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 //===----------------------------------------------------------------------===//
15 //===----------------------------------------------------------------------===//
16 // NEON-specific Operands.
17 //===----------------------------------------------------------------------===//
18 def nModImm : Operand<i32> {
19 let PrintMethod = "printNEONModImmOperand";
22 def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
23 def nImmSplatI8 : Operand<i32> {
24 let PrintMethod = "printNEONModImmOperand";
25 let ParserMatchClass = nImmSplatI8AsmOperand;
27 def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
28 def nImmSplatI16 : Operand<i32> {
29 let PrintMethod = "printNEONModImmOperand";
30 let ParserMatchClass = nImmSplatI16AsmOperand;
32 def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
33 def nImmSplatI32 : Operand<i32> {
34 let PrintMethod = "printNEONModImmOperand";
35 let ParserMatchClass = nImmSplatI32AsmOperand;
37 def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
38 def nImmVMOVI32 : Operand<i32> {
39 let PrintMethod = "printNEONModImmOperand";
40 let ParserMatchClass = nImmVMOVI32AsmOperand;
42 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
43 def nImmSplatI64 : Operand<i32> {
44 let PrintMethod = "printNEONModImmOperand";
45 let ParserMatchClass = nImmSplatI64AsmOperand;
48 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
49 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
50 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
51 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
52 return ((uint64_t)Imm) < 8;
54 let ParserMatchClass = VectorIndex8Operand;
55 let PrintMethod = "printVectorIndex";
56 let MIOperandInfo = (ops i32imm);
58 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
59 return ((uint64_t)Imm) < 4;
61 let ParserMatchClass = VectorIndex16Operand;
62 let PrintMethod = "printVectorIndex";
63 let MIOperandInfo = (ops i32imm);
65 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
66 return ((uint64_t)Imm) < 2;
68 let ParserMatchClass = VectorIndex32Operand;
69 let PrintMethod = "printVectorIndex";
70 let MIOperandInfo = (ops i32imm);
73 def VecListOneDAsmOperand : AsmOperandClass {
74 let Name = "VecListOneD";
75 let ParserMethod = "parseVectorList";
77 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
78 let ParserMatchClass = VecListOneDAsmOperand;
80 // Register list of two sequential D registers.
81 def VecListTwoDAsmOperand : AsmOperandClass {
82 let Name = "VecListTwoD";
83 let ParserMethod = "parseVectorList";
85 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
86 let ParserMatchClass = VecListTwoDAsmOperand;
88 // Register list of three sequential D registers.
89 def VecListThreeDAsmOperand : AsmOperandClass {
90 let Name = "VecListThreeD";
91 let ParserMethod = "parseVectorList";
93 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
94 let ParserMatchClass = VecListThreeDAsmOperand;
97 //===----------------------------------------------------------------------===//
98 // NEON-specific DAG Nodes.
99 //===----------------------------------------------------------------------===//
101 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
102 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
104 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
105 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
106 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
107 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
108 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
109 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
110 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
111 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
112 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
113 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
114 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
116 // Types for vector shift by immediates. The "SHX" version is for long and
117 // narrow operations where the source and destination vectors have different
118 // types. The "SHINS" version is for shift and insert operations.
119 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
121 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
123 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
124 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
126 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
127 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
128 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
129 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
130 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
131 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
132 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
134 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
135 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
136 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
138 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
139 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
140 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
141 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
142 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
143 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
145 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
146 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
147 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
149 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
150 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
152 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
154 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
155 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
157 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
158 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
159 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
161 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
163 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
164 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
166 def NEONvbsl : SDNode<"ARMISD::VBSL",
167 SDTypeProfile<1, 3, [SDTCisVec<0>,
170 SDTCisSameAs<0, 3>]>>;
172 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
174 // VDUPLANE can produce a quad-register result from a double-register source,
175 // so the result is not constrained to match the source.
176 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
177 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
180 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
181 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
182 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
184 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
185 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
186 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
187 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
189 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
191 SDTCisSameAs<0, 3>]>;
192 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
193 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
194 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
196 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
197 SDTCisSameAs<1, 2>]>;
198 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
199 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
201 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
202 SDTCisSameAs<0, 2>]>;
203 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
204 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
206 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
207 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
208 unsigned EltBits = 0;
209 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
210 return (EltBits == 32 && EltVal == 0);
213 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
214 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
215 unsigned EltBits = 0;
216 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
217 return (EltBits == 8 && EltVal == 0xff);
220 //===----------------------------------------------------------------------===//
221 // NEON load / store instructions
222 //===----------------------------------------------------------------------===//
224 // Use VLDM to load a Q register as a D register pair.
225 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
227 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
229 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
231 // Use VSTM to store a Q register as a D register pair.
232 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
234 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
236 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
238 // Classes for VLD* pseudo-instructions with multi-register operands.
239 // These are expanded to real instructions after register allocation.
240 class VLDQPseudo<InstrItinClass itin>
241 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
242 class VLDQWBPseudo<InstrItinClass itin>
243 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
244 (ins addrmode6:$addr, am6offset:$offset), itin,
246 class VLDQQPseudo<InstrItinClass itin>
247 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
248 class VLDQQWBPseudo<InstrItinClass itin>
249 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
250 (ins addrmode6:$addr, am6offset:$offset), itin,
252 class VLDQQQQPseudo<InstrItinClass itin>
253 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
255 class VLDQQQQWBPseudo<InstrItinClass itin>
256 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
257 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
258 "$addr.addr = $wb, $src = $dst">;
260 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
262 // VLD1 : Vector Load (multiple single elements)
263 class VLD1D<bits<4> op7_4, string Dt>
264 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
265 (ins addrmode6:$Rn), IIC_VLD1,
266 "vld1", Dt, "$Vd, $Rn", "", []> {
269 let DecoderMethod = "DecodeVLDInstruction";
271 class VLD1Q<bits<4> op7_4, string Dt>
272 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
273 (ins addrmode6:$Rn), IIC_VLD1x2,
274 "vld1", Dt, "$Vd, $Rn", "", []> {
276 let Inst{5-4} = Rn{5-4};
277 let DecoderMethod = "DecodeVLDInstruction";
280 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
281 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
282 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
283 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
285 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
286 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
287 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
288 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
290 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
291 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
292 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
293 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
295 // ...with address register writeback:
296 class VLD1DWB<bits<4> op7_4, string Dt>
297 : NLdSt<0,0b10,0b0111,op7_4, (outs DPR:$Vd, GPR:$wb),
298 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1u,
299 "vld1", Dt, "\\{$Vd\\}, $Rn$Rm",
300 "$Rn.addr = $wb", []> {
302 let DecoderMethod = "DecodeVLDInstruction";
304 class VLD1QWB<bits<4> op7_4, string Dt>
305 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
306 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x2u,
307 "vld1", Dt, "$Vd, $Rn$Rm",
308 "$Rn.addr = $wb", []> {
309 let Inst{5-4} = Rn{5-4};
310 let DecoderMethod = "DecodeVLDInstruction";
313 def VLD1d8_UPD : VLD1DWB<{0,0,0,?}, "8">;
314 def VLD1d16_UPD : VLD1DWB<{0,1,0,?}, "16">;
315 def VLD1d32_UPD : VLD1DWB<{1,0,0,?}, "32">;
316 def VLD1d64_UPD : VLD1DWB<{1,1,0,?}, "64">;
318 def VLD1q8_UPD : VLD1QWB<{0,0,?,?}, "8">;
319 def VLD1q16_UPD : VLD1QWB<{0,1,?,?}, "16">;
320 def VLD1q32_UPD : VLD1QWB<{1,0,?,?}, "32">;
321 def VLD1q64_UPD : VLD1QWB<{1,1,?,?}, "64">;
323 def VLD1q8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
324 def VLD1q16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
325 def VLD1q32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
326 def VLD1q64Pseudo_UPD : VLDQWBPseudo<IIC_VLD1x2u>;
328 // ...with 3 registers
329 class VLD1D3<bits<4> op7_4, string Dt>
330 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
331 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
332 "$Vd, $Rn", "", []> {
335 let DecoderMethod = "DecodeVLDInstruction";
337 class VLD1D3WB<bits<4> op7_4, string Dt>
338 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
339 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
340 "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
342 let DecoderMethod = "DecodeVLDInstruction";
345 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
346 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
347 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
348 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
350 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
351 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
352 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
353 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
355 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
356 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
358 // ...with 4 registers
359 class VLD1D4<bits<4> op7_4, string Dt>
360 : NLdSt<0,0b10,0b0010,op7_4,(outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
361 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
362 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
364 let Inst{5-4} = Rn{5-4};
365 let DecoderMethod = "DecodeVLDInstruction";
367 class VLD1D4WB<bits<4> op7_4, string Dt>
368 : NLdSt<0,0b10,0b0010,op7_4,
369 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
370 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x4u, "vld1", Dt,
371 "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm", "$Rn.addr = $wb",
373 let Inst{5-4} = Rn{5-4};
374 let DecoderMethod = "DecodeVLDInstruction";
377 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
378 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
379 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
380 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
382 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
383 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
384 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
385 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
387 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
388 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
390 // VLD2 : Vector Load (multiple 2-element structures)
391 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt>
392 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
393 (ins addrmode6:$Rn), IIC_VLD2,
394 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn", "", []> {
396 let Inst{5-4} = Rn{5-4};
397 let DecoderMethod = "DecodeVLDInstruction";
399 class VLD2Q<bits<4> op7_4, string Dt>
400 : NLdSt<0, 0b10, 0b0011, op7_4,
401 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
402 (ins addrmode6:$Rn), IIC_VLD2x2,
403 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
405 let Inst{5-4} = Rn{5-4};
406 let DecoderMethod = "DecodeVLDInstruction";
409 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8">;
410 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16">;
411 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32">;
413 def VLD2q8 : VLD2Q<{0,0,?,?}, "8">;
414 def VLD2q16 : VLD2Q<{0,1,?,?}, "16">;
415 def VLD2q32 : VLD2Q<{1,0,?,?}, "32">;
417 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
418 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
419 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
421 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
422 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
423 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
425 // ...with address register writeback:
426 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
427 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
428 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
429 "vld2", Dt, "\\{$Vd, $dst2\\}, $Rn$Rm",
430 "$Rn.addr = $wb", []> {
431 let Inst{5-4} = Rn{5-4};
432 let DecoderMethod = "DecodeVLDInstruction";
434 class VLD2QWB<bits<4> op7_4, string Dt>
435 : NLdSt<0, 0b10, 0b0011, op7_4,
436 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
437 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
438 "vld2", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
439 "$Rn.addr = $wb", []> {
440 let Inst{5-4} = Rn{5-4};
441 let DecoderMethod = "DecodeVLDInstruction";
444 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8">;
445 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16">;
446 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32">;
448 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8">;
449 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16">;
450 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32">;
452 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
453 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
454 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
456 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
457 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
458 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
460 // ...with double-spaced registers
461 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8">;
462 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16">;
463 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32">;
464 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8">;
465 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16">;
466 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32">;
468 // VLD3 : Vector Load (multiple 3-element structures)
469 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
470 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
471 (ins addrmode6:$Rn), IIC_VLD3,
472 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
475 let DecoderMethod = "DecodeVLDInstruction";
478 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
479 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
480 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
482 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
483 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
484 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
486 // ...with address register writeback:
487 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
488 : NLdSt<0, 0b10, op11_8, op7_4,
489 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
490 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
491 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
492 "$Rn.addr = $wb", []> {
494 let DecoderMethod = "DecodeVLDInstruction";
497 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
498 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
499 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
501 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
502 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
503 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
505 // ...with double-spaced registers:
506 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
507 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
508 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
509 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
510 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
511 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
513 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
514 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
515 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
517 // ...alternate versions to be allocated odd register numbers:
518 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
519 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
520 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
522 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
523 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
524 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
526 // VLD4 : Vector Load (multiple 4-element structures)
527 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
528 : NLdSt<0, 0b10, op11_8, op7_4,
529 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
530 (ins addrmode6:$Rn), IIC_VLD4,
531 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
533 let Inst{5-4} = Rn{5-4};
534 let DecoderMethod = "DecodeVLDInstruction";
537 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
538 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
539 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
541 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
542 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
543 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
545 // ...with address register writeback:
546 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
547 : NLdSt<0, 0b10, op11_8, op7_4,
548 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
549 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
550 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
551 "$Rn.addr = $wb", []> {
552 let Inst{5-4} = Rn{5-4};
553 let DecoderMethod = "DecodeVLDInstruction";
556 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
557 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
558 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
560 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
561 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
562 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
564 // ...with double-spaced registers:
565 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
566 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
567 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
568 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
569 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
570 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
572 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
573 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
574 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
576 // ...alternate versions to be allocated odd register numbers:
577 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
578 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
579 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
581 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
582 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
583 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
585 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
587 // Classes for VLD*LN pseudo-instructions with multi-register operands.
588 // These are expanded to real instructions after register allocation.
589 class VLDQLNPseudo<InstrItinClass itin>
590 : PseudoNLdSt<(outs QPR:$dst),
591 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
592 itin, "$src = $dst">;
593 class VLDQLNWBPseudo<InstrItinClass itin>
594 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
595 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
596 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
597 class VLDQQLNPseudo<InstrItinClass itin>
598 : PseudoNLdSt<(outs QQPR:$dst),
599 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
600 itin, "$src = $dst">;
601 class VLDQQLNWBPseudo<InstrItinClass itin>
602 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
603 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
604 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
605 class VLDQQQQLNPseudo<InstrItinClass itin>
606 : PseudoNLdSt<(outs QQQQPR:$dst),
607 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
608 itin, "$src = $dst">;
609 class VLDQQQQLNWBPseudo<InstrItinClass itin>
610 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
611 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
612 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
614 // VLD1LN : Vector Load (single element to one lane)
615 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
617 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
618 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
619 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
621 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
622 (i32 (LoadOp addrmode6:$Rn)),
625 let DecoderMethod = "DecodeVLD1LN";
627 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
629 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
630 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
631 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
633 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
634 (i32 (LoadOp addrmode6oneL32:$Rn)),
637 let DecoderMethod = "DecodeVLD1LN";
639 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
640 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
641 (i32 (LoadOp addrmode6:$addr)),
645 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
646 let Inst{7-5} = lane{2-0};
648 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
649 let Inst{7-6} = lane{1-0};
652 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
653 let Inst{7} = lane{0};
658 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
659 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
660 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
662 def : Pat<(vector_insert (v2f32 DPR:$src),
663 (f32 (load addrmode6:$addr)), imm:$lane),
664 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
665 def : Pat<(vector_insert (v4f32 QPR:$src),
666 (f32 (load addrmode6:$addr)), imm:$lane),
667 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
669 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
671 // ...with address register writeback:
672 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
673 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
674 (ins addrmode6:$Rn, am6offset:$Rm,
675 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
676 "\\{$Vd[$lane]\\}, $Rn$Rm",
677 "$src = $Vd, $Rn.addr = $wb", []> {
678 let DecoderMethod = "DecodeVLD1LN";
681 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
682 let Inst{7-5} = lane{2-0};
684 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
685 let Inst{7-6} = lane{1-0};
688 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
689 let Inst{7} = lane{0};
694 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
695 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
696 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
698 // VLD2LN : Vector Load (single 2-element structure to one lane)
699 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
700 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
701 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
702 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
703 "$src1 = $Vd, $src2 = $dst2", []> {
706 let DecoderMethod = "DecodeVLD2LN";
709 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
710 let Inst{7-5} = lane{2-0};
712 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
713 let Inst{7-6} = lane{1-0};
715 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
716 let Inst{7} = lane{0};
719 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
720 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
721 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
723 // ...with double-spaced registers:
724 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
725 let Inst{7-6} = lane{1-0};
727 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
728 let Inst{7} = lane{0};
731 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
732 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
734 // ...with address register writeback:
735 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
736 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
737 (ins addrmode6:$Rn, am6offset:$Rm,
738 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
739 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
740 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
742 let DecoderMethod = "DecodeVLD2LN";
745 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
746 let Inst{7-5} = lane{2-0};
748 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
749 let Inst{7-6} = lane{1-0};
751 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
752 let Inst{7} = lane{0};
755 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
756 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
757 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
759 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
760 let Inst{7-6} = lane{1-0};
762 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
763 let Inst{7} = lane{0};
766 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
767 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
769 // VLD3LN : Vector Load (single 3-element structure to one lane)
770 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
771 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
772 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
773 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
774 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
775 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
777 let DecoderMethod = "DecodeVLD3LN";
780 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
781 let Inst{7-5} = lane{2-0};
783 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
784 let Inst{7-6} = lane{1-0};
786 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
787 let Inst{7} = lane{0};
790 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
791 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
792 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
794 // ...with double-spaced registers:
795 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
796 let Inst{7-6} = lane{1-0};
798 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
799 let Inst{7} = lane{0};
802 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
803 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
805 // ...with address register writeback:
806 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
807 : NLdStLn<1, 0b10, op11_8, op7_4,
808 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
809 (ins addrmode6:$Rn, am6offset:$Rm,
810 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
811 IIC_VLD3lnu, "vld3", Dt,
812 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
813 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
815 let DecoderMethod = "DecodeVLD3LN";
818 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
819 let Inst{7-5} = lane{2-0};
821 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
822 let Inst{7-6} = lane{1-0};
824 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
825 let Inst{7} = lane{0};
828 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
829 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
830 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
832 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
833 let Inst{7-6} = lane{1-0};
835 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
836 let Inst{7} = lane{0};
839 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
840 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
842 // VLD4LN : Vector Load (single 4-element structure to one lane)
843 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
844 : NLdStLn<1, 0b10, op11_8, op7_4,
845 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
846 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
847 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
848 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
849 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
852 let DecoderMethod = "DecodeVLD4LN";
855 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
856 let Inst{7-5} = lane{2-0};
858 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
859 let Inst{7-6} = lane{1-0};
861 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
862 let Inst{7} = lane{0};
866 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
867 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
868 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
870 // ...with double-spaced registers:
871 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
872 let Inst{7-6} = lane{1-0};
874 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
875 let Inst{7} = lane{0};
879 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
880 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
882 // ...with address register writeback:
883 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
884 : NLdStLn<1, 0b10, op11_8, op7_4,
885 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
886 (ins addrmode6:$Rn, am6offset:$Rm,
887 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
888 IIC_VLD4lnu, "vld4", Dt,
889 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
890 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
893 let DecoderMethod = "DecodeVLD4LN" ;
896 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
897 let Inst{7-5} = lane{2-0};
899 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
900 let Inst{7-6} = lane{1-0};
902 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
903 let Inst{7} = lane{0};
907 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
908 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
909 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
911 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
912 let Inst{7-6} = lane{1-0};
914 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
915 let Inst{7} = lane{0};
919 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
920 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
922 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
924 // VLD1DUP : Vector Load (single element to all lanes)
925 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
926 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
927 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
928 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
931 let DecoderMethod = "DecodeVLD1DupInstruction";
933 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
934 let Pattern = [(set QPR:$dst,
935 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
938 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
939 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
940 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
942 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
943 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
944 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
946 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
947 (VLD1DUPd32 addrmode6:$addr)>;
948 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
949 (VLD1DUPq32Pseudo addrmode6:$addr)>;
951 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
953 class VLD1QDUP<bits<4> op7_4, string Dt>
954 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
955 (ins addrmode6dup:$Rn), IIC_VLD1dup,
956 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
959 let DecoderMethod = "DecodeVLD1DupInstruction";
962 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
963 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
964 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
966 // ...with address register writeback:
967 class VLD1DUPWB<bits<4> op7_4, string Dt>
968 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
969 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
970 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
972 let DecoderMethod = "DecodeVLD1DupInstruction";
974 class VLD1QDUPWB<bits<4> op7_4, string Dt>
975 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
976 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
977 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
979 let DecoderMethod = "DecodeVLD1DupInstruction";
982 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
983 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
984 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
986 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
987 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
988 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
990 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
991 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
992 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
994 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
995 class VLD2DUP<bits<4> op7_4, string Dt>
996 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
997 (ins addrmode6dup:$Rn), IIC_VLD2dup,
998 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1000 let Inst{4} = Rn{4};
1001 let DecoderMethod = "DecodeVLD2DupInstruction";
1004 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1005 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1006 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1008 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1009 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1010 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1012 // ...with double-spaced registers (not used for codegen):
1013 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1014 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1015 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1017 // ...with address register writeback:
1018 class VLD2DUPWB<bits<4> op7_4, string Dt>
1019 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1020 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1021 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1022 let Inst{4} = Rn{4};
1023 let DecoderMethod = "DecodeVLD2DupInstruction";
1026 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1027 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1028 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1030 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1031 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1032 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1034 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1035 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1036 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1038 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1039 class VLD3DUP<bits<4> op7_4, string Dt>
1040 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1041 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1042 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1045 let DecoderMethod = "DecodeVLD3DupInstruction";
1048 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1049 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1050 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1052 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1053 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1054 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1056 // ...with double-spaced registers (not used for codegen):
1057 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1058 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1059 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1061 // ...with address register writeback:
1062 class VLD3DUPWB<bits<4> op7_4, string Dt>
1063 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1064 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1065 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1066 "$Rn.addr = $wb", []> {
1068 let DecoderMethod = "DecodeVLD3DupInstruction";
1071 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1072 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1073 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1075 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1076 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1077 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1079 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1080 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1081 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1083 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1084 class VLD4DUP<bits<4> op7_4, string Dt>
1085 : NLdSt<1, 0b10, 0b1111, op7_4,
1086 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1087 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1088 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1090 let Inst{4} = Rn{4};
1091 let DecoderMethod = "DecodeVLD4DupInstruction";
1094 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1095 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1096 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1098 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1099 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1100 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1102 // ...with double-spaced registers (not used for codegen):
1103 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1104 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1105 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1107 // ...with address register writeback:
1108 class VLD4DUPWB<bits<4> op7_4, string Dt>
1109 : NLdSt<1, 0b10, 0b1111, op7_4,
1110 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1111 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1112 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1113 "$Rn.addr = $wb", []> {
1114 let Inst{4} = Rn{4};
1115 let DecoderMethod = "DecodeVLD4DupInstruction";
1118 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1119 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1120 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1122 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1123 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1124 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1126 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1127 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1128 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1130 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1132 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1134 // Classes for VST* pseudo-instructions with multi-register operands.
1135 // These are expanded to real instructions after register allocation.
1136 class VSTQPseudo<InstrItinClass itin>
1137 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1138 class VSTQWBPseudo<InstrItinClass itin>
1139 : PseudoNLdSt<(outs GPR:$wb),
1140 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1141 "$addr.addr = $wb">;
1142 class VSTQQPseudo<InstrItinClass itin>
1143 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1144 class VSTQQWBPseudo<InstrItinClass itin>
1145 : PseudoNLdSt<(outs GPR:$wb),
1146 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1147 "$addr.addr = $wb">;
1148 class VSTQQQQPseudo<InstrItinClass itin>
1149 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1150 class VSTQQQQWBPseudo<InstrItinClass itin>
1151 : PseudoNLdSt<(outs GPR:$wb),
1152 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1153 "$addr.addr = $wb">;
1155 // VST1 : Vector Store (multiple single elements)
1156 class VST1D<bits<4> op7_4, string Dt>
1157 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1158 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1160 let Inst{4} = Rn{4};
1161 let DecoderMethod = "DecodeVSTInstruction";
1163 class VST1Q<bits<4> op7_4, string Dt>
1164 : NLdSt<0,0b00,0b1010,op7_4, (outs),
1165 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
1166 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1168 let Inst{5-4} = Rn{5-4};
1169 let DecoderMethod = "DecodeVSTInstruction";
1172 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1173 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1174 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1175 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1177 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1178 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1179 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1180 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1182 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1183 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1184 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1185 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1187 // ...with address register writeback:
1188 class VST1DWB<bits<4> op7_4, string Dt>
1189 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
1190 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1191 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1192 let Inst{4} = Rn{4};
1193 let DecoderMethod = "DecodeVSTInstruction";
1195 class VST1QWB<bits<4> op7_4, string Dt>
1196 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1197 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1198 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1199 "$Rn.addr = $wb", []> {
1200 let Inst{5-4} = Rn{5-4};
1201 let DecoderMethod = "DecodeVSTInstruction";
1204 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1205 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1206 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1207 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1209 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1210 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1211 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1212 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1214 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1215 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1216 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1217 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1219 // ...with 3 registers
1220 class VST1D3<bits<4> op7_4, string Dt>
1221 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1222 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1223 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1225 let Inst{4} = Rn{4};
1226 let DecoderMethod = "DecodeVSTInstruction";
1228 class VST1D3WB<bits<4> op7_4, string Dt>
1229 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1230 (ins addrmode6:$Rn, am6offset:$Rm,
1231 DPR:$Vd, DPR:$src2, DPR:$src3),
1232 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1233 "$Rn.addr = $wb", []> {
1234 let Inst{4} = Rn{4};
1235 let DecoderMethod = "DecodeVSTInstruction";
1238 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1239 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1240 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1241 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1243 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1244 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1245 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1246 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1248 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1249 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1251 // ...with 4 registers
1252 class VST1D4<bits<4> op7_4, string Dt>
1253 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1254 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1255 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1258 let Inst{5-4} = Rn{5-4};
1259 let DecoderMethod = "DecodeVSTInstruction";
1261 class VST1D4WB<bits<4> op7_4, string Dt>
1262 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1263 (ins addrmode6:$Rn, am6offset:$Rm,
1264 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1265 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1266 "$Rn.addr = $wb", []> {
1267 let Inst{5-4} = Rn{5-4};
1268 let DecoderMethod = "DecodeVSTInstruction";
1271 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1272 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1273 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1274 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1276 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1277 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1278 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1279 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1281 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1282 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1284 // VST2 : Vector Store (multiple 2-element structures)
1285 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1286 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1287 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1288 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1290 let Inst{5-4} = Rn{5-4};
1291 let DecoderMethod = "DecodeVSTInstruction";
1293 class VST2Q<bits<4> op7_4, string Dt>
1294 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1295 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1296 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1299 let Inst{5-4} = Rn{5-4};
1300 let DecoderMethod = "DecodeVSTInstruction";
1303 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1304 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1305 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1307 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1308 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1309 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1311 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1312 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1313 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1315 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1316 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1317 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1319 // ...with address register writeback:
1320 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1321 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1322 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1323 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1324 "$Rn.addr = $wb", []> {
1325 let Inst{5-4} = Rn{5-4};
1326 let DecoderMethod = "DecodeVSTInstruction";
1328 class VST2QWB<bits<4> op7_4, string Dt>
1329 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1330 (ins addrmode6:$Rn, am6offset:$Rm,
1331 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1332 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1333 "$Rn.addr = $wb", []> {
1334 let Inst{5-4} = Rn{5-4};
1335 let DecoderMethod = "DecodeVSTInstruction";
1338 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1339 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1340 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1342 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1343 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1344 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1346 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1347 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1348 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1350 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1351 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1352 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1354 // ...with double-spaced registers
1355 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1356 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1357 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1358 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1359 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1360 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1362 // VST3 : Vector Store (multiple 3-element structures)
1363 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1364 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1365 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1366 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1368 let Inst{4} = Rn{4};
1369 let DecoderMethod = "DecodeVSTInstruction";
1372 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1373 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1374 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1376 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1377 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1378 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1380 // ...with address register writeback:
1381 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1382 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1383 (ins addrmode6:$Rn, am6offset:$Rm,
1384 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1385 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1386 "$Rn.addr = $wb", []> {
1387 let Inst{4} = Rn{4};
1388 let DecoderMethod = "DecodeVSTInstruction";
1391 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1392 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1393 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1395 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1396 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1397 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1399 // ...with double-spaced registers:
1400 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1401 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1402 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1403 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1404 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1405 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1407 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1408 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1409 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1411 // ...alternate versions to be allocated odd register numbers:
1412 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1413 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1414 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1416 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1417 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1418 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1420 // VST4 : Vector Store (multiple 4-element structures)
1421 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1422 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1423 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1424 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1427 let Inst{5-4} = Rn{5-4};
1428 let DecoderMethod = "DecodeVSTInstruction";
1431 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1432 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1433 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1435 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1436 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1437 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1439 // ...with address register writeback:
1440 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1441 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1442 (ins addrmode6:$Rn, am6offset:$Rm,
1443 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1444 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1445 "$Rn.addr = $wb", []> {
1446 let Inst{5-4} = Rn{5-4};
1447 let DecoderMethod = "DecodeVSTInstruction";
1450 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1451 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1452 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1454 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1455 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1456 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1458 // ...with double-spaced registers:
1459 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1460 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1461 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1462 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1463 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1464 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1466 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1467 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1468 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1470 // ...alternate versions to be allocated odd register numbers:
1471 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1472 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1473 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1475 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1476 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1477 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1479 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1481 // Classes for VST*LN pseudo-instructions with multi-register operands.
1482 // These are expanded to real instructions after register allocation.
1483 class VSTQLNPseudo<InstrItinClass itin>
1484 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1486 class VSTQLNWBPseudo<InstrItinClass itin>
1487 : PseudoNLdSt<(outs GPR:$wb),
1488 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1489 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1490 class VSTQQLNPseudo<InstrItinClass itin>
1491 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1493 class VSTQQLNWBPseudo<InstrItinClass itin>
1494 : PseudoNLdSt<(outs GPR:$wb),
1495 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1496 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1497 class VSTQQQQLNPseudo<InstrItinClass itin>
1498 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1500 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1501 : PseudoNLdSt<(outs GPR:$wb),
1502 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1503 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1505 // VST1LN : Vector Store (single element from one lane)
1506 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1507 PatFrag StoreOp, SDNode ExtractOp>
1508 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1509 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1510 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1511 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1513 let DecoderMethod = "DecodeVST1LN";
1515 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1516 PatFrag StoreOp, SDNode ExtractOp>
1517 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1518 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1519 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1520 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1522 let DecoderMethod = "DecodeVST1LN";
1524 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1525 : VSTQLNPseudo<IIC_VST1ln> {
1526 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1530 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1532 let Inst{7-5} = lane{2-0};
1534 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1536 let Inst{7-6} = lane{1-0};
1537 let Inst{4} = Rn{5};
1540 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1541 let Inst{7} = lane{0};
1542 let Inst{5-4} = Rn{5-4};
1545 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1546 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1547 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1549 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1550 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1551 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1552 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1554 // ...with address register writeback:
1555 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1556 PatFrag StoreOp, SDNode ExtractOp>
1557 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1558 (ins addrmode6:$Rn, am6offset:$Rm,
1559 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1560 "\\{$Vd[$lane]\\}, $Rn$Rm",
1562 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1563 addrmode6:$Rn, am6offset:$Rm))]> {
1564 let DecoderMethod = "DecodeVST1LN";
1566 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1567 : VSTQLNWBPseudo<IIC_VST1lnu> {
1568 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1569 addrmode6:$addr, am6offset:$offset))];
1572 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1574 let Inst{7-5} = lane{2-0};
1576 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1578 let Inst{7-6} = lane{1-0};
1579 let Inst{4} = Rn{5};
1581 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1583 let Inst{7} = lane{0};
1584 let Inst{5-4} = Rn{5-4};
1587 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1588 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1589 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1591 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1593 // VST2LN : Vector Store (single 2-element structure from one lane)
1594 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1595 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1596 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1597 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1600 let Inst{4} = Rn{4};
1601 let DecoderMethod = "DecodeVST2LN";
1604 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1605 let Inst{7-5} = lane{2-0};
1607 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1608 let Inst{7-6} = lane{1-0};
1610 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1611 let Inst{7} = lane{0};
1614 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1615 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1616 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1618 // ...with double-spaced registers:
1619 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1620 let Inst{7-6} = lane{1-0};
1621 let Inst{4} = Rn{4};
1623 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1624 let Inst{7} = lane{0};
1625 let Inst{4} = Rn{4};
1628 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1629 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1631 // ...with address register writeback:
1632 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1633 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1634 (ins addrmode6:$addr, am6offset:$offset,
1635 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1636 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1637 "$addr.addr = $wb", []> {
1638 let Inst{4} = Rn{4};
1639 let DecoderMethod = "DecodeVST2LN";
1642 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1643 let Inst{7-5} = lane{2-0};
1645 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1646 let Inst{7-6} = lane{1-0};
1648 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1649 let Inst{7} = lane{0};
1652 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1653 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1654 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1656 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1657 let Inst{7-6} = lane{1-0};
1659 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1660 let Inst{7} = lane{0};
1663 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1664 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1666 // VST3LN : Vector Store (single 3-element structure from one lane)
1667 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1668 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1669 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1670 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1671 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1673 let DecoderMethod = "DecodeVST3LN";
1676 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1677 let Inst{7-5} = lane{2-0};
1679 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1680 let Inst{7-6} = lane{1-0};
1682 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1683 let Inst{7} = lane{0};
1686 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1687 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1688 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1690 // ...with double-spaced registers:
1691 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1692 let Inst{7-6} = lane{1-0};
1694 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1695 let Inst{7} = lane{0};
1698 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1699 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1701 // ...with address register writeback:
1702 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1703 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1704 (ins addrmode6:$Rn, am6offset:$Rm,
1705 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1706 IIC_VST3lnu, "vst3", Dt,
1707 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1708 "$Rn.addr = $wb", []> {
1709 let DecoderMethod = "DecodeVST3LN";
1712 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1713 let Inst{7-5} = lane{2-0};
1715 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1716 let Inst{7-6} = lane{1-0};
1718 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1719 let Inst{7} = lane{0};
1722 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1723 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1724 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1726 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1727 let Inst{7-6} = lane{1-0};
1729 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1730 let Inst{7} = lane{0};
1733 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1734 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1736 // VST4LN : Vector Store (single 4-element structure from one lane)
1737 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1738 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1739 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1740 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1741 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1744 let Inst{4} = Rn{4};
1745 let DecoderMethod = "DecodeVST4LN";
1748 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1749 let Inst{7-5} = lane{2-0};
1751 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1752 let Inst{7-6} = lane{1-0};
1754 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1755 let Inst{7} = lane{0};
1756 let Inst{5} = Rn{5};
1759 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1760 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1761 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1763 // ...with double-spaced registers:
1764 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1765 let Inst{7-6} = lane{1-0};
1767 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1768 let Inst{7} = lane{0};
1769 let Inst{5} = Rn{5};
1772 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1773 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1775 // ...with address register writeback:
1776 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1777 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1778 (ins addrmode6:$Rn, am6offset:$Rm,
1779 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1780 IIC_VST4lnu, "vst4", Dt,
1781 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1782 "$Rn.addr = $wb", []> {
1783 let Inst{4} = Rn{4};
1784 let DecoderMethod = "DecodeVST4LN";
1787 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1788 let Inst{7-5} = lane{2-0};
1790 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1791 let Inst{7-6} = lane{1-0};
1793 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1794 let Inst{7} = lane{0};
1795 let Inst{5} = Rn{5};
1798 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1799 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1800 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1802 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1803 let Inst{7-6} = lane{1-0};
1805 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1806 let Inst{7} = lane{0};
1807 let Inst{5} = Rn{5};
1810 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1811 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1813 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1816 //===----------------------------------------------------------------------===//
1817 // NEON pattern fragments
1818 //===----------------------------------------------------------------------===//
1820 // Extract D sub-registers of Q registers.
1821 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1822 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1823 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1825 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1826 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1827 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1829 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1830 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1831 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1833 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1834 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1835 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1838 // Extract S sub-registers of Q/D registers.
1839 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1840 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1841 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1844 // Translate lane numbers from Q registers to D subregs.
1845 def SubReg_i8_lane : SDNodeXForm<imm, [{
1846 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1848 def SubReg_i16_lane : SDNodeXForm<imm, [{
1849 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1851 def SubReg_i32_lane : SDNodeXForm<imm, [{
1852 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1855 //===----------------------------------------------------------------------===//
1856 // Instruction Classes
1857 //===----------------------------------------------------------------------===//
1859 // Basic 2-register operations: double- and quad-register.
1860 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1861 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1862 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1863 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1864 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1865 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1866 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1867 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1868 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1869 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1870 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1871 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1873 // Basic 2-register intrinsics, both double- and quad-register.
1874 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1875 bits<2> op17_16, bits<5> op11_7, bit op4,
1876 InstrItinClass itin, string OpcodeStr, string Dt,
1877 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1878 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1879 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1880 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1881 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1882 bits<2> op17_16, bits<5> op11_7, bit op4,
1883 InstrItinClass itin, string OpcodeStr, string Dt,
1884 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1885 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1886 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1887 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1889 // Narrow 2-register operations.
1890 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1891 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1892 InstrItinClass itin, string OpcodeStr, string Dt,
1893 ValueType TyD, ValueType TyQ, SDNode OpNode>
1894 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1895 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1896 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1898 // Narrow 2-register intrinsics.
1899 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1900 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1901 InstrItinClass itin, string OpcodeStr, string Dt,
1902 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1903 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1904 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1905 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1907 // Long 2-register operations (currently only used for VMOVL).
1908 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1909 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1910 InstrItinClass itin, string OpcodeStr, string Dt,
1911 ValueType TyQ, ValueType TyD, SDNode OpNode>
1912 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1913 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1914 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1916 // Long 2-register intrinsics.
1917 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1918 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1919 InstrItinClass itin, string OpcodeStr, string Dt,
1920 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1921 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1922 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1923 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
1925 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1926 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1927 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1928 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1929 OpcodeStr, Dt, "$Vd, $Vm",
1930 "$src1 = $Vd, $src2 = $Vm", []>;
1931 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1932 InstrItinClass itin, string OpcodeStr, string Dt>
1933 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
1934 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
1935 "$src1 = $Vd, $src2 = $Vm", []>;
1937 // Basic 3-register operations: double- and quad-register.
1938 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1939 InstrItinClass itin, string OpcodeStr, string Dt,
1940 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1941 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1942 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1943 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1944 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1945 let isCommutable = Commutable;
1947 // Same as N3VD but no data type.
1948 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1949 InstrItinClass itin, string OpcodeStr,
1950 ValueType ResTy, ValueType OpTy,
1951 SDNode OpNode, bit Commutable>
1952 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
1953 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1954 OpcodeStr, "$Vd, $Vn, $Vm", "",
1955 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
1956 let isCommutable = Commutable;
1959 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
1960 InstrItinClass itin, string OpcodeStr, string Dt,
1961 ValueType Ty, SDNode ShOp>
1962 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
1963 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
1964 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
1966 (Ty (ShOp (Ty DPR:$Vn),
1967 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
1968 let isCommutable = 0;
1970 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
1971 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
1972 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
1973 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
1974 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
1976 (Ty (ShOp (Ty DPR:$Vn),
1977 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
1978 let isCommutable = 0;
1981 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1982 InstrItinClass itin, string OpcodeStr, string Dt,
1983 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1984 : N3V<op24, op23, op21_20, op11_8, 1, op4,
1985 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1986 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1987 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
1988 let isCommutable = Commutable;
1990 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1991 InstrItinClass itin, string OpcodeStr,
1992 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1993 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
1994 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
1995 OpcodeStr, "$Vd, $Vn, $Vm", "",
1996 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
1997 let isCommutable = Commutable;
1999 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2000 InstrItinClass itin, string OpcodeStr, string Dt,
2001 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2002 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2003 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2004 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2005 [(set (ResTy QPR:$Vd),
2006 (ResTy (ShOp (ResTy QPR:$Vn),
2007 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2009 let isCommutable = 0;
2011 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2012 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2013 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2014 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2015 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2016 [(set (ResTy QPR:$Vd),
2017 (ResTy (ShOp (ResTy QPR:$Vn),
2018 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2020 let isCommutable = 0;
2023 // Basic 3-register intrinsics, both double- and quad-register.
2024 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2025 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2026 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2027 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2028 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2029 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2030 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2031 let isCommutable = Commutable;
2033 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2034 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2035 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2036 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2037 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2039 (Ty (IntOp (Ty DPR:$Vn),
2040 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2042 let isCommutable = 0;
2044 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2045 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2046 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2047 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2048 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2050 (Ty (IntOp (Ty DPR:$Vn),
2051 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2052 let isCommutable = 0;
2054 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2055 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2056 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2057 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2058 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2059 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2060 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2061 let isCommutable = 0;
2064 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2065 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2066 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2067 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2068 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2069 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2070 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2071 let isCommutable = Commutable;
2073 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2074 string OpcodeStr, string Dt,
2075 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2076 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2077 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2078 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2079 [(set (ResTy QPR:$Vd),
2080 (ResTy (IntOp (ResTy QPR:$Vn),
2081 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2083 let isCommutable = 0;
2085 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2086 string OpcodeStr, string Dt,
2087 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2088 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2089 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2090 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2091 [(set (ResTy QPR:$Vd),
2092 (ResTy (IntOp (ResTy QPR:$Vn),
2093 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2095 let isCommutable = 0;
2097 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2098 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2099 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2100 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2101 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2102 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2103 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2104 let isCommutable = 0;
2107 // Multiply-Add/Sub operations: double- and quad-register.
2108 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2109 InstrItinClass itin, string OpcodeStr, string Dt,
2110 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2111 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2112 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2113 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2114 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2115 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2117 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2118 string OpcodeStr, string Dt,
2119 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2120 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2122 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2124 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2126 (Ty (ShOp (Ty DPR:$src1),
2128 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2130 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2131 string OpcodeStr, string Dt,
2132 ValueType Ty, SDNode MulOp, SDNode ShOp>
2133 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2135 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2137 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2139 (Ty (ShOp (Ty DPR:$src1),
2141 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2144 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2145 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2146 SDPatternOperator MulOp, SDPatternOperator OpNode>
2147 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2148 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2149 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2150 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2151 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2152 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2153 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2154 SDPatternOperator MulOp, SDPatternOperator ShOp>
2155 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2157 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2159 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2160 [(set (ResTy QPR:$Vd),
2161 (ResTy (ShOp (ResTy QPR:$src1),
2162 (ResTy (MulOp QPR:$Vn,
2163 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2165 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2166 string OpcodeStr, string Dt,
2167 ValueType ResTy, ValueType OpTy,
2168 SDNode MulOp, SDNode ShOp>
2169 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2171 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2173 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2174 [(set (ResTy QPR:$Vd),
2175 (ResTy (ShOp (ResTy QPR:$src1),
2176 (ResTy (MulOp QPR:$Vn,
2177 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2180 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2181 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2182 InstrItinClass itin, string OpcodeStr, string Dt,
2183 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2184 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2185 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2186 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2187 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2188 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2189 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2190 InstrItinClass itin, string OpcodeStr, string Dt,
2191 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2192 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2193 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2194 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2195 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2196 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2198 // Neon 3-argument intrinsics, both double- and quad-register.
2199 // The destination register is also used as the first source operand register.
2200 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2201 InstrItinClass itin, string OpcodeStr, string Dt,
2202 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2203 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2204 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2205 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2206 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2207 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2208 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2209 InstrItinClass itin, string OpcodeStr, string Dt,
2210 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2211 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2212 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2213 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2214 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2215 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2217 // Long Multiply-Add/Sub operations.
2218 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2219 InstrItinClass itin, string OpcodeStr, string Dt,
2220 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2221 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2222 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2223 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2224 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2225 (TyQ (MulOp (TyD DPR:$Vn),
2226 (TyD DPR:$Vm)))))]>;
2227 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2228 InstrItinClass itin, string OpcodeStr, string Dt,
2229 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2230 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2231 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2233 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2235 (OpNode (TyQ QPR:$src1),
2236 (TyQ (MulOp (TyD DPR:$Vn),
2237 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2239 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2240 InstrItinClass itin, string OpcodeStr, string Dt,
2241 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2242 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2243 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2245 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2247 (OpNode (TyQ QPR:$src1),
2248 (TyQ (MulOp (TyD DPR:$Vn),
2249 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2252 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2253 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2254 InstrItinClass itin, string OpcodeStr, string Dt,
2255 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2257 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2258 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2259 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2260 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2261 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2262 (TyD DPR:$Vm)))))))]>;
2264 // Neon Long 3-argument intrinsic. The destination register is
2265 // a quad-register and is also used as the first source operand register.
2266 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2267 InstrItinClass itin, string OpcodeStr, string Dt,
2268 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2269 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2270 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2271 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2273 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2274 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2275 string OpcodeStr, string Dt,
2276 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2277 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2279 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2281 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2282 [(set (ResTy QPR:$Vd),
2283 (ResTy (IntOp (ResTy QPR:$src1),
2285 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2287 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2288 InstrItinClass itin, string OpcodeStr, string Dt,
2289 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2290 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2292 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2294 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2295 [(set (ResTy QPR:$Vd),
2296 (ResTy (IntOp (ResTy QPR:$src1),
2298 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2301 // Narrowing 3-register intrinsics.
2302 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2303 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2304 Intrinsic IntOp, bit Commutable>
2305 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2306 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2307 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2308 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2309 let isCommutable = Commutable;
2312 // Long 3-register operations.
2313 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2314 InstrItinClass itin, string OpcodeStr, string Dt,
2315 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2316 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2317 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2318 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2319 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2320 let isCommutable = Commutable;
2322 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2323 InstrItinClass itin, string OpcodeStr, string Dt,
2324 ValueType TyQ, ValueType TyD, SDNode OpNode>
2325 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2326 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2327 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2329 (TyQ (OpNode (TyD DPR:$Vn),
2330 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2331 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2332 InstrItinClass itin, string OpcodeStr, string Dt,
2333 ValueType TyQ, ValueType TyD, SDNode OpNode>
2334 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2335 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2336 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2338 (TyQ (OpNode (TyD DPR:$Vn),
2339 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2341 // Long 3-register operations with explicitly extended operands.
2342 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2343 InstrItinClass itin, string OpcodeStr, string Dt,
2344 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2346 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2347 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2348 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2349 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2350 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2351 let isCommutable = Commutable;
2354 // Long 3-register intrinsics with explicit extend (VABDL).
2355 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2356 InstrItinClass itin, string OpcodeStr, string Dt,
2357 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2359 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2360 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2361 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2362 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2363 (TyD DPR:$Vm))))))]> {
2364 let isCommutable = Commutable;
2367 // Long 3-register intrinsics.
2368 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2369 InstrItinClass itin, string OpcodeStr, string Dt,
2370 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2371 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2372 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2373 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2374 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2375 let isCommutable = Commutable;
2377 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2378 string OpcodeStr, string Dt,
2379 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2380 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2381 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2382 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2383 [(set (ResTy QPR:$Vd),
2384 (ResTy (IntOp (OpTy DPR:$Vn),
2385 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2387 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2388 InstrItinClass itin, string OpcodeStr, string Dt,
2389 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2390 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2391 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2392 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2393 [(set (ResTy QPR:$Vd),
2394 (ResTy (IntOp (OpTy DPR:$Vn),
2395 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2398 // Wide 3-register operations.
2399 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2400 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2401 SDNode OpNode, SDNode ExtOp, bit Commutable>
2402 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2403 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2404 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2405 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2406 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2407 let isCommutable = Commutable;
2410 // Pairwise long 2-register intrinsics, both double- and quad-register.
2411 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2412 bits<2> op17_16, bits<5> op11_7, bit op4,
2413 string OpcodeStr, string Dt,
2414 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2415 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2416 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2417 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2418 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2419 bits<2> op17_16, bits<5> op11_7, bit op4,
2420 string OpcodeStr, string Dt,
2421 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2422 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2423 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2424 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2426 // Pairwise long 2-register accumulate intrinsics,
2427 // both double- and quad-register.
2428 // The destination register is also used as the first source operand register.
2429 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2430 bits<2> op17_16, bits<5> op11_7, bit op4,
2431 string OpcodeStr, string Dt,
2432 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2433 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2434 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2435 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2436 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2437 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2438 bits<2> op17_16, bits<5> op11_7, bit op4,
2439 string OpcodeStr, string Dt,
2440 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2441 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2442 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2443 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2444 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2446 // Shift by immediate,
2447 // both double- and quad-register.
2448 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2449 Format f, InstrItinClass itin, Operand ImmTy,
2450 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2451 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2452 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2453 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2454 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2455 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2456 Format f, InstrItinClass itin, Operand ImmTy,
2457 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2458 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2459 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2460 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2461 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2463 // Long shift by immediate.
2464 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2465 string OpcodeStr, string Dt,
2466 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2467 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2468 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2469 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2470 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2471 (i32 imm:$SIMM))))]>;
2473 // Narrow shift by immediate.
2474 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2475 InstrItinClass itin, string OpcodeStr, string Dt,
2476 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2477 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2478 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2479 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2480 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2481 (i32 imm:$SIMM))))]>;
2483 // Shift right by immediate and accumulate,
2484 // both double- and quad-register.
2485 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2486 Operand ImmTy, string OpcodeStr, string Dt,
2487 ValueType Ty, SDNode ShOp>
2488 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2489 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2490 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2491 [(set DPR:$Vd, (Ty (add DPR:$src1,
2492 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2493 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2494 Operand ImmTy, string OpcodeStr, string Dt,
2495 ValueType Ty, SDNode ShOp>
2496 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2497 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2498 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2499 [(set QPR:$Vd, (Ty (add QPR:$src1,
2500 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2502 // Shift by immediate and insert,
2503 // both double- and quad-register.
2504 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2505 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2506 ValueType Ty,SDNode ShOp>
2507 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2508 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2509 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2510 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2511 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2512 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2513 ValueType Ty,SDNode ShOp>
2514 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2515 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2516 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2517 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2519 // Convert, with fractional bits immediate,
2520 // both double- and quad-register.
2521 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2522 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2524 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2525 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2526 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2527 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2528 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2529 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2531 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2532 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2533 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2534 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2536 //===----------------------------------------------------------------------===//
2538 //===----------------------------------------------------------------------===//
2540 // Abbreviations used in multiclass suffixes:
2541 // Q = quarter int (8 bit) elements
2542 // H = half int (16 bit) elements
2543 // S = single int (32 bit) elements
2544 // D = double int (64 bit) elements
2546 // Neon 2-register vector operations and intrinsics.
2548 // Neon 2-register comparisons.
2549 // source operand element sizes of 8, 16 and 32 bits:
2550 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2551 bits<5> op11_7, bit op4, string opc, string Dt,
2552 string asm, SDNode OpNode> {
2553 // 64-bit vector types.
2554 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2555 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2556 opc, !strconcat(Dt, "8"), asm, "",
2557 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2558 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2559 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2560 opc, !strconcat(Dt, "16"), asm, "",
2561 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2562 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2563 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2564 opc, !strconcat(Dt, "32"), asm, "",
2565 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2566 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2567 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2568 opc, "f32", asm, "",
2569 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2570 let Inst{10} = 1; // overwrite F = 1
2573 // 128-bit vector types.
2574 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2575 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2576 opc, !strconcat(Dt, "8"), asm, "",
2577 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2578 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2579 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2580 opc, !strconcat(Dt, "16"), asm, "",
2581 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2582 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2583 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2584 opc, !strconcat(Dt, "32"), asm, "",
2585 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2586 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2587 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2588 opc, "f32", asm, "",
2589 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2590 let Inst{10} = 1; // overwrite F = 1
2595 // Neon 2-register vector intrinsics,
2596 // element sizes of 8, 16 and 32 bits:
2597 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2598 bits<5> op11_7, bit op4,
2599 InstrItinClass itinD, InstrItinClass itinQ,
2600 string OpcodeStr, string Dt, Intrinsic IntOp> {
2601 // 64-bit vector types.
2602 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2603 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2604 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2605 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2606 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2607 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2609 // 128-bit vector types.
2610 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2611 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2612 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2613 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2614 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2615 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2619 // Neon Narrowing 2-register vector operations,
2620 // source operand element sizes of 16, 32 and 64 bits:
2621 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2622 bits<5> op11_7, bit op6, bit op4,
2623 InstrItinClass itin, string OpcodeStr, string Dt,
2625 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2626 itin, OpcodeStr, !strconcat(Dt, "16"),
2627 v8i8, v8i16, OpNode>;
2628 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2629 itin, OpcodeStr, !strconcat(Dt, "32"),
2630 v4i16, v4i32, OpNode>;
2631 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2632 itin, OpcodeStr, !strconcat(Dt, "64"),
2633 v2i32, v2i64, OpNode>;
2636 // Neon Narrowing 2-register vector intrinsics,
2637 // source operand element sizes of 16, 32 and 64 bits:
2638 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2639 bits<5> op11_7, bit op6, bit op4,
2640 InstrItinClass itin, string OpcodeStr, string Dt,
2642 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2643 itin, OpcodeStr, !strconcat(Dt, "16"),
2644 v8i8, v8i16, IntOp>;
2645 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2646 itin, OpcodeStr, !strconcat(Dt, "32"),
2647 v4i16, v4i32, IntOp>;
2648 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2649 itin, OpcodeStr, !strconcat(Dt, "64"),
2650 v2i32, v2i64, IntOp>;
2654 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2655 // source operand element sizes of 16, 32 and 64 bits:
2656 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2657 string OpcodeStr, string Dt, SDNode OpNode> {
2658 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2659 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2660 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2661 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2662 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2663 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2667 // Neon 3-register vector operations.
2669 // First with only element sizes of 8, 16 and 32 bits:
2670 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2671 InstrItinClass itinD16, InstrItinClass itinD32,
2672 InstrItinClass itinQ16, InstrItinClass itinQ32,
2673 string OpcodeStr, string Dt,
2674 SDNode OpNode, bit Commutable = 0> {
2675 // 64-bit vector types.
2676 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2677 OpcodeStr, !strconcat(Dt, "8"),
2678 v8i8, v8i8, OpNode, Commutable>;
2679 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2680 OpcodeStr, !strconcat(Dt, "16"),
2681 v4i16, v4i16, OpNode, Commutable>;
2682 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2683 OpcodeStr, !strconcat(Dt, "32"),
2684 v2i32, v2i32, OpNode, Commutable>;
2686 // 128-bit vector types.
2687 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2688 OpcodeStr, !strconcat(Dt, "8"),
2689 v16i8, v16i8, OpNode, Commutable>;
2690 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2691 OpcodeStr, !strconcat(Dt, "16"),
2692 v8i16, v8i16, OpNode, Commutable>;
2693 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2694 OpcodeStr, !strconcat(Dt, "32"),
2695 v4i32, v4i32, OpNode, Commutable>;
2698 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2699 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2701 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2703 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2704 v8i16, v4i16, ShOp>;
2705 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2706 v4i32, v2i32, ShOp>;
2709 // ....then also with element size 64 bits:
2710 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2711 InstrItinClass itinD, InstrItinClass itinQ,
2712 string OpcodeStr, string Dt,
2713 SDNode OpNode, bit Commutable = 0>
2714 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2715 OpcodeStr, Dt, OpNode, Commutable> {
2716 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2717 OpcodeStr, !strconcat(Dt, "64"),
2718 v1i64, v1i64, OpNode, Commutable>;
2719 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2720 OpcodeStr, !strconcat(Dt, "64"),
2721 v2i64, v2i64, OpNode, Commutable>;
2725 // Neon 3-register vector intrinsics.
2727 // First with only element sizes of 16 and 32 bits:
2728 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2729 InstrItinClass itinD16, InstrItinClass itinD32,
2730 InstrItinClass itinQ16, InstrItinClass itinQ32,
2731 string OpcodeStr, string Dt,
2732 Intrinsic IntOp, bit Commutable = 0> {
2733 // 64-bit vector types.
2734 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2735 OpcodeStr, !strconcat(Dt, "16"),
2736 v4i16, v4i16, IntOp, Commutable>;
2737 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2738 OpcodeStr, !strconcat(Dt, "32"),
2739 v2i32, v2i32, IntOp, Commutable>;
2741 // 128-bit vector types.
2742 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2743 OpcodeStr, !strconcat(Dt, "16"),
2744 v8i16, v8i16, IntOp, Commutable>;
2745 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2746 OpcodeStr, !strconcat(Dt, "32"),
2747 v4i32, v4i32, IntOp, Commutable>;
2749 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2750 InstrItinClass itinD16, InstrItinClass itinD32,
2751 InstrItinClass itinQ16, InstrItinClass itinQ32,
2752 string OpcodeStr, string Dt,
2754 // 64-bit vector types.
2755 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2756 OpcodeStr, !strconcat(Dt, "16"),
2757 v4i16, v4i16, IntOp>;
2758 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2759 OpcodeStr, !strconcat(Dt, "32"),
2760 v2i32, v2i32, IntOp>;
2762 // 128-bit vector types.
2763 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2764 OpcodeStr, !strconcat(Dt, "16"),
2765 v8i16, v8i16, IntOp>;
2766 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2767 OpcodeStr, !strconcat(Dt, "32"),
2768 v4i32, v4i32, IntOp>;
2771 multiclass N3VIntSL_HS<bits<4> op11_8,
2772 InstrItinClass itinD16, InstrItinClass itinD32,
2773 InstrItinClass itinQ16, InstrItinClass itinQ32,
2774 string OpcodeStr, string Dt, Intrinsic IntOp> {
2775 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2776 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2777 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2778 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2779 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2780 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2781 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2782 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2785 // ....then also with element size of 8 bits:
2786 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2787 InstrItinClass itinD16, InstrItinClass itinD32,
2788 InstrItinClass itinQ16, InstrItinClass itinQ32,
2789 string OpcodeStr, string Dt,
2790 Intrinsic IntOp, bit Commutable = 0>
2791 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2792 OpcodeStr, Dt, IntOp, Commutable> {
2793 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2794 OpcodeStr, !strconcat(Dt, "8"),
2795 v8i8, v8i8, IntOp, Commutable>;
2796 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2797 OpcodeStr, !strconcat(Dt, "8"),
2798 v16i8, v16i8, IntOp, Commutable>;
2800 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2801 InstrItinClass itinD16, InstrItinClass itinD32,
2802 InstrItinClass itinQ16, InstrItinClass itinQ32,
2803 string OpcodeStr, string Dt,
2805 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2806 OpcodeStr, Dt, IntOp> {
2807 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2808 OpcodeStr, !strconcat(Dt, "8"),
2810 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2811 OpcodeStr, !strconcat(Dt, "8"),
2812 v16i8, v16i8, IntOp>;
2816 // ....then also with element size of 64 bits:
2817 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2818 InstrItinClass itinD16, InstrItinClass itinD32,
2819 InstrItinClass itinQ16, InstrItinClass itinQ32,
2820 string OpcodeStr, string Dt,
2821 Intrinsic IntOp, bit Commutable = 0>
2822 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2823 OpcodeStr, Dt, IntOp, Commutable> {
2824 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2825 OpcodeStr, !strconcat(Dt, "64"),
2826 v1i64, v1i64, IntOp, Commutable>;
2827 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2828 OpcodeStr, !strconcat(Dt, "64"),
2829 v2i64, v2i64, IntOp, Commutable>;
2831 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2832 InstrItinClass itinD16, InstrItinClass itinD32,
2833 InstrItinClass itinQ16, InstrItinClass itinQ32,
2834 string OpcodeStr, string Dt,
2836 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2837 OpcodeStr, Dt, IntOp> {
2838 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2839 OpcodeStr, !strconcat(Dt, "64"),
2840 v1i64, v1i64, IntOp>;
2841 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2842 OpcodeStr, !strconcat(Dt, "64"),
2843 v2i64, v2i64, IntOp>;
2846 // Neon Narrowing 3-register vector intrinsics,
2847 // source operand element sizes of 16, 32 and 64 bits:
2848 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2849 string OpcodeStr, string Dt,
2850 Intrinsic IntOp, bit Commutable = 0> {
2851 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2852 OpcodeStr, !strconcat(Dt, "16"),
2853 v8i8, v8i16, IntOp, Commutable>;
2854 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2855 OpcodeStr, !strconcat(Dt, "32"),
2856 v4i16, v4i32, IntOp, Commutable>;
2857 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2858 OpcodeStr, !strconcat(Dt, "64"),
2859 v2i32, v2i64, IntOp, Commutable>;
2863 // Neon Long 3-register vector operations.
2865 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2866 InstrItinClass itin16, InstrItinClass itin32,
2867 string OpcodeStr, string Dt,
2868 SDNode OpNode, bit Commutable = 0> {
2869 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2870 OpcodeStr, !strconcat(Dt, "8"),
2871 v8i16, v8i8, OpNode, Commutable>;
2872 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2873 OpcodeStr, !strconcat(Dt, "16"),
2874 v4i32, v4i16, OpNode, Commutable>;
2875 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2876 OpcodeStr, !strconcat(Dt, "32"),
2877 v2i64, v2i32, OpNode, Commutable>;
2880 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2881 InstrItinClass itin, string OpcodeStr, string Dt,
2883 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2884 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2885 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2886 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2889 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2890 InstrItinClass itin16, InstrItinClass itin32,
2891 string OpcodeStr, string Dt,
2892 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2893 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2894 OpcodeStr, !strconcat(Dt, "8"),
2895 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2896 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2897 OpcodeStr, !strconcat(Dt, "16"),
2898 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2899 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2900 OpcodeStr, !strconcat(Dt, "32"),
2901 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2904 // Neon Long 3-register vector intrinsics.
2906 // First with only element sizes of 16 and 32 bits:
2907 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2908 InstrItinClass itin16, InstrItinClass itin32,
2909 string OpcodeStr, string Dt,
2910 Intrinsic IntOp, bit Commutable = 0> {
2911 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2912 OpcodeStr, !strconcat(Dt, "16"),
2913 v4i32, v4i16, IntOp, Commutable>;
2914 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2915 OpcodeStr, !strconcat(Dt, "32"),
2916 v2i64, v2i32, IntOp, Commutable>;
2919 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2920 InstrItinClass itin, string OpcodeStr, string Dt,
2922 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2923 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2924 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2925 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2928 // ....then also with element size of 8 bits:
2929 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2930 InstrItinClass itin16, InstrItinClass itin32,
2931 string OpcodeStr, string Dt,
2932 Intrinsic IntOp, bit Commutable = 0>
2933 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2934 IntOp, Commutable> {
2935 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2936 OpcodeStr, !strconcat(Dt, "8"),
2937 v8i16, v8i8, IntOp, Commutable>;
2940 // ....with explicit extend (VABDL).
2941 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2942 InstrItinClass itin, string OpcodeStr, string Dt,
2943 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2944 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2945 OpcodeStr, !strconcat(Dt, "8"),
2946 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2947 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2948 OpcodeStr, !strconcat(Dt, "16"),
2949 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2950 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2951 OpcodeStr, !strconcat(Dt, "32"),
2952 v2i64, v2i32, IntOp, ExtOp, Commutable>;
2956 // Neon Wide 3-register vector intrinsics,
2957 // source operand element sizes of 8, 16 and 32 bits:
2958 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2959 string OpcodeStr, string Dt,
2960 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2961 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
2962 OpcodeStr, !strconcat(Dt, "8"),
2963 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2964 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
2965 OpcodeStr, !strconcat(Dt, "16"),
2966 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2967 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
2968 OpcodeStr, !strconcat(Dt, "32"),
2969 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2973 // Neon Multiply-Op vector operations,
2974 // element sizes of 8, 16 and 32 bits:
2975 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2976 InstrItinClass itinD16, InstrItinClass itinD32,
2977 InstrItinClass itinQ16, InstrItinClass itinQ32,
2978 string OpcodeStr, string Dt, SDNode OpNode> {
2979 // 64-bit vector types.
2980 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
2981 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
2982 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
2983 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
2984 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
2985 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
2987 // 128-bit vector types.
2988 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
2989 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
2990 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
2991 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
2992 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
2993 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
2996 multiclass N3VMulOpSL_HS<bits<4> op11_8,
2997 InstrItinClass itinD16, InstrItinClass itinD32,
2998 InstrItinClass itinQ16, InstrItinClass itinQ32,
2999 string OpcodeStr, string Dt, SDNode ShOp> {
3000 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3001 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3002 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3003 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3004 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3005 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3007 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3008 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3012 // Neon Intrinsic-Op vector operations,
3013 // element sizes of 8, 16 and 32 bits:
3014 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3015 InstrItinClass itinD, InstrItinClass itinQ,
3016 string OpcodeStr, string Dt, Intrinsic IntOp,
3018 // 64-bit vector types.
3019 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3020 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3021 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3022 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3023 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3024 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3026 // 128-bit vector types.
3027 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3028 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3029 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3030 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3031 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3032 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3035 // Neon 3-argument intrinsics,
3036 // element sizes of 8, 16 and 32 bits:
3037 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3038 InstrItinClass itinD, InstrItinClass itinQ,
3039 string OpcodeStr, string Dt, Intrinsic IntOp> {
3040 // 64-bit vector types.
3041 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3042 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3043 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3044 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3045 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3046 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3048 // 128-bit vector types.
3049 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3050 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3051 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3052 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3053 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3054 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3058 // Neon Long Multiply-Op vector operations,
3059 // element sizes of 8, 16 and 32 bits:
3060 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3061 InstrItinClass itin16, InstrItinClass itin32,
3062 string OpcodeStr, string Dt, SDNode MulOp,
3064 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3065 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3066 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3067 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3068 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3069 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3072 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3073 string Dt, SDNode MulOp, SDNode OpNode> {
3074 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3075 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3076 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3077 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3081 // Neon Long 3-argument intrinsics.
3083 // First with only element sizes of 16 and 32 bits:
3084 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3085 InstrItinClass itin16, InstrItinClass itin32,
3086 string OpcodeStr, string Dt, Intrinsic IntOp> {
3087 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3088 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3089 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3090 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3093 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3094 string OpcodeStr, string Dt, Intrinsic IntOp> {
3095 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3096 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3097 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3098 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3101 // ....then also with element size of 8 bits:
3102 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3103 InstrItinClass itin16, InstrItinClass itin32,
3104 string OpcodeStr, string Dt, Intrinsic IntOp>
3105 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3106 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3107 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3110 // ....with explicit extend (VABAL).
3111 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3112 InstrItinClass itin, string OpcodeStr, string Dt,
3113 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3114 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3115 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3116 IntOp, ExtOp, OpNode>;
3117 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3118 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3119 IntOp, ExtOp, OpNode>;
3120 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3121 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3122 IntOp, ExtOp, OpNode>;
3126 // Neon Pairwise long 2-register intrinsics,
3127 // element sizes of 8, 16 and 32 bits:
3128 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3129 bits<5> op11_7, bit op4,
3130 string OpcodeStr, string Dt, Intrinsic IntOp> {
3131 // 64-bit vector types.
3132 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3133 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3134 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3135 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3136 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3137 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3139 // 128-bit vector types.
3140 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3141 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3142 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3143 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3144 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3145 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3149 // Neon Pairwise long 2-register accumulate intrinsics,
3150 // element sizes of 8, 16 and 32 bits:
3151 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3152 bits<5> op11_7, bit op4,
3153 string OpcodeStr, string Dt, Intrinsic IntOp> {
3154 // 64-bit vector types.
3155 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3156 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3157 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3158 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3159 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3160 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3162 // 128-bit vector types.
3163 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3164 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3165 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3166 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3167 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3168 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3172 // Neon 2-register vector shift by immediate,
3173 // with f of either N2RegVShLFrm or N2RegVShRFrm
3174 // element sizes of 8, 16, 32 and 64 bits:
3175 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3176 InstrItinClass itin, string OpcodeStr, string Dt,
3178 // 64-bit vector types.
3179 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3180 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3181 let Inst{21-19} = 0b001; // imm6 = 001xxx
3183 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3184 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3185 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3187 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3188 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3189 let Inst{21} = 0b1; // imm6 = 1xxxxx
3191 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3192 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3195 // 128-bit vector types.
3196 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3197 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3198 let Inst{21-19} = 0b001; // imm6 = 001xxx
3200 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3201 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3202 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3204 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3205 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3206 let Inst{21} = 0b1; // imm6 = 1xxxxx
3208 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3209 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3212 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3213 InstrItinClass itin, string OpcodeStr, string Dt,
3215 // 64-bit vector types.
3216 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3217 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3218 let Inst{21-19} = 0b001; // imm6 = 001xxx
3220 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3221 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3222 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3224 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3225 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3226 let Inst{21} = 0b1; // imm6 = 1xxxxx
3228 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3229 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3232 // 128-bit vector types.
3233 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3234 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3235 let Inst{21-19} = 0b001; // imm6 = 001xxx
3237 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3238 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3239 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3241 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3242 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3243 let Inst{21} = 0b1; // imm6 = 1xxxxx
3245 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3246 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3250 // Neon Shift-Accumulate vector operations,
3251 // element sizes of 8, 16, 32 and 64 bits:
3252 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3253 string OpcodeStr, string Dt, SDNode ShOp> {
3254 // 64-bit vector types.
3255 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3256 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3257 let Inst{21-19} = 0b001; // imm6 = 001xxx
3259 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3260 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3261 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3263 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3264 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3265 let Inst{21} = 0b1; // imm6 = 1xxxxx
3267 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3268 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3271 // 128-bit vector types.
3272 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3273 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3274 let Inst{21-19} = 0b001; // imm6 = 001xxx
3276 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3277 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3278 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3280 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3281 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3282 let Inst{21} = 0b1; // imm6 = 1xxxxx
3284 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3285 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3289 // Neon Shift-Insert vector operations,
3290 // with f of either N2RegVShLFrm or N2RegVShRFrm
3291 // element sizes of 8, 16, 32 and 64 bits:
3292 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3294 // 64-bit vector types.
3295 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3296 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3297 let Inst{21-19} = 0b001; // imm6 = 001xxx
3299 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3300 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3301 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3303 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3304 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3305 let Inst{21} = 0b1; // imm6 = 1xxxxx
3307 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3308 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3311 // 128-bit vector types.
3312 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3313 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3314 let Inst{21-19} = 0b001; // imm6 = 001xxx
3316 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3317 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3318 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3320 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3321 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3322 let Inst{21} = 0b1; // imm6 = 1xxxxx
3324 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3325 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3328 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3330 // 64-bit vector types.
3331 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3332 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3333 let Inst{21-19} = 0b001; // imm6 = 001xxx
3335 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3336 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3337 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3339 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3340 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3341 let Inst{21} = 0b1; // imm6 = 1xxxxx
3343 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3344 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3347 // 128-bit vector types.
3348 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3349 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3350 let Inst{21-19} = 0b001; // imm6 = 001xxx
3352 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3353 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3354 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3356 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3357 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3358 let Inst{21} = 0b1; // imm6 = 1xxxxx
3360 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3361 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3365 // Neon Shift Long operations,
3366 // element sizes of 8, 16, 32 bits:
3367 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3368 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3369 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3370 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3371 let Inst{21-19} = 0b001; // imm6 = 001xxx
3373 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3374 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3375 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3377 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3378 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3379 let Inst{21} = 0b1; // imm6 = 1xxxxx
3383 // Neon Shift Narrow operations,
3384 // element sizes of 16, 32, 64 bits:
3385 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3386 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3388 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3389 OpcodeStr, !strconcat(Dt, "16"),
3390 v8i8, v8i16, shr_imm8, OpNode> {
3391 let Inst{21-19} = 0b001; // imm6 = 001xxx
3393 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3394 OpcodeStr, !strconcat(Dt, "32"),
3395 v4i16, v4i32, shr_imm16, OpNode> {
3396 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3398 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3399 OpcodeStr, !strconcat(Dt, "64"),
3400 v2i32, v2i64, shr_imm32, OpNode> {
3401 let Inst{21} = 0b1; // imm6 = 1xxxxx
3405 //===----------------------------------------------------------------------===//
3406 // Instruction Definitions.
3407 //===----------------------------------------------------------------------===//
3409 // Vector Add Operations.
3411 // VADD : Vector Add (integer and floating-point)
3412 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3414 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3415 v2f32, v2f32, fadd, 1>;
3416 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3417 v4f32, v4f32, fadd, 1>;
3418 // VADDL : Vector Add Long (Q = D + D)
3419 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3420 "vaddl", "s", add, sext, 1>;
3421 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3422 "vaddl", "u", add, zext, 1>;
3423 // VADDW : Vector Add Wide (Q = Q + D)
3424 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3425 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3426 // VHADD : Vector Halving Add
3427 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3428 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3429 "vhadd", "s", int_arm_neon_vhadds, 1>;
3430 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3431 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3432 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3433 // VRHADD : Vector Rounding Halving Add
3434 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3435 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3436 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3437 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3438 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3439 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3440 // VQADD : Vector Saturating Add
3441 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3442 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3443 "vqadd", "s", int_arm_neon_vqadds, 1>;
3444 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3445 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3446 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3447 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3448 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3449 int_arm_neon_vaddhn, 1>;
3450 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3451 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3452 int_arm_neon_vraddhn, 1>;
3454 // Vector Multiply Operations.
3456 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3457 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3458 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3459 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3460 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3461 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3462 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3463 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3464 v2f32, v2f32, fmul, 1>;
3465 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3466 v4f32, v4f32, fmul, 1>;
3467 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3468 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3469 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3472 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3473 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3474 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3475 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3476 (DSubReg_i16_reg imm:$lane))),
3477 (SubReg_i16_lane imm:$lane)))>;
3478 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3479 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3480 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3481 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3482 (DSubReg_i32_reg imm:$lane))),
3483 (SubReg_i32_lane imm:$lane)))>;
3484 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3485 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3486 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3487 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3488 (DSubReg_i32_reg imm:$lane))),
3489 (SubReg_i32_lane imm:$lane)))>;
3491 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3492 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3493 IIC_VMULi16Q, IIC_VMULi32Q,
3494 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3495 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3496 IIC_VMULi16Q, IIC_VMULi32Q,
3497 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3498 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3499 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3501 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3502 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3503 (DSubReg_i16_reg imm:$lane))),
3504 (SubReg_i16_lane imm:$lane)))>;
3505 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3506 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3508 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3509 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3510 (DSubReg_i32_reg imm:$lane))),
3511 (SubReg_i32_lane imm:$lane)))>;
3513 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3514 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3515 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3516 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3517 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3518 IIC_VMULi16Q, IIC_VMULi32Q,
3519 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3520 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3521 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3523 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3524 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3525 (DSubReg_i16_reg imm:$lane))),
3526 (SubReg_i16_lane imm:$lane)))>;
3527 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3528 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3530 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3531 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3532 (DSubReg_i32_reg imm:$lane))),
3533 (SubReg_i32_lane imm:$lane)))>;
3535 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3536 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3537 "vmull", "s", NEONvmulls, 1>;
3538 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3539 "vmull", "u", NEONvmullu, 1>;
3540 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3541 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3542 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3543 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3545 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3546 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3547 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3548 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3549 "vqdmull", "s", int_arm_neon_vqdmull>;
3551 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3553 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3554 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3555 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3556 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3557 v2f32, fmul_su, fadd_mlx>,
3558 Requires<[HasNEON, UseFPVMLx]>;
3559 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3560 v4f32, fmul_su, fadd_mlx>,
3561 Requires<[HasNEON, UseFPVMLx]>;
3562 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3563 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3564 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3565 v2f32, fmul_su, fadd_mlx>,
3566 Requires<[HasNEON, UseFPVMLx]>;
3567 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3568 v4f32, v2f32, fmul_su, fadd_mlx>,
3569 Requires<[HasNEON, UseFPVMLx]>;
3571 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3572 (mul (v8i16 QPR:$src2),
3573 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3574 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3575 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3576 (DSubReg_i16_reg imm:$lane))),
3577 (SubReg_i16_lane imm:$lane)))>;
3579 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3580 (mul (v4i32 QPR:$src2),
3581 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3582 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3583 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3584 (DSubReg_i32_reg imm:$lane))),
3585 (SubReg_i32_lane imm:$lane)))>;
3587 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3588 (fmul_su (v4f32 QPR:$src2),
3589 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3590 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3592 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3593 (DSubReg_i32_reg imm:$lane))),
3594 (SubReg_i32_lane imm:$lane)))>,
3595 Requires<[HasNEON, UseFPVMLx]>;
3597 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3598 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3599 "vmlal", "s", NEONvmulls, add>;
3600 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3601 "vmlal", "u", NEONvmullu, add>;
3603 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3604 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3606 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3607 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3608 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3609 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3611 // VMLS : Vector Multiply Subtract (integer and floating-point)
3612 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3613 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3614 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3615 v2f32, fmul_su, fsub_mlx>,
3616 Requires<[HasNEON, UseFPVMLx]>;
3617 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3618 v4f32, fmul_su, fsub_mlx>,
3619 Requires<[HasNEON, UseFPVMLx]>;
3620 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3621 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3622 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3623 v2f32, fmul_su, fsub_mlx>,
3624 Requires<[HasNEON, UseFPVMLx]>;
3625 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3626 v4f32, v2f32, fmul_su, fsub_mlx>,
3627 Requires<[HasNEON, UseFPVMLx]>;
3629 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3630 (mul (v8i16 QPR:$src2),
3631 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3632 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3633 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3634 (DSubReg_i16_reg imm:$lane))),
3635 (SubReg_i16_lane imm:$lane)))>;
3637 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3638 (mul (v4i32 QPR:$src2),
3639 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3640 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3641 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3642 (DSubReg_i32_reg imm:$lane))),
3643 (SubReg_i32_lane imm:$lane)))>;
3645 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3646 (fmul_su (v4f32 QPR:$src2),
3647 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3648 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3649 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3650 (DSubReg_i32_reg imm:$lane))),
3651 (SubReg_i32_lane imm:$lane)))>,
3652 Requires<[HasNEON, UseFPVMLx]>;
3654 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3655 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3656 "vmlsl", "s", NEONvmulls, sub>;
3657 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3658 "vmlsl", "u", NEONvmullu, sub>;
3660 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3661 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3663 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3664 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3665 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3666 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3668 // Vector Subtract Operations.
3670 // VSUB : Vector Subtract (integer and floating-point)
3671 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3672 "vsub", "i", sub, 0>;
3673 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3674 v2f32, v2f32, fsub, 0>;
3675 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3676 v4f32, v4f32, fsub, 0>;
3677 // VSUBL : Vector Subtract Long (Q = D - D)
3678 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3679 "vsubl", "s", sub, sext, 0>;
3680 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3681 "vsubl", "u", sub, zext, 0>;
3682 // VSUBW : Vector Subtract Wide (Q = Q - D)
3683 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3684 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3685 // VHSUB : Vector Halving Subtract
3686 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3687 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3688 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3689 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3690 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3691 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3692 // VQSUB : Vector Saturing Subtract
3693 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3694 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3695 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3696 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3697 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3698 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3699 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3700 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3701 int_arm_neon_vsubhn, 0>;
3702 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3703 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3704 int_arm_neon_vrsubhn, 0>;
3706 // Vector Comparisons.
3708 // VCEQ : Vector Compare Equal
3709 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3710 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3711 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3713 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3716 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3717 "$Vd, $Vm, #0", NEONvceqz>;
3719 // VCGE : Vector Compare Greater Than or Equal
3720 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3721 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3722 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3723 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3724 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3726 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3729 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3730 "$Vd, $Vm, #0", NEONvcgez>;
3731 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3732 "$Vd, $Vm, #0", NEONvclez>;
3734 // VCGT : Vector Compare Greater Than
3735 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3736 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3737 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3738 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3739 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3741 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3744 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3745 "$Vd, $Vm, #0", NEONvcgtz>;
3746 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3747 "$Vd, $Vm, #0", NEONvcltz>;
3749 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3750 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3751 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3752 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3753 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3754 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3755 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3756 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3757 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3758 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3759 // VTST : Vector Test Bits
3760 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3761 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3763 // Vector Bitwise Operations.
3765 def vnotd : PatFrag<(ops node:$in),
3766 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3767 def vnotq : PatFrag<(ops node:$in),
3768 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3771 // VAND : Vector Bitwise AND
3772 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3773 v2i32, v2i32, and, 1>;
3774 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3775 v4i32, v4i32, and, 1>;
3777 // VEOR : Vector Bitwise Exclusive OR
3778 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3779 v2i32, v2i32, xor, 1>;
3780 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3781 v4i32, v4i32, xor, 1>;
3783 // VORR : Vector Bitwise OR
3784 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3785 v2i32, v2i32, or, 1>;
3786 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3787 v4i32, v4i32, or, 1>;
3789 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3790 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3792 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3794 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3795 let Inst{9} = SIMM{9};
3798 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3799 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3801 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3803 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3804 let Inst{10-9} = SIMM{10-9};
3807 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3808 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3810 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3812 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3813 let Inst{9} = SIMM{9};
3816 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3817 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3819 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3821 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3822 let Inst{10-9} = SIMM{10-9};
3826 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3827 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3828 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3829 "vbic", "$Vd, $Vn, $Vm", "",
3830 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3831 (vnotd DPR:$Vm))))]>;
3832 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3833 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3834 "vbic", "$Vd, $Vn, $Vm", "",
3835 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3836 (vnotq QPR:$Vm))))]>;
3838 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3839 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3841 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3843 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3844 let Inst{9} = SIMM{9};
3847 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3848 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3850 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3852 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3853 let Inst{10-9} = SIMM{10-9};
3856 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3857 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3859 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3861 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3862 let Inst{9} = SIMM{9};
3865 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3866 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3868 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3870 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3871 let Inst{10-9} = SIMM{10-9};
3874 // VORN : Vector Bitwise OR NOT
3875 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3876 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3877 "vorn", "$Vd, $Vn, $Vm", "",
3878 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3879 (vnotd DPR:$Vm))))]>;
3880 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3881 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3882 "vorn", "$Vd, $Vn, $Vm", "",
3883 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3884 (vnotq QPR:$Vm))))]>;
3886 // VMVN : Vector Bitwise NOT (Immediate)
3888 let isReMaterializable = 1 in {
3890 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3891 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3892 "vmvn", "i16", "$Vd, $SIMM", "",
3893 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3894 let Inst{9} = SIMM{9};
3897 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3898 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3899 "vmvn", "i16", "$Vd, $SIMM", "",
3900 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3901 let Inst{9} = SIMM{9};
3904 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3905 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3906 "vmvn", "i32", "$Vd, $SIMM", "",
3907 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3908 let Inst{11-8} = SIMM{11-8};
3911 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3912 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3913 "vmvn", "i32", "$Vd, $SIMM", "",
3914 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3915 let Inst{11-8} = SIMM{11-8};
3919 // VMVN : Vector Bitwise NOT
3920 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3921 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3922 "vmvn", "$Vd, $Vm", "",
3923 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3924 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3925 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3926 "vmvn", "$Vd, $Vm", "",
3927 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3928 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3929 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3931 // VBSL : Vector Bitwise Select
3932 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3933 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3934 N3RegFrm, IIC_VCNTiD,
3935 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3937 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
3939 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
3940 (and DPR:$Vm, (vnotd DPR:$Vd)))),
3941 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
3943 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3944 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3945 N3RegFrm, IIC_VCNTiQ,
3946 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3948 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
3950 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
3951 (and QPR:$Vm, (vnotq QPR:$Vd)))),
3952 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
3954 // VBIF : Vector Bitwise Insert if False
3955 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
3956 // FIXME: This instruction's encoding MAY NOT BE correct.
3957 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
3958 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3959 N3RegFrm, IIC_VBINiD,
3960 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3962 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
3963 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3964 N3RegFrm, IIC_VBINiQ,
3965 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3968 // VBIT : Vector Bitwise Insert if True
3969 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
3970 // FIXME: This instruction's encoding MAY NOT BE correct.
3971 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
3972 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3973 N3RegFrm, IIC_VBINiD,
3974 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3976 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
3977 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3978 N3RegFrm, IIC_VBINiQ,
3979 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3982 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
3983 // for equivalent operations with different register constraints; it just
3986 // Vector Absolute Differences.
3988 // VABD : Vector Absolute Difference
3989 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
3990 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3991 "vabd", "s", int_arm_neon_vabds, 1>;
3992 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
3993 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3994 "vabd", "u", int_arm_neon_vabdu, 1>;
3995 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
3996 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
3997 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
3998 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4000 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4001 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4002 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4003 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4004 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4006 // VABA : Vector Absolute Difference and Accumulate
4007 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4008 "vaba", "s", int_arm_neon_vabds, add>;
4009 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4010 "vaba", "u", int_arm_neon_vabdu, add>;
4012 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4013 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4014 "vabal", "s", int_arm_neon_vabds, zext, add>;
4015 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4016 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4018 // Vector Maximum and Minimum.
4020 // VMAX : Vector Maximum
4021 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4022 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4023 "vmax", "s", int_arm_neon_vmaxs, 1>;
4024 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4025 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4026 "vmax", "u", int_arm_neon_vmaxu, 1>;
4027 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4029 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4030 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4032 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4034 // VMIN : Vector Minimum
4035 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4036 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4037 "vmin", "s", int_arm_neon_vmins, 1>;
4038 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4039 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4040 "vmin", "u", int_arm_neon_vminu, 1>;
4041 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4043 v2f32, v2f32, int_arm_neon_vmins, 1>;
4044 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4046 v4f32, v4f32, int_arm_neon_vmins, 1>;
4048 // Vector Pairwise Operations.
4050 // VPADD : Vector Pairwise Add
4051 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4053 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4054 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4056 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4057 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4059 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4060 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4061 IIC_VPBIND, "vpadd", "f32",
4062 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4064 // VPADDL : Vector Pairwise Add Long
4065 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4066 int_arm_neon_vpaddls>;
4067 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4068 int_arm_neon_vpaddlu>;
4070 // VPADAL : Vector Pairwise Add and Accumulate Long
4071 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4072 int_arm_neon_vpadals>;
4073 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4074 int_arm_neon_vpadalu>;
4076 // VPMAX : Vector Pairwise Maximum
4077 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4078 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4079 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4080 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4081 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4082 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4083 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4084 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4085 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4086 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4087 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4088 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4089 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4090 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4092 // VPMIN : Vector Pairwise Minimum
4093 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4094 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4095 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4096 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4097 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4098 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4099 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4100 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4101 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4102 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4103 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4104 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4105 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4106 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4108 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4110 // VRECPE : Vector Reciprocal Estimate
4111 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4112 IIC_VUNAD, "vrecpe", "u32",
4113 v2i32, v2i32, int_arm_neon_vrecpe>;
4114 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4115 IIC_VUNAQ, "vrecpe", "u32",
4116 v4i32, v4i32, int_arm_neon_vrecpe>;
4117 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4118 IIC_VUNAD, "vrecpe", "f32",
4119 v2f32, v2f32, int_arm_neon_vrecpe>;
4120 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4121 IIC_VUNAQ, "vrecpe", "f32",
4122 v4f32, v4f32, int_arm_neon_vrecpe>;
4124 // VRECPS : Vector Reciprocal Step
4125 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4126 IIC_VRECSD, "vrecps", "f32",
4127 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4128 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4129 IIC_VRECSQ, "vrecps", "f32",
4130 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4132 // VRSQRTE : Vector Reciprocal Square Root Estimate
4133 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4134 IIC_VUNAD, "vrsqrte", "u32",
4135 v2i32, v2i32, int_arm_neon_vrsqrte>;
4136 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4137 IIC_VUNAQ, "vrsqrte", "u32",
4138 v4i32, v4i32, int_arm_neon_vrsqrte>;
4139 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4140 IIC_VUNAD, "vrsqrte", "f32",
4141 v2f32, v2f32, int_arm_neon_vrsqrte>;
4142 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4143 IIC_VUNAQ, "vrsqrte", "f32",
4144 v4f32, v4f32, int_arm_neon_vrsqrte>;
4146 // VRSQRTS : Vector Reciprocal Square Root Step
4147 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4148 IIC_VRECSD, "vrsqrts", "f32",
4149 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4150 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4151 IIC_VRECSQ, "vrsqrts", "f32",
4152 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4156 // VSHL : Vector Shift
4157 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4158 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4159 "vshl", "s", int_arm_neon_vshifts>;
4160 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4161 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4162 "vshl", "u", int_arm_neon_vshiftu>;
4164 // VSHL : Vector Shift Left (Immediate)
4165 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4167 // VSHR : Vector Shift Right (Immediate)
4168 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4169 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4171 // VSHLL : Vector Shift Left Long
4172 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4173 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4175 // VSHLL : Vector Shift Left Long (with maximum shift count)
4176 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4177 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4178 ValueType OpTy, SDNode OpNode>
4179 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4180 ResTy, OpTy, OpNode> {
4181 let Inst{21-16} = op21_16;
4182 let DecoderMethod = "DecodeVSHLMaxInstruction";
4184 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4185 v8i16, v8i8, NEONvshlli>;
4186 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4187 v4i32, v4i16, NEONvshlli>;
4188 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4189 v2i64, v2i32, NEONvshlli>;
4191 // VSHRN : Vector Shift Right and Narrow
4192 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4195 // VRSHL : Vector Rounding Shift
4196 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4197 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4198 "vrshl", "s", int_arm_neon_vrshifts>;
4199 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4200 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4201 "vrshl", "u", int_arm_neon_vrshiftu>;
4202 // VRSHR : Vector Rounding Shift Right
4203 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4204 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4206 // VRSHRN : Vector Rounding Shift Right and Narrow
4207 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4210 // VQSHL : Vector Saturating Shift
4211 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4212 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4213 "vqshl", "s", int_arm_neon_vqshifts>;
4214 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4215 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4216 "vqshl", "u", int_arm_neon_vqshiftu>;
4217 // VQSHL : Vector Saturating Shift Left (Immediate)
4218 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4219 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4221 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4222 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4224 // VQSHRN : Vector Saturating Shift Right and Narrow
4225 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4227 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4230 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4231 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4234 // VQRSHL : Vector Saturating Rounding Shift
4235 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4236 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4237 "vqrshl", "s", int_arm_neon_vqrshifts>;
4238 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4239 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4240 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4242 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4243 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4245 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4248 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4249 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4252 // VSRA : Vector Shift Right and Accumulate
4253 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4254 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4255 // VRSRA : Vector Rounding Shift Right and Accumulate
4256 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4257 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4259 // VSLI : Vector Shift Left and Insert
4260 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4262 // VSRI : Vector Shift Right and Insert
4263 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4265 // Vector Absolute and Saturating Absolute.
4267 // VABS : Vector Absolute Value
4268 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4269 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4271 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4272 IIC_VUNAD, "vabs", "f32",
4273 v2f32, v2f32, int_arm_neon_vabs>;
4274 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4275 IIC_VUNAQ, "vabs", "f32",
4276 v4f32, v4f32, int_arm_neon_vabs>;
4278 // VQABS : Vector Saturating Absolute Value
4279 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4280 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4281 int_arm_neon_vqabs>;
4285 def vnegd : PatFrag<(ops node:$in),
4286 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4287 def vnegq : PatFrag<(ops node:$in),
4288 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4290 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4291 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4292 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4293 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4294 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4295 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4296 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4297 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4299 // VNEG : Vector Negate (integer)
4300 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4301 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4302 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4303 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4304 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4305 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4307 // VNEG : Vector Negate (floating-point)
4308 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4309 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4310 "vneg", "f32", "$Vd, $Vm", "",
4311 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4312 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4313 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4314 "vneg", "f32", "$Vd, $Vm", "",
4315 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4317 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4318 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4319 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4320 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4321 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4322 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4324 // VQNEG : Vector Saturating Negate
4325 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4326 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4327 int_arm_neon_vqneg>;
4329 // Vector Bit Counting Operations.
4331 // VCLS : Vector Count Leading Sign Bits
4332 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4333 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4335 // VCLZ : Vector Count Leading Zeros
4336 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4337 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4339 // VCNT : Vector Count One Bits
4340 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4341 IIC_VCNTiD, "vcnt", "8",
4342 v8i8, v8i8, int_arm_neon_vcnt>;
4343 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4344 IIC_VCNTiQ, "vcnt", "8",
4345 v16i8, v16i8, int_arm_neon_vcnt>;
4348 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4349 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4350 "vswp", "$Vd, $Vm", "", []>;
4351 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4352 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4353 "vswp", "$Vd, $Vm", "", []>;
4355 // Vector Move Operations.
4357 // VMOV : Vector Move (Register)
4358 def : InstAlias<"vmov${p} $Vd, $Vm",
4359 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4360 def : InstAlias<"vmov${p} $Vd, $Vm",
4361 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4363 // VMOV : Vector Move (Immediate)
4365 let isReMaterializable = 1 in {
4366 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4367 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4368 "vmov", "i8", "$Vd, $SIMM", "",
4369 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4370 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4371 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4372 "vmov", "i8", "$Vd, $SIMM", "",
4373 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4375 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4376 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4377 "vmov", "i16", "$Vd, $SIMM", "",
4378 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4379 let Inst{9} = SIMM{9};
4382 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4383 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4384 "vmov", "i16", "$Vd, $SIMM", "",
4385 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4386 let Inst{9} = SIMM{9};
4389 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4390 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4391 "vmov", "i32", "$Vd, $SIMM", "",
4392 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4393 let Inst{11-8} = SIMM{11-8};
4396 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4397 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4398 "vmov", "i32", "$Vd, $SIMM", "",
4399 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4400 let Inst{11-8} = SIMM{11-8};
4403 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4404 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4405 "vmov", "i64", "$Vd, $SIMM", "",
4406 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4407 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4408 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4409 "vmov", "i64", "$Vd, $SIMM", "",
4410 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4411 } // isReMaterializable
4413 // VMOV : Vector Get Lane (move scalar to ARM core register)
4415 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4416 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4417 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4418 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4420 let Inst{21} = lane{2};
4421 let Inst{6-5} = lane{1-0};
4423 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4424 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4425 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4426 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4428 let Inst{21} = lane{1};
4429 let Inst{6} = lane{0};
4431 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4432 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4433 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4434 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4436 let Inst{21} = lane{2};
4437 let Inst{6-5} = lane{1-0};
4439 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4440 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4441 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4442 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4444 let Inst{21} = lane{1};
4445 let Inst{6} = lane{0};
4447 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4448 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4449 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4450 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4452 let Inst{21} = lane{0};
4454 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4455 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4456 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4457 (DSubReg_i8_reg imm:$lane))),
4458 (SubReg_i8_lane imm:$lane))>;
4459 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4460 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4461 (DSubReg_i16_reg imm:$lane))),
4462 (SubReg_i16_lane imm:$lane))>;
4463 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4464 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4465 (DSubReg_i8_reg imm:$lane))),
4466 (SubReg_i8_lane imm:$lane))>;
4467 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4468 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4469 (DSubReg_i16_reg imm:$lane))),
4470 (SubReg_i16_lane imm:$lane))>;
4471 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4472 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4473 (DSubReg_i32_reg imm:$lane))),
4474 (SubReg_i32_lane imm:$lane))>;
4475 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4476 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4477 (SSubReg_f32_reg imm:$src2))>;
4478 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4479 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4480 (SSubReg_f32_reg imm:$src2))>;
4481 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4482 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4483 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4484 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4487 // VMOV : Vector Set Lane (move ARM core register to scalar)
4489 let Constraints = "$src1 = $V" in {
4490 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4491 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4492 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4493 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4494 GPR:$R, imm:$lane))]> {
4495 let Inst{21} = lane{2};
4496 let Inst{6-5} = lane{1-0};
4498 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4499 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4500 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4501 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4502 GPR:$R, imm:$lane))]> {
4503 let Inst{21} = lane{1};
4504 let Inst{6} = lane{0};
4506 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4507 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4508 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4509 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4510 GPR:$R, imm:$lane))]> {
4511 let Inst{21} = lane{0};
4514 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4515 (v16i8 (INSERT_SUBREG QPR:$src1,
4516 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4517 (DSubReg_i8_reg imm:$lane))),
4518 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4519 (DSubReg_i8_reg imm:$lane)))>;
4520 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4521 (v8i16 (INSERT_SUBREG QPR:$src1,
4522 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4523 (DSubReg_i16_reg imm:$lane))),
4524 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4525 (DSubReg_i16_reg imm:$lane)))>;
4526 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4527 (v4i32 (INSERT_SUBREG QPR:$src1,
4528 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4529 (DSubReg_i32_reg imm:$lane))),
4530 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4531 (DSubReg_i32_reg imm:$lane)))>;
4533 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4534 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4535 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4536 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4537 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4538 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4540 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4541 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4542 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4543 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4545 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4546 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4547 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4548 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4549 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4550 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4552 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4553 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4554 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4555 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4556 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4557 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4559 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4560 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4561 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4563 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4564 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4565 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4567 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4568 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4569 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4572 // VDUP : Vector Duplicate (from ARM core register to all elements)
4574 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4575 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4576 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4577 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4578 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4579 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4580 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4581 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4583 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4584 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4585 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4586 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4587 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4588 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4590 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4591 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4593 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4595 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4596 ValueType Ty, Operand IdxTy>
4597 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4598 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4599 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4601 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4602 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4603 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4604 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4605 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4606 VectorIndex32:$lane)))]>;
4608 // Inst{19-16} is partially specified depending on the element size.
4610 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4612 let Inst{19-17} = lane{2-0};
4614 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4616 let Inst{19-18} = lane{1-0};
4618 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4620 let Inst{19} = lane{0};
4622 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4624 let Inst{19-17} = lane{2-0};
4626 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4628 let Inst{19-18} = lane{1-0};
4630 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4632 let Inst{19} = lane{0};
4635 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4636 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4638 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4639 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4641 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4642 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4643 (DSubReg_i8_reg imm:$lane))),
4644 (SubReg_i8_lane imm:$lane)))>;
4645 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4646 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4647 (DSubReg_i16_reg imm:$lane))),
4648 (SubReg_i16_lane imm:$lane)))>;
4649 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4650 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4651 (DSubReg_i32_reg imm:$lane))),
4652 (SubReg_i32_lane imm:$lane)))>;
4653 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4654 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4655 (DSubReg_i32_reg imm:$lane))),
4656 (SubReg_i32_lane imm:$lane)))>;
4658 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4659 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4660 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4661 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4663 // VMOVN : Vector Narrowing Move
4664 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4665 "vmovn", "i", trunc>;
4666 // VQMOVN : Vector Saturating Narrowing Move
4667 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4668 "vqmovn", "s", int_arm_neon_vqmovns>;
4669 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4670 "vqmovn", "u", int_arm_neon_vqmovnu>;
4671 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4672 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4673 // VMOVL : Vector Lengthening Move
4674 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4675 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4677 // Vector Conversions.
4679 // VCVT : Vector Convert Between Floating-Point and Integers
4680 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4681 v2i32, v2f32, fp_to_sint>;
4682 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4683 v2i32, v2f32, fp_to_uint>;
4684 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4685 v2f32, v2i32, sint_to_fp>;
4686 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4687 v2f32, v2i32, uint_to_fp>;
4689 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4690 v4i32, v4f32, fp_to_sint>;
4691 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4692 v4i32, v4f32, fp_to_uint>;
4693 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4694 v4f32, v4i32, sint_to_fp>;
4695 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4696 v4f32, v4i32, uint_to_fp>;
4698 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4699 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4700 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4701 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4702 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4703 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4704 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4705 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4706 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4708 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4709 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4710 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4711 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4712 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4713 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4714 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4715 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4717 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4718 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4719 IIC_VUNAQ, "vcvt", "f16.f32",
4720 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4721 Requires<[HasNEON, HasFP16]>;
4722 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4723 IIC_VUNAQ, "vcvt", "f32.f16",
4724 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4725 Requires<[HasNEON, HasFP16]>;
4729 // VREV64 : Vector Reverse elements within 64-bit doublewords
4731 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4732 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4733 (ins DPR:$Vm), IIC_VMOVD,
4734 OpcodeStr, Dt, "$Vd, $Vm", "",
4735 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4736 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4737 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4738 (ins QPR:$Vm), IIC_VMOVQ,
4739 OpcodeStr, Dt, "$Vd, $Vm", "",
4740 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4742 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4743 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4744 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4745 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4747 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4748 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4749 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4750 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4752 // VREV32 : Vector Reverse elements within 32-bit words
4754 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4755 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4756 (ins DPR:$Vm), IIC_VMOVD,
4757 OpcodeStr, Dt, "$Vd, $Vm", "",
4758 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4759 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4760 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4761 (ins QPR:$Vm), IIC_VMOVQ,
4762 OpcodeStr, Dt, "$Vd, $Vm", "",
4763 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4765 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4766 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4768 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4769 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4771 // VREV16 : Vector Reverse elements within 16-bit halfwords
4773 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4774 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4775 (ins DPR:$Vm), IIC_VMOVD,
4776 OpcodeStr, Dt, "$Vd, $Vm", "",
4777 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4778 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4779 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4780 (ins QPR:$Vm), IIC_VMOVQ,
4781 OpcodeStr, Dt, "$Vd, $Vm", "",
4782 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4784 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4785 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4787 // Other Vector Shuffles.
4789 // Aligned extractions: really just dropping registers
4791 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4792 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4793 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4795 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4797 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4799 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4801 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4803 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4806 // VEXT : Vector Extract
4808 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4809 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4810 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4811 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4812 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4813 (Ty DPR:$Vm), imm:$index)))]> {
4815 let Inst{11-8} = index{3-0};
4818 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4819 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4820 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4821 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4822 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4823 (Ty QPR:$Vm), imm:$index)))]> {
4825 let Inst{11-8} = index{3-0};
4828 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4829 let Inst{11-8} = index{3-0};
4831 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4832 let Inst{11-9} = index{2-0};
4835 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4836 let Inst{11-10} = index{1-0};
4837 let Inst{9-8} = 0b00;
4839 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4842 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
4844 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4845 let Inst{11-8} = index{3-0};
4847 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4848 let Inst{11-9} = index{2-0};
4851 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4852 let Inst{11-10} = index{1-0};
4853 let Inst{9-8} = 0b00;
4855 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
4858 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
4860 // VTRN : Vector Transpose
4862 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4863 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4864 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4866 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4867 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4868 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4870 // VUZP : Vector Unzip (Deinterleave)
4872 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4873 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4874 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4876 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4877 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4878 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4880 // VZIP : Vector Zip (Interleave)
4882 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4883 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4884 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4886 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4887 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4888 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4890 // Vector Table Lookup and Table Extension.
4892 // VTBL : Vector Table Lookup
4893 let DecoderMethod = "DecodeTBLInstruction" in {
4895 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4896 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4897 "vtbl", "8", "$Vd, $Vn, $Vm", "",
4898 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
4899 let hasExtraSrcRegAllocReq = 1 in {
4901 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4902 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4903 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4905 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4906 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4907 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4909 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4910 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4912 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4913 } // hasExtraSrcRegAllocReq = 1
4916 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4918 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4920 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4922 // VTBX : Vector Table Extension
4924 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4925 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4926 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
4927 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4928 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
4929 let hasExtraSrcRegAllocReq = 1 in {
4931 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4932 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4933 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4935 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4936 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4937 NVTBLFrm, IIC_VTBX3,
4938 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4941 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4942 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4943 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4945 } // hasExtraSrcRegAllocReq = 1
4948 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4949 IIC_VTBX2, "$orig = $dst", []>;
4951 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4952 IIC_VTBX3, "$orig = $dst", []>;
4954 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
4955 IIC_VTBX4, "$orig = $dst", []>;
4956 } // DecoderMethod = "DecodeTBLInstruction"
4958 //===----------------------------------------------------------------------===//
4959 // NEON instructions for single-precision FP math
4960 //===----------------------------------------------------------------------===//
4962 class N2VSPat<SDNode OpNode, NeonI Inst>
4963 : NEONFPPat<(f32 (OpNode SPR:$a)),
4965 (v2f32 (COPY_TO_REGCLASS (Inst
4967 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4968 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
4970 class N3VSPat<SDNode OpNode, NeonI Inst>
4971 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
4973 (v2f32 (COPY_TO_REGCLASS (Inst
4975 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4978 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4979 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4981 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
4982 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
4984 (v2f32 (COPY_TO_REGCLASS (Inst
4986 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4989 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4992 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
4993 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
4995 def : N3VSPat<fadd, VADDfd>;
4996 def : N3VSPat<fsub, VSUBfd>;
4997 def : N3VSPat<fmul, VMULfd>;
4998 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
4999 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5000 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5001 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5002 def : N2VSPat<fabs, VABSfd>;
5003 def : N2VSPat<fneg, VNEGfd>;
5004 def : N3VSPat<NEONfmax, VMAXfd>;
5005 def : N3VSPat<NEONfmin, VMINfd>;
5006 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5007 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5008 def : N2VSPat<arm_sitof, VCVTs2fd>;
5009 def : N2VSPat<arm_uitof, VCVTu2fd>;
5011 //===----------------------------------------------------------------------===//
5012 // Non-Instruction Patterns
5013 //===----------------------------------------------------------------------===//
5016 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5017 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5018 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5019 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5020 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5021 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5022 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5023 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5024 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5025 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5026 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5027 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5028 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5029 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5030 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5031 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5032 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5033 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5034 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5035 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5036 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5037 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5038 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5039 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5040 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5041 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5042 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5043 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5044 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5045 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5047 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5048 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5049 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5050 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5051 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5052 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5053 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5054 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5055 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5056 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5057 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5058 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5059 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5060 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5061 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5062 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5063 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5064 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5065 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5066 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5067 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5068 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5069 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5070 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5071 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5072 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5073 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5074 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5075 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5076 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;