 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32">;
1242 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1243 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1244 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1246 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16">;
1247 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32">;
1249 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1250 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1252 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1255 //===----------------------------------------------------------------------===//
1256 // NEON pattern fragments
1257 //===----------------------------------------------------------------------===//
1259 // Extract D sub-registers of Q registers.
1260 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1261 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1262 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1264 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1265 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1266 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1268 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1269 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1270 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1272 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1273 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1274 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1277 // Extract S sub-registers of Q/D registers.
1278 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1279 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1280 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1283 // Translate lane numbers from Q registers to D subregs.
1284 def SubReg_i8_lane : SDNodeXForm<imm, [{
1285 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1287 def SubReg_i16_lane : SDNodeXForm<imm, [{
1288 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1290 def SubReg_i32_lane : SDNodeXForm<imm, [{
1291 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1294 //===----------------------------------------------------------------------===//
1295 // Instruction Classes
1296 //===----------------------------------------------------------------------===//
1298 // Basic 2-register operations: single-, double- and quad-register.
1299 class N2VS<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1300 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1301 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1302 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1303 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src),
1304 IIC_VUNAD, OpcodeStr, Dt, "$dst, $src", "", []>;
1305 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1306 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1307 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1308 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1309 (ins DPR:$src), IIC_VUNAD, OpcodeStr, Dt,"$dst, $src", "",
1310 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src))))]>;
1311 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1312 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1313 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1314 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1315 (ins QPR:$src), IIC_VUNAQ, OpcodeStr, Dt,"$dst, $src", "",
1316 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src))))]>;
1318 // Basic 2-register intrinsics, both double- and quad-register.
1319 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1320 bits<2> op17_16, bits<5> op11_7, bit op4,
1321 InstrItinClass itin, string OpcodeStr, string Dt,
1322 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1323 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1324 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1325 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1326 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1327 bits<2> op17_16, bits<5> op11_7, bit op4,
1328 InstrItinClass itin, string OpcodeStr, string Dt,
1329 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1330 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1331 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1332 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1334 // Narrow 2-register operations.
1335 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1336 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1337 InstrItinClass itin, string OpcodeStr, string Dt,
1338 ValueType TyD, ValueType TyQ, SDNode OpNode>
1339 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1340 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1341 [(set DPR:$dst, (TyD (OpNode (TyQ QPR:$src))))]>;
1343 // Narrow 2-register intrinsics.
1344 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1345 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1346 InstrItinClass itin, string OpcodeStr, string Dt,
1347 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1348 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$dst),
1349 (ins QPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1350 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src))))]>;
1352 // Long 2-register operations (currently only used for VMOVL).
1353 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1354 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1355 InstrItinClass itin, string OpcodeStr, string Dt,
1356 ValueType TyQ, ValueType TyD, SDNode OpNode>
1357 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$dst),
1358 (ins DPR:$src), itin, OpcodeStr, Dt, "$dst, $src", "",
1359 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src))))]>;
1361 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1362 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1363 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$dst1, DPR:$dst2),
1364 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1365 OpcodeStr, Dt, "$dst1, $dst2",
1366 "$src1 = $dst1, $src2 = $dst2", []>;
1367 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1368 InstrItinClass itin, string OpcodeStr, string Dt>
1369 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$dst1, QPR:$dst2),
1370 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$dst1, $dst2",
1371 "$src1 = $dst1, $src2 = $dst2", []>;
1373 // Basic 3-register operations: single-, double- and quad-register.
1374 class N3VS<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1375 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1376 SDNode OpNode, bit Commutable>
1377 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1378 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm,
1379 IIC_VBIND, OpcodeStr, Dt, "$dst, $src1, $src2", "", []> {
1380 let isCommutable = Commutable;
1383 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1384 InstrItinClass itin, string OpcodeStr, string Dt,
1385 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1386 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1387 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1388 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1389 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1390 let isCommutable = Commutable;
1392 // Same as N3VD but no data type.
1393 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1394 InstrItinClass itin, string OpcodeStr,
1395 ValueType ResTy, ValueType OpTy,
1396 SDNode OpNode, bit Commutable>
1397 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1398 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1399 OpcodeStr, "$dst, $src1, $src2", "",
1400 [(set DPR:$dst, (ResTy (OpNode (OpTy DPR:$src1), (OpTy DPR:$src2))))]>{
1401 let isCommutable = Commutable;
1404 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1405 InstrItinClass itin, string OpcodeStr, string Dt,
1406 ValueType Ty, SDNode ShOp>
1407 : N3V<0, 1, op21_20, op11_8, 1, 0,
1408 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1409 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1410 [(set (Ty DPR:$dst),
1411 (Ty (ShOp (Ty DPR:$src1),
1412 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),imm:$lane)))))]> {
1413 let isCommutable = 0;
1415 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1416 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1417 : N3V<0, 1, op21_20, op11_8, 1, 0,
1418 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1419 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1420 [(set (Ty DPR:$dst),
1421 (Ty (ShOp (Ty DPR:$src1),
1422 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1423 let isCommutable = 0;
1426 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1427 InstrItinClass itin, string OpcodeStr, string Dt,
1428 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1429 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1430 (outs QPR:$Qd), (ins QPR:$Qn, QPR:$Qm), N3RegFrm, itin,
1431 OpcodeStr, Dt, "$Qd, $Qn, $Qm", "",
1432 [(set QPR:$Qd, (ResTy (OpNode (OpTy QPR:$Qn), (OpTy QPR:$Qm))))]> {
1433 let isCommutable = Commutable;
1435 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1436 InstrItinClass itin, string OpcodeStr,
1437 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1438 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1439 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, itin,
1440 OpcodeStr, "$dst, $src1, $src2", "",
1441 [(set QPR:$dst, (ResTy (OpNode (OpTy QPR:$src1), (OpTy QPR:$src2))))]>{
1442 let isCommutable = Commutable;
1444 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
1445 InstrItinClass itin, string OpcodeStr, string Dt,
1446 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1447 : N3V<1, 1, op21_20, op11_8, 1, 0,
1448 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1449 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1450 [(set (ResTy QPR:$dst),
1451 (ResTy (ShOp (ResTy QPR:$src1),
1452 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1454 let isCommutable = 0;
1456 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
1457 ValueType ResTy, ValueType OpTy, SDNode ShOp>
1458 : N3V<1, 1, op21_20, op11_8, 1, 0,
1459 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1460 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$dst, $src1, $src2[$lane]","",
1461 [(set (ResTy QPR:$dst),
1462 (ResTy (ShOp (ResTy QPR:$src1),
1463 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1465 let isCommutable = 0;
1468 // Basic 3-register intrinsics, both double- and quad-register.
1469 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1470 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1471 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1472 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1473 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
1474 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1475 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1476 let isCommutable = Commutable;
1478 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1479 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1480 : N3V<0, 1, op21_20, op11_8, 1, 0,
1481 (outs DPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1482 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1483 [(set (Ty DPR:$dst),
1484 (Ty (IntOp (Ty DPR:$src1),
1485 (Ty (NEONvduplane (Ty DPR_VFP2:$src2),
1487 let isCommutable = 0;
1489 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1490 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
1491 : N3V<0, 1, op21_20, op11_8, 1, 0,
1492 (outs DPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1493 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1494 [(set (Ty DPR:$dst),
1495 (Ty (IntOp (Ty DPR:$src1),
1496 (Ty (NEONvduplane (Ty DPR_8:$src2), imm:$lane)))))]> {
1497 let isCommutable = 0;
1499 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1500 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1501 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1502 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1503 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
1504 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1505 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
1506 let isCommutable = 0;
1509 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1510 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1511 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
1512 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1513 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
1514 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1515 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1516 let isCommutable = Commutable;
1518 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1519 string OpcodeStr, string Dt,
1520 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1521 : N3V<1, 1, op21_20, op11_8, 1, 0,
1522 (outs QPR:$dst), (ins QPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1523 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1524 [(set (ResTy QPR:$dst),
1525 (ResTy (IntOp (ResTy QPR:$src1),
1526 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1528 let isCommutable = 0;
1530 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1531 string OpcodeStr, string Dt,
1532 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1533 : N3V<1, 1, op21_20, op11_8, 1, 0,
1534 (outs QPR:$dst), (ins QPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1535 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1536 [(set (ResTy QPR:$dst),
1537 (ResTy (IntOp (ResTy QPR:$src1),
1538 (ResTy (NEONvduplane (OpTy DPR_8:$src2),
1540 let isCommutable = 0;
1542 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1543 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1544 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1545 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1546 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
1547 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
1548 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
1549 let isCommutable = 0;
1552 // Multiply-Add/Sub operations: single-, double- and quad-register.
1553 class N3VSMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1554 InstrItinClass itin, string OpcodeStr, string Dt,
1555 ValueType Ty, SDNode MulOp, SDNode OpNode>
1556 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1557 (outs DPR_VFP2:$dst),
1558 (ins DPR_VFP2:$src1, DPR_VFP2:$src2, DPR_VFP2:$src3), N3RegFrm, itin,
1559 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst", []>;
1561 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1562 InstrItinClass itin, string OpcodeStr, string Dt,
1563 ValueType Ty, SDNode MulOp, SDNode OpNode>
1564 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1565 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1566 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1567 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1568 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
1570 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1571 string OpcodeStr, string Dt,
1572 ValueType Ty, SDNode MulOp, SDNode ShOp>
1573 : N3V<0, 1, op21_20, op11_8, 1, 0,
1575 (ins DPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1577 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1578 [(set (Ty DPR:$dst),
1579 (Ty (ShOp (Ty DPR:$src1),
1580 (Ty (MulOp DPR:$src2,
1581 (Ty (NEONvduplane (Ty DPR_VFP2:$src3),
1583 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1584 string OpcodeStr, string Dt,
1585 ValueType Ty, SDNode MulOp, SDNode ShOp>
1586 : N3V<0, 1, op21_20, op11_8, 1, 0,
1588 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, nohash_imm:$lane),
1590 OpcodeStr, Dt, "$Vd, $Vn, $Vm[$lane]", "$src1 = $Vd",
1592 (Ty (ShOp (Ty DPR:$src1),
1594 (Ty (NEONvduplane (Ty DPR_8:$Vm),
1597 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1598 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
1599 SDNode MulOp, SDNode OpNode>
1600 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1601 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1602 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1603 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1604 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
1605 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1606 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1607 SDNode MulOp, SDNode ShOp>
1608 : N3V<1, 1, op21_20, op11_8, 1, 0,
1610 (ins QPR:$src1, QPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1612 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1613 [(set (ResTy QPR:$dst),
1614 (ResTy (ShOp (ResTy QPR:$src1),
1615 (ResTy (MulOp QPR:$src2,
1616 (ResTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1618 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1619 string OpcodeStr, string Dt,
1620 ValueType ResTy, ValueType OpTy,
1621 SDNode MulOp, SDNode ShOp>
1622 : N3V<1, 1, op21_20, op11_8, 1, 0,
1624 (ins QPR:$src1, QPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1626 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1627 [(set (ResTy QPR:$dst),
1628 (ResTy (ShOp (ResTy QPR:$src1),
1629 (ResTy (MulOp QPR:$src2,
1630 (ResTy (NEONvduplane (OpTy DPR_8:$src3),
1633 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
1634 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1635 InstrItinClass itin, string OpcodeStr, string Dt,
1636 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1637 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1638 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1639 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1640 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
1641 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
1642 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1643 InstrItinClass itin, string OpcodeStr, string Dt,
1644 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
1645 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1646 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1647 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1648 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
1649 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
1651 // Neon 3-argument intrinsics, both double- and quad-register.
1652 // The destination register is also used as the first source operand register.
1653 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1654 InstrItinClass itin, string OpcodeStr, string Dt,
1655 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1656 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1657 (outs DPR:$dst), (ins DPR:$src1, DPR:$src2, DPR:$src3), N3RegFrm, itin,
1658 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1659 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src1),
1660 (OpTy DPR:$src2), (OpTy DPR:$src3))))]>;
1661 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1662 InstrItinClass itin, string OpcodeStr, string Dt,
1663 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1664 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1665 (outs QPR:$dst), (ins QPR:$src1, QPR:$src2, QPR:$src3), N3RegFrm, itin,
1666 OpcodeStr, Dt, "$dst, $src2, $src3", "$src1 = $dst",
1667 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src1),
1668 (OpTy QPR:$src2), (OpTy QPR:$src3))))]>;
1670 // Long Multiply-Add/Sub operations.
1671 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1672 InstrItinClass itin, string OpcodeStr, string Dt,
1673 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1674 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1675 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1676 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1677 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1678 (TyQ (MulOp (TyD DPR:$Vn),
1679 (TyD DPR:$Vm)))))]>;
1680 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1681 InstrItinClass itin, string OpcodeStr, string Dt,
1682 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1683 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1684 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1686 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1688 (OpNode (TyQ QPR:$src1),
1689 (TyQ (MulOp (TyD DPR:$src2),
1690 (TyD (NEONvduplane (TyD DPR_VFP2:$src3),
1692 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1693 InstrItinClass itin, string OpcodeStr, string Dt,
1694 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
1695 : N3V<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$dst),
1696 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1698 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1700 (OpNode (TyQ QPR:$src1),
1701 (TyQ (MulOp (TyD DPR:$src2),
1702 (TyD (NEONvduplane (TyD DPR_8:$src3),
1705 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
1706 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1707 InstrItinClass itin, string OpcodeStr, string Dt,
1708 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
1710 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1711 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1712 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1713 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
1714 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
1715 (TyD DPR:$Vm)))))))]>;
1717 // Neon Long 3-argument intrinsic. The destination register is
1718 // a quad-register and is also used as the first source operand register.
1719 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1720 InstrItinClass itin, string OpcodeStr, string Dt,
1721 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1722 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1723 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1724 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
1726 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
1727 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1728 string OpcodeStr, string Dt,
1729 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1730 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1732 (ins QPR:$src1, DPR:$src2, DPR_VFP2:$src3, nohash_imm:$lane),
1734 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1735 [(set (ResTy QPR:$dst),
1736 (ResTy (IntOp (ResTy QPR:$src1),
1738 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src3),
1740 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1741 InstrItinClass itin, string OpcodeStr, string Dt,
1742 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1743 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1745 (ins QPR:$src1, DPR:$src2, DPR_8:$src3, nohash_imm:$lane),
1747 OpcodeStr, Dt, "$dst, $src2, $src3[$lane]", "$src1 = $dst",
1748 [(set (ResTy QPR:$dst),
1749 (ResTy (IntOp (ResTy QPR:$src1),
1751 (OpTy (NEONvduplane (OpTy DPR_8:$src3),
1754 // Narrowing 3-register intrinsics.
1755 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1756 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
1757 Intrinsic IntOp, bit Commutable>
1758 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1759 (outs DPR:$dst), (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINi4D,
1760 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1761 [(set DPR:$dst, (TyD (IntOp (TyQ QPR:$src1), (TyQ QPR:$src2))))]> {
1762 let isCommutable = Commutable;
1765 // Long 3-register operations.
1766 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1767 InstrItinClass itin, string OpcodeStr, string Dt,
1768 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
1769 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1770 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1771 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1772 [(set QPR:$dst, (TyQ (OpNode (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1773 let isCommutable = Commutable;
1775 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
1776 InstrItinClass itin, string OpcodeStr, string Dt,
1777 ValueType TyQ, ValueType TyD, SDNode OpNode>
1778 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1779 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1780 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1782 (TyQ (OpNode (TyD DPR:$src1),
1783 (TyD (NEONvduplane (TyD DPR_VFP2:$src2),imm:$lane)))))]>;
1784 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1785 InstrItinClass itin, string OpcodeStr, string Dt,
1786 ValueType TyQ, ValueType TyD, SDNode OpNode>
1787 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1788 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1789 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1791 (TyQ (OpNode (TyD DPR:$src1),
1792 (TyD (NEONvduplane (TyD DPR_8:$src2), imm:$lane)))))]>;
1794 // Long 3-register operations with explicitly extended operands.
1795 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1796 InstrItinClass itin, string OpcodeStr, string Dt,
1797 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
1799 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1800 (outs QPR:$Qd), (ins DPR:$Dn, DPR:$Dm), N3RegFrm, itin,
1801 OpcodeStr, Dt, "$Qd, $Dn, $Dm", "",
1802 [(set QPR:$Qd, (OpNode (TyQ (ExtOp (TyD DPR:$Dn))),
1803 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1804 let isCommutable = Commutable;
1807 // Long 3-register intrinsics with explicit extend (VABDL).
1808 class N3VLIntExt<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, Intrinsic IntOp, SDNode ExtOp,
1812 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1813 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1814 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1815 [(set QPR:$dst, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$src1),
1816 (TyD DPR:$src2))))))]> {
1817 let isCommutable = Commutable;
1820 // Long 3-register intrinsics.
1821 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1822 InstrItinClass itin, string OpcodeStr, string Dt,
1823 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
1824 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1825 (outs QPR:$dst), (ins DPR:$src1, DPR:$src2), N3RegFrm, itin,
1826 OpcodeStr, Dt, "$dst, $src1, $src2", "",
1827 [(set QPR:$dst, (TyQ (IntOp (TyD DPR:$src1), (TyD DPR:$src2))))]> {
1828 let isCommutable = Commutable;
1830 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
1831 string OpcodeStr, string Dt,
1832 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1833 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1834 (outs QPR:$dst), (ins DPR:$src1, DPR_VFP2:$src2, nohash_imm:$lane),
1835 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1836 [(set (ResTy QPR:$dst),
1837 (ResTy (IntOp (OpTy DPR:$src1),
1838 (OpTy (NEONvduplane (OpTy DPR_VFP2:$src2),
1840 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
1841 InstrItinClass itin, string OpcodeStr, string Dt,
1842 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1843 : N3V<op24, 1, op21_20, op11_8, 1, 0,
1844 (outs QPR:$dst), (ins DPR:$src1, DPR_8:$src2, nohash_imm:$lane),
1845 NVMulSLFrm, itin, OpcodeStr, Dt, "$dst, $src1, $src2[$lane]", "",
1846 [(set (ResTy QPR:$dst),
1847 (ResTy (IntOp (OpTy DPR:$src1),
1848 (OpTy (NEONvduplane (OpTy DPR_8:$src2),
1851 // Wide 3-register operations.
1852 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1853 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
1854 SDNode OpNode, SDNode ExtOp, bit Commutable>
1855 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1856 (outs QPR:$Qd), (ins QPR:$Qn, DPR:$Dm), N3RegFrm, IIC_VSUBiD,
1857 OpcodeStr, Dt, "$Qd, $Qn, $Dm", "",
1858 [(set QPR:$Qd, (OpNode (TyQ QPR:$Qn),
1859 (TyQ (ExtOp (TyD DPR:$Dm)))))]> {
1860 let isCommutable = Commutable;
1863 // Pairwise long 2-register intrinsics, both double- and quad-register.
1864 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1865 bits<2> op17_16, bits<5> op11_7, bit op4,
1866 string OpcodeStr, string Dt,
1867 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1868 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$dst),
1869 (ins DPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1870 [(set DPR:$dst, (ResTy (IntOp (OpTy DPR:$src))))]>;
1871 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1872 bits<2> op17_16, bits<5> op11_7, bit op4,
1873 string OpcodeStr, string Dt,
1874 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1875 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$dst),
1876 (ins QPR:$src), IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
1877 [(set QPR:$dst, (ResTy (IntOp (OpTy QPR:$src))))]>;
1879 // Pairwise long 2-register accumulate intrinsics,
1880 // both double- and quad-register.
1881 // The destination register is also used as the first source operand register.
1882 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1883 bits<2> op17_16, bits<5> op11_7, bit op4,
1884 string OpcodeStr, string Dt,
1885 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1886 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
1887 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
1888 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
1889 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
1890 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1891 bits<2> op17_16, bits<5> op11_7, bit op4,
1892 string OpcodeStr, string Dt,
1893 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1894 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
1895 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
1896 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
1897 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
1899 // Shift by immediate,
1900 // both double- and quad-register.
1901 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1902 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1903 ValueType Ty, SDNode OpNode>
1904 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1905 (outs DPR:$dst), (ins DPR:$src, i32imm:$SIMM), f, itin,
1906 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1907 [(set DPR:$dst, (Ty (OpNode (Ty DPR:$src), (i32 imm:$SIMM))))]>;
1908 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1909 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
1910 ValueType Ty, SDNode OpNode>
1911 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1912 (outs QPR:$dst), (ins QPR:$src, i32imm:$SIMM), f, itin,
1913 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1914 [(set QPR:$dst, (Ty (OpNode (Ty QPR:$src), (i32 imm:$SIMM))))]>;
1916 // Long shift by immediate.
1917 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1918 string OpcodeStr, string Dt,
1919 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1920 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1921 (outs QPR:$dst), (ins DPR:$src, i32imm:$SIMM), N2RegVShLFrm,
1922 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1923 [(set QPR:$dst, (ResTy (OpNode (OpTy DPR:$src),
1924 (i32 imm:$SIMM))))]>;
1926 // Narrow shift by immediate.
1927 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
1928 InstrItinClass itin, string OpcodeStr, string Dt,
1929 ValueType ResTy, ValueType OpTy, SDNode OpNode>
1930 : N2VImm<op24, op23, op11_8, op7, op6, op4,
1931 (outs DPR:$dst), (ins QPR:$src, i32imm:$SIMM), N2RegVShRFrm, itin,
1932 OpcodeStr, Dt, "$dst, $src, $SIMM", "",
1933 [(set DPR:$dst, (ResTy (OpNode (OpTy QPR:$src),
1934 (i32 imm:$SIMM))))]>;
1936 // Shift right by immediate and accumulate,
1937 // both double- and quad-register.
1938 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1939 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1940 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
1941 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1942 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1943 [(set DPR:$Vd, (Ty (add DPR:$src1,
1944 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
1945 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1946 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1947 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
1948 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), N2RegVShRFrm, IIC_VPALiD,
1949 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1950 [(set QPR:$Vd, (Ty (add QPR:$src1,
1951 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
1953 // Shift by immediate and insert,
1954 // both double- and quad-register.
1955 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1956 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1957 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
1958 (ins DPR:$src1, DPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiD,
1959 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1960 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
1961 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1962 Format f, string OpcodeStr, string Dt, ValueType Ty,SDNode ShOp>
1963 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
1964 (ins QPR:$src1, QPR:$Vm, i32imm:$SIMM), f, IIC_VSHLiQ,
1965 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
1966 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
1968 // Convert, with fractional bits immediate,
1969 // both double- and quad-register.
1970 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1971 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1973 : N2VImm<op24, op23, op11_8, op7, 0, op4,
1974 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
1975 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
1976 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
1977 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
1978 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
1980 : N2VImm<op24, op23, op11_8, op7, 1, op4,
1981 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
1982 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
1983 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
1985 //===----------------------------------------------------------------------===//
1987 //===----------------------------------------------------------------------===//
1989 // Abbreviations used in multiclass suffixes:
1990 // Q = quarter int (8 bit) elements
1991 // H = half int (16 bit) elements
1992 // S = single int (32 bit) elements
1993 // D = double int (64 bit) elements
1995 // Neon 2-register vector operations -- for disassembly only.
1997 // First with only element sizes of 8, 16 and 32 bits:
1998 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
1999 bits<5> op11_7, bit op4, string opc, string Dt,
2001 // 64-bit vector types.
2002 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2003 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2004 opc, !strconcat(Dt, "8"), asm, "", []>;
2005 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2006 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2007 opc, !strconcat(Dt, "16"), asm, "", []>;
2008 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2009 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2010 opc, !strconcat(Dt, "32"), asm, "", []>;
2011 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2012 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
2013 opc, "f32", asm, "", []> {
2014 let Inst{10} = 1; // overwrite F = 1
2017 // 128-bit vector types.
2018 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2019 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2020 opc, !strconcat(Dt, "8"), asm, "", []>;
2021 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2022 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2023 opc, !strconcat(Dt, "16"), asm, "", []>;
2024 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2025 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2026 opc, !strconcat(Dt, "32"), asm, "", []>;
2027 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2028 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
2029 opc, "f32", asm, "", []> {
2030 let Inst{10} = 1; // overwrite F = 1
2034 // Neon 3-register vector operations.
2036 // First with only element sizes of 8, 16 and 32 bits:
2037 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2038 InstrItinClass itinD16, InstrItinClass itinD32,
2039 InstrItinClass itinQ16, InstrItinClass itinQ32,
2040 string OpcodeStr, string Dt,
2041 SDNode OpNode, bit Commutable = 0> {
2042 // 64-bit vector types.
2043 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2044 OpcodeStr, !strconcat(Dt, "8"),
2045 v8i8, v8i8, OpNode, Commutable>;
2046 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2047 OpcodeStr, !strconcat(Dt, "16"),
2048 v4i16, v4i16, OpNode, Commutable>;
2049 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2050 OpcodeStr, !strconcat(Dt, "32"),
2051 v2i32, v2i32, OpNode, Commutable>;
2053 // 128-bit vector types.
2054 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2055 OpcodeStr, !strconcat(Dt, "8"),
2056 v16i8, v16i8, OpNode, Commutable>;
2057 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2058 OpcodeStr, !strconcat(Dt, "16"),
2059 v8i16, v8i16, OpNode, Commutable>;
2060 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2061 OpcodeStr, !strconcat(Dt, "32"),
2062 v4i32, v4i32, OpNode, Commutable>;
2065 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2066 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2068 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2070 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2071 v8i16, v4i16, ShOp>;
2072 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2073 v4i32, v2i32, ShOp>;
2076 // ....then also with element size 64 bits:
2077 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2078 InstrItinClass itinD, InstrItinClass itinQ,
2079 string OpcodeStr, string Dt,
2080 SDNode OpNode, bit Commutable = 0>
2081 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2082 OpcodeStr, Dt, OpNode, Commutable> {
2083 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2084 OpcodeStr, !strconcat(Dt, "64"),
2085 v1i64, v1i64, OpNode, Commutable>;
2086 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2087 OpcodeStr, !strconcat(Dt, "64"),
2088 v2i64, v2i64, OpNode, Commutable>;
2092 // Neon Narrowing 2-register vector operations,
2093 // source operand element sizes of 16, 32 and 64 bits:
2094 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2095 bits<5> op11_7, bit op6, bit op4,
2096 InstrItinClass itin, string OpcodeStr, string Dt,
2098 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2099 itin, OpcodeStr, !strconcat(Dt, "16"),
2100 v8i8, v8i16, OpNode>;
2101 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2102 itin, OpcodeStr, !strconcat(Dt, "32"),
2103 v4i16, v4i32, OpNode>;
2104 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2105 itin, OpcodeStr, !strconcat(Dt, "64"),
2106 v2i32, v2i64, OpNode>;
2109 // Neon Narrowing 2-register vector intrinsics,
2110 // source operand element sizes of 16, 32 and 64 bits:
2111 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2112 bits<5> op11_7, bit op6, bit op4,
2113 InstrItinClass itin, string OpcodeStr, string Dt,
2115 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2116 itin, OpcodeStr, !strconcat(Dt, "16"),
2117 v8i8, v8i16, IntOp>;
2118 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2119 itin, OpcodeStr, !strconcat(Dt, "32"),
2120 v4i16, v4i32, IntOp>;
2121 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2122 itin, OpcodeStr, !strconcat(Dt, "64"),
2123 v2i32, v2i64, IntOp>;
2127 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2128 // source operand element sizes of 16, 32 and 64 bits:
2129 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2130 string OpcodeStr, string Dt, SDNode OpNode> {
2131 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2132 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2133 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2134 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2135 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2136 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2140 // Neon 3-register vector intrinsics.
2142 // First with only element sizes of 16 and 32 bits:
2143 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2144 InstrItinClass itinD16, InstrItinClass itinD32,
2145 InstrItinClass itinQ16, InstrItinClass itinQ32,
2146 string OpcodeStr, string Dt,
2147 Intrinsic IntOp, bit Commutable = 0> {
2148 // 64-bit vector types.
2149 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2150 OpcodeStr, !strconcat(Dt, "16"),
2151 v4i16, v4i16, IntOp, Commutable>;
2152 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2153 OpcodeStr, !strconcat(Dt, "32"),
2154 v2i32, v2i32, IntOp, Commutable>;
2156 // 128-bit vector types.
2157 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2158 OpcodeStr, !strconcat(Dt, "16"),
2159 v8i16, v8i16, IntOp, Commutable>;
2160 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2161 OpcodeStr, !strconcat(Dt, "32"),
2162 v4i32, v4i32, IntOp, Commutable>;
2164 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2165 InstrItinClass itinD16, InstrItinClass itinD32,
2166 InstrItinClass itinQ16, InstrItinClass itinQ32,
2167 string OpcodeStr, string Dt,
2169 // 64-bit vector types.
2170 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2171 OpcodeStr, !strconcat(Dt, "16"),
2172 v4i16, v4i16, IntOp>;
2173 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2174 OpcodeStr, !strconcat(Dt, "32"),
2175 v2i32, v2i32, IntOp>;
2177 // 128-bit vector types.
2178 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2179 OpcodeStr, !strconcat(Dt, "16"),
2180 v8i16, v8i16, IntOp>;
2181 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2182 OpcodeStr, !strconcat(Dt, "32"),
2183 v4i32, v4i32, IntOp>;
2186 multiclass N3VIntSL_HS<bits<4> op11_8,
2187 InstrItinClass itinD16, InstrItinClass itinD32,
2188 InstrItinClass itinQ16, InstrItinClass itinQ32,
2189 string OpcodeStr, string Dt, Intrinsic IntOp> {
2190 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2191 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2192 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2193 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2194 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2195 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2196 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2197 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2200 // ....then also with element size of 8 bits:
2201 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2202 InstrItinClass itinD16, InstrItinClass itinD32,
2203 InstrItinClass itinQ16, InstrItinClass itinQ32,
2204 string OpcodeStr, string Dt,
2205 Intrinsic IntOp, bit Commutable = 0>
2206 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2207 OpcodeStr, Dt, IntOp, Commutable> {
2208 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2209 OpcodeStr, !strconcat(Dt, "8"),
2210 v8i8, v8i8, IntOp, Commutable>;
2211 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2212 OpcodeStr, !strconcat(Dt, "8"),
2213 v16i8, v16i8, IntOp, Commutable>;
2215 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2216 InstrItinClass itinD16, InstrItinClass itinD32,
2217 InstrItinClass itinQ16, InstrItinClass itinQ32,
2218 string OpcodeStr, string Dt,
2220 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2221 OpcodeStr, Dt, IntOp> {
2222 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2223 OpcodeStr, !strconcat(Dt, "8"),
2225 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2226 OpcodeStr, !strconcat(Dt, "8"),
2227 v16i8, v16i8, IntOp>;
2231 // ....then also with element size of 64 bits:
2232 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2233 InstrItinClass itinD16, InstrItinClass itinD32,
2234 InstrItinClass itinQ16, InstrItinClass itinQ32,
2235 string OpcodeStr, string Dt,
2236 Intrinsic IntOp, bit Commutable = 0>
2237 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2238 OpcodeStr, Dt, IntOp, Commutable> {
2239 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2240 OpcodeStr, !strconcat(Dt, "64"),
2241 v1i64, v1i64, IntOp, Commutable>;
2242 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2243 OpcodeStr, !strconcat(Dt, "64"),
2244 v2i64, v2i64, IntOp, Commutable>;
2246 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2247 InstrItinClass itinD16, InstrItinClass itinD32,
2248 InstrItinClass itinQ16, InstrItinClass itinQ32,
2249 string OpcodeStr, string Dt,
2251 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2252 OpcodeStr, Dt, IntOp> {
2253 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2254 OpcodeStr, !strconcat(Dt, "64"),
2255 v1i64, v1i64, IntOp>;
2256 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2257 OpcodeStr, !strconcat(Dt, "64"),
2258 v2i64, v2i64, IntOp>;
2261 // Neon Narrowing 3-register vector intrinsics,
2262 // source operand element sizes of 16, 32 and 64 bits:
2263 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2264 string OpcodeStr, string Dt,
2265 Intrinsic IntOp, bit Commutable = 0> {
2266 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2267 OpcodeStr, !strconcat(Dt, "16"),
2268 v8i8, v8i16, IntOp, Commutable>;
2269 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2270 OpcodeStr, !strconcat(Dt, "32"),
2271 v4i16, v4i32, IntOp, Commutable>;
2272 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2273 OpcodeStr, !strconcat(Dt, "64"),
2274 v2i32, v2i64, IntOp, Commutable>;
2278 // Neon Long 3-register vector operations.
2280 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2281 InstrItinClass itin16, InstrItinClass itin32,
2282 string OpcodeStr, string Dt,
2283 SDNode OpNode, bit Commutable = 0> {
2284 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2285 OpcodeStr, !strconcat(Dt, "8"),
2286 v8i16, v8i8, OpNode, Commutable>;
2287 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2288 OpcodeStr, !strconcat(Dt, "16"),
2289 v4i32, v4i16, OpNode, Commutable>;
2290 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2291 OpcodeStr, !strconcat(Dt, "32"),
2292 v2i64, v2i32, OpNode, Commutable>;
2295 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2296 InstrItinClass itin, string OpcodeStr, string Dt,
2298 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2299 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2300 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2301 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2304 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2305 InstrItinClass itin16, InstrItinClass itin32,
2306 string OpcodeStr, string Dt,
2307 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2308 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2309 OpcodeStr, !strconcat(Dt, "8"),
2310 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2311 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2312 OpcodeStr, !strconcat(Dt, "16"),
2313 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2314 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2315 OpcodeStr, !strconcat(Dt, "32"),
2316 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2319 // Neon Long 3-register vector intrinsics.
2321 // First with only element sizes of 16 and 32 bits:
2322 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2323 InstrItinClass itin16, InstrItinClass itin32,
2324 string OpcodeStr, string Dt,
2325 Intrinsic IntOp, bit Commutable = 0> {
2326 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2327 OpcodeStr, !strconcat(Dt, "16"),
2328 v4i32, v4i16, IntOp, Commutable>;
2329 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2330 OpcodeStr, !strconcat(Dt, "32"),
2331 v2i64, v2i32, IntOp, Commutable>;
2334 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2335 InstrItinClass itin, string OpcodeStr, string Dt,
2337 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2338 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2339 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2340 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2343 // ....then also with element size of 8 bits:
2344 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2345 InstrItinClass itin16, InstrItinClass itin32,
2346 string OpcodeStr, string Dt,
2347 Intrinsic IntOp, bit Commutable = 0>
2348 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2349 IntOp, Commutable> {
2350 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2351 OpcodeStr, !strconcat(Dt, "8"),
2352 v8i16, v8i8, IntOp, Commutable>;
2355 // ....with explicit extend (VABDL).
2356 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2357 InstrItinClass itin, string OpcodeStr, string Dt,
2358 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2359 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2360 OpcodeStr, !strconcat(Dt, "8"),
2361 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2362 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2363 OpcodeStr, !strconcat(Dt, "16"),
2364 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2365 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2366 OpcodeStr, !strconcat(Dt, "32"),
2367 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2371 // Neon Wide 3-register vector intrinsics,
2372 // source operand element sizes of 8, 16 and 32 bits:
2373 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2374 string OpcodeStr, string Dt,
2375 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2376 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2377 OpcodeStr, !strconcat(Dt, "8"),
2378 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2379 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2380 OpcodeStr, !strconcat(Dt, "16"),
2381 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2382 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2383 OpcodeStr, !strconcat(Dt, "32"),
2384 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2388 // Neon Multiply-Op vector operations,
2389 // element sizes of 8, 16 and 32 bits:
2390 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2391 InstrItinClass itinD16, InstrItinClass itinD32,
2392 InstrItinClass itinQ16, InstrItinClass itinQ32,
2393 string OpcodeStr, string Dt, SDNode OpNode> {
2394 // 64-bit vector types.
2395 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2396 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2397 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2398 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2399 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2400 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2402 // 128-bit vector types.
2403 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2404 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2405 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2406 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2407 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2408 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2411 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2412 InstrItinClass itinD16, InstrItinClass itinD32,
2413 InstrItinClass itinQ16, InstrItinClass itinQ32,
2414 string OpcodeStr, string Dt, SDNode ShOp> {
2415 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
2416 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
2417 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
2418 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
2419 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
2420 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
2422 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
2423 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
2427 // Neon Intrinsic-Op vector operations,
2428 // element sizes of 8, 16 and 32 bits:
2429 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2430 InstrItinClass itinD, InstrItinClass itinQ,
2431 string OpcodeStr, string Dt, Intrinsic IntOp,
2433 // 64-bit vector types.
2434 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
2435 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
2436 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
2437 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
2438 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
2439 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
2441 // 128-bit vector types.
2442 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
2443 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
2444 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
2445 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
2446 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
2447 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
2450 // Neon 3-argument intrinsics,
2451 // element sizes of 8, 16 and 32 bits:
2452 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2453 InstrItinClass itinD, InstrItinClass itinQ,
2454 string OpcodeStr, string Dt, Intrinsic IntOp> {
2455 // 64-bit vector types.
2456 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
2457 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2458 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
2459 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
2460 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
2461 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
2463 // 128-bit vector types.
2464 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
2465 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
2466 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
2467 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
2468 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
2469 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
2473 // Neon Long Multiply-Op vector operations,
2474 // element sizes of 8, 16 and 32 bits:
2475 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2476 InstrItinClass itin16, InstrItinClass itin32,
2477 string OpcodeStr, string Dt, SDNode MulOp,
2479 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
2480 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
2481 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
2482 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
2483 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
2484 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2487 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
2488 string Dt, SDNode MulOp, SDNode OpNode> {
2489 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
2490 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
2491 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
2492 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
2496 // Neon Long 3-argument intrinsics.
2498 // First with only element sizes of 16 and 32 bits:
2499 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2500 InstrItinClass itin16, InstrItinClass itin32,
2501 string OpcodeStr, string Dt, Intrinsic IntOp> {
2502 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
2503 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2504 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
2505 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2508 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
2509 string OpcodeStr, string Dt, Intrinsic IntOp> {
2510 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
2511 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
2512 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
2513 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2516 // ....then also with element size of 8 bits:
2517 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2518 InstrItinClass itin16, InstrItinClass itin32,
2519 string OpcodeStr, string Dt, Intrinsic IntOp>
2520 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
2521 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
2522 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
2525 // ....with explicit extend (VABAL).
2526 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2527 InstrItinClass itin, string OpcodeStr, string Dt,
2528 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
2529 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
2530 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
2531 IntOp, ExtOp, OpNode>;
2532 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
2533 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
2534 IntOp, ExtOp, OpNode>;
2535 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
2536 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
2537 IntOp, ExtOp, OpNode>;
2541 // Neon 2-register vector intrinsics,
2542 // element sizes of 8, 16 and 32 bits:
2543 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2544 bits<5> op11_7, bit op4,
2545 InstrItinClass itinD, InstrItinClass itinQ,
2546 string OpcodeStr, string Dt, Intrinsic IntOp> {
2547 // 64-bit vector types.
2548 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2549 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2550 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2551 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2552 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2553 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2555 // 128-bit vector types.
2556 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2557 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2558 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2559 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2560 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2561 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2565 // Neon Pairwise long 2-register intrinsics,
2566 // element sizes of 8, 16 and 32 bits:
2567 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2568 bits<5> op11_7, bit op4,
2569 string OpcodeStr, string Dt, Intrinsic IntOp> {
2570 // 64-bit vector types.
2571 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2572 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2573 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2574 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2575 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2576 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2578 // 128-bit vector types.
2579 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2580 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2581 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2582 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2583 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2584 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2588 // Neon Pairwise long 2-register accumulate intrinsics,
2589 // element sizes of 8, 16 and 32 bits:
2590 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2591 bits<5> op11_7, bit op4,
2592 string OpcodeStr, string Dt, Intrinsic IntOp> {
2593 // 64-bit vector types.
2594 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2595 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
2596 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2597 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
2598 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2599 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
2601 // 128-bit vector types.
2602 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2603 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
2604 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2605 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
2606 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2607 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
2611 // Neon 2-register vector shift by immediate,
2612 // with f of either N2RegVShLFrm or N2RegVShRFrm
2613 // element sizes of 8, 16, 32 and 64 bits:
2614 multiclass N2VSh_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2615 InstrItinClass itin, string OpcodeStr, string Dt,
2616 SDNode OpNode, Format f> {
2617 // 64-bit vector types.
2618 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2619 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
2620 let Inst{21-19} = 0b001; // imm6 = 001xxx
2622 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2623 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
2624 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2626 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, f, itin,
2627 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
2628 let Inst{21} = 0b1; // imm6 = 1xxxxx
2630 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, f, itin,
2631 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
2634 // 128-bit vector types.
2635 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2636 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
2637 let Inst{21-19} = 0b001; // imm6 = 001xxx
2639 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2640 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
2641 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2643 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, f, itin,
2644 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
2645 let Inst{21} = 0b1; // imm6 = 1xxxxx
2647 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, f, itin,
2648 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
2652 // Neon Shift-Accumulate vector operations,
2653 // element sizes of 8, 16, 32 and 64 bits:
2654 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2655 string OpcodeStr, string Dt, SDNode ShOp> {
2656 // 64-bit vector types.
2657 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2658 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
2659 let Inst{21-19} = 0b001; // imm6 = 001xxx
2661 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2662 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
2663 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2665 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4,
2666 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
2667 let Inst{21} = 0b1; // imm6 = 1xxxxx
2669 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4,
2670 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
2673 // 128-bit vector types.
2674 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2675 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
2676 let Inst{21-19} = 0b001; // imm6 = 001xxx
2678 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2679 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
2680 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2682 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4,
2683 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
2684 let Inst{21} = 0b1; // imm6 = 1xxxxx
2686 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4,
2687 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
2692 // Neon Shift-Insert vector operations,
2693 // with f of either N2RegVShLFrm or N2RegVShRFrm
2694 // element sizes of 8, 16, 32 and 64 bits:
2695 multiclass N2VShIns_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2696 string OpcodeStr, SDNode ShOp,
2698 // 64-bit vector types.
2699 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4,
2700 f, OpcodeStr, "8", v8i8, ShOp> {
2701 let Inst{21-19} = 0b001; // imm6 = 001xxx
2703 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4,
2704 f, OpcodeStr, "16", v4i16, ShOp> {
2705 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2707 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4,
2708 f, OpcodeStr, "32", v2i32, ShOp> {
2709 let Inst{21} = 0b1; // imm6 = 1xxxxx
2711 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4,
2712 f, OpcodeStr, "64", v1i64, ShOp>;
2715 // 128-bit vector types.
2716 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4,
2717 f, OpcodeStr, "8", v16i8, ShOp> {
2718 let Inst{21-19} = 0b001; // imm6 = 001xxx
2720 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4,
2721 f, OpcodeStr, "16", v8i16, ShOp> {
2722 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2724 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4,
2725 f, OpcodeStr, "32", v4i32, ShOp> {
2726 let Inst{21} = 0b1; // imm6 = 1xxxxx
2728 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4,
2729 f, OpcodeStr, "64", v2i64, ShOp>;
2733 // Neon Shift Long operations,
2734 // element sizes of 8, 16, 32 bits:
2735 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2736 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
2737 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2738 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
2739 let Inst{21-19} = 0b001; // imm6 = 001xxx
2741 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2742 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
2743 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2745 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
2746 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
2747 let Inst{21} = 0b1; // imm6 = 1xxxxx
2751 // Neon Shift Narrow operations,
2752 // element sizes of 16, 32, 64 bits:
2753 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
2754 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
2756 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2757 OpcodeStr, !strconcat(Dt, "16"), v8i8, v8i16, OpNode> {
2758 let Inst{21-19} = 0b001; // imm6 = 001xxx
2760 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2761 OpcodeStr, !strconcat(Dt, "32"), v4i16, v4i32, OpNode> {
2762 let Inst{21-20} = 0b01; // imm6 = 01xxxx
2764 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
2765 OpcodeStr, !strconcat(Dt, "64"), v2i32, v2i64, OpNode> {
2766 let Inst{21} = 0b1; // imm6 = 1xxxxx
2770 //===----------------------------------------------------------------------===//
2771 // Instruction Definitions.
2772 //===----------------------------------------------------------------------===//
2774 // Vector Add Operations.
2776 // VADD : Vector Add (integer and floating-point)
2777 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
2779 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
2780 v2f32, v2f32, fadd, 1>;
2781 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
2782 v4f32, v4f32, fadd, 1>;
2783 // VADDL : Vector Add Long (Q = D + D)
2784 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2785 "vaddl", "s", add, sext, 1>;
2786 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
2787 "vaddl", "u", add, zext, 1>;
2788 // VADDW : Vector Add Wide (Q = Q + D)
2789 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
2790 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
2791 // VHADD : Vector Halving Add
2792 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
2793 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2794 "vhadd", "s", int_arm_neon_vhadds, 1>;
2795 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
2796 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2797 "vhadd", "u", int_arm_neon_vhaddu, 1>;
2798 // VRHADD : Vector Rounding Halving Add
2799 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
2800 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2801 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
2802 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
2803 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2804 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
2805 // VQADD : Vector Saturating Add
2806 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
2807 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2808 "vqadd", "s", int_arm_neon_vqadds, 1>;
2809 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
2810 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
2811 "vqadd", "u", int_arm_neon_vqaddu, 1>;
2812 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
2813 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
2814 int_arm_neon_vaddhn, 1>;
2815 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
2816 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
2817 int_arm_neon_vraddhn, 1>;
2819 // Vector Multiply Operations.
2821 // VMUL : Vector Multiply (integer, polynomial and floating-point)
2822 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
2823 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
2824 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
2825 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
2826 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
2827 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
2828 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
2829 v2f32, v2f32, fmul, 1>;
2830 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
2831 v4f32, v4f32, fmul, 1>;
2832 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
2833 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
2834 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
2837 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
2838 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
2839 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
2840 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2841 (DSubReg_i16_reg imm:$lane))),
2842 (SubReg_i16_lane imm:$lane)))>;
2843 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
2844 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
2845 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
2846 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2847 (DSubReg_i32_reg imm:$lane))),
2848 (SubReg_i32_lane imm:$lane)))>;
2849 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
2850 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
2851 (v4f32 (VMULslfq (v4f32 QPR:$src1),
2852 (v2f32 (EXTRACT_SUBREG QPR:$src2,
2853 (DSubReg_i32_reg imm:$lane))),
2854 (SubReg_i32_lane imm:$lane)))>;
2856 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
2857 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
2858 IIC_VMULi16Q, IIC_VMULi32Q,
2859 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
2860 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
2861 IIC_VMULi16Q, IIC_VMULi32Q,
2862 "vqdmulh", "s", int_arm_neon_vqdmulh>;
2863 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
2864 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2866 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
2867 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2868 (DSubReg_i16_reg imm:$lane))),
2869 (SubReg_i16_lane imm:$lane)))>;
2870 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
2871 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2873 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
2874 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2875 (DSubReg_i32_reg imm:$lane))),
2876 (SubReg_i32_lane imm:$lane)))>;
2878 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
2879 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
2880 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
2881 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
2882 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
2883 IIC_VMULi16Q, IIC_VMULi32Q,
2884 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
2885 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
2886 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
2888 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
2889 (v4i16 (EXTRACT_SUBREG QPR:$src2,
2890 (DSubReg_i16_reg imm:$lane))),
2891 (SubReg_i16_lane imm:$lane)))>;
2892 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
2893 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
2895 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
2896 (v2i32 (EXTRACT_SUBREG QPR:$src2,
2897 (DSubReg_i32_reg imm:$lane))),
2898 (SubReg_i32_lane imm:$lane)))>;
2900 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
2901 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2902 "vmull", "s", NEONvmulls, 1>;
2903 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
2904 "vmull", "u", NEONvmullu, 1>;
2905 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
2906 v8i16, v8i8, int_arm_neon_vmullp, 1>;
2907 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
2908 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
2910 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
2911 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
2912 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
2913 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
2914 "vqdmull", "s", int_arm_neon_vqdmull>;
2916 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
2918 // VMLA : Vector Multiply Accumulate (integer and floating-point)
2919 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2920 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2921 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
2923 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
2925 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
2926 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
2927 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
2929 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
2930 v4f32, v2f32, fmul, fadd>;
2932 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
2933 (mul (v8i16 QPR:$src2),
2934 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2935 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2936 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2937 (DSubReg_i16_reg imm:$lane))),
2938 (SubReg_i16_lane imm:$lane)))>;
2940 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
2941 (mul (v4i32 QPR:$src2),
2942 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2943 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2944 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2945 (DSubReg_i32_reg imm:$lane))),
2946 (SubReg_i32_lane imm:$lane)))>;
2948 def : Pat<(v4f32 (fadd (v4f32 QPR:$src1),
2949 (fmul (v4f32 QPR:$src2),
2950 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
2951 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
2953 (v2f32 (EXTRACT_SUBREG QPR:$src3,
2954 (DSubReg_i32_reg imm:$lane))),
2955 (SubReg_i32_lane imm:$lane)))>;
2957 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
2958 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2959 "vmlal", "s", NEONvmulls, add>;
2960 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
2961 "vmlal", "u", NEONvmullu, add>;
2963 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
2964 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
2966 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
2967 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2968 "vqdmlal", "s", int_arm_neon_vqdmlal>;
2969 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
2971 // VMLS : Vector Multiply Subtract (integer and floating-point)
2972 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
2973 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2974 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
2976 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
2978 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
2979 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
2980 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
2982 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
2983 v4f32, v2f32, fmul, fsub>;
2985 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
2986 (mul (v8i16 QPR:$src2),
2987 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
2988 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
2989 (v4i16 (EXTRACT_SUBREG QPR:$src3,
2990 (DSubReg_i16_reg imm:$lane))),
2991 (SubReg_i16_lane imm:$lane)))>;
2993 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
2994 (mul (v4i32 QPR:$src2),
2995 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
2996 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
2997 (v2i32 (EXTRACT_SUBREG QPR:$src3,
2998 (DSubReg_i32_reg imm:$lane))),
2999 (SubReg_i32_lane imm:$lane)))>;
3001 def : Pat<(v4f32 (fsub (v4f32 QPR:$src1),
3002 (fmul (v4f32 QPR:$src2),
3003 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3004 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3005 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3006 (DSubReg_i32_reg imm:$lane))),
3007 (SubReg_i32_lane imm:$lane)))>;
3009 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3010 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3011 "vmlsl", "s", NEONvmulls, sub>;
3012 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3013 "vmlsl", "u", NEONvmullu, sub>;
3015 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3016 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3018 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3019 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3020 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3021 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3023 // Vector Subtract Operations.
3025 // VSUB : Vector Subtract (integer and floating-point)
3026 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3027 "vsub", "i", sub, 0>;
3028 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3029 v2f32, v2f32, fsub, 0>;
3030 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3031 v4f32, v4f32, fsub, 0>;
3032 // VSUBL : Vector Subtract Long (Q = D - D)
3033 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3034 "vsubl", "s", sub, sext, 0>;
3035 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3036 "vsubl", "u", sub, zext, 0>;
3037 // VSUBW : Vector Subtract Wide (Q = Q - D)
3038 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3039 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3040 // VHSUB : Vector Halving Subtract
3041 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3042 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3043 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3044 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3045 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3046 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3047 // VQSUB : Vector Saturing Subtract
3048 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3049 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3050 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3051 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3052 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3053 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3054 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3055 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3056 int_arm_neon_vsubhn, 0>;
3057 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3058 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3059 int_arm_neon_vrsubhn, 0>;
3061 // Vector Comparisons.
3063 // VCEQ : Vector Compare Equal
3064 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3065 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3066 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3068 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3070 // For disassembly only.
3071 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3074 // VCGE : Vector Compare Greater Than or Equal
3075 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3076 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3077 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3078 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3079 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3081 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3083 // For disassembly only.
3084 // FIXME: This instruction's encoding MAY NOT BE correct.
3085 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3087 // For disassembly only.
3088 // FIXME: This instruction's encoding MAY NOT BE correct.
3089 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3092 // VCGT : Vector Compare Greater Than
3093 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3094 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3095 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3096 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3097 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3099 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3101 // For disassembly only.
3102 // FIXME: This instruction's encoding MAY NOT BE correct.
3103 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3105 // For disassembly only.
3106 // FIXME: This instruction's encoding MAY NOT BE correct.
3107 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3110 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3111 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3112 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3113 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3114 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3115 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3116 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3117 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3118 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3119 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3120 // VTST : Vector Test Bits
3121 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3122 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3124 // Vector Bitwise Operations.
3126 def vnotd : PatFrag<(ops node:$in),
3127 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3128 def vnotq : PatFrag<(ops node:$in),
3129 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3132 // VAND : Vector Bitwise AND
3133 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3134 v2i32, v2i32, and, 1>;
3135 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3136 v4i32, v4i32, and, 1>;
3138 // VEOR : Vector Bitwise Exclusive OR
3139 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3140 v2i32, v2i32, xor, 1>;
3141 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3142 v4i32, v4i32, xor, 1>;
3144 // VORR : Vector Bitwise OR
3145 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3146 v2i32, v2i32, or, 1>;
3147 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3148 v4i32, v4i32, or, 1>;
3150 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3151 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$dst),
3152 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3153 "vbic", "$dst, $src1, $src2", "",
3154 [(set DPR:$dst, (v2i32 (and DPR:$src1,
3155 (vnotd DPR:$src2))))]>;
3156 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$dst),
3157 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3158 "vbic", "$dst, $src1, $src2", "",
3159 [(set QPR:$dst, (v4i32 (and QPR:$src1,
3160 (vnotq QPR:$src2))))]>;
3162 // VORN : Vector Bitwise OR NOT
3163 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$dst),
3164 (ins DPR:$src1, DPR:$src2), N3RegFrm, IIC_VBINiD,
3165 "vorn", "$dst, $src1, $src2", "",
3166 [(set DPR:$dst, (v2i32 (or DPR:$src1,
3167 (vnotd DPR:$src2))))]>;
3168 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$dst),
3169 (ins QPR:$src1, QPR:$src2), N3RegFrm, IIC_VBINiQ,
3170 "vorn", "$dst, $src1, $src2", "",
3171 [(set QPR:$dst, (v4i32 (or QPR:$src1,
3172 (vnotq QPR:$src2))))]>;
3174 // VMVN : Vector Bitwise NOT (Immediate)
3176 let isReMaterializable = 1 in {
3178 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$dst),
3179 (ins nModImm:$SIMM), IIC_VMOVImm,
3180 "vmvn", "i16", "$dst, $SIMM", "",
3181 [(set DPR:$dst, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3182 let Inst{9} = SIMM{9};
3185 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$dst),
3186 (ins nModImm:$SIMM), IIC_VMOVImm,
3187 "vmvn", "i16", "$dst, $SIMM", "",
3188 [(set QPR:$dst, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3189 let Inst{9} = SIMM{9};
3192 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$dst),
3193 (ins nModImm:$SIMM), IIC_VMOVImm,
3194 "vmvn", "i32", "$dst, $SIMM", "",
3195 [(set DPR:$dst, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3196 let Inst{11-8} = SIMM{11-8};
3199 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$dst),
3200 (ins nModImm:$SIMM), IIC_VMOVImm,
3201 "vmvn", "i32", "$dst, $SIMM", "",
3202 [(set QPR:$dst, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3203 let Inst{11-8} = SIMM{11-8};
3207 // VMVN : Vector Bitwise NOT
3208 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3209 (outs DPR:$dst), (ins DPR:$src), IIC_VSUBiD,
3210 "vmvn", "$dst, $src", "",
3211 [(set DPR:$dst, (v2i32 (vnotd DPR:$src)))]>;
3212 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3213 (outs QPR:$dst), (ins QPR:$src), IIC_VSUBiD,
3214 "vmvn", "$dst, $src", "",
3215 [(set QPR:$dst, (v4i32 (vnotq QPR:$src)))]>;
3216 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3217 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3219 // VBSL : Vector Bitwise Select
3220 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3221 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3222 N3RegFrm, IIC_VCNTiD,
3223 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3225 (v2i32 (or (and DPR:$Vn, DPR:$src1),
3226 (and DPR:$Vm, (vnotd DPR:$src1)))))]>;
3227 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3228 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3229 N3RegFrm, IIC_VCNTiQ,
3230 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3232 (v4i32 (or (and QPR:$Vn, QPR:$src1),
3233 (and QPR:$Vm, (vnotq QPR:$src1)))))]>;
3235 // VBIF : Vector Bitwise Insert if False
3236 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3237 // FIXME: This instruction's encoding MAY NOT BE correct.
3238 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3239 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3240 N3RegFrm, IIC_VBINiD,
3241 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3242 [/* For disassembly only; pattern left blank */]>;
3243 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3244 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3245 N3RegFrm, IIC_VBINiQ,
3246 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3247 [/* For disassembly only; pattern left blank */]>;
3249 // VBIT : Vector Bitwise Insert if True
3250 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3251 // FIXME: This instruction's encoding MAY NOT BE correct.
3252 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3253 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3254 N3RegFrm, IIC_VBINiD,
3255 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3256 [/* For disassembly only; pattern left blank */]>;
3257 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3258 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3259 N3RegFrm, IIC_VBINiQ,
3260 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3261 [/* For disassembly only; pattern left blank */]>;
3263 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3264 // for equivalent operations with different register constraints; it just
3267 // Vector Absolute Differences.
3269 // VABD : Vector Absolute Difference
3270 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3271 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3272 "vabd", "s", int_arm_neon_vabds, 1>;
3273 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3274 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3275 "vabd", "u", int_arm_neon_vabdu, 1>;
3276 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3277 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3278 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3279 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
3281 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
3282 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
3283 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
3284 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
3285 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
3287 // VABA : Vector Absolute Difference and Accumulate
3288 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3289 "vaba", "s", int_arm_neon_vabds, add>;
3290 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
3291 "vaba", "u", int_arm_neon_vabdu, add>;
3293 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
3294 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
3295 "vabal", "s", int_arm_neon_vabds, zext, add>;
3296 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
3297 "vabal", "u", int_arm_neon_vabdu, zext, add>;
3299 // Vector Maximum and Minimum.
3301 // VMAX : Vector Maximum
3302 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
3303 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3304 "vmax", "s", int_arm_neon_vmaxs, 1>;
3305 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
3306 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3307 "vmax", "u", int_arm_neon_vmaxu, 1>;
3308 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
3310 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
3311 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3313 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
3315 // VMIN : Vector Minimum
3316 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
3317 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3318 "vmin", "s", int_arm_neon_vmins, 1>;
3319 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
3320 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3321 "vmin", "u", int_arm_neon_vminu, 1>;
3322 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
3324 v2f32, v2f32, int_arm_neon_vmins, 1>;
3325 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
3327 v4f32, v4f32, int_arm_neon_vmins, 1>;
3329 // Vector Pairwise Operations.
3331 // VPADD : Vector Pairwise Add
3332 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3334 v8i8, v8i8, int_arm_neon_vpadd, 0>;
3335 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3337 v4i16, v4i16, int_arm_neon_vpadd, 0>;
3338 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
3340 v2i32, v2i32, int_arm_neon_vpadd, 0>;
3341 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
3342 IIC_VPBIND, "vpadd", "f32",
3343 v2f32, v2f32, int_arm_neon_vpadd, 0>;
3345 // VPADDL : Vector Pairwise Add Long
3346 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
3347 int_arm_neon_vpaddls>;
3348 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
3349 int_arm_neon_vpaddlu>;
3351 // VPADAL : Vector Pairwise Add and Accumulate Long
3352 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
3353 int_arm_neon_vpadals>;
3354 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
3355 int_arm_neon_vpadalu>;
3357 // VPMAX : Vector Pairwise Maximum
3358 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3359 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
3360 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3361 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
3362 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3363 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
3364 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3365 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
3366 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3367 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
3368 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
3369 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
3370 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
3371 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
3373 // VPMIN : Vector Pairwise Minimum
3374 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3375 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
3376 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3377 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
3378 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3379 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
3380 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3381 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
3382 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3383 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
3384 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
3385 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
3386 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
3387 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
3389 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
3391 // VRECPE : Vector Reciprocal Estimate
3392 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3393 IIC_VUNAD, "vrecpe", "u32",
3394 v2i32, v2i32, int_arm_neon_vrecpe>;
3395 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
3396 IIC_VUNAQ, "vrecpe", "u32",
3397 v4i32, v4i32, int_arm_neon_vrecpe>;
3398 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3399 IIC_VUNAD, "vrecpe", "f32",
3400 v2f32, v2f32, int_arm_neon_vrecpe>;
3401 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
3402 IIC_VUNAQ, "vrecpe", "f32",
3403 v4f32, v4f32, int_arm_neon_vrecpe>;
3405 // VRECPS : Vector Reciprocal Step
3406 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3407 IIC_VRECSD, "vrecps", "f32",
3408 v2f32, v2f32, int_arm_neon_vrecps, 1>;
3409 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
3410 IIC_VRECSQ, "vrecps", "f32",
3411 v4f32, v4f32, int_arm_neon_vrecps, 1>;
3413 // VRSQRTE : Vector Reciprocal Square Root Estimate
3414 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3415 IIC_VUNAD, "vrsqrte", "u32",
3416 v2i32, v2i32, int_arm_neon_vrsqrte>;
3417 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
3418 IIC_VUNAQ, "vrsqrte", "u32",
3419 v4i32, v4i32, int_arm_neon_vrsqrte>;
3420 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3421 IIC_VUNAD, "vrsqrte", "f32",
3422 v2f32, v2f32, int_arm_neon_vrsqrte>;
3423 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
3424 IIC_VUNAQ, "vrsqrte", "f32",
3425 v4f32, v4f32, int_arm_neon_vrsqrte>;
3427 // VRSQRTS : Vector Reciprocal Square Root Step
3428 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3429 IIC_VRECSD, "vrsqrts", "f32",
3430 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
3431 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
3432 IIC_VRECSQ, "vrsqrts", "f32",
3433 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
3437 // VSHL : Vector Shift
3438 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
3439 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3440 "vshl", "s", int_arm_neon_vshifts>;
3441 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
3442 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
3443 "vshl", "u", int_arm_neon_vshiftu>;
3444 // VSHL : Vector Shift Left (Immediate)
3445 defm VSHLi : N2VSh_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl,
3447 // VSHR : Vector Shift Right (Immediate)
3448 defm VSHRs : N2VSh_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s", NEONvshrs,
3450 defm VSHRu : N2VSh_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u", NEONvshru,
3453 // VSHLL : Vector Shift Left Long
3454 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
3455 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
3457 // VSHLL : Vector Shift Left Long (with maximum shift count)
3458 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
3459 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
3460 ValueType OpTy, SDNode OpNode>
3461 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
3462 ResTy, OpTy, OpNode> {
3463 let Inst{21-16} = op21_16;
3465 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
3466 v8i16, v8i8, NEONvshlli>;
3467 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
3468 v4i32, v4i16, NEONvshlli>;
3469 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
3470 v2i64, v2i32, NEONvshlli>;
3472 // VSHRN : Vector Shift Right and Narrow
3473 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
3476 // VRSHL : Vector Rounding Shift
3477 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
3478 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3479 "vrshl", "s", int_arm_neon_vrshifts>;
3480 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
3481 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3482 "vrshl", "u", int_arm_neon_vrshiftu>;
3483 // VRSHR : Vector Rounding Shift Right
3484 defm VRSHRs : N2VSh_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s", NEONvrshrs,
3486 defm VRSHRu : N2VSh_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u", NEONvrshru,
3489 // VRSHRN : Vector Rounding Shift Right and Narrow
3490 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
3493 // VQSHL : Vector Saturating Shift
3494 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
3495 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3496 "vqshl", "s", int_arm_neon_vqshifts>;
3497 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
3498 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3499 "vqshl", "u", int_arm_neon_vqshiftu>;
3500 // VQSHL : Vector Saturating Shift Left (Immediate)
3501 defm VQSHLsi : N2VSh_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls,
3503 defm VQSHLui : N2VSh_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu,
3505 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
3506 defm VQSHLsu : N2VSh_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu,
3509 // VQSHRN : Vector Saturating Shift Right and Narrow
3510 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
3512 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
3515 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
3516 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
3519 // VQRSHL : Vector Saturating Rounding Shift
3520 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
3521 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3522 "vqrshl", "s", int_arm_neon_vqrshifts>;
3523 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
3524 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
3525 "vqrshl", "u", int_arm_neon_vqrshiftu>;
3527 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
3528 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
3530 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
3533 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
3534 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
3537 // VSRA : Vector Shift Right and Accumulate
3538 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
3539 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
3540 // VRSRA : Vector Rounding Shift Right and Accumulate
3541 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
3542 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
3544 // VSLI : Vector Shift Left and Insert
3545 defm VSLI : N2VShIns_QHSD<1, 1, 0b0101, 1, "vsli", NEONvsli, N2RegVShLFrm>;
3546 // VSRI : Vector Shift Right and Insert
3547 defm VSRI : N2VShIns_QHSD<1, 1, 0b0100, 1, "vsri", NEONvsri, N2RegVShRFrm>;
3549 // Vector Absolute and Saturating Absolute.
3551 // VABS : Vector Absolute Value
3552 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
3553 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
3555 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3556 IIC_VUNAD, "vabs", "f32",
3557 v2f32, v2f32, int_arm_neon_vabs>;
3558 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
3559 IIC_VUNAQ, "vabs", "f32",
3560 v4f32, v4f32, int_arm_neon_vabs>;
3562 // VQABS : Vector Saturating Absolute Value
3563 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
3564 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
3565 int_arm_neon_vqabs>;
3569 def vnegd : PatFrag<(ops node:$in),
3570 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
3571 def vnegq : PatFrag<(ops node:$in),
3572 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
3574 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3575 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$dst), (ins DPR:$src),
3576 IIC_VSHLiD, OpcodeStr, Dt, "$dst, $src", "",
3577 [(set DPR:$dst, (Ty (vnegd DPR:$src)))]>;
3578 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
3579 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$dst), (ins QPR:$src),
3580 IIC_VSHLiQ, OpcodeStr, Dt, "$dst, $src", "",
3581 [(set QPR:$dst, (Ty (vnegq QPR:$src)))]>;
3583 // VNEG : Vector Negate (integer)
3584 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
3585 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
3586 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
3587 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
3588 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
3589 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
3591 // VNEG : Vector Negate (floating-point)
3592 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
3593 (outs DPR:$dst), (ins DPR:$src), IIC_VUNAD,
3594 "vneg", "f32", "$dst, $src", "",
3595 [(set DPR:$dst, (v2f32 (fneg DPR:$src)))]>;
3596 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
3597 (outs QPR:$dst), (ins QPR:$src), IIC_VUNAQ,
3598 "vneg", "f32", "$dst, $src", "",
3599 [(set QPR:$dst, (v4f32 (fneg QPR:$src)))]>;
3601 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
3602 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
3603 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
3604 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
3605 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
3606 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
3608 // VQNEG : Vector Saturating Negate
3609 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
3610 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
3611 int_arm_neon_vqneg>;
3613 // Vector Bit Counting Operations.
3615 // VCLS : Vector Count Leading Sign Bits
3616 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
3617 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
3619 // VCLZ : Vector Count Leading Zeros
3620 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
3621 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
3623 // VCNT : Vector Count One Bits
3624 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3625 IIC_VCNTiD, "vcnt", "8",
3626 v8i8, v8i8, int_arm_neon_vcnt>;
3627 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
3628 IIC_VCNTiQ, "vcnt", "8",
3629 v16i8, v16i8, int_arm_neon_vcnt>;
3631 // Vector Swap -- for disassembly only.
3632 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
3633 (outs DPR:$dst), (ins DPR:$src), NoItinerary,
3634 "vswp", "$dst, $src", "", []>;
3635 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
3636 (outs QPR:$dst), (ins QPR:$src), NoItinerary,
3637 "vswp", "$dst, $src", "", []>;
3639 // Vector Move Operations.
3641 // VMOV : Vector Move (Register)
3643 let neverHasSideEffects = 1 in {
3644 def VMOVDneon: N3VX<0, 0, 0b10, 0b0001, 0, 1, (outs DPR:$dst), (ins DPR:$src),
3645 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3646 def VMOVQ : N3VX<0, 0, 0b10, 0b0001, 1, 1, (outs QPR:$dst), (ins QPR:$src),
3647 N3RegFrm, IIC_VMOV, "vmov", "$dst, $src", "", []>;
3649 // Pseudo vector move instructions for QQ and QQQQ registers. This should
3650 // be expanded after register allocation is completed.
3651 def VMOVQQ : PseudoInst<(outs QQPR:$dst), (ins QQPR:$src),
3652 NoItinerary, "", []>;
3654 def VMOVQQQQ : PseudoInst<(outs QQQQPR:$dst), (ins QQQQPR:$src),
3655 NoItinerary, "", []>;
3656 } // neverHasSideEffects
3658 // VMOV : Vector Move (Immediate)
3660 let isReMaterializable = 1 in {
3661 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$dst),
3662 (ins nModImm:$SIMM), IIC_VMOVImm,
3663 "vmov", "i8", "$dst, $SIMM", "",
3664 [(set DPR:$dst, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
3665 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$dst),
3666 (ins nModImm:$SIMM), IIC_VMOVImm,
3667 "vmov", "i8", "$dst, $SIMM", "",
3668 [(set QPR:$dst, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
3670 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$dst),
3671 (ins nModImm:$SIMM), IIC_VMOVImm,
3672 "vmov", "i16", "$dst, $SIMM", "",
3673 [(set DPR:$dst, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
3674 let Inst{9} = SIMM{9};
3677 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$dst),
3678 (ins nModImm:$SIMM), IIC_VMOVImm,
3679 "vmov", "i16", "$dst, $SIMM", "",
3680 [(set QPR:$dst, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
3681 let Inst{9} = SIMM{9};
3684 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$dst),
3685 (ins nModImm:$SIMM), IIC_VMOVImm,
3686 "vmov", "i32", "$dst, $SIMM", "",
3687 [(set DPR:$dst, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
3688 let Inst{11-8} = SIMM{11-8};
3691 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$dst),
3692 (ins nModImm:$SIMM), IIC_VMOVImm,
3693 "vmov", "i32", "$dst, $SIMM", "",
3694 [(set QPR:$dst, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
3695 let Inst{11-8} = SIMM{11-8};
3698 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$dst),
3699 (ins nModImm:$SIMM), IIC_VMOVImm,
3700 "vmov", "i64", "$dst, $SIMM", "",
3701 [(set DPR:$dst, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
3702 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$dst),
3703 (ins nModImm:$SIMM), IIC_VMOVImm,
3704 "vmov", "i64", "$dst, $SIMM", "",
3705 [(set QPR:$dst, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
3706 } // isReMaterializable
3708 // VMOV : Vector Get Lane (move scalar to ARM core register)
3710 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
3711 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3712 IIC_VMOVSI, "vmov", "s8", "$R, $V[$lane]",
3713 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
3715 let Inst{21} = lane{2};
3716 let Inst{6-5} = lane{1-0};
3718 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
3719 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3720 IIC_VMOVSI, "vmov", "s16", "$R, $V[$lane]",
3721 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
3723 let Inst{21} = lane{1};
3724 let Inst{6} = lane{0};
3726 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
3727 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3728 IIC_VMOVSI, "vmov", "u8", "$R, $V[$lane]",
3729 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
3731 let Inst{21} = lane{2};
3732 let Inst{6-5} = lane{1-0};
3734 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
3735 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3736 IIC_VMOVSI, "vmov", "u16", "$R, $V[$lane]",
3737 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
3739 let Inst{21} = lane{1};
3740 let Inst{6} = lane{0};
3742 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
3743 (outs GPR:$R), (ins DPR:$V, nohash_imm:$lane),
3744 IIC_VMOVSI, "vmov", "32", "$R, $V[$lane]",
3745 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
3747 let Inst{21} = lane{0};
3749 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
3750 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
3751 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3752 (DSubReg_i8_reg imm:$lane))),
3753 (SubReg_i8_lane imm:$lane))>;
3754 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
3755 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3756 (DSubReg_i16_reg imm:$lane))),
3757 (SubReg_i16_lane imm:$lane))>;
3758 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
3759 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
3760 (DSubReg_i8_reg imm:$lane))),
3761 (SubReg_i8_lane imm:$lane))>;
3762 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
3763 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
3764 (DSubReg_i16_reg imm:$lane))),
3765 (SubReg_i16_lane imm:$lane))>;
3766 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
3767 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
3768 (DSubReg_i32_reg imm:$lane))),
3769 (SubReg_i32_lane imm:$lane))>;
3770 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
3771 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
3772 (SSubReg_f32_reg imm:$src2))>;
3773 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
3774 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
3775 (SSubReg_f32_reg imm:$src2))>;
3776 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
3777 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3778 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
3779 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
3782 // VMOV : Vector Set Lane (move ARM core register to scalar)
3784 let Constraints = "$src1 = $V" in {
3785 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
3786 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3787 IIC_VMOVISL, "vmov", "8", "$V[$lane], $R",
3788 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
3789 GPR:$R, imm:$lane))]> {
3790 let Inst{21} = lane{2};
3791 let Inst{6-5} = lane{1-0};
3793 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
3794 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3795 IIC_VMOVISL, "vmov", "16", "$V[$lane], $R",
3796 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
3797 GPR:$R, imm:$lane))]> {
3798 let Inst{21} = lane{1};
3799 let Inst{6} = lane{0};
3801 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
3802 (ins DPR:$src1, GPR:$R, nohash_imm:$lane),
3803 IIC_VMOVISL, "vmov", "32", "$V[$lane], $R",
3804 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
3805 GPR:$R, imm:$lane))]> {
3806 let Inst{21} = lane{0};
3809 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
3810 (v16i8 (INSERT_SUBREG QPR:$src1,
3811 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
3812 (DSubReg_i8_reg imm:$lane))),
3813 GPR:$src2, (SubReg_i8_lane imm:$lane))),
3814 (DSubReg_i8_reg imm:$lane)))>;
3815 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
3816 (v8i16 (INSERT_SUBREG QPR:$src1,
3817 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
3818 (DSubReg_i16_reg imm:$lane))),
3819 GPR:$src2, (SubReg_i16_lane imm:$lane))),
3820 (DSubReg_i16_reg imm:$lane)))>;
3821 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
3822 (v4i32 (INSERT_SUBREG QPR:$src1,
3823 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
3824 (DSubReg_i32_reg imm:$lane))),
3825 GPR:$src2, (SubReg_i32_lane imm:$lane))),
3826 (DSubReg_i32_reg imm:$lane)))>;
3828 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
3829 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
3830 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3831 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
3832 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
3833 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
3835 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3836 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3837 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
3838 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
3840 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
3841 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3842 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
3843 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
3844 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
3845 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
3847 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
3848 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3849 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
3850 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3851 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
3852 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
3854 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
3855 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
3856 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3858 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
3859 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
3860 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3862 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
3863 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
3864 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
3867 // VDUP : Vector Duplicate (from ARM core register to all elements)
3869 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3870 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$dst), (ins GPR:$src),
3871 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3872 [(set DPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3873 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
3874 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$dst), (ins GPR:$src),
3875 IIC_VMOVIS, "vdup", Dt, "$dst, $src",
3876 [(set QPR:$dst, (Ty (NEONvdup (i32 GPR:$src))))]>;
3878 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
3879 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
3880 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
3881 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
3882 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
3883 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
3885 def VDUPfd : NVDup<0b11101000, 0b1011, 0b00, (outs DPR:$dst), (ins GPR:$src),
3886 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3887 [(set DPR:$dst, (v2f32 (NEONvdup
3888 (f32 (bitconvert GPR:$src)))))]>;
3889 def VDUPfq : NVDup<0b11101010, 0b1011, 0b00, (outs QPR:$dst), (ins GPR:$src),
3890 IIC_VMOVIS, "vdup", "32", "$dst, $src",
3891 [(set QPR:$dst, (v4f32 (NEONvdup
3892 (f32 (bitconvert GPR:$src)))))]>;
3894 // VDUP : Vector Duplicate Lane (from scalar to all elements)
3896 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
3898 : NVDupLane<op19_16, 0, (outs DPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3899 IIC_VMOVD, OpcodeStr, Dt, "$dst, $src[$lane]",
3900 [(set DPR:$dst, (Ty (NEONvduplane (Ty DPR:$src), imm:$lane)))]>;
3902 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
3903 ValueType ResTy, ValueType OpTy>
3904 : NVDupLane<op19_16, 1, (outs QPR:$dst), (ins DPR:$src, nohash_imm:$lane),
3905 IIC_VMOVQ, OpcodeStr, Dt, "$dst, $src[$lane]",
3906 [(set QPR:$dst, (ResTy (NEONvduplane (OpTy DPR:$src),
3909 // Inst{19-16} is partially specified depending on the element size.
3911 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8> {
3912 let Inst{19-17} = lane{2-0};
3914 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16> {
3915 let Inst{19-18} = lane{1-0};
3917 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32> {
3918 let Inst{19} = lane{0};
3920 def VDUPLNfd : VDUPLND<{?,1,0,0}, "vdup", "32", v2f32> {
3921 let Inst{19} = lane{0};
3923 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8> {
3924 let Inst{19-17} = lane{2-0};
3926 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16> {
3927 let Inst{19-18} = lane{1-0};
3929 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32> {
3930 let Inst{19} = lane{0};
3932 def VDUPLNfq : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4f32, v2f32> {
3933 let Inst{19} = lane{0};
3936 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
3937 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
3938 (DSubReg_i8_reg imm:$lane))),
3939 (SubReg_i8_lane imm:$lane)))>;
3940 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
3941 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
3942 (DSubReg_i16_reg imm:$lane))),
3943 (SubReg_i16_lane imm:$lane)))>;
3944 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
3945 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
3946 (DSubReg_i32_reg imm:$lane))),
3947 (SubReg_i32_lane imm:$lane)))>;
3948 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
3949 (v4f32 (VDUPLNfq (v2f32 (EXTRACT_SUBREG QPR:$src,
3950 (DSubReg_i32_reg imm:$lane))),
3951 (SubReg_i32_lane imm:$lane)))>;
3953 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
3954 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
3955 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
3956 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
3958 // VMOVN : Vector Narrowing Move
3959 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
3960 "vmovn", "i", trunc>;
3961 // VQMOVN : Vector Saturating Narrowing Move
3962 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
3963 "vqmovn", "s", int_arm_neon_vqmovns>;
3964 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
3965 "vqmovn", "u", int_arm_neon_vqmovnu>;
3966 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
3967 "vqmovun", "s", int_arm_neon_vqmovnsu>;
3968 // VMOVL : Vector Lengthening Move
3969 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
3970 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
3972 // Vector Conversions.
3974 // VCVT : Vector Convert Between Floating-Point and Integers
3975 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3976 v2i32, v2f32, fp_to_sint>;
3977 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3978 v2i32, v2f32, fp_to_uint>;
3979 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3980 v2f32, v2i32, sint_to_fp>;
3981 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3982 v2f32, v2i32, uint_to_fp>;
3984 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
3985 v4i32, v4f32, fp_to_sint>;
3986 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
3987 v4i32, v4f32, fp_to_uint>;
3988 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
3989 v4f32, v4i32, sint_to_fp>;
3990 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
3991 v4f32, v4i32, uint_to_fp>;
3993 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
3994 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
3995 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
3996 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
3997 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
3998 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
3999 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4000 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4001 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4003 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4004 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4005 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4006 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4007 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4008 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4009 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4010 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4014 // VREV64 : Vector Reverse elements within 64-bit doublewords
4016 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4017 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$dst),
4018 (ins DPR:$src), IIC_VMOVD,
4019 OpcodeStr, Dt, "$dst, $src", "",
4020 [(set DPR:$dst, (Ty (NEONvrev64 (Ty DPR:$src))))]>;
4021 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4022 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$dst),
4023 (ins QPR:$src), IIC_VMOVQ,
4024 OpcodeStr, Dt, "$dst, $src", "",
4025 [(set QPR:$dst, (Ty (NEONvrev64 (Ty QPR:$src))))]>;
4027 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4028 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4029 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4030 def VREV64df : VREV64D<0b10, "vrev64", "32", v2f32>;
4032 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4033 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4034 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4035 def VREV64qf : VREV64Q<0b10, "vrev64", "32", v4f32>;
4037 // VREV32 : Vector Reverse elements within 32-bit words
4039 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4040 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$dst),
4041 (ins DPR:$src), IIC_VMOVD,
4042 OpcodeStr, Dt, "$dst, $src", "",
4043 [(set DPR:$dst, (Ty (NEONvrev32 (Ty DPR:$src))))]>;
4044 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4045 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$dst),
4046 (ins QPR:$src), IIC_VMOVQ,
4047 OpcodeStr, Dt, "$dst, $src", "",
4048 [(set QPR:$dst, (Ty (NEONvrev32 (Ty QPR:$src))))]>;
4050 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4051 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4053 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4054 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4056 // VREV16 : Vector Reverse elements within 16-bit halfwords
4058 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4059 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$dst),
4060 (ins DPR:$src), IIC_VMOVD,
4061 OpcodeStr, Dt, "$dst, $src", "",
4062 [(set DPR:$dst, (Ty (NEONvrev16 (Ty DPR:$src))))]>;
4063 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4064 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$dst),
4065 (ins QPR:$src), IIC_VMOVQ,
4066 OpcodeStr, Dt, "$dst, $src", "",
4067 [(set QPR:$dst, (Ty (NEONvrev16 (Ty QPR:$src))))]>;
4069 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4070 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4072 // Other Vector Shuffles.
4074 // VEXT : Vector Extract
4076 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4077 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$dst),
4078 (ins DPR:$lhs, DPR:$rhs, i32imm:$index), NVExtFrm,
4079 IIC_VEXTD, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4080 [(set DPR:$dst, (Ty (NEONvext (Ty DPR:$lhs),
4081 (Ty DPR:$rhs), imm:$index)))]> {
4083 let Inst{11-8} = index{3-0};
4086 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4087 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$dst),
4088 (ins QPR:$lhs, QPR:$rhs, i32imm:$index), NVExtFrm,
4089 IIC_VEXTQ, OpcodeStr, Dt, "$dst, $lhs, $rhs, $index", "",
4090 [(set QPR:$dst, (Ty (NEONvext (Ty QPR:$lhs),
4091 (Ty QPR:$rhs), imm:$index)))]> {
4093 let Inst{11-8} = index{3-0};
4096 def VEXTd8 : VEXTd<"vext", "8", v8i8>;
4097 def VEXTd16 : VEXTd<"vext", "16", v4i16>;
4098 def VEXTd32 : VEXTd<"vext", "32", v2i32>;
4099 def VEXTdf : VEXTd<"vext", "32", v2f32>;
4101 def VEXTq8 : VEXTq<"vext", "8", v16i8>;
4102 def VEXTq16 : VEXTq<"vext", "16", v8i16>;
4103 def VEXTq32 : VEXTq<"vext", "32", v4i32>;
4104 def VEXTqf : VEXTq<"vext", "32", v4f32>;
4106 // VTRN : Vector Transpose
4108 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4109 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4110 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4112 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4113 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4114 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4116 // VUZP : Vector Unzip (Deinterleave)
4118 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4119 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4120 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4122 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4123 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4124 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4126 // VZIP : Vector Zip (Interleave)
4128 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4129 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4130 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4132 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4133 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4134 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4136 // Vector Table Lookup and Table Extension.
4138 // VTBL : Vector Table Lookup
4140 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4141 (ins DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4142 "vtbl", "8", "$Vd, \\{$Vn\\}, $Vm", "",
4143 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 DPR:$Vn, DPR:$Vm)))]>;
4144 let hasExtraSrcRegAllocReq = 1 in {
4146 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4147 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4148 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4150 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4151 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4152 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4154 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4155 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4157 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4158 } // hasExtraSrcRegAllocReq = 1
4161 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4163 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4165 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4167 // VTBX : Vector Table Extension
4169 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4170 (ins DPR:$orig, DPR:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4171 "vtbx", "8", "$Vd, \\{$Vn\\}, $Vm", "$orig = $Vd",
4172 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4173 DPR:$orig, DPR:$Vn, DPR:$Vm)))]>;
4174 let hasExtraSrcRegAllocReq = 1 in {
4176 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4177 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4178 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4180 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4181 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4182 NVTBLFrm, IIC_VTBX3,
4183 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4186 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4187 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4188 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4190 } // hasExtraSrcRegAllocReq = 1
4193 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4194 IIC_VTBX2, "$orig = $dst", []>;
4196 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4197 IIC_VTBX3, "$orig = $dst", []>;
4199 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4200 IIC_VTBX4, "$orig = $dst", []>;
4202 //===----------------------------------------------------------------------===//
4203 // NEON instructions for single-precision FP math
4204 //===----------------------------------------------------------------------===//
4206 class N2VSPat<SDNode OpNode, ValueType ResTy, ValueType OpTy, NeonI Inst>
4207 : NEONFPPat<(ResTy (OpNode SPR:$a)),
4208 (EXTRACT_SUBREG (OpTy (Inst (INSERT_SUBREG (OpTy (IMPLICIT_DEF)),
4212 class N3VSPat<SDNode OpNode, NeonI Inst>
4213 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4214 (EXTRACT_SUBREG (v2f32
4215 (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4217 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4221 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4222 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4223 (EXTRACT_SUBREG (Inst (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4225 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4227 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)),
4231 // These need separate instructions because they must use DPR_VFP2 register
4232 // class which have SPR sub-registers.
4234 // Vector Add Operations used for single-precision FP
4235 let neverHasSideEffects = 1 in
4236 def VADDfd_sfp : N3VS<0,0,0b00,0b1101,0, "vadd", "f32", v2f32, v2f32, fadd, 1>;
4237 def : N3VSPat<fadd, VADDfd_sfp>;
4239 // Vector Sub Operations used for single-precision FP
4240 let neverHasSideEffects = 1 in
4241 def VSUBfd_sfp : N3VS<0,0,0b10,0b1101,0, "vsub", "f32", v2f32, v2f32, fsub, 0>;
4242 def : N3VSPat<fsub, VSUBfd_sfp>;
4244 // Vector Multiply Operations used for single-precision FP
4245 let neverHasSideEffects = 1 in
4246 def VMULfd_sfp : N3VS<1,0,0b00,0b1101,1, "vmul", "f32", v2f32, v2f32, fmul, 1>;
4247 def : N3VSPat<fmul, VMULfd_sfp>;
4249 // Vector Multiply-Accumulate/Subtract used for single-precision FP
4250 // vml[as].f32 can cause 4-8 cycle stalls in following ASIMD instructions, so
4251 // we want to avoid them for now. e.g., alternating vmla/vadd instructions.
4253 //let neverHasSideEffects = 1 in
4254 //def VMLAfd_sfp : N3VSMulOp<0,0,0b00,0b1101,1, IIC_VMACD, "vmla", "f32",
4255 // v2f32, fmul, fadd>;
4256 //def : N3VSMulOpPat<fmul, fadd, VMLAfd_sfp>;
4258 //let neverHasSideEffects = 1 in
4259 //def VMLSfd_sfp : N3VSMulOp<0,0,0b10,0b1101,1, IIC_VMACD, "vmls", "f32",
4260 // v2f32, fmul, fsub>;
4261 //def : N3VSMulOpPat<fmul, fsub, VMLSfd_sfp>;
4263 // Vector Absolute used for single-precision FP
4264 let neverHasSideEffects = 1 in
4265 def VABSfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01110, 0, 0,
4266 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4267 "vabs", "f32", "$dst, $src", "", []>;
4268 def : N2VSPat<fabs, f32, v2f32, VABSfd_sfp>;
4270 // Vector Negate used for single-precision FP
4271 let neverHasSideEffects = 1 in
4272 def VNEGfd_sfp : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4273 (outs DPR_VFP2:$dst), (ins DPR_VFP2:$src), IIC_VUNAD,
4274 "vneg", "f32", "$dst, $src", "", []>;
4275 def : N2VSPat<fneg, f32, v2f32, VNEGfd_sfp>;
4277 // Vector Maximum used for single-precision FP
4278 let neverHasSideEffects = 1 in
4279 def VMAXfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4280 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4281 "vmax", "f32", "$dst, $src1, $src2", "", []>;
4282 def : N3VSPat<NEONfmax, VMAXfd_sfp>;
4284 // Vector Minimum used for single-precision FP
4285 let neverHasSideEffects = 1 in
4286 def VMINfd_sfp : N3V<0, 0, 0b00, 0b1111, 0, 0, (outs DPR_VFP2:$dst),
4287 (ins DPR_VFP2:$src1, DPR_VFP2:$src2), N3RegFrm, IIC_VBIND,
4288 "vmin", "f32", "$dst, $src1, $src2", "", []>;
4289 def : N3VSPat<NEONfmin, VMINfd_sfp>;
4291 // Vector Convert between single-precision FP and integer
4292 let neverHasSideEffects = 1 in
4293 def VCVTf2sd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4294 v2i32, v2f32, fp_to_sint>;
4295 def : N2VSPat<arm_ftosi, f32, v2f32, VCVTf2sd_sfp>;
4297 let neverHasSideEffects = 1 in
4298 def VCVTf2ud_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4299 v2i32, v2f32, fp_to_uint>;
4300 def : N2VSPat<arm_ftoui, f32, v2f32, VCVTf2ud_sfp>;
4302 let neverHasSideEffects = 1 in
4303 def VCVTs2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4304 v2f32, v2i32, sint_to_fp>;
4305 def : N2VSPat<arm_sitof, f32, v2i32, VCVTs2fd_sfp>;
4307 let neverHasSideEffects = 1 in
4308 def VCVTu2fd_sfp : N2VS<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4309 v2f32, v2i32, uint_to_fp>;
4310 def : N2VSPat<arm_uitof, f32, v2i32, VCVTu2fd_sfp>;
4312 //===----------------------------------------------------------------------===//
4313 // Non-Instruction Patterns
4314 //===----------------------------------------------------------------------===//
4317 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
4318 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
4319 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
4320 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
4321 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
4322 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
4323 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
4324 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
4325 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
4326 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
4327 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
4328 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
4329 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
4330 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
4331 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
4332 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
4333 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
4334 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
4335 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
4336 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
4337 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
4338 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
4339 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
4340 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
4341 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
4342 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
4343 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
4344 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
4345 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
4346 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
4348 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
4349 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
4350 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
4351 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
4352 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
4353 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
4354 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
4355 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
4356 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
4357 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
4358 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
4359 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
4360 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
4361 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
4362 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
4363 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
4364 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
4365 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
4366 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
4367 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
4368 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
4369 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
4370 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
4371 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
4372 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
4373 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
4374 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
4375 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
4376 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
4377 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;