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 nImmVMOVF32 : Operand<i32> {
43 let PrintMethod = "printFPImmOperand";
44 let ParserMatchClass = FPImmOperand;
46 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
47 def nImmSplatI64 : Operand<i32> {
48 let PrintMethod = "printNEONModImmOperand";
49 let ParserMatchClass = nImmSplatI64AsmOperand;
52 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
53 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
54 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
55 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
56 return ((uint64_t)Imm) < 8;
58 let ParserMatchClass = VectorIndex8Operand;
59 let PrintMethod = "printVectorIndex";
60 let MIOperandInfo = (ops i32imm);
62 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
63 return ((uint64_t)Imm) < 4;
65 let ParserMatchClass = VectorIndex16Operand;
66 let PrintMethod = "printVectorIndex";
67 let MIOperandInfo = (ops i32imm);
69 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
70 return ((uint64_t)Imm) < 2;
72 let ParserMatchClass = VectorIndex32Operand;
73 let PrintMethod = "printVectorIndex";
74 let MIOperandInfo = (ops i32imm);
77 // Register list of one D register.
78 def VecListOneDAsmOperand : AsmOperandClass {
79 let Name = "VecListOneD";
80 let ParserMethod = "parseVectorList";
81 let RenderMethod = "addVecListOperands";
83 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
84 let ParserMatchClass = VecListOneDAsmOperand;
86 // Register list of two sequential D registers.
87 def VecListTwoDAsmOperand : AsmOperandClass {
88 let Name = "VecListTwoD";
89 let ParserMethod = "parseVectorList";
90 let RenderMethod = "addVecListOperands";
92 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
93 let ParserMatchClass = VecListTwoDAsmOperand;
95 // Register list of three sequential D registers.
96 def VecListThreeDAsmOperand : AsmOperandClass {
97 let Name = "VecListThreeD";
98 let ParserMethod = "parseVectorList";
99 let RenderMethod = "addVecListOperands";
101 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
102 let ParserMatchClass = VecListThreeDAsmOperand;
104 // Register list of four sequential D registers.
105 def VecListFourDAsmOperand : AsmOperandClass {
106 let Name = "VecListFourD";
107 let ParserMethod = "parseVectorList";
108 let RenderMethod = "addVecListOperands";
110 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
111 let ParserMatchClass = VecListFourDAsmOperand;
113 // Register list of two D registers spaced by 2 (two sequential Q registers).
114 def VecListTwoQAsmOperand : AsmOperandClass {
115 let Name = "VecListTwoQ";
116 let ParserMethod = "parseVectorList";
117 let RenderMethod = "addVecListOperands";
119 def VecListTwoQ : RegisterOperand<DPR, "printVectorListTwo"> {
120 let ParserMatchClass = VecListTwoQAsmOperand;
123 // Register list of one D register, with "all lanes" subscripting.
124 def VecListOneDAllLanesAsmOperand : AsmOperandClass {
125 let Name = "VecListOneDAllLanes";
126 let ParserMethod = "parseVectorList";
127 let RenderMethod = "addVecListOperands";
129 def VecListOneDAllLanes : RegisterOperand<DPR, "printVectorListOneAllLanes"> {
130 let ParserMatchClass = VecListOneDAllLanesAsmOperand;
132 // Register list of two D registers, with "all lanes" subscripting.
133 def VecListTwoDAllLanesAsmOperand : AsmOperandClass {
134 let Name = "VecListTwoDAllLanes";
135 let ParserMethod = "parseVectorList";
136 let RenderMethod = "addVecListOperands";
138 def VecListTwoDAllLanes : RegisterOperand<DPR, "printVectorListTwoAllLanes"> {
139 let ParserMatchClass = VecListTwoDAllLanesAsmOperand;
142 // Register list of one D register, with byte lane subscripting.
143 def VecListOneDByteIndexAsmOperand : AsmOperandClass {
144 let Name = "VecListOneDByteIndexed";
145 let ParserMethod = "parseVectorList";
146 let RenderMethod = "addVecListIndexedOperands";
148 def VecListOneDByteIndexed : Operand<i32> {
149 let ParserMatchClass = VecListOneDByteIndexAsmOperand;
150 let MIOperandInfo = (ops DPR:$Vd, i32imm:$idx);
153 //===----------------------------------------------------------------------===//
154 // NEON-specific DAG Nodes.
155 //===----------------------------------------------------------------------===//
157 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
158 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
160 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
161 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
162 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
163 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
164 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
165 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
166 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
167 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
168 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
169 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
170 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
172 // Types for vector shift by immediates. The "SHX" version is for long and
173 // narrow operations where the source and destination vectors have different
174 // types. The "SHINS" version is for shift and insert operations.
175 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
177 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
179 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
180 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
182 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
183 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
184 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
185 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
186 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
187 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
188 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
190 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
191 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
192 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
194 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
195 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
196 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
197 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
198 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
199 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
201 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
202 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
203 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
205 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
206 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
208 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
210 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
211 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
213 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
214 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
215 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
216 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
218 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
220 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
221 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
223 def NEONvbsl : SDNode<"ARMISD::VBSL",
224 SDTypeProfile<1, 3, [SDTCisVec<0>,
227 SDTCisSameAs<0, 3>]>>;
229 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
231 // VDUPLANE can produce a quad-register result from a double-register source,
232 // so the result is not constrained to match the source.
233 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
234 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
237 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
238 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
239 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
241 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
242 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
243 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
244 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
246 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
248 SDTCisSameAs<0, 3>]>;
249 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
250 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
251 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
253 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
254 SDTCisSameAs<1, 2>]>;
255 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
256 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
258 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
259 SDTCisSameAs<0, 2>]>;
260 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
261 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
263 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
264 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
265 unsigned EltBits = 0;
266 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
267 return (EltBits == 32 && EltVal == 0);
270 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
271 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
272 unsigned EltBits = 0;
273 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
274 return (EltBits == 8 && EltVal == 0xff);
277 //===----------------------------------------------------------------------===//
278 // NEON load / store instructions
279 //===----------------------------------------------------------------------===//
281 // Use VLDM to load a Q register as a D register pair.
282 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
284 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
286 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
288 // Use VSTM to store a Q register as a D register pair.
289 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
291 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
293 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
295 // Classes for VLD* pseudo-instructions with multi-register operands.
296 // These are expanded to real instructions after register allocation.
297 class VLDQPseudo<InstrItinClass itin>
298 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
299 class VLDQWBPseudo<InstrItinClass itin>
300 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
301 (ins addrmode6:$addr, am6offset:$offset), itin,
303 class VLDQWBfixedPseudo<InstrItinClass itin>
304 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
305 (ins addrmode6:$addr), itin,
307 class VLDQWBregisterPseudo<InstrItinClass itin>
308 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
309 (ins addrmode6:$addr, rGPR:$offset), itin,
311 class VLDQQPseudo<InstrItinClass itin>
312 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
313 class VLDQQWBPseudo<InstrItinClass itin>
314 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
315 (ins addrmode6:$addr, am6offset:$offset), itin,
317 class VLDQQQQPseudo<InstrItinClass itin>
318 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
320 class VLDQQQQWBPseudo<InstrItinClass itin>
321 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
322 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
323 "$addr.addr = $wb, $src = $dst">;
325 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
327 // VLD1 : Vector Load (multiple single elements)
328 class VLD1D<bits<4> op7_4, string Dt>
329 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
330 (ins addrmode6:$Rn), IIC_VLD1,
331 "vld1", Dt, "$Vd, $Rn", "", []> {
334 let DecoderMethod = "DecodeVLDInstruction";
336 class VLD1Q<bits<4> op7_4, string Dt>
337 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
338 (ins addrmode6:$Rn), IIC_VLD1x2,
339 "vld1", Dt, "$Vd, $Rn", "", []> {
341 let Inst{5-4} = Rn{5-4};
342 let DecoderMethod = "DecodeVLDInstruction";
345 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
346 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
347 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
348 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
350 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
351 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
352 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
353 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
355 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
356 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
357 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
358 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
360 // ...with address register writeback:
361 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
362 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
363 (ins addrmode6:$Rn), IIC_VLD1u,
364 "vld1", Dt, "$Vd, $Rn!",
365 "$Rn.addr = $wb", []> {
366 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
368 let DecoderMethod = "DecodeVLDInstruction";
369 let AsmMatchConverter = "cvtVLDwbFixed";
371 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
372 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
373 "vld1", Dt, "$Vd, $Rn, $Rm",
374 "$Rn.addr = $wb", []> {
376 let DecoderMethod = "DecodeVLDInstruction";
377 let AsmMatchConverter = "cvtVLDwbRegister";
380 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
381 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
382 (ins addrmode6:$Rn), IIC_VLD1x2u,
383 "vld1", Dt, "$Vd, $Rn!",
384 "$Rn.addr = $wb", []> {
385 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
386 let Inst{5-4} = Rn{5-4};
387 let DecoderMethod = "DecodeVLDInstruction";
388 let AsmMatchConverter = "cvtVLDwbFixed";
390 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
391 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
392 "vld1", Dt, "$Vd, $Rn, $Rm",
393 "$Rn.addr = $wb", []> {
394 let Inst{5-4} = Rn{5-4};
395 let DecoderMethod = "DecodeVLDInstruction";
396 let AsmMatchConverter = "cvtVLDwbRegister";
400 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
401 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
402 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
403 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
404 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
405 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
406 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
407 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
409 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
410 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
411 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
412 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
413 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
414 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
415 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
416 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
418 // ...with 3 registers
419 class VLD1D3<bits<4> op7_4, string Dt>
420 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
421 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
422 "$Vd, $Rn", "", []> {
425 let DecoderMethod = "DecodeVLDInstruction";
427 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
428 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
429 (ins addrmode6:$Rn), IIC_VLD1x2u,
430 "vld1", Dt, "$Vd, $Rn!",
431 "$Rn.addr = $wb", []> {
432 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
434 let DecoderMethod = "DecodeVLDInstruction";
435 let AsmMatchConverter = "cvtVLDwbFixed";
437 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
438 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
439 "vld1", Dt, "$Vd, $Rn, $Rm",
440 "$Rn.addr = $wb", []> {
442 let DecoderMethod = "DecodeVLDInstruction";
443 let AsmMatchConverter = "cvtVLDwbRegister";
447 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
448 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
449 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
450 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
452 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
453 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
454 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
455 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
457 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
459 // ...with 4 registers
460 class VLD1D4<bits<4> op7_4, string Dt>
461 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
462 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
463 "$Vd, $Rn", "", []> {
465 let Inst{5-4} = Rn{5-4};
466 let DecoderMethod = "DecodeVLDInstruction";
468 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
469 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
470 (ins addrmode6:$Rn), IIC_VLD1x2u,
471 "vld1", Dt, "$Vd, $Rn!",
472 "$Rn.addr = $wb", []> {
473 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
474 let Inst{5-4} = Rn{5-4};
475 let DecoderMethod = "DecodeVLDInstruction";
476 let AsmMatchConverter = "cvtVLDwbFixed";
478 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
479 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
480 "vld1", Dt, "$Vd, $Rn, $Rm",
481 "$Rn.addr = $wb", []> {
482 let Inst{5-4} = Rn{5-4};
483 let DecoderMethod = "DecodeVLDInstruction";
484 let AsmMatchConverter = "cvtVLDwbRegister";
488 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
489 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
490 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
491 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
493 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
494 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
495 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
496 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
498 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
500 // VLD2 : Vector Load (multiple 2-element structures)
501 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
502 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
503 (ins addrmode6:$Rn), IIC_VLD2,
504 "vld2", Dt, "$Vd, $Rn", "", []> {
506 let Inst{5-4} = Rn{5-4};
507 let DecoderMethod = "DecodeVLDInstruction";
509 class VLD2Q<bits<4> op7_4, string Dt, RegisterOperand VdTy>
510 : NLdSt<0, 0b10, 0b0011, op7_4,
512 (ins addrmode6:$Rn), IIC_VLD2x2,
513 "vld2", Dt, "$Vd, $Rn", "", []> {
515 let Inst{5-4} = Rn{5-4};
516 let DecoderMethod = "DecodeVLDInstruction";
519 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
520 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
521 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
523 def VLD2q8 : VLD2Q<{0,0,?,?}, "8", VecListFourD>;
524 def VLD2q16 : VLD2Q<{0,1,?,?}, "16", VecListFourD>;
525 def VLD2q32 : VLD2Q<{1,0,?,?}, "32", VecListFourD>;
527 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
528 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
529 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
531 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
532 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
533 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
535 // ...with address register writeback:
536 class VLD2WB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy,
538 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
539 (ins addrmode6:$Rn, am6offset:$Rm), itin,
540 "vld2", Dt, "$Vd, $Rn$Rm",
541 "$Rn.addr = $wb", []> {
542 let Inst{5-4} = Rn{5-4};
543 let DecoderMethod = "DecodeVLDInstruction";
546 def VLD2d8_UPD : VLD2WB<0b1000, {0,0,?,?}, "8", VecListTwoD, IIC_VLD2u>;
547 def VLD2d16_UPD : VLD2WB<0b1000, {0,1,?,?}, "16", VecListTwoD, IIC_VLD2u>;
548 def VLD2d32_UPD : VLD2WB<0b1000, {1,0,?,?}, "32", VecListTwoD, IIC_VLD2u>;
550 def VLD2q8_UPD : VLD2WB<0b0011, {0,0,?,?}, "8", VecListFourD, IIC_VLD2x2u>;
551 def VLD2q16_UPD : VLD2WB<0b0011, {0,1,?,?}, "16", VecListFourD, IIC_VLD2x2u>;
552 def VLD2q32_UPD : VLD2WB<0b0011, {1,0,?,?}, "32", VecListFourD, IIC_VLD2x2u>;
554 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
555 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
556 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
558 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
559 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
560 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
562 // ...with double-spaced registers
563 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
564 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
565 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
566 def VLD2b8_UPD : VLD2WB<0b1001, {0,0,?,?}, "8", VecListTwoQ, IIC_VLD2u>;
567 def VLD2b16_UPD : VLD2WB<0b1001, {0,1,?,?}, "16", VecListTwoQ, IIC_VLD2u>;
568 def VLD2b32_UPD : VLD2WB<0b1001, {1,0,?,?}, "32", VecListTwoQ, IIC_VLD2u>;
570 // VLD3 : Vector Load (multiple 3-element structures)
571 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
572 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
573 (ins addrmode6:$Rn), IIC_VLD3,
574 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
577 let DecoderMethod = "DecodeVLDInstruction";
580 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
581 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
582 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
584 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
585 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
586 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
588 // ...with address register writeback:
589 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
590 : NLdSt<0, 0b10, op11_8, op7_4,
591 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
592 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
593 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
594 "$Rn.addr = $wb", []> {
596 let DecoderMethod = "DecodeVLDInstruction";
599 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
600 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
601 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
603 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
604 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
605 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
607 // ...with double-spaced registers:
608 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
609 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
610 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
611 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
612 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
613 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
615 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
616 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
617 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
619 // ...alternate versions to be allocated odd register numbers:
620 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
621 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
622 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
624 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
625 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
626 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
628 // VLD4 : Vector Load (multiple 4-element structures)
629 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
630 : NLdSt<0, 0b10, op11_8, op7_4,
631 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
632 (ins addrmode6:$Rn), IIC_VLD4,
633 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
635 let Inst{5-4} = Rn{5-4};
636 let DecoderMethod = "DecodeVLDInstruction";
639 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
640 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
641 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
643 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
644 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
645 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
647 // ...with address register writeback:
648 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
649 : NLdSt<0, 0b10, op11_8, op7_4,
650 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
651 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
652 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
653 "$Rn.addr = $wb", []> {
654 let Inst{5-4} = Rn{5-4};
655 let DecoderMethod = "DecodeVLDInstruction";
658 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
659 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
660 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
662 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
663 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
664 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
666 // ...with double-spaced registers:
667 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
668 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
669 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
670 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
671 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
672 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
674 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
675 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
676 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
678 // ...alternate versions to be allocated odd register numbers:
679 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
680 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
681 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
683 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
684 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
685 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
687 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
689 // Classes for VLD*LN pseudo-instructions with multi-register operands.
690 // These are expanded to real instructions after register allocation.
691 class VLDQLNPseudo<InstrItinClass itin>
692 : PseudoNLdSt<(outs QPR:$dst),
693 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
694 itin, "$src = $dst">;
695 class VLDQLNWBPseudo<InstrItinClass itin>
696 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
697 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
698 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
699 class VLDQQLNPseudo<InstrItinClass itin>
700 : PseudoNLdSt<(outs QQPR:$dst),
701 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
702 itin, "$src = $dst">;
703 class VLDQQLNWBPseudo<InstrItinClass itin>
704 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
705 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
706 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
707 class VLDQQQQLNPseudo<InstrItinClass itin>
708 : PseudoNLdSt<(outs QQQQPR:$dst),
709 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
710 itin, "$src = $dst">;
711 class VLDQQQQLNWBPseudo<InstrItinClass itin>
712 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
713 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
714 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
716 // VLD1LN : Vector Load (single element to one lane)
717 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
719 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
720 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
721 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
723 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
724 (i32 (LoadOp addrmode6:$Rn)),
727 let DecoderMethod = "DecodeVLD1LN";
729 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
731 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
732 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
733 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
735 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
736 (i32 (LoadOp addrmode6oneL32:$Rn)),
739 let DecoderMethod = "DecodeVLD1LN";
741 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
742 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
743 (i32 (LoadOp addrmode6:$addr)),
747 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
748 let Inst{7-5} = lane{2-0};
750 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
751 let Inst{7-6} = lane{1-0};
754 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
755 let Inst{7} = lane{0};
760 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
761 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
762 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
764 def : Pat<(vector_insert (v2f32 DPR:$src),
765 (f32 (load addrmode6:$addr)), imm:$lane),
766 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
767 def : Pat<(vector_insert (v4f32 QPR:$src),
768 (f32 (load addrmode6:$addr)), imm:$lane),
769 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
771 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
773 // ...with address register writeback:
774 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
775 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
776 (ins addrmode6:$Rn, am6offset:$Rm,
777 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
778 "\\{$Vd[$lane]\\}, $Rn$Rm",
779 "$src = $Vd, $Rn.addr = $wb", []> {
780 let DecoderMethod = "DecodeVLD1LN";
783 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
784 let Inst{7-5} = lane{2-0};
786 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
787 let Inst{7-6} = lane{1-0};
790 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
791 let Inst{7} = lane{0};
796 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
797 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
798 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
800 // VLD2LN : Vector Load (single 2-element structure to one lane)
801 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
802 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
803 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
804 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
805 "$src1 = $Vd, $src2 = $dst2", []> {
808 let DecoderMethod = "DecodeVLD2LN";
811 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
812 let Inst{7-5} = lane{2-0};
814 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
815 let Inst{7-6} = lane{1-0};
817 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
818 let Inst{7} = lane{0};
821 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
822 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
823 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
825 // ...with double-spaced registers:
826 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
827 let Inst{7-6} = lane{1-0};
829 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
830 let Inst{7} = lane{0};
833 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
834 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
836 // ...with address register writeback:
837 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
838 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
839 (ins addrmode6:$Rn, am6offset:$Rm,
840 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
841 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
842 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
844 let DecoderMethod = "DecodeVLD2LN";
847 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
848 let Inst{7-5} = lane{2-0};
850 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
851 let Inst{7-6} = lane{1-0};
853 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
854 let Inst{7} = lane{0};
857 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
858 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
859 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
861 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
862 let Inst{7-6} = lane{1-0};
864 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
865 let Inst{7} = lane{0};
868 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
869 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
871 // VLD3LN : Vector Load (single 3-element structure to one lane)
872 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
873 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
874 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
875 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
876 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
877 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
879 let DecoderMethod = "DecodeVLD3LN";
882 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
883 let Inst{7-5} = lane{2-0};
885 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
886 let Inst{7-6} = lane{1-0};
888 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
889 let Inst{7} = lane{0};
892 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
893 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
894 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
896 // ...with double-spaced registers:
897 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
898 let Inst{7-6} = lane{1-0};
900 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
901 let Inst{7} = lane{0};
904 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
905 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
907 // ...with address register writeback:
908 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
909 : NLdStLn<1, 0b10, op11_8, op7_4,
910 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
911 (ins addrmode6:$Rn, am6offset:$Rm,
912 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
913 IIC_VLD3lnu, "vld3", Dt,
914 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
915 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
917 let DecoderMethod = "DecodeVLD3LN";
920 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
921 let Inst{7-5} = lane{2-0};
923 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
924 let Inst{7-6} = lane{1-0};
926 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
927 let Inst{7} = lane{0};
930 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
931 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
932 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
934 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
935 let Inst{7-6} = lane{1-0};
937 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
938 let Inst{7} = lane{0};
941 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
942 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
944 // VLD4LN : Vector Load (single 4-element structure to one lane)
945 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
946 : NLdStLn<1, 0b10, op11_8, op7_4,
947 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
948 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
949 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
950 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
951 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
954 let DecoderMethod = "DecodeVLD4LN";
957 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
958 let Inst{7-5} = lane{2-0};
960 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
961 let Inst{7-6} = lane{1-0};
963 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
964 let Inst{7} = lane{0};
968 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
969 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
970 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
972 // ...with double-spaced registers:
973 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
974 let Inst{7-6} = lane{1-0};
976 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
977 let Inst{7} = lane{0};
981 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
982 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
984 // ...with address register writeback:
985 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
986 : NLdStLn<1, 0b10, op11_8, op7_4,
987 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
988 (ins addrmode6:$Rn, am6offset:$Rm,
989 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
990 IIC_VLD4lnu, "vld4", Dt,
991 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
992 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
995 let DecoderMethod = "DecodeVLD4LN" ;
998 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
999 let Inst{7-5} = lane{2-0};
1001 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1002 let Inst{7-6} = lane{1-0};
1004 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1005 let Inst{7} = lane{0};
1006 let Inst{5} = Rn{5};
1009 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1010 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1011 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1013 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1014 let Inst{7-6} = lane{1-0};
1016 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1017 let Inst{7} = lane{0};
1018 let Inst{5} = Rn{5};
1021 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1022 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1024 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1026 // VLD1DUP : Vector Load (single element to all lanes)
1027 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1028 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1029 (ins addrmode6dup:$Rn),
1030 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1031 [(set VecListOneDAllLanes:$Vd,
1032 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1034 let Inst{4} = Rn{4};
1035 let DecoderMethod = "DecodeVLD1DupInstruction";
1037 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1038 let Pattern = [(set QPR:$dst,
1039 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1042 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1043 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1044 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1046 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1047 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1048 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1050 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1051 (VLD1DUPd32 addrmode6:$addr)>;
1052 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1053 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1055 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1057 class VLD1QDUP<bits<4> op7_4, string Dt>
1058 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1059 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1060 "vld1", Dt, "$Vd, $Rn", "", []> {
1062 let Inst{4} = Rn{4};
1063 let DecoderMethod = "DecodeVLD1DupInstruction";
1066 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1067 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1068 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1070 // ...with address register writeback:
1071 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1072 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1073 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1074 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1075 "vld1", Dt, "$Vd, $Rn!",
1076 "$Rn.addr = $wb", []> {
1077 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1078 let Inst{4} = Rn{4};
1079 let DecoderMethod = "DecodeVLD1DupInstruction";
1080 let AsmMatchConverter = "cvtVLDwbFixed";
1082 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1083 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1084 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1085 "vld1", Dt, "$Vd, $Rn, $Rm",
1086 "$Rn.addr = $wb", []> {
1087 let Inst{4} = Rn{4};
1088 let DecoderMethod = "DecodeVLD1DupInstruction";
1089 let AsmMatchConverter = "cvtVLDwbRegister";
1092 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1093 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1094 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1095 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1096 "vld1", Dt, "$Vd, $Rn!",
1097 "$Rn.addr = $wb", []> {
1098 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1099 let Inst{4} = Rn{4};
1100 let DecoderMethod = "DecodeVLD1DupInstruction";
1101 let AsmMatchConverter = "cvtVLDwbFixed";
1103 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1104 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1105 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1106 "vld1", Dt, "$Vd, $Rn, $Rm",
1107 "$Rn.addr = $wb", []> {
1108 let Inst{4} = Rn{4};
1109 let DecoderMethod = "DecodeVLD1DupInstruction";
1110 let AsmMatchConverter = "cvtVLDwbRegister";
1114 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1115 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1116 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1118 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1119 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1120 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1122 def VLD1DUPq8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1123 def VLD1DUPq16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1124 def VLD1DUPq32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1125 def VLD1DUPq8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1126 def VLD1DUPq16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1127 def VLD1DUPq32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1129 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1130 class VLD2DUP<bits<4> op7_4, string Dt>
1131 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1132 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1133 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1135 let Inst{4} = Rn{4};
1136 let DecoderMethod = "DecodeVLD2DupInstruction";
1139 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1140 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1141 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1143 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1144 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1145 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1147 // ...with double-spaced registers (not used for codegen):
1148 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1149 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1150 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1152 // ...with address register writeback:
1153 class VLD2DUPWB<bits<4> op7_4, string Dt>
1154 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1155 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1156 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1157 let Inst{4} = Rn{4};
1158 let DecoderMethod = "DecodeVLD2DupInstruction";
1161 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1162 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1163 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1165 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1166 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1167 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1169 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1170 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1171 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1173 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1174 class VLD3DUP<bits<4> op7_4, string Dt>
1175 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1176 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1177 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1180 let DecoderMethod = "DecodeVLD3DupInstruction";
1183 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1184 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1185 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1187 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1188 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1189 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1191 // ...with double-spaced registers (not used for codegen):
1192 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1193 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1194 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1196 // ...with address register writeback:
1197 class VLD3DUPWB<bits<4> op7_4, string Dt>
1198 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1199 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1200 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1201 "$Rn.addr = $wb", []> {
1203 let DecoderMethod = "DecodeVLD3DupInstruction";
1206 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1207 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1208 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1210 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1211 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1212 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1214 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1215 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1216 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1218 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1219 class VLD4DUP<bits<4> op7_4, string Dt>
1220 : NLdSt<1, 0b10, 0b1111, op7_4,
1221 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1222 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1223 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1225 let Inst{4} = Rn{4};
1226 let DecoderMethod = "DecodeVLD4DupInstruction";
1229 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1230 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1231 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1233 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1234 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1235 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1237 // ...with double-spaced registers (not used for codegen):
1238 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1239 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1240 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1242 // ...with address register writeback:
1243 class VLD4DUPWB<bits<4> op7_4, string Dt>
1244 : NLdSt<1, 0b10, 0b1111, op7_4,
1245 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1246 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1247 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1248 "$Rn.addr = $wb", []> {
1249 let Inst{4} = Rn{4};
1250 let DecoderMethod = "DecodeVLD4DupInstruction";
1253 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1254 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1255 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1257 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1258 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1259 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1261 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1262 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1263 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1265 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1267 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1269 // Classes for VST* pseudo-instructions with multi-register operands.
1270 // These are expanded to real instructions after register allocation.
1271 class VSTQPseudo<InstrItinClass itin>
1272 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1273 class VSTQWBPseudo<InstrItinClass itin>
1274 : PseudoNLdSt<(outs GPR:$wb),
1275 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1276 "$addr.addr = $wb">;
1277 class VSTQWBfixedPseudo<InstrItinClass itin>
1278 : PseudoNLdSt<(outs GPR:$wb),
1279 (ins addrmode6:$addr, QPR:$src), itin,
1280 "$addr.addr = $wb">;
1281 class VSTQWBregisterPseudo<InstrItinClass itin>
1282 : PseudoNLdSt<(outs GPR:$wb),
1283 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1284 "$addr.addr = $wb">;
1285 class VSTQQPseudo<InstrItinClass itin>
1286 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1287 class VSTQQWBPseudo<InstrItinClass itin>
1288 : PseudoNLdSt<(outs GPR:$wb),
1289 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1290 "$addr.addr = $wb">;
1291 class VSTQQQQPseudo<InstrItinClass itin>
1292 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1293 class VSTQQQQWBPseudo<InstrItinClass itin>
1294 : PseudoNLdSt<(outs GPR:$wb),
1295 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1296 "$addr.addr = $wb">;
1298 // VST1 : Vector Store (multiple single elements)
1299 class VST1D<bits<4> op7_4, string Dt>
1300 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1301 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1303 let Inst{4} = Rn{4};
1304 let DecoderMethod = "DecodeVSTInstruction";
1306 class VST1Q<bits<4> op7_4, string Dt>
1307 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1308 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1310 let Inst{5-4} = Rn{5-4};
1311 let DecoderMethod = "DecodeVSTInstruction";
1314 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1315 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1316 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1317 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1319 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1320 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1321 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1322 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1324 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1325 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1326 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1327 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1329 // ...with address register writeback:
1330 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1331 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1332 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1333 "vst1", Dt, "$Vd, $Rn!",
1334 "$Rn.addr = $wb", []> {
1335 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1336 let Inst{4} = Rn{4};
1337 let DecoderMethod = "DecodeVSTInstruction";
1338 let AsmMatchConverter = "cvtVSTwbFixed";
1340 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1341 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1343 "vst1", Dt, "$Vd, $Rn, $Rm",
1344 "$Rn.addr = $wb", []> {
1345 let Inst{4} = Rn{4};
1346 let DecoderMethod = "DecodeVSTInstruction";
1347 let AsmMatchConverter = "cvtVSTwbRegister";
1350 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1351 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1352 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1353 "vst1", Dt, "$Vd, $Rn!",
1354 "$Rn.addr = $wb", []> {
1355 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1356 let Inst{5-4} = Rn{5-4};
1357 let DecoderMethod = "DecodeVSTInstruction";
1358 let AsmMatchConverter = "cvtVSTwbFixed";
1360 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1361 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1363 "vst1", Dt, "$Vd, $Rn, $Rm",
1364 "$Rn.addr = $wb", []> {
1365 let Inst{5-4} = Rn{5-4};
1366 let DecoderMethod = "DecodeVSTInstruction";
1367 let AsmMatchConverter = "cvtVSTwbRegister";
1371 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1372 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1373 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1374 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1376 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1377 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1378 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1379 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1381 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1382 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1383 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1384 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1385 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1386 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1387 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1388 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1390 // ...with 3 registers
1391 class VST1D3<bits<4> op7_4, string Dt>
1392 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1393 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1394 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1396 let Inst{4} = Rn{4};
1397 let DecoderMethod = "DecodeVSTInstruction";
1399 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1400 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1401 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1402 "vst1", Dt, "$Vd, $Rn!",
1403 "$Rn.addr = $wb", []> {
1404 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1405 let Inst{5-4} = Rn{5-4};
1406 let DecoderMethod = "DecodeVSTInstruction";
1407 let AsmMatchConverter = "cvtVSTwbFixed";
1409 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1410 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1412 "vst1", Dt, "$Vd, $Rn, $Rm",
1413 "$Rn.addr = $wb", []> {
1414 let Inst{5-4} = Rn{5-4};
1415 let DecoderMethod = "DecodeVSTInstruction";
1416 let AsmMatchConverter = "cvtVSTwbRegister";
1420 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1421 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1422 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1423 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1425 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1426 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1427 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1428 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1430 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1431 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1432 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1434 // ...with 4 registers
1435 class VST1D4<bits<4> op7_4, string Dt>
1436 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1437 (ins addrmode6:$Rn, VecListFourD:$Vd),
1438 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1441 let Inst{5-4} = Rn{5-4};
1442 let DecoderMethod = "DecodeVSTInstruction";
1444 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1445 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1446 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1447 "vst1", Dt, "$Vd, $Rn!",
1448 "$Rn.addr = $wb", []> {
1449 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1450 let Inst{5-4} = Rn{5-4};
1451 let DecoderMethod = "DecodeVSTInstruction";
1452 let AsmMatchConverter = "cvtVSTwbFixed";
1454 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1455 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1457 "vst1", Dt, "$Vd, $Rn, $Rm",
1458 "$Rn.addr = $wb", []> {
1459 let Inst{5-4} = Rn{5-4};
1460 let DecoderMethod = "DecodeVSTInstruction";
1461 let AsmMatchConverter = "cvtVSTwbRegister";
1465 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1466 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1467 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1468 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1470 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1471 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1472 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1473 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1475 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1476 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1477 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1479 // VST2 : Vector Store (multiple 2-element structures)
1480 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1481 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1482 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1483 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1485 let Inst{5-4} = Rn{5-4};
1486 let DecoderMethod = "DecodeVSTInstruction";
1488 class VST2Q<bits<4> op7_4, string Dt>
1489 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1490 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1491 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1494 let Inst{5-4} = Rn{5-4};
1495 let DecoderMethod = "DecodeVSTInstruction";
1498 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1499 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1500 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1502 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1503 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1504 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1506 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1507 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1508 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1510 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1511 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1512 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1514 // ...with address register writeback:
1515 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1516 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1517 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1518 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1519 "$Rn.addr = $wb", []> {
1520 let Inst{5-4} = Rn{5-4};
1521 let DecoderMethod = "DecodeVSTInstruction";
1523 class VST2QWB<bits<4> op7_4, string Dt>
1524 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1525 (ins addrmode6:$Rn, am6offset:$Rm,
1526 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1527 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1528 "$Rn.addr = $wb", []> {
1529 let Inst{5-4} = Rn{5-4};
1530 let DecoderMethod = "DecodeVSTInstruction";
1533 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1534 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1535 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1537 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1538 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1539 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1541 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1542 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1543 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1545 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1546 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1547 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1549 // ...with double-spaced registers
1550 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1551 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1552 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1553 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1554 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1555 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1557 // VST3 : Vector Store (multiple 3-element structures)
1558 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1559 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1560 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1561 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1563 let Inst{4} = Rn{4};
1564 let DecoderMethod = "DecodeVSTInstruction";
1567 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1568 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1569 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1571 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1572 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1573 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1575 // ...with address register writeback:
1576 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1577 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1578 (ins addrmode6:$Rn, am6offset:$Rm,
1579 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1580 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1581 "$Rn.addr = $wb", []> {
1582 let Inst{4} = Rn{4};
1583 let DecoderMethod = "DecodeVSTInstruction";
1586 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1587 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1588 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1590 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1591 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1592 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1594 // ...with double-spaced registers:
1595 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1596 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1597 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1598 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1599 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1600 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1602 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1603 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1604 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1606 // ...alternate versions to be allocated odd register numbers:
1607 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1608 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1609 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1611 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1612 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1613 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1615 // VST4 : Vector Store (multiple 4-element structures)
1616 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1617 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1618 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1619 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1622 let Inst{5-4} = Rn{5-4};
1623 let DecoderMethod = "DecodeVSTInstruction";
1626 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1627 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1628 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1630 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1631 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1632 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1634 // ...with address register writeback:
1635 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1636 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1637 (ins addrmode6:$Rn, am6offset:$Rm,
1638 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1639 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1640 "$Rn.addr = $wb", []> {
1641 let Inst{5-4} = Rn{5-4};
1642 let DecoderMethod = "DecodeVSTInstruction";
1645 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1646 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1647 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1649 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1650 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1651 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1653 // ...with double-spaced registers:
1654 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1655 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1656 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1657 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1658 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1659 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1661 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1662 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1663 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1665 // ...alternate versions to be allocated odd register numbers:
1666 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1667 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1668 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1670 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1671 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1672 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1674 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1676 // Classes for VST*LN pseudo-instructions with multi-register operands.
1677 // These are expanded to real instructions after register allocation.
1678 class VSTQLNPseudo<InstrItinClass itin>
1679 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1681 class VSTQLNWBPseudo<InstrItinClass itin>
1682 : PseudoNLdSt<(outs GPR:$wb),
1683 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1684 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1685 class VSTQQLNPseudo<InstrItinClass itin>
1686 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1688 class VSTQQLNWBPseudo<InstrItinClass itin>
1689 : PseudoNLdSt<(outs GPR:$wb),
1690 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1691 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1692 class VSTQQQQLNPseudo<InstrItinClass itin>
1693 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1695 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1696 : PseudoNLdSt<(outs GPR:$wb),
1697 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1698 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1700 // VST1LN : Vector Store (single element from one lane)
1701 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1702 PatFrag StoreOp, SDNode ExtractOp>
1703 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1704 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1705 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1706 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1708 let DecoderMethod = "DecodeVST1LN";
1710 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1711 PatFrag StoreOp, SDNode ExtractOp>
1712 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1713 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1714 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1715 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1717 let DecoderMethod = "DecodeVST1LN";
1719 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1720 : VSTQLNPseudo<IIC_VST1ln> {
1721 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1725 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1727 let Inst{7-5} = lane{2-0};
1729 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1731 let Inst{7-6} = lane{1-0};
1732 let Inst{4} = Rn{5};
1735 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1736 let Inst{7} = lane{0};
1737 let Inst{5-4} = Rn{5-4};
1740 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1741 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1742 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1744 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1745 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1746 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1747 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1749 // ...with address register writeback:
1750 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1751 PatFrag StoreOp, SDNode ExtractOp>
1752 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1753 (ins addrmode6:$Rn, am6offset:$Rm,
1754 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1755 "\\{$Vd[$lane]\\}, $Rn$Rm",
1757 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1758 addrmode6:$Rn, am6offset:$Rm))]> {
1759 let DecoderMethod = "DecodeVST1LN";
1761 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1762 : VSTQLNWBPseudo<IIC_VST1lnu> {
1763 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1764 addrmode6:$addr, am6offset:$offset))];
1767 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1769 let Inst{7-5} = lane{2-0};
1771 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1773 let Inst{7-6} = lane{1-0};
1774 let Inst{4} = Rn{5};
1776 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1778 let Inst{7} = lane{0};
1779 let Inst{5-4} = Rn{5-4};
1782 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1783 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1784 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1786 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1788 // VST2LN : Vector Store (single 2-element structure from one lane)
1789 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1790 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1791 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1792 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1795 let Inst{4} = Rn{4};
1796 let DecoderMethod = "DecodeVST2LN";
1799 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1800 let Inst{7-5} = lane{2-0};
1802 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1803 let Inst{7-6} = lane{1-0};
1805 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1806 let Inst{7} = lane{0};
1809 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1810 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1811 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1813 // ...with double-spaced registers:
1814 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1815 let Inst{7-6} = lane{1-0};
1816 let Inst{4} = Rn{4};
1818 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1819 let Inst{7} = lane{0};
1820 let Inst{4} = Rn{4};
1823 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1824 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1826 // ...with address register writeback:
1827 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1828 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1829 (ins addrmode6:$addr, am6offset:$offset,
1830 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1831 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1832 "$addr.addr = $wb", []> {
1833 let Inst{4} = Rn{4};
1834 let DecoderMethod = "DecodeVST2LN";
1837 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1838 let Inst{7-5} = lane{2-0};
1840 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1841 let Inst{7-6} = lane{1-0};
1843 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1844 let Inst{7} = lane{0};
1847 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1848 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1849 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1851 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1852 let Inst{7-6} = lane{1-0};
1854 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1855 let Inst{7} = lane{0};
1858 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1859 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1861 // VST3LN : Vector Store (single 3-element structure from one lane)
1862 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1863 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1864 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1865 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1866 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1868 let DecoderMethod = "DecodeVST3LN";
1871 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1872 let Inst{7-5} = lane{2-0};
1874 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1875 let Inst{7-6} = lane{1-0};
1877 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1878 let Inst{7} = lane{0};
1881 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1882 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1883 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1885 // ...with double-spaced registers:
1886 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1887 let Inst{7-6} = lane{1-0};
1889 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1890 let Inst{7} = lane{0};
1893 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1894 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1896 // ...with address register writeback:
1897 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1898 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1899 (ins addrmode6:$Rn, am6offset:$Rm,
1900 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1901 IIC_VST3lnu, "vst3", Dt,
1902 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1903 "$Rn.addr = $wb", []> {
1904 let DecoderMethod = "DecodeVST3LN";
1907 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1908 let Inst{7-5} = lane{2-0};
1910 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1911 let Inst{7-6} = lane{1-0};
1913 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1914 let Inst{7} = lane{0};
1917 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1918 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1919 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1921 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1922 let Inst{7-6} = lane{1-0};
1924 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1925 let Inst{7} = lane{0};
1928 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1929 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1931 // VST4LN : Vector Store (single 4-element structure from one lane)
1932 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1933 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1934 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1935 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1936 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1939 let Inst{4} = Rn{4};
1940 let DecoderMethod = "DecodeVST4LN";
1943 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1944 let Inst{7-5} = lane{2-0};
1946 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1947 let Inst{7-6} = lane{1-0};
1949 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1950 let Inst{7} = lane{0};
1951 let Inst{5} = Rn{5};
1954 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1955 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1956 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1958 // ...with double-spaced registers:
1959 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1960 let Inst{7-6} = lane{1-0};
1962 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1963 let Inst{7} = lane{0};
1964 let Inst{5} = Rn{5};
1967 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1968 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1970 // ...with address register writeback:
1971 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1972 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1973 (ins addrmode6:$Rn, am6offset:$Rm,
1974 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1975 IIC_VST4lnu, "vst4", Dt,
1976 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1977 "$Rn.addr = $wb", []> {
1978 let Inst{4} = Rn{4};
1979 let DecoderMethod = "DecodeVST4LN";
1982 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1983 let Inst{7-5} = lane{2-0};
1985 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1986 let Inst{7-6} = lane{1-0};
1988 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1989 let Inst{7} = lane{0};
1990 let Inst{5} = Rn{5};
1993 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1994 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1995 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1997 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1998 let Inst{7-6} = lane{1-0};
2000 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2001 let Inst{7} = lane{0};
2002 let Inst{5} = Rn{5};
2005 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2006 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2008 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2011 //===----------------------------------------------------------------------===//
2012 // NEON pattern fragments
2013 //===----------------------------------------------------------------------===//
2015 // Extract D sub-registers of Q registers.
2016 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2017 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2018 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2020 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2021 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2022 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2024 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2025 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2026 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2028 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2029 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2030 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2033 // Extract S sub-registers of Q/D registers.
2034 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2035 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2036 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2039 // Translate lane numbers from Q registers to D subregs.
2040 def SubReg_i8_lane : SDNodeXForm<imm, [{
2041 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2043 def SubReg_i16_lane : SDNodeXForm<imm, [{
2044 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2046 def SubReg_i32_lane : SDNodeXForm<imm, [{
2047 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2050 //===----------------------------------------------------------------------===//
2051 // Instruction Classes
2052 //===----------------------------------------------------------------------===//
2054 // Basic 2-register operations: double- and quad-register.
2055 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2056 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2057 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2058 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2059 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2060 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2061 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2062 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2063 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2064 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2065 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2066 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2068 // Basic 2-register intrinsics, both double- and quad-register.
2069 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2070 bits<2> op17_16, bits<5> op11_7, bit op4,
2071 InstrItinClass itin, string OpcodeStr, string Dt,
2072 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2073 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2074 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2075 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2076 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2077 bits<2> op17_16, bits<5> op11_7, bit op4,
2078 InstrItinClass itin, string OpcodeStr, string Dt,
2079 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2080 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2081 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2082 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2084 // Narrow 2-register operations.
2085 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2086 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2087 InstrItinClass itin, string OpcodeStr, string Dt,
2088 ValueType TyD, ValueType TyQ, SDNode OpNode>
2089 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2090 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2091 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2093 // Narrow 2-register intrinsics.
2094 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2095 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2096 InstrItinClass itin, string OpcodeStr, string Dt,
2097 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2098 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2099 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2100 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2102 // Long 2-register operations (currently only used for VMOVL).
2103 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2104 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2105 InstrItinClass itin, string OpcodeStr, string Dt,
2106 ValueType TyQ, ValueType TyD, SDNode OpNode>
2107 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2108 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2109 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2111 // Long 2-register intrinsics.
2112 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2113 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2114 InstrItinClass itin, string OpcodeStr, string Dt,
2115 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2116 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2117 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2118 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2120 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2121 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2122 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2123 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2124 OpcodeStr, Dt, "$Vd, $Vm",
2125 "$src1 = $Vd, $src2 = $Vm", []>;
2126 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2127 InstrItinClass itin, string OpcodeStr, string Dt>
2128 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2129 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2130 "$src1 = $Vd, $src2 = $Vm", []>;
2132 // Basic 3-register operations: double- and quad-register.
2133 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2134 InstrItinClass itin, string OpcodeStr, string Dt,
2135 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2136 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2137 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2138 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2139 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2140 let isCommutable = Commutable;
2142 // Same as N3VD but no data type.
2143 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2144 InstrItinClass itin, string OpcodeStr,
2145 ValueType ResTy, ValueType OpTy,
2146 SDNode OpNode, bit Commutable>
2147 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2148 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2149 OpcodeStr, "$Vd, $Vn, $Vm", "",
2150 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2151 let isCommutable = Commutable;
2154 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2155 InstrItinClass itin, string OpcodeStr, string Dt,
2156 ValueType Ty, SDNode ShOp>
2157 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2158 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2159 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2161 (Ty (ShOp (Ty DPR:$Vn),
2162 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2163 let isCommutable = 0;
2165 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2166 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2167 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2168 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2169 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2171 (Ty (ShOp (Ty DPR:$Vn),
2172 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2173 let isCommutable = 0;
2176 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2177 InstrItinClass itin, string OpcodeStr, string Dt,
2178 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2179 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2180 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2181 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2182 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2183 let isCommutable = Commutable;
2185 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2186 InstrItinClass itin, string OpcodeStr,
2187 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2188 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2189 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2190 OpcodeStr, "$Vd, $Vn, $Vm", "",
2191 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2192 let isCommutable = Commutable;
2194 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2195 InstrItinClass itin, string OpcodeStr, string Dt,
2196 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2197 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2198 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2199 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2200 [(set (ResTy QPR:$Vd),
2201 (ResTy (ShOp (ResTy QPR:$Vn),
2202 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2204 let isCommutable = 0;
2206 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2207 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2208 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2209 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2210 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2211 [(set (ResTy QPR:$Vd),
2212 (ResTy (ShOp (ResTy QPR:$Vn),
2213 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2215 let isCommutable = 0;
2218 // Basic 3-register intrinsics, both double- and quad-register.
2219 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2220 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2221 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2222 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2223 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2224 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2225 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2226 let isCommutable = Commutable;
2228 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2229 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2230 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2231 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2232 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2234 (Ty (IntOp (Ty DPR:$Vn),
2235 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2237 let isCommutable = 0;
2239 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2240 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2241 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2242 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2243 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2245 (Ty (IntOp (Ty DPR:$Vn),
2246 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2247 let isCommutable = 0;
2249 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2250 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2251 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2252 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2253 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2254 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2255 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2256 let isCommutable = 0;
2259 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2260 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2261 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2262 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2263 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2264 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2265 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2266 let isCommutable = Commutable;
2268 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2269 string OpcodeStr, string Dt,
2270 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2271 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2272 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2273 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2274 [(set (ResTy QPR:$Vd),
2275 (ResTy (IntOp (ResTy QPR:$Vn),
2276 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2278 let isCommutable = 0;
2280 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2281 string OpcodeStr, string Dt,
2282 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2283 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2284 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2285 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2286 [(set (ResTy QPR:$Vd),
2287 (ResTy (IntOp (ResTy QPR:$Vn),
2288 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2290 let isCommutable = 0;
2292 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2293 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2294 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2295 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2296 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2297 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2298 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2299 let isCommutable = 0;
2302 // Multiply-Add/Sub operations: double- and quad-register.
2303 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2304 InstrItinClass itin, string OpcodeStr, string Dt,
2305 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2306 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2307 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2308 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2309 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2310 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2312 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2313 string OpcodeStr, string Dt,
2314 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2315 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2317 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2319 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2321 (Ty (ShOp (Ty DPR:$src1),
2323 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2325 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2326 string OpcodeStr, string Dt,
2327 ValueType Ty, SDNode MulOp, SDNode ShOp>
2328 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2330 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2332 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2334 (Ty (ShOp (Ty DPR:$src1),
2336 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2339 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2340 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2341 SDPatternOperator MulOp, SDPatternOperator OpNode>
2342 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2343 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2344 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2345 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2346 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2347 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2348 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2349 SDPatternOperator MulOp, SDPatternOperator ShOp>
2350 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2352 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2354 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2355 [(set (ResTy QPR:$Vd),
2356 (ResTy (ShOp (ResTy QPR:$src1),
2357 (ResTy (MulOp QPR:$Vn,
2358 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2360 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2361 string OpcodeStr, string Dt,
2362 ValueType ResTy, ValueType OpTy,
2363 SDNode MulOp, SDNode ShOp>
2364 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2366 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2368 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2369 [(set (ResTy QPR:$Vd),
2370 (ResTy (ShOp (ResTy QPR:$src1),
2371 (ResTy (MulOp QPR:$Vn,
2372 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2375 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2376 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2377 InstrItinClass itin, string OpcodeStr, string Dt,
2378 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2379 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2380 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2381 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2382 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2383 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2384 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2385 InstrItinClass itin, string OpcodeStr, string Dt,
2386 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2387 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2388 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2389 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2390 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2391 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2393 // Neon 3-argument intrinsics, both double- and quad-register.
2394 // The destination register is also used as the first source operand register.
2395 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2396 InstrItinClass itin, string OpcodeStr, string Dt,
2397 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2398 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2399 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2400 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2401 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2402 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2403 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2404 InstrItinClass itin, string OpcodeStr, string Dt,
2405 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2406 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2407 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2408 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2409 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2410 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2412 // Long Multiply-Add/Sub operations.
2413 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2414 InstrItinClass itin, string OpcodeStr, string Dt,
2415 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2416 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2417 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2418 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2419 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2420 (TyQ (MulOp (TyD DPR:$Vn),
2421 (TyD DPR:$Vm)))))]>;
2422 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2423 InstrItinClass itin, string OpcodeStr, string Dt,
2424 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2425 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2426 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2428 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2430 (OpNode (TyQ QPR:$src1),
2431 (TyQ (MulOp (TyD DPR:$Vn),
2432 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2434 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2435 InstrItinClass itin, string OpcodeStr, string Dt,
2436 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2437 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2438 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2440 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2442 (OpNode (TyQ QPR:$src1),
2443 (TyQ (MulOp (TyD DPR:$Vn),
2444 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2447 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2448 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2449 InstrItinClass itin, string OpcodeStr, string Dt,
2450 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2452 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2453 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2454 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2455 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2456 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2457 (TyD DPR:$Vm)))))))]>;
2459 // Neon Long 3-argument intrinsic. The destination register is
2460 // a quad-register and is also used as the first source operand register.
2461 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2462 InstrItinClass itin, string OpcodeStr, string Dt,
2463 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2464 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2465 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2466 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2468 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2469 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2470 string OpcodeStr, string Dt,
2471 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2472 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2474 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2476 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2477 [(set (ResTy QPR:$Vd),
2478 (ResTy (IntOp (ResTy QPR:$src1),
2480 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2482 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2483 InstrItinClass itin, string OpcodeStr, string Dt,
2484 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2485 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2487 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2489 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2490 [(set (ResTy QPR:$Vd),
2491 (ResTy (IntOp (ResTy QPR:$src1),
2493 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2496 // Narrowing 3-register intrinsics.
2497 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2498 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2499 Intrinsic IntOp, bit Commutable>
2500 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2501 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2502 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2503 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2504 let isCommutable = Commutable;
2507 // Long 3-register operations.
2508 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2509 InstrItinClass itin, string OpcodeStr, string Dt,
2510 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2511 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2512 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2513 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2514 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2515 let isCommutable = Commutable;
2517 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2518 InstrItinClass itin, string OpcodeStr, string Dt,
2519 ValueType TyQ, ValueType TyD, SDNode OpNode>
2520 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2521 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2522 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2524 (TyQ (OpNode (TyD DPR:$Vn),
2525 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2526 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2527 InstrItinClass itin, string OpcodeStr, string Dt,
2528 ValueType TyQ, ValueType TyD, SDNode OpNode>
2529 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2530 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2531 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2533 (TyQ (OpNode (TyD DPR:$Vn),
2534 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2536 // Long 3-register operations with explicitly extended operands.
2537 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2538 InstrItinClass itin, string OpcodeStr, string Dt,
2539 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2541 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2542 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2543 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2544 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2545 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2546 let isCommutable = Commutable;
2549 // Long 3-register intrinsics with explicit extend (VABDL).
2550 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2551 InstrItinClass itin, string OpcodeStr, string Dt,
2552 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2554 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2555 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2556 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2557 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2558 (TyD DPR:$Vm))))))]> {
2559 let isCommutable = Commutable;
2562 // Long 3-register intrinsics.
2563 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2564 InstrItinClass itin, string OpcodeStr, string Dt,
2565 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2566 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2567 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2568 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2569 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2570 let isCommutable = Commutable;
2572 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2573 string OpcodeStr, string Dt,
2574 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2575 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2576 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2577 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2578 [(set (ResTy QPR:$Vd),
2579 (ResTy (IntOp (OpTy DPR:$Vn),
2580 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2582 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2583 InstrItinClass itin, string OpcodeStr, string Dt,
2584 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2585 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2586 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2587 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2588 [(set (ResTy QPR:$Vd),
2589 (ResTy (IntOp (OpTy DPR:$Vn),
2590 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2593 // Wide 3-register operations.
2594 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2595 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2596 SDNode OpNode, SDNode ExtOp, bit Commutable>
2597 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2598 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2599 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2600 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2601 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2602 let isCommutable = Commutable;
2605 // Pairwise long 2-register intrinsics, both double- and quad-register.
2606 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2607 bits<2> op17_16, bits<5> op11_7, bit op4,
2608 string OpcodeStr, string Dt,
2609 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2610 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2611 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2612 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2613 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2614 bits<2> op17_16, bits<5> op11_7, bit op4,
2615 string OpcodeStr, string Dt,
2616 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2617 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2618 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2619 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2621 // Pairwise long 2-register accumulate intrinsics,
2622 // both double- and quad-register.
2623 // The destination register is also used as the first source operand register.
2624 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2625 bits<2> op17_16, bits<5> op11_7, bit op4,
2626 string OpcodeStr, string Dt,
2627 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2628 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2629 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2630 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2631 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2632 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2633 bits<2> op17_16, bits<5> op11_7, bit op4,
2634 string OpcodeStr, string Dt,
2635 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2636 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2637 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2638 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2639 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2641 // Shift by immediate,
2642 // both double- and quad-register.
2643 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2644 Format f, InstrItinClass itin, Operand ImmTy,
2645 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2646 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2647 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2648 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2649 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2650 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2651 Format f, InstrItinClass itin, Operand ImmTy,
2652 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2653 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2654 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2655 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2656 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2658 // Long shift by immediate.
2659 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2660 string OpcodeStr, string Dt,
2661 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2662 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2663 (outs QPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), N2RegVShLFrm,
2664 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2665 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2666 (i32 imm:$SIMM))))]>;
2668 // Narrow shift by immediate.
2669 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2670 InstrItinClass itin, string OpcodeStr, string Dt,
2671 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2672 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2673 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2674 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2675 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2676 (i32 imm:$SIMM))))]>;
2678 // Shift right by immediate and accumulate,
2679 // both double- and quad-register.
2680 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2681 Operand ImmTy, string OpcodeStr, string Dt,
2682 ValueType Ty, SDNode ShOp>
2683 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2684 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2685 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2686 [(set DPR:$Vd, (Ty (add DPR:$src1,
2687 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2688 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2689 Operand ImmTy, string OpcodeStr, string Dt,
2690 ValueType Ty, SDNode ShOp>
2691 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2692 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2693 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2694 [(set QPR:$Vd, (Ty (add QPR:$src1,
2695 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2697 // Shift by immediate and insert,
2698 // both double- and quad-register.
2699 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2700 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2701 ValueType Ty,SDNode ShOp>
2702 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2703 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2704 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2705 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2706 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2707 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2708 ValueType Ty,SDNode ShOp>
2709 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2710 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2711 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2712 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2714 // Convert, with fractional bits immediate,
2715 // both double- and quad-register.
2716 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2717 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2719 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2720 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2721 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2722 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2723 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2724 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2726 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2727 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2728 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2729 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2731 //===----------------------------------------------------------------------===//
2733 //===----------------------------------------------------------------------===//
2735 // Abbreviations used in multiclass suffixes:
2736 // Q = quarter int (8 bit) elements
2737 // H = half int (16 bit) elements
2738 // S = single int (32 bit) elements
2739 // D = double int (64 bit) elements
2741 // Neon 2-register vector operations and intrinsics.
2743 // Neon 2-register comparisons.
2744 // source operand element sizes of 8, 16 and 32 bits:
2745 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2746 bits<5> op11_7, bit op4, string opc, string Dt,
2747 string asm, SDNode OpNode> {
2748 // 64-bit vector types.
2749 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2750 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2751 opc, !strconcat(Dt, "8"), asm, "",
2752 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2753 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2754 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2755 opc, !strconcat(Dt, "16"), asm, "",
2756 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2757 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2758 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2759 opc, !strconcat(Dt, "32"), asm, "",
2760 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2761 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2762 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2763 opc, "f32", asm, "",
2764 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2765 let Inst{10} = 1; // overwrite F = 1
2768 // 128-bit vector types.
2769 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2770 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2771 opc, !strconcat(Dt, "8"), asm, "",
2772 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2773 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2774 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2775 opc, !strconcat(Dt, "16"), asm, "",
2776 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2777 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2778 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2779 opc, !strconcat(Dt, "32"), asm, "",
2780 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2781 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2782 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2783 opc, "f32", asm, "",
2784 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2785 let Inst{10} = 1; // overwrite F = 1
2790 // Neon 2-register vector intrinsics,
2791 // element sizes of 8, 16 and 32 bits:
2792 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2793 bits<5> op11_7, bit op4,
2794 InstrItinClass itinD, InstrItinClass itinQ,
2795 string OpcodeStr, string Dt, Intrinsic IntOp> {
2796 // 64-bit vector types.
2797 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2798 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2799 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2800 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2801 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2802 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2804 // 128-bit vector types.
2805 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2806 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2807 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2808 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2809 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2810 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2814 // Neon Narrowing 2-register vector operations,
2815 // source operand element sizes of 16, 32 and 64 bits:
2816 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2817 bits<5> op11_7, bit op6, bit op4,
2818 InstrItinClass itin, string OpcodeStr, string Dt,
2820 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2821 itin, OpcodeStr, !strconcat(Dt, "16"),
2822 v8i8, v8i16, OpNode>;
2823 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2824 itin, OpcodeStr, !strconcat(Dt, "32"),
2825 v4i16, v4i32, OpNode>;
2826 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2827 itin, OpcodeStr, !strconcat(Dt, "64"),
2828 v2i32, v2i64, OpNode>;
2831 // Neon Narrowing 2-register vector intrinsics,
2832 // source operand element sizes of 16, 32 and 64 bits:
2833 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2834 bits<5> op11_7, bit op6, bit op4,
2835 InstrItinClass itin, string OpcodeStr, string Dt,
2837 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2838 itin, OpcodeStr, !strconcat(Dt, "16"),
2839 v8i8, v8i16, IntOp>;
2840 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2841 itin, OpcodeStr, !strconcat(Dt, "32"),
2842 v4i16, v4i32, IntOp>;
2843 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2844 itin, OpcodeStr, !strconcat(Dt, "64"),
2845 v2i32, v2i64, IntOp>;
2849 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2850 // source operand element sizes of 16, 32 and 64 bits:
2851 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2852 string OpcodeStr, string Dt, SDNode OpNode> {
2853 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2854 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2855 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2856 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2857 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2858 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2862 // Neon 3-register vector operations.
2864 // First with only element sizes of 8, 16 and 32 bits:
2865 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2866 InstrItinClass itinD16, InstrItinClass itinD32,
2867 InstrItinClass itinQ16, InstrItinClass itinQ32,
2868 string OpcodeStr, string Dt,
2869 SDNode OpNode, bit Commutable = 0> {
2870 // 64-bit vector types.
2871 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2872 OpcodeStr, !strconcat(Dt, "8"),
2873 v8i8, v8i8, OpNode, Commutable>;
2874 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2875 OpcodeStr, !strconcat(Dt, "16"),
2876 v4i16, v4i16, OpNode, Commutable>;
2877 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2878 OpcodeStr, !strconcat(Dt, "32"),
2879 v2i32, v2i32, OpNode, Commutable>;
2881 // 128-bit vector types.
2882 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2883 OpcodeStr, !strconcat(Dt, "8"),
2884 v16i8, v16i8, OpNode, Commutable>;
2885 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2886 OpcodeStr, !strconcat(Dt, "16"),
2887 v8i16, v8i16, OpNode, Commutable>;
2888 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2889 OpcodeStr, !strconcat(Dt, "32"),
2890 v4i32, v4i32, OpNode, Commutable>;
2893 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, SDNode ShOp> {
2894 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, "i16", v4i16, ShOp>;
2895 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, "i32", v2i32, ShOp>;
2896 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, "i16", v8i16, v4i16, ShOp>;
2897 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, "i32",
2898 v4i32, v2i32, ShOp>;
2901 // ....then also with element size 64 bits:
2902 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2903 InstrItinClass itinD, InstrItinClass itinQ,
2904 string OpcodeStr, string Dt,
2905 SDNode OpNode, bit Commutable = 0>
2906 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2907 OpcodeStr, Dt, OpNode, Commutable> {
2908 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2909 OpcodeStr, !strconcat(Dt, "64"),
2910 v1i64, v1i64, OpNode, Commutable>;
2911 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2912 OpcodeStr, !strconcat(Dt, "64"),
2913 v2i64, v2i64, OpNode, Commutable>;
2917 // Neon 3-register vector intrinsics.
2919 // First with only element sizes of 16 and 32 bits:
2920 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2921 InstrItinClass itinD16, InstrItinClass itinD32,
2922 InstrItinClass itinQ16, InstrItinClass itinQ32,
2923 string OpcodeStr, string Dt,
2924 Intrinsic IntOp, bit Commutable = 0> {
2925 // 64-bit vector types.
2926 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2927 OpcodeStr, !strconcat(Dt, "16"),
2928 v4i16, v4i16, IntOp, Commutable>;
2929 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2930 OpcodeStr, !strconcat(Dt, "32"),
2931 v2i32, v2i32, IntOp, Commutable>;
2933 // 128-bit vector types.
2934 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2935 OpcodeStr, !strconcat(Dt, "16"),
2936 v8i16, v8i16, IntOp, Commutable>;
2937 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2938 OpcodeStr, !strconcat(Dt, "32"),
2939 v4i32, v4i32, IntOp, Commutable>;
2941 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2942 InstrItinClass itinD16, InstrItinClass itinD32,
2943 InstrItinClass itinQ16, InstrItinClass itinQ32,
2944 string OpcodeStr, string Dt,
2946 // 64-bit vector types.
2947 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2948 OpcodeStr, !strconcat(Dt, "16"),
2949 v4i16, v4i16, IntOp>;
2950 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2951 OpcodeStr, !strconcat(Dt, "32"),
2952 v2i32, v2i32, IntOp>;
2954 // 128-bit vector types.
2955 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2956 OpcodeStr, !strconcat(Dt, "16"),
2957 v8i16, v8i16, IntOp>;
2958 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2959 OpcodeStr, !strconcat(Dt, "32"),
2960 v4i32, v4i32, IntOp>;
2963 multiclass N3VIntSL_HS<bits<4> op11_8,
2964 InstrItinClass itinD16, InstrItinClass itinD32,
2965 InstrItinClass itinQ16, InstrItinClass itinQ32,
2966 string OpcodeStr, string Dt, Intrinsic IntOp> {
2967 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2968 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2969 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2970 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2971 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2972 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2973 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2974 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2977 // ....then also with element size of 8 bits:
2978 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2979 InstrItinClass itinD16, InstrItinClass itinD32,
2980 InstrItinClass itinQ16, InstrItinClass itinQ32,
2981 string OpcodeStr, string Dt,
2982 Intrinsic IntOp, bit Commutable = 0>
2983 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2984 OpcodeStr, Dt, IntOp, Commutable> {
2985 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2986 OpcodeStr, !strconcat(Dt, "8"),
2987 v8i8, v8i8, IntOp, Commutable>;
2988 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2989 OpcodeStr, !strconcat(Dt, "8"),
2990 v16i8, v16i8, IntOp, Commutable>;
2992 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2993 InstrItinClass itinD16, InstrItinClass itinD32,
2994 InstrItinClass itinQ16, InstrItinClass itinQ32,
2995 string OpcodeStr, string Dt,
2997 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2998 OpcodeStr, Dt, IntOp> {
2999 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3000 OpcodeStr, !strconcat(Dt, "8"),
3002 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3003 OpcodeStr, !strconcat(Dt, "8"),
3004 v16i8, v16i8, IntOp>;
3008 // ....then also with element size of 64 bits:
3009 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3010 InstrItinClass itinD16, InstrItinClass itinD32,
3011 InstrItinClass itinQ16, InstrItinClass itinQ32,
3012 string OpcodeStr, string Dt,
3013 Intrinsic IntOp, bit Commutable = 0>
3014 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3015 OpcodeStr, Dt, IntOp, Commutable> {
3016 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3017 OpcodeStr, !strconcat(Dt, "64"),
3018 v1i64, v1i64, IntOp, Commutable>;
3019 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3020 OpcodeStr, !strconcat(Dt, "64"),
3021 v2i64, v2i64, IntOp, Commutable>;
3023 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3024 InstrItinClass itinD16, InstrItinClass itinD32,
3025 InstrItinClass itinQ16, InstrItinClass itinQ32,
3026 string OpcodeStr, string Dt,
3028 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3029 OpcodeStr, Dt, IntOp> {
3030 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3031 OpcodeStr, !strconcat(Dt, "64"),
3032 v1i64, v1i64, IntOp>;
3033 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3034 OpcodeStr, !strconcat(Dt, "64"),
3035 v2i64, v2i64, IntOp>;
3038 // Neon Narrowing 3-register vector intrinsics,
3039 // source operand element sizes of 16, 32 and 64 bits:
3040 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3041 string OpcodeStr, string Dt,
3042 Intrinsic IntOp, bit Commutable = 0> {
3043 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3044 OpcodeStr, !strconcat(Dt, "16"),
3045 v8i8, v8i16, IntOp, Commutable>;
3046 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3047 OpcodeStr, !strconcat(Dt, "32"),
3048 v4i16, v4i32, IntOp, Commutable>;
3049 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3050 OpcodeStr, !strconcat(Dt, "64"),
3051 v2i32, v2i64, IntOp, Commutable>;
3055 // Neon Long 3-register vector operations.
3057 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3058 InstrItinClass itin16, InstrItinClass itin32,
3059 string OpcodeStr, string Dt,
3060 SDNode OpNode, bit Commutable = 0> {
3061 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3062 OpcodeStr, !strconcat(Dt, "8"),
3063 v8i16, v8i8, OpNode, Commutable>;
3064 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3065 OpcodeStr, !strconcat(Dt, "16"),
3066 v4i32, v4i16, OpNode, Commutable>;
3067 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3068 OpcodeStr, !strconcat(Dt, "32"),
3069 v2i64, v2i32, OpNode, Commutable>;
3072 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3073 InstrItinClass itin, string OpcodeStr, string Dt,
3075 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3076 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3077 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3078 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3081 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3082 InstrItinClass itin16, InstrItinClass itin32,
3083 string OpcodeStr, string Dt,
3084 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3085 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3086 OpcodeStr, !strconcat(Dt, "8"),
3087 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3088 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3089 OpcodeStr, !strconcat(Dt, "16"),
3090 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3091 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3092 OpcodeStr, !strconcat(Dt, "32"),
3093 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3096 // Neon Long 3-register vector intrinsics.
3098 // First with only element sizes of 16 and 32 bits:
3099 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3100 InstrItinClass itin16, InstrItinClass itin32,
3101 string OpcodeStr, string Dt,
3102 Intrinsic IntOp, bit Commutable = 0> {
3103 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3104 OpcodeStr, !strconcat(Dt, "16"),
3105 v4i32, v4i16, IntOp, Commutable>;
3106 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3107 OpcodeStr, !strconcat(Dt, "32"),
3108 v2i64, v2i32, IntOp, Commutable>;
3111 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3112 InstrItinClass itin, string OpcodeStr, string Dt,
3114 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3115 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3116 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3117 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3120 // ....then also with element size of 8 bits:
3121 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3122 InstrItinClass itin16, InstrItinClass itin32,
3123 string OpcodeStr, string Dt,
3124 Intrinsic IntOp, bit Commutable = 0>
3125 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3126 IntOp, Commutable> {
3127 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3128 OpcodeStr, !strconcat(Dt, "8"),
3129 v8i16, v8i8, IntOp, Commutable>;
3132 // ....with explicit extend (VABDL).
3133 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3134 InstrItinClass itin, string OpcodeStr, string Dt,
3135 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3136 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3137 OpcodeStr, !strconcat(Dt, "8"),
3138 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3139 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3140 OpcodeStr, !strconcat(Dt, "16"),
3141 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3142 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3143 OpcodeStr, !strconcat(Dt, "32"),
3144 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3148 // Neon Wide 3-register vector intrinsics,
3149 // source operand element sizes of 8, 16 and 32 bits:
3150 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3151 string OpcodeStr, string Dt,
3152 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3153 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3154 OpcodeStr, !strconcat(Dt, "8"),
3155 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3156 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3157 OpcodeStr, !strconcat(Dt, "16"),
3158 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3159 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3160 OpcodeStr, !strconcat(Dt, "32"),
3161 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3165 // Neon Multiply-Op vector operations,
3166 // element sizes of 8, 16 and 32 bits:
3167 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3168 InstrItinClass itinD16, InstrItinClass itinD32,
3169 InstrItinClass itinQ16, InstrItinClass itinQ32,
3170 string OpcodeStr, string Dt, SDNode OpNode> {
3171 // 64-bit vector types.
3172 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3173 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3174 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3175 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3176 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3177 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3179 // 128-bit vector types.
3180 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3181 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3182 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3183 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3184 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3185 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3188 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3189 InstrItinClass itinD16, InstrItinClass itinD32,
3190 InstrItinClass itinQ16, InstrItinClass itinQ32,
3191 string OpcodeStr, string Dt, SDNode ShOp> {
3192 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3193 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3194 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3195 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3196 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3197 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3199 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3200 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3204 // Neon Intrinsic-Op vector operations,
3205 // element sizes of 8, 16 and 32 bits:
3206 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3207 InstrItinClass itinD, InstrItinClass itinQ,
3208 string OpcodeStr, string Dt, Intrinsic IntOp,
3210 // 64-bit vector types.
3211 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3212 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3213 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3214 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3215 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3216 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3218 // 128-bit vector types.
3219 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3220 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3221 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3222 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3223 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3224 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3227 // Neon 3-argument intrinsics,
3228 // element sizes of 8, 16 and 32 bits:
3229 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3230 InstrItinClass itinD, InstrItinClass itinQ,
3231 string OpcodeStr, string Dt, Intrinsic IntOp> {
3232 // 64-bit vector types.
3233 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3234 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3235 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3236 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3237 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3238 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3240 // 128-bit vector types.
3241 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3242 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3243 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3244 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3245 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3246 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3250 // Neon Long Multiply-Op vector operations,
3251 // element sizes of 8, 16 and 32 bits:
3252 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3253 InstrItinClass itin16, InstrItinClass itin32,
3254 string OpcodeStr, string Dt, SDNode MulOp,
3256 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3257 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3258 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3259 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3260 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3261 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3264 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3265 string Dt, SDNode MulOp, SDNode OpNode> {
3266 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3267 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3268 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3269 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3273 // Neon Long 3-argument intrinsics.
3275 // First with only element sizes of 16 and 32 bits:
3276 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3277 InstrItinClass itin16, InstrItinClass itin32,
3278 string OpcodeStr, string Dt, Intrinsic IntOp> {
3279 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3280 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3281 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3282 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3285 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3286 string OpcodeStr, string Dt, Intrinsic IntOp> {
3287 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3288 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3289 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3290 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3293 // ....then also with element size of 8 bits:
3294 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3295 InstrItinClass itin16, InstrItinClass itin32,
3296 string OpcodeStr, string Dt, Intrinsic IntOp>
3297 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3298 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3299 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3302 // ....with explicit extend (VABAL).
3303 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3304 InstrItinClass itin, string OpcodeStr, string Dt,
3305 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3306 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3307 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3308 IntOp, ExtOp, OpNode>;
3309 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3310 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3311 IntOp, ExtOp, OpNode>;
3312 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3313 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3314 IntOp, ExtOp, OpNode>;
3318 // Neon Pairwise long 2-register intrinsics,
3319 // element sizes of 8, 16 and 32 bits:
3320 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3321 bits<5> op11_7, bit op4,
3322 string OpcodeStr, string Dt, Intrinsic IntOp> {
3323 // 64-bit vector types.
3324 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3325 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3326 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3327 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3328 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3329 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3331 // 128-bit vector types.
3332 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3333 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3334 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3335 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3336 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3337 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3341 // Neon Pairwise long 2-register accumulate intrinsics,
3342 // element sizes of 8, 16 and 32 bits:
3343 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3344 bits<5> op11_7, bit op4,
3345 string OpcodeStr, string Dt, Intrinsic IntOp> {
3346 // 64-bit vector types.
3347 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3348 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3349 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3350 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3351 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3352 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3354 // 128-bit vector types.
3355 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3356 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3357 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3358 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3359 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3360 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3364 // Neon 2-register vector shift by immediate,
3365 // with f of either N2RegVShLFrm or N2RegVShRFrm
3366 // element sizes of 8, 16, 32 and 64 bits:
3367 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3368 InstrItinClass itin, string OpcodeStr, string Dt,
3370 // 64-bit vector types.
3371 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3372 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3373 let Inst{21-19} = 0b001; // imm6 = 001xxx
3375 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3376 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3377 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3379 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3380 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3381 let Inst{21} = 0b1; // imm6 = 1xxxxx
3383 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3384 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3387 // 128-bit vector types.
3388 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3389 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3390 let Inst{21-19} = 0b001; // imm6 = 001xxx
3392 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3393 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3394 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3396 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3397 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3398 let Inst{21} = 0b1; // imm6 = 1xxxxx
3400 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3401 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3404 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3405 InstrItinClass itin, string OpcodeStr, string Dt,
3407 // 64-bit vector types.
3408 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3409 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3410 let Inst{21-19} = 0b001; // imm6 = 001xxx
3412 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3413 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3414 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3416 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3417 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3418 let Inst{21} = 0b1; // imm6 = 1xxxxx
3420 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3421 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3424 // 128-bit vector types.
3425 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3426 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3427 let Inst{21-19} = 0b001; // imm6 = 001xxx
3429 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3430 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3431 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3433 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3434 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3435 let Inst{21} = 0b1; // imm6 = 1xxxxx
3437 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3438 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3442 // Neon Shift-Accumulate vector operations,
3443 // element sizes of 8, 16, 32 and 64 bits:
3444 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3445 string OpcodeStr, string Dt, SDNode ShOp> {
3446 // 64-bit vector types.
3447 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3448 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3449 let Inst{21-19} = 0b001; // imm6 = 001xxx
3451 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3452 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3453 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3455 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3456 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3457 let Inst{21} = 0b1; // imm6 = 1xxxxx
3459 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3460 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3463 // 128-bit vector types.
3464 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3465 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3466 let Inst{21-19} = 0b001; // imm6 = 001xxx
3468 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3469 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3470 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3472 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3473 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3474 let Inst{21} = 0b1; // imm6 = 1xxxxx
3476 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3477 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3481 // Neon Shift-Insert vector operations,
3482 // with f of either N2RegVShLFrm or N2RegVShRFrm
3483 // element sizes of 8, 16, 32 and 64 bits:
3484 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3486 // 64-bit vector types.
3487 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3488 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3489 let Inst{21-19} = 0b001; // imm6 = 001xxx
3491 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3492 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3493 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3495 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3496 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3497 let Inst{21} = 0b1; // imm6 = 1xxxxx
3499 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3500 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3503 // 128-bit vector types.
3504 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3505 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3506 let Inst{21-19} = 0b001; // imm6 = 001xxx
3508 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3509 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3510 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3512 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3513 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3514 let Inst{21} = 0b1; // imm6 = 1xxxxx
3516 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3517 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3520 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3522 // 64-bit vector types.
3523 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3524 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3525 let Inst{21-19} = 0b001; // imm6 = 001xxx
3527 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3528 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3529 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3531 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3532 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3533 let Inst{21} = 0b1; // imm6 = 1xxxxx
3535 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3536 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3539 // 128-bit vector types.
3540 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3541 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3542 let Inst{21-19} = 0b001; // imm6 = 001xxx
3544 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3545 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3546 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3548 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3549 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3550 let Inst{21} = 0b1; // imm6 = 1xxxxx
3552 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3553 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3557 // Neon Shift Long operations,
3558 // element sizes of 8, 16, 32 bits:
3559 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3560 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3561 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3562 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, imm1_7, OpNode> {
3563 let Inst{21-19} = 0b001; // imm6 = 001xxx
3565 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3566 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, imm1_15, OpNode> {
3567 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3569 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3570 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, imm1_31, OpNode> {
3571 let Inst{21} = 0b1; // imm6 = 1xxxxx
3575 // Neon Shift Narrow operations,
3576 // element sizes of 16, 32, 64 bits:
3577 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3578 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3580 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3581 OpcodeStr, !strconcat(Dt, "16"),
3582 v8i8, v8i16, shr_imm8, OpNode> {
3583 let Inst{21-19} = 0b001; // imm6 = 001xxx
3585 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3586 OpcodeStr, !strconcat(Dt, "32"),
3587 v4i16, v4i32, shr_imm16, OpNode> {
3588 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3590 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3591 OpcodeStr, !strconcat(Dt, "64"),
3592 v2i32, v2i64, shr_imm32, OpNode> {
3593 let Inst{21} = 0b1; // imm6 = 1xxxxx
3597 //===----------------------------------------------------------------------===//
3598 // Instruction Definitions.
3599 //===----------------------------------------------------------------------===//
3601 // Vector Add Operations.
3603 // VADD : Vector Add (integer and floating-point)
3604 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3606 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3607 v2f32, v2f32, fadd, 1>;
3608 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3609 v4f32, v4f32, fadd, 1>;
3610 // VADDL : Vector Add Long (Q = D + D)
3611 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3612 "vaddl", "s", add, sext, 1>;
3613 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3614 "vaddl", "u", add, zext, 1>;
3615 // VADDW : Vector Add Wide (Q = Q + D)
3616 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3617 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3618 // VHADD : Vector Halving Add
3619 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3620 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3621 "vhadd", "s", int_arm_neon_vhadds, 1>;
3622 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3623 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3624 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3625 // VRHADD : Vector Rounding Halving Add
3626 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3627 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3628 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3629 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3630 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3631 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3632 // VQADD : Vector Saturating Add
3633 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3634 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3635 "vqadd", "s", int_arm_neon_vqadds, 1>;
3636 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3637 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3638 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3639 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3640 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3641 int_arm_neon_vaddhn, 1>;
3642 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3643 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3644 int_arm_neon_vraddhn, 1>;
3646 // Vector Multiply Operations.
3648 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3649 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3650 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3651 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3652 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3653 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3654 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3655 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3656 v2f32, v2f32, fmul, 1>;
3657 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3658 v4f32, v4f32, fmul, 1>;
3659 defm VMULsl : N3VSL_HS<0b1000, "vmul", mul>;
3660 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3661 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3664 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3665 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3666 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3667 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3668 (DSubReg_i16_reg imm:$lane))),
3669 (SubReg_i16_lane imm:$lane)))>;
3670 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3671 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3672 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3673 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3674 (DSubReg_i32_reg imm:$lane))),
3675 (SubReg_i32_lane imm:$lane)))>;
3676 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3677 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3678 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3679 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3680 (DSubReg_i32_reg imm:$lane))),
3681 (SubReg_i32_lane imm:$lane)))>;
3683 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3684 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3685 IIC_VMULi16Q, IIC_VMULi32Q,
3686 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3687 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3688 IIC_VMULi16Q, IIC_VMULi32Q,
3689 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3690 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3691 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3693 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3694 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3695 (DSubReg_i16_reg imm:$lane))),
3696 (SubReg_i16_lane imm:$lane)))>;
3697 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3698 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3700 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3701 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3702 (DSubReg_i32_reg imm:$lane))),
3703 (SubReg_i32_lane imm:$lane)))>;
3705 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3706 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3707 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3708 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3709 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3710 IIC_VMULi16Q, IIC_VMULi32Q,
3711 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3712 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3713 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3715 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3716 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3717 (DSubReg_i16_reg imm:$lane))),
3718 (SubReg_i16_lane imm:$lane)))>;
3719 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3720 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3722 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3723 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3724 (DSubReg_i32_reg imm:$lane))),
3725 (SubReg_i32_lane imm:$lane)))>;
3727 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3728 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3729 "vmull", "s", NEONvmulls, 1>;
3730 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3731 "vmull", "u", NEONvmullu, 1>;
3732 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3733 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3734 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3735 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3737 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3738 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3739 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3740 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3741 "vqdmull", "s", int_arm_neon_vqdmull>;
3743 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3745 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3746 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3747 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3748 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3749 v2f32, fmul_su, fadd_mlx>,
3750 Requires<[HasNEON, UseFPVMLx]>;
3751 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3752 v4f32, fmul_su, fadd_mlx>,
3753 Requires<[HasNEON, UseFPVMLx]>;
3754 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3755 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3756 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3757 v2f32, fmul_su, fadd_mlx>,
3758 Requires<[HasNEON, UseFPVMLx]>;
3759 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3760 v4f32, v2f32, fmul_su, fadd_mlx>,
3761 Requires<[HasNEON, UseFPVMLx]>;
3763 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3764 (mul (v8i16 QPR:$src2),
3765 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3766 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3767 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3768 (DSubReg_i16_reg imm:$lane))),
3769 (SubReg_i16_lane imm:$lane)))>;
3771 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3772 (mul (v4i32 QPR:$src2),
3773 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3774 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3775 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3776 (DSubReg_i32_reg imm:$lane))),
3777 (SubReg_i32_lane imm:$lane)))>;
3779 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3780 (fmul_su (v4f32 QPR:$src2),
3781 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3782 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3784 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3785 (DSubReg_i32_reg imm:$lane))),
3786 (SubReg_i32_lane imm:$lane)))>,
3787 Requires<[HasNEON, UseFPVMLx]>;
3789 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3790 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3791 "vmlal", "s", NEONvmulls, add>;
3792 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3793 "vmlal", "u", NEONvmullu, add>;
3795 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3796 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3798 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3799 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3800 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3801 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3803 // VMLS : Vector Multiply Subtract (integer and floating-point)
3804 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3805 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3806 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3807 v2f32, fmul_su, fsub_mlx>,
3808 Requires<[HasNEON, UseFPVMLx]>;
3809 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3810 v4f32, fmul_su, fsub_mlx>,
3811 Requires<[HasNEON, UseFPVMLx]>;
3812 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3813 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3814 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3815 v2f32, fmul_su, fsub_mlx>,
3816 Requires<[HasNEON, UseFPVMLx]>;
3817 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3818 v4f32, v2f32, fmul_su, fsub_mlx>,
3819 Requires<[HasNEON, UseFPVMLx]>;
3821 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3822 (mul (v8i16 QPR:$src2),
3823 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3824 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3825 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3826 (DSubReg_i16_reg imm:$lane))),
3827 (SubReg_i16_lane imm:$lane)))>;
3829 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3830 (mul (v4i32 QPR:$src2),
3831 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3832 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3833 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3834 (DSubReg_i32_reg imm:$lane))),
3835 (SubReg_i32_lane imm:$lane)))>;
3837 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3838 (fmul_su (v4f32 QPR:$src2),
3839 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3840 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3841 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3842 (DSubReg_i32_reg imm:$lane))),
3843 (SubReg_i32_lane imm:$lane)))>,
3844 Requires<[HasNEON, UseFPVMLx]>;
3846 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3847 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3848 "vmlsl", "s", NEONvmulls, sub>;
3849 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3850 "vmlsl", "u", NEONvmullu, sub>;
3852 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3853 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3855 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3856 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3857 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3858 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3860 // Vector Subtract Operations.
3862 // VSUB : Vector Subtract (integer and floating-point)
3863 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3864 "vsub", "i", sub, 0>;
3865 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3866 v2f32, v2f32, fsub, 0>;
3867 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3868 v4f32, v4f32, fsub, 0>;
3869 // VSUBL : Vector Subtract Long (Q = D - D)
3870 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3871 "vsubl", "s", sub, sext, 0>;
3872 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3873 "vsubl", "u", sub, zext, 0>;
3874 // VSUBW : Vector Subtract Wide (Q = Q - D)
3875 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3876 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3877 // VHSUB : Vector Halving Subtract
3878 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3879 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3880 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3881 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3882 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3883 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3884 // VQSUB : Vector Saturing Subtract
3885 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3886 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3887 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3888 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3889 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3890 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3891 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3892 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3893 int_arm_neon_vsubhn, 0>;
3894 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3895 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3896 int_arm_neon_vrsubhn, 0>;
3898 // Vector Comparisons.
3900 // VCEQ : Vector Compare Equal
3901 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3902 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3903 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3905 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3908 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3909 "$Vd, $Vm, #0", NEONvceqz>;
3911 // VCGE : Vector Compare Greater Than or Equal
3912 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3913 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3914 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3915 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3916 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3918 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3921 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3922 "$Vd, $Vm, #0", NEONvcgez>;
3923 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3924 "$Vd, $Vm, #0", NEONvclez>;
3926 // VCGT : Vector Compare Greater Than
3927 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3928 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3929 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3930 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3931 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3933 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3936 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3937 "$Vd, $Vm, #0", NEONvcgtz>;
3938 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3939 "$Vd, $Vm, #0", NEONvcltz>;
3941 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3942 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3943 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3944 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3945 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3946 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3947 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3948 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3949 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3950 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3951 // VTST : Vector Test Bits
3952 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3953 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3955 // Vector Bitwise Operations.
3957 def vnotd : PatFrag<(ops node:$in),
3958 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3959 def vnotq : PatFrag<(ops node:$in),
3960 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3963 // VAND : Vector Bitwise AND
3964 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3965 v2i32, v2i32, and, 1>;
3966 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3967 v4i32, v4i32, and, 1>;
3969 // VEOR : Vector Bitwise Exclusive OR
3970 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3971 v2i32, v2i32, xor, 1>;
3972 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3973 v4i32, v4i32, xor, 1>;
3975 // VORR : Vector Bitwise OR
3976 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3977 v2i32, v2i32, or, 1>;
3978 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3979 v4i32, v4i32, or, 1>;
3981 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3982 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3984 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3986 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3987 let Inst{9} = SIMM{9};
3990 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3991 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3993 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3995 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3996 let Inst{10-9} = SIMM{10-9};
3999 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4000 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4002 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4004 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4005 let Inst{9} = SIMM{9};
4008 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4009 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4011 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4013 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4014 let Inst{10-9} = SIMM{10-9};
4018 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4019 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4020 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4021 "vbic", "$Vd, $Vn, $Vm", "",
4022 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4023 (vnotd DPR:$Vm))))]>;
4024 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4025 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4026 "vbic", "$Vd, $Vn, $Vm", "",
4027 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4028 (vnotq QPR:$Vm))))]>;
4030 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4031 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4033 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4035 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4036 let Inst{9} = SIMM{9};
4039 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4040 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4042 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4044 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4045 let Inst{10-9} = SIMM{10-9};
4048 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4049 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4051 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4053 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4054 let Inst{9} = SIMM{9};
4057 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4058 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4060 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4062 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4063 let Inst{10-9} = SIMM{10-9};
4066 // VORN : Vector Bitwise OR NOT
4067 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4068 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4069 "vorn", "$Vd, $Vn, $Vm", "",
4070 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4071 (vnotd DPR:$Vm))))]>;
4072 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4073 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4074 "vorn", "$Vd, $Vn, $Vm", "",
4075 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4076 (vnotq QPR:$Vm))))]>;
4078 // VMVN : Vector Bitwise NOT (Immediate)
4080 let isReMaterializable = 1 in {
4082 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4083 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4084 "vmvn", "i16", "$Vd, $SIMM", "",
4085 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4086 let Inst{9} = SIMM{9};
4089 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4090 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4091 "vmvn", "i16", "$Vd, $SIMM", "",
4092 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4093 let Inst{9} = SIMM{9};
4096 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4097 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4098 "vmvn", "i32", "$Vd, $SIMM", "",
4099 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4100 let Inst{11-8} = SIMM{11-8};
4103 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4104 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4105 "vmvn", "i32", "$Vd, $SIMM", "",
4106 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4107 let Inst{11-8} = SIMM{11-8};
4111 // VMVN : Vector Bitwise NOT
4112 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4113 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4114 "vmvn", "$Vd, $Vm", "",
4115 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4116 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4117 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4118 "vmvn", "$Vd, $Vm", "",
4119 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4120 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4121 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4123 // VBSL : Vector Bitwise Select
4124 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4125 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4126 N3RegFrm, IIC_VCNTiD,
4127 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4129 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4131 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4132 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4133 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4135 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4136 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4137 N3RegFrm, IIC_VCNTiQ,
4138 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4140 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4142 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4143 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4144 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4146 // VBIF : Vector Bitwise Insert if False
4147 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4148 // FIXME: This instruction's encoding MAY NOT BE correct.
4149 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4150 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4151 N3RegFrm, IIC_VBINiD,
4152 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4154 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4155 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4156 N3RegFrm, IIC_VBINiQ,
4157 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4160 // VBIT : Vector Bitwise Insert if True
4161 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4162 // FIXME: This instruction's encoding MAY NOT BE correct.
4163 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4164 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4165 N3RegFrm, IIC_VBINiD,
4166 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4168 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4169 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4170 N3RegFrm, IIC_VBINiQ,
4171 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4174 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4175 // for equivalent operations with different register constraints; it just
4178 // Vector Absolute Differences.
4180 // VABD : Vector Absolute Difference
4181 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4182 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4183 "vabd", "s", int_arm_neon_vabds, 1>;
4184 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4185 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4186 "vabd", "u", int_arm_neon_vabdu, 1>;
4187 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4188 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4189 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4190 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4192 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4193 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4194 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4195 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4196 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4198 // VABA : Vector Absolute Difference and Accumulate
4199 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4200 "vaba", "s", int_arm_neon_vabds, add>;
4201 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4202 "vaba", "u", int_arm_neon_vabdu, add>;
4204 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4205 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4206 "vabal", "s", int_arm_neon_vabds, zext, add>;
4207 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4208 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4210 // Vector Maximum and Minimum.
4212 // VMAX : Vector Maximum
4213 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4214 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4215 "vmax", "s", int_arm_neon_vmaxs, 1>;
4216 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4217 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4218 "vmax", "u", int_arm_neon_vmaxu, 1>;
4219 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4221 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4222 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4224 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4226 // VMIN : Vector Minimum
4227 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4228 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4229 "vmin", "s", int_arm_neon_vmins, 1>;
4230 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4231 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4232 "vmin", "u", int_arm_neon_vminu, 1>;
4233 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4235 v2f32, v2f32, int_arm_neon_vmins, 1>;
4236 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4238 v4f32, v4f32, int_arm_neon_vmins, 1>;
4240 // Vector Pairwise Operations.
4242 // VPADD : Vector Pairwise Add
4243 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4245 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4246 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4248 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4249 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4251 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4252 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4253 IIC_VPBIND, "vpadd", "f32",
4254 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4256 // VPADDL : Vector Pairwise Add Long
4257 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4258 int_arm_neon_vpaddls>;
4259 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4260 int_arm_neon_vpaddlu>;
4262 // VPADAL : Vector Pairwise Add and Accumulate Long
4263 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4264 int_arm_neon_vpadals>;
4265 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4266 int_arm_neon_vpadalu>;
4268 // VPMAX : Vector Pairwise Maximum
4269 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4270 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4271 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4272 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4273 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4274 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4275 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4276 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4277 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4278 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4279 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4280 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4281 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4282 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4284 // VPMIN : Vector Pairwise Minimum
4285 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4286 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4287 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4288 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4289 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4290 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4291 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4292 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4293 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4294 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4295 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4296 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4297 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4298 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4300 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4302 // VRECPE : Vector Reciprocal Estimate
4303 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4304 IIC_VUNAD, "vrecpe", "u32",
4305 v2i32, v2i32, int_arm_neon_vrecpe>;
4306 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4307 IIC_VUNAQ, "vrecpe", "u32",
4308 v4i32, v4i32, int_arm_neon_vrecpe>;
4309 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4310 IIC_VUNAD, "vrecpe", "f32",
4311 v2f32, v2f32, int_arm_neon_vrecpe>;
4312 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4313 IIC_VUNAQ, "vrecpe", "f32",
4314 v4f32, v4f32, int_arm_neon_vrecpe>;
4316 // VRECPS : Vector Reciprocal Step
4317 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4318 IIC_VRECSD, "vrecps", "f32",
4319 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4320 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4321 IIC_VRECSQ, "vrecps", "f32",
4322 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4324 // VRSQRTE : Vector Reciprocal Square Root Estimate
4325 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4326 IIC_VUNAD, "vrsqrte", "u32",
4327 v2i32, v2i32, int_arm_neon_vrsqrte>;
4328 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4329 IIC_VUNAQ, "vrsqrte", "u32",
4330 v4i32, v4i32, int_arm_neon_vrsqrte>;
4331 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4332 IIC_VUNAD, "vrsqrte", "f32",
4333 v2f32, v2f32, int_arm_neon_vrsqrte>;
4334 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4335 IIC_VUNAQ, "vrsqrte", "f32",
4336 v4f32, v4f32, int_arm_neon_vrsqrte>;
4338 // VRSQRTS : Vector Reciprocal Square Root Step
4339 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4340 IIC_VRECSD, "vrsqrts", "f32",
4341 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4342 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4343 IIC_VRECSQ, "vrsqrts", "f32",
4344 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4348 // VSHL : Vector Shift
4349 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4350 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4351 "vshl", "s", int_arm_neon_vshifts>;
4352 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4353 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4354 "vshl", "u", int_arm_neon_vshiftu>;
4356 // VSHL : Vector Shift Left (Immediate)
4357 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4359 // VSHR : Vector Shift Right (Immediate)
4360 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4361 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4363 // VSHLL : Vector Shift Left Long
4364 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4365 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4367 // VSHLL : Vector Shift Left Long (with maximum shift count)
4368 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4369 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4370 ValueType OpTy, Operand ImmTy, SDNode OpNode>
4371 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4372 ResTy, OpTy, ImmTy, OpNode> {
4373 let Inst{21-16} = op21_16;
4374 let DecoderMethod = "DecodeVSHLMaxInstruction";
4376 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4377 v8i16, v8i8, imm8, NEONvshlli>;
4378 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4379 v4i32, v4i16, imm16, NEONvshlli>;
4380 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4381 v2i64, v2i32, imm32, NEONvshlli>;
4383 // VSHRN : Vector Shift Right and Narrow
4384 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4387 // VRSHL : Vector Rounding Shift
4388 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4389 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4390 "vrshl", "s", int_arm_neon_vrshifts>;
4391 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4392 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4393 "vrshl", "u", int_arm_neon_vrshiftu>;
4394 // VRSHR : Vector Rounding Shift Right
4395 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4396 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4398 // VRSHRN : Vector Rounding Shift Right and Narrow
4399 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4402 // VQSHL : Vector Saturating Shift
4403 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4404 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4405 "vqshl", "s", int_arm_neon_vqshifts>;
4406 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4407 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4408 "vqshl", "u", int_arm_neon_vqshiftu>;
4409 // VQSHL : Vector Saturating Shift Left (Immediate)
4410 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4411 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4413 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4414 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4416 // VQSHRN : Vector Saturating Shift Right and Narrow
4417 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4419 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4422 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4423 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4426 // VQRSHL : Vector Saturating Rounding Shift
4427 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4428 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4429 "vqrshl", "s", int_arm_neon_vqrshifts>;
4430 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4431 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4432 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4434 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4435 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4437 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4440 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4441 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4444 // VSRA : Vector Shift Right and Accumulate
4445 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4446 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4447 // VRSRA : Vector Rounding Shift Right and Accumulate
4448 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4449 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4451 // VSLI : Vector Shift Left and Insert
4452 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4454 // VSRI : Vector Shift Right and Insert
4455 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4457 // Vector Absolute and Saturating Absolute.
4459 // VABS : Vector Absolute Value
4460 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4461 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4463 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4464 IIC_VUNAD, "vabs", "f32",
4465 v2f32, v2f32, int_arm_neon_vabs>;
4466 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4467 IIC_VUNAQ, "vabs", "f32",
4468 v4f32, v4f32, int_arm_neon_vabs>;
4470 // VQABS : Vector Saturating Absolute Value
4471 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4472 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4473 int_arm_neon_vqabs>;
4477 def vnegd : PatFrag<(ops node:$in),
4478 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4479 def vnegq : PatFrag<(ops node:$in),
4480 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4482 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4483 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4484 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4485 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4486 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4487 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4488 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4489 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4491 // VNEG : Vector Negate (integer)
4492 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4493 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4494 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4495 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4496 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4497 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4499 // VNEG : Vector Negate (floating-point)
4500 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4501 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4502 "vneg", "f32", "$Vd, $Vm", "",
4503 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4504 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4505 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4506 "vneg", "f32", "$Vd, $Vm", "",
4507 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4509 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4510 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4511 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4512 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4513 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4514 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4516 // VQNEG : Vector Saturating Negate
4517 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4518 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4519 int_arm_neon_vqneg>;
4521 // Vector Bit Counting Operations.
4523 // VCLS : Vector Count Leading Sign Bits
4524 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4525 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4527 // VCLZ : Vector Count Leading Zeros
4528 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4529 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4531 // VCNT : Vector Count One Bits
4532 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4533 IIC_VCNTiD, "vcnt", "8",
4534 v8i8, v8i8, int_arm_neon_vcnt>;
4535 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4536 IIC_VCNTiQ, "vcnt", "8",
4537 v16i8, v16i8, int_arm_neon_vcnt>;
4540 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4541 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4542 "vswp", "$Vd, $Vm", "", []>;
4543 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4544 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4545 "vswp", "$Vd, $Vm", "", []>;
4547 // Vector Move Operations.
4549 // VMOV : Vector Move (Register)
4550 def : InstAlias<"vmov${p} $Vd, $Vm",
4551 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4552 def : InstAlias<"vmov${p} $Vd, $Vm",
4553 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4555 // VMOV : Vector Move (Immediate)
4557 let isReMaterializable = 1 in {
4558 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4559 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4560 "vmov", "i8", "$Vd, $SIMM", "",
4561 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4562 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4563 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4564 "vmov", "i8", "$Vd, $SIMM", "",
4565 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4567 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4568 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4569 "vmov", "i16", "$Vd, $SIMM", "",
4570 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4571 let Inst{9} = SIMM{9};
4574 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4575 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4576 "vmov", "i16", "$Vd, $SIMM", "",
4577 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4578 let Inst{9} = SIMM{9};
4581 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4582 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4583 "vmov", "i32", "$Vd, $SIMM", "",
4584 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4585 let Inst{11-8} = SIMM{11-8};
4588 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4589 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4590 "vmov", "i32", "$Vd, $SIMM", "",
4591 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4592 let Inst{11-8} = SIMM{11-8};
4595 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4596 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4597 "vmov", "i64", "$Vd, $SIMM", "",
4598 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4599 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4600 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4601 "vmov", "i64", "$Vd, $SIMM", "",
4602 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4604 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4605 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4606 "vmov", "f32", "$Vd, $SIMM", "",
4607 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4608 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4609 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4610 "vmov", "f32", "$Vd, $SIMM", "",
4611 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4612 } // isReMaterializable
4614 // VMOV : Vector Get Lane (move scalar to ARM core register)
4616 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4617 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4618 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4619 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4621 let Inst{21} = lane{2};
4622 let Inst{6-5} = lane{1-0};
4624 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4625 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4626 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4627 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4629 let Inst{21} = lane{1};
4630 let Inst{6} = lane{0};
4632 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4633 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4634 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4635 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4637 let Inst{21} = lane{2};
4638 let Inst{6-5} = lane{1-0};
4640 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4641 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4642 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4643 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4645 let Inst{21} = lane{1};
4646 let Inst{6} = lane{0};
4648 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4649 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4650 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4651 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4653 let Inst{21} = lane{0};
4655 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4656 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4657 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4658 (DSubReg_i8_reg imm:$lane))),
4659 (SubReg_i8_lane imm:$lane))>;
4660 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4661 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4662 (DSubReg_i16_reg imm:$lane))),
4663 (SubReg_i16_lane imm:$lane))>;
4664 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4665 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4666 (DSubReg_i8_reg imm:$lane))),
4667 (SubReg_i8_lane imm:$lane))>;
4668 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4669 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4670 (DSubReg_i16_reg imm:$lane))),
4671 (SubReg_i16_lane imm:$lane))>;
4672 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4673 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4674 (DSubReg_i32_reg imm:$lane))),
4675 (SubReg_i32_lane imm:$lane))>;
4676 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4677 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4678 (SSubReg_f32_reg imm:$src2))>;
4679 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4680 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4681 (SSubReg_f32_reg imm:$src2))>;
4682 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4683 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4684 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4685 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4688 // VMOV : Vector Set Lane (move ARM core register to scalar)
4690 let Constraints = "$src1 = $V" in {
4691 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4692 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4693 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4694 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4695 GPR:$R, imm:$lane))]> {
4696 let Inst{21} = lane{2};
4697 let Inst{6-5} = lane{1-0};
4699 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4700 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4701 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4702 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4703 GPR:$R, imm:$lane))]> {
4704 let Inst{21} = lane{1};
4705 let Inst{6} = lane{0};
4707 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4708 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4709 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4710 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4711 GPR:$R, imm:$lane))]> {
4712 let Inst{21} = lane{0};
4715 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4716 (v16i8 (INSERT_SUBREG QPR:$src1,
4717 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4718 (DSubReg_i8_reg imm:$lane))),
4719 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4720 (DSubReg_i8_reg imm:$lane)))>;
4721 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4722 (v8i16 (INSERT_SUBREG QPR:$src1,
4723 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4724 (DSubReg_i16_reg imm:$lane))),
4725 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4726 (DSubReg_i16_reg imm:$lane)))>;
4727 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4728 (v4i32 (INSERT_SUBREG QPR:$src1,
4729 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4730 (DSubReg_i32_reg imm:$lane))),
4731 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4732 (DSubReg_i32_reg imm:$lane)))>;
4734 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4735 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4736 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4737 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4738 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4739 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4741 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4742 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4743 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4744 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4746 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4747 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4748 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4749 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4750 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4751 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4753 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4754 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4755 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4756 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4757 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4758 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4760 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4761 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4762 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4764 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4765 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4766 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4768 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4769 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4770 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4773 // VDUP : Vector Duplicate (from ARM core register to all elements)
4775 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4776 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4777 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4778 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4779 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4780 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4781 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4782 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4784 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4785 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4786 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4787 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4788 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4789 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4791 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4792 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4794 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4796 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4797 ValueType Ty, Operand IdxTy>
4798 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4799 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4800 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4802 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4803 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4804 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4805 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4806 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4807 VectorIndex32:$lane)))]>;
4809 // Inst{19-16} is partially specified depending on the element size.
4811 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4813 let Inst{19-17} = lane{2-0};
4815 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4817 let Inst{19-18} = lane{1-0};
4819 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4821 let Inst{19} = lane{0};
4823 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4825 let Inst{19-17} = lane{2-0};
4827 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4829 let Inst{19-18} = lane{1-0};
4831 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4833 let Inst{19} = lane{0};
4836 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4837 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4839 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4840 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4842 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4843 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4844 (DSubReg_i8_reg imm:$lane))),
4845 (SubReg_i8_lane imm:$lane)))>;
4846 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4847 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4848 (DSubReg_i16_reg imm:$lane))),
4849 (SubReg_i16_lane imm:$lane)))>;
4850 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4851 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4852 (DSubReg_i32_reg imm:$lane))),
4853 (SubReg_i32_lane imm:$lane)))>;
4854 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4855 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4856 (DSubReg_i32_reg imm:$lane))),
4857 (SubReg_i32_lane imm:$lane)))>;
4859 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4860 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4861 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4862 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4864 // VMOVN : Vector Narrowing Move
4865 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4866 "vmovn", "i", trunc>;
4867 // VQMOVN : Vector Saturating Narrowing Move
4868 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4869 "vqmovn", "s", int_arm_neon_vqmovns>;
4870 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4871 "vqmovn", "u", int_arm_neon_vqmovnu>;
4872 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4873 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4874 // VMOVL : Vector Lengthening Move
4875 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4876 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4878 // Vector Conversions.
4880 // VCVT : Vector Convert Between Floating-Point and Integers
4881 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4882 v2i32, v2f32, fp_to_sint>;
4883 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4884 v2i32, v2f32, fp_to_uint>;
4885 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4886 v2f32, v2i32, sint_to_fp>;
4887 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4888 v2f32, v2i32, uint_to_fp>;
4890 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4891 v4i32, v4f32, fp_to_sint>;
4892 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4893 v4i32, v4f32, fp_to_uint>;
4894 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4895 v4f32, v4i32, sint_to_fp>;
4896 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4897 v4f32, v4i32, uint_to_fp>;
4899 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4900 let DecoderMethod = "DecodeVCVTD" in {
4901 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4902 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4903 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4904 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4905 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4906 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4907 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4908 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4911 let DecoderMethod = "DecodeVCVTQ" in {
4912 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4913 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4914 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4915 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4916 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4917 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4918 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4919 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4922 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4923 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4924 IIC_VUNAQ, "vcvt", "f16.f32",
4925 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4926 Requires<[HasNEON, HasFP16]>;
4927 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4928 IIC_VUNAQ, "vcvt", "f32.f16",
4929 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4930 Requires<[HasNEON, HasFP16]>;
4934 // VREV64 : Vector Reverse elements within 64-bit doublewords
4936 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4937 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4938 (ins DPR:$Vm), IIC_VMOVD,
4939 OpcodeStr, Dt, "$Vd, $Vm", "",
4940 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4941 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4942 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4943 (ins QPR:$Vm), IIC_VMOVQ,
4944 OpcodeStr, Dt, "$Vd, $Vm", "",
4945 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4947 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4948 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4949 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4950 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4952 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4953 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4954 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4955 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4957 // VREV32 : Vector Reverse elements within 32-bit words
4959 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4960 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4961 (ins DPR:$Vm), IIC_VMOVD,
4962 OpcodeStr, Dt, "$Vd, $Vm", "",
4963 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4964 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4965 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4966 (ins QPR:$Vm), IIC_VMOVQ,
4967 OpcodeStr, Dt, "$Vd, $Vm", "",
4968 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4970 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4971 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4973 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4974 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4976 // VREV16 : Vector Reverse elements within 16-bit halfwords
4978 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4979 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4980 (ins DPR:$Vm), IIC_VMOVD,
4981 OpcodeStr, Dt, "$Vd, $Vm", "",
4982 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4983 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4984 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4985 (ins QPR:$Vm), IIC_VMOVQ,
4986 OpcodeStr, Dt, "$Vd, $Vm", "",
4987 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4989 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4990 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4992 // Other Vector Shuffles.
4994 // Aligned extractions: really just dropping registers
4996 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4997 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4998 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5000 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5002 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5004 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5006 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5008 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5011 // VEXT : Vector Extract
5013 class VEXTd<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5014 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5015 (ins DPR:$Vn, DPR:$Vm, immTy:$index), NVExtFrm,
5016 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5017 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5018 (Ty DPR:$Vm), imm:$index)))]> {
5020 let Inst{11-8} = index{3-0};
5023 class VEXTq<string OpcodeStr, string Dt, ValueType Ty, Operand immTy>
5024 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5025 (ins QPR:$Vn, QPR:$Vm, imm0_15:$index), NVExtFrm,
5026 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5027 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5028 (Ty QPR:$Vm), imm:$index)))]> {
5030 let Inst{11-8} = index{3-0};
5033 def VEXTd8 : VEXTd<"vext", "8", v8i8, imm0_7> {
5034 let Inst{11-8} = index{3-0};
5036 def VEXTd16 : VEXTd<"vext", "16", v4i16, imm0_3> {
5037 let Inst{11-9} = index{2-0};
5040 def VEXTd32 : VEXTd<"vext", "32", v2i32, imm0_1> {
5041 let Inst{11-10} = index{1-0};
5042 let Inst{9-8} = 0b00;
5044 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5047 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5049 def VEXTq8 : VEXTq<"vext", "8", v16i8, imm0_15> {
5050 let Inst{11-8} = index{3-0};
5052 def VEXTq16 : VEXTq<"vext", "16", v8i16, imm0_7> {
5053 let Inst{11-9} = index{2-0};
5056 def VEXTq32 : VEXTq<"vext", "32", v4i32, imm0_3> {
5057 let Inst{11-10} = index{1-0};
5058 let Inst{9-8} = 0b00;
5060 def VEXTq64 : VEXTq<"vext", "64", v2i64, imm0_1> {
5061 let Inst{11} = index{0};
5062 let Inst{10-8} = 0b000;
5064 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5067 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5069 // VTRN : Vector Transpose
5071 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5072 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5073 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5075 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5076 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5077 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5079 // VUZP : Vector Unzip (Deinterleave)
5081 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5082 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5083 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5085 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5086 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5087 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5089 // VZIP : Vector Zip (Interleave)
5091 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5092 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5093 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5095 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5096 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5097 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5099 // Vector Table Lookup and Table Extension.
5101 // VTBL : Vector Table Lookup
5102 let DecoderMethod = "DecodeTBLInstruction" in {
5104 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5105 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5106 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5107 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5108 let hasExtraSrcRegAllocReq = 1 in {
5110 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5111 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5112 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
5114 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5115 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5116 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
5118 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5119 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
5121 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
5122 } // hasExtraSrcRegAllocReq = 1
5125 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5127 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5129 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5131 // VTBX : Vector Table Extension
5133 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5134 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5135 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5136 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5137 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5138 let hasExtraSrcRegAllocReq = 1 in {
5140 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5141 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5142 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5144 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5145 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5146 NVTBLFrm, IIC_VTBX3,
5147 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5150 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5151 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5152 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5154 } // hasExtraSrcRegAllocReq = 1
5157 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5158 IIC_VTBX2, "$orig = $dst", []>;
5160 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5161 IIC_VTBX3, "$orig = $dst", []>;
5163 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5164 IIC_VTBX4, "$orig = $dst", []>;
5165 } // DecoderMethod = "DecodeTBLInstruction"
5167 //===----------------------------------------------------------------------===//
5168 // NEON instructions for single-precision FP math
5169 //===----------------------------------------------------------------------===//
5171 class N2VSPat<SDNode OpNode, NeonI Inst>
5172 : NEONFPPat<(f32 (OpNode SPR:$a)),
5174 (v2f32 (COPY_TO_REGCLASS (Inst
5176 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5177 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5179 class N3VSPat<SDNode OpNode, NeonI Inst>
5180 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5182 (v2f32 (COPY_TO_REGCLASS (Inst
5184 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5187 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5188 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5190 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5191 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5193 (v2f32 (COPY_TO_REGCLASS (Inst
5195 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5198 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5201 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5202 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5204 def : N3VSPat<fadd, VADDfd>;
5205 def : N3VSPat<fsub, VSUBfd>;
5206 def : N3VSPat<fmul, VMULfd>;
5207 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5208 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5209 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5210 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5211 def : N2VSPat<fabs, VABSfd>;
5212 def : N2VSPat<fneg, VNEGfd>;
5213 def : N3VSPat<NEONfmax, VMAXfd>;
5214 def : N3VSPat<NEONfmin, VMINfd>;
5215 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5216 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5217 def : N2VSPat<arm_sitof, VCVTs2fd>;
5218 def : N2VSPat<arm_uitof, VCVTu2fd>;
5220 //===----------------------------------------------------------------------===//
5221 // Non-Instruction Patterns
5222 //===----------------------------------------------------------------------===//
5225 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5226 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5227 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5228 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5229 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5230 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5231 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5232 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5233 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5234 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5235 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5236 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5237 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5238 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5239 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5240 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5241 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5242 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5243 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5244 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5245 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5246 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5247 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5248 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5249 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5250 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5251 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5252 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5253 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5254 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5256 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5257 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5258 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5259 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5260 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5261 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5262 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5263 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5264 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5265 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5266 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5267 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5268 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5269 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5270 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5271 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5272 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5273 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5274 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5275 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5276 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5277 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5278 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5279 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5280 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5281 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5282 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5283 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5284 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5285 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5288 //===----------------------------------------------------------------------===//
5289 // Assembler aliases
5292 // VADD two-operand aliases.
5293 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5294 (VADDv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5295 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5296 (VADDv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5297 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5298 (VADDv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5299 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5300 (VADDv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5302 def : NEONInstAlias<"vadd${p}.i8 $Vdn, $Vm",
5303 (VADDv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5304 def : NEONInstAlias<"vadd${p}.i16 $Vdn, $Vm",
5305 (VADDv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5306 def : NEONInstAlias<"vadd${p}.i32 $Vdn, $Vm",
5307 (VADDv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5308 def : NEONInstAlias<"vadd${p}.i64 $Vdn, $Vm",
5309 (VADDv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5311 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5312 (VADDfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5313 def : NEONInstAlias<"vadd${p}.f32 $Vdn, $Vm",
5314 (VADDfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5316 // VSUB two-operand aliases.
5317 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5318 (VSUBv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5319 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5320 (VSUBv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5321 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5322 (VSUBv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5323 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5324 (VSUBv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5326 def : NEONInstAlias<"vsub${p}.i8 $Vdn, $Vm",
5327 (VSUBv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5328 def : NEONInstAlias<"vsub${p}.i16 $Vdn, $Vm",
5329 (VSUBv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5330 def : NEONInstAlias<"vsub${p}.i32 $Vdn, $Vm",
5331 (VSUBv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5332 def : NEONInstAlias<"vsub${p}.i64 $Vdn, $Vm",
5333 (VSUBv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5335 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5336 (VSUBfd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5337 def : NEONInstAlias<"vsub${p}.f32 $Vdn, $Vm",
5338 (VSUBfq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5340 // VADDW two-operand aliases.
5341 def : NEONInstAlias<"vaddw${p}.s8 $Vdn, $Vm",
5342 (VADDWsv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5343 def : NEONInstAlias<"vaddw${p}.s16 $Vdn, $Vm",
5344 (VADDWsv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5345 def : NEONInstAlias<"vaddw${p}.s32 $Vdn, $Vm",
5346 (VADDWsv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5347 def : NEONInstAlias<"vaddw${p}.u8 $Vdn, $Vm",
5348 (VADDWuv8i16 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5349 def : NEONInstAlias<"vaddw${p}.u16 $Vdn, $Vm",
5350 (VADDWuv4i32 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5351 def : NEONInstAlias<"vaddw${p}.u32 $Vdn, $Vm",
5352 (VADDWuv2i64 QPR:$Vdn, QPR:$Vdn, DPR:$Vm, pred:$p)>;
5354 // VAND/VEOR/VORR accept but do not require a type suffix.
5355 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5356 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5357 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5358 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5359 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5360 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5361 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5362 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5363 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5364 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5365 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5366 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5367 // ... two-operand aliases
5368 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5369 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5370 def : NEONInstAlias<"vand${p} $Vdn, $Vm",
5371 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5372 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5373 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5374 def : NEONInstAlias<"veor${p} $Vdn, $Vm",
5375 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5376 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5377 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5378 def : NEONInstAlias<"vorr${p} $Vdn, $Vm",
5379 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5381 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5382 (VANDd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5383 defm : VFPDTAnyInstAlias<"vand${p}", "$Vdn, $Vm",
5384 (VANDq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5385 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5386 (VEORd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5387 defm : VFPDTAnyInstAlias<"veor${p}", "$Vdn, $Vm",
5388 (VEORq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5389 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5390 (VORRd DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5391 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vdn, $Vm",
5392 (VORRq QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5394 // VMUL two-operand aliases.
5395 def : NEONInstAlias<"vmul${p}.p8 $Qdn, $Qm",
5396 (VMULpq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5397 def : NEONInstAlias<"vmul${p}.i8 $Qdn, $Qm",
5398 (VMULv16i8 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5399 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Qm",
5400 (VMULv8i16 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5401 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Qm",
5402 (VMULv4i32 QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5404 def : NEONInstAlias<"vmul${p}.p8 $Ddn, $Dm",
5405 (VMULpd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5406 def : NEONInstAlias<"vmul${p}.i8 $Ddn, $Dm",
5407 (VMULv8i8 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5408 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm",
5409 (VMULv4i16 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5410 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm",
5411 (VMULv2i32 DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5413 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Qm",
5414 (VMULfq QPR:$Qdn, QPR:$Qdn, QPR:$Qm, pred:$p)>;
5415 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm",
5416 (VMULfd DPR:$Ddn, DPR:$Ddn, DPR:$Dm, pred:$p)>;
5418 def : NEONInstAlias<"vmul${p}.i16 $Ddn, $Dm$lane",
5419 (VMULslv4i16 DPR:$Ddn, DPR:$Ddn, DPR_8:$Dm,
5420 VectorIndex16:$lane, pred:$p)>;
5421 def : NEONInstAlias<"vmul${p}.i16 $Qdn, $Dm$lane",
5422 (VMULslv8i16 QPR:$Qdn, QPR:$Qdn, DPR_8:$Dm,
5423 VectorIndex16:$lane, pred:$p)>;
5425 def : NEONInstAlias<"vmul${p}.i32 $Ddn, $Dm$lane",
5426 (VMULslv2i32 DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5427 VectorIndex32:$lane, pred:$p)>;
5428 def : NEONInstAlias<"vmul${p}.i32 $Qdn, $Dm$lane",
5429 (VMULslv4i32 QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5430 VectorIndex32:$lane, pred:$p)>;
5432 def : NEONInstAlias<"vmul${p}.f32 $Ddn, $Dm$lane",
5433 (VMULslfd DPR:$Ddn, DPR:$Ddn, DPR_VFP2:$Dm,
5434 VectorIndex32:$lane, pred:$p)>;
5435 def : NEONInstAlias<"vmul${p}.f32 $Qdn, $Dm$lane",
5436 (VMULslfq QPR:$Qdn, QPR:$Qdn, DPR_VFP2:$Dm,
5437 VectorIndex32:$lane, pred:$p)>;
5439 // VQADD (register) two-operand aliases.
5440 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5441 (VQADDsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5442 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5443 (VQADDsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5444 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5445 (VQADDsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5446 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5447 (VQADDsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5448 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5449 (VQADDuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5450 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5451 (VQADDuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5452 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5453 (VQADDuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5454 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5455 (VQADDuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5457 def : NEONInstAlias<"vqadd${p}.s8 $Vdn, $Vm",
5458 (VQADDsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5459 def : NEONInstAlias<"vqadd${p}.s16 $Vdn, $Vm",
5460 (VQADDsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5461 def : NEONInstAlias<"vqadd${p}.s32 $Vdn, $Vm",
5462 (VQADDsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5463 def : NEONInstAlias<"vqadd${p}.s64 $Vdn, $Vm",
5464 (VQADDsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5465 def : NEONInstAlias<"vqadd${p}.u8 $Vdn, $Vm",
5466 (VQADDuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5467 def : NEONInstAlias<"vqadd${p}.u16 $Vdn, $Vm",
5468 (VQADDuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5469 def : NEONInstAlias<"vqadd${p}.u32 $Vdn, $Vm",
5470 (VQADDuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5471 def : NEONInstAlias<"vqadd${p}.u64 $Vdn, $Vm",
5472 (VQADDuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5474 // VSHL (immediate) two-operand aliases.
5475 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5476 (VSHLiv8i8 DPR:$Vdn, DPR:$Vdn, imm0_7:$imm, pred:$p)>;
5477 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5478 (VSHLiv4i16 DPR:$Vdn, DPR:$Vdn, imm0_15:$imm, pred:$p)>;
5479 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5480 (VSHLiv2i32 DPR:$Vdn, DPR:$Vdn, imm0_31:$imm, pred:$p)>;
5481 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5482 (VSHLiv1i64 DPR:$Vdn, DPR:$Vdn, imm0_63:$imm, pred:$p)>;
5484 def : NEONInstAlias<"vshl${p}.i8 $Vdn, $imm",
5485 (VSHLiv16i8 QPR:$Vdn, QPR:$Vdn, imm0_7:$imm, pred:$p)>;
5486 def : NEONInstAlias<"vshl${p}.i16 $Vdn, $imm",
5487 (VSHLiv8i16 QPR:$Vdn, QPR:$Vdn, imm0_15:$imm, pred:$p)>;
5488 def : NEONInstAlias<"vshl${p}.i32 $Vdn, $imm",
5489 (VSHLiv4i32 QPR:$Vdn, QPR:$Vdn, imm0_31:$imm, pred:$p)>;
5490 def : NEONInstAlias<"vshl${p}.i64 $Vdn, $imm",
5491 (VSHLiv2i64 QPR:$Vdn, QPR:$Vdn, imm0_63:$imm, pred:$p)>;
5493 // VSHL (register) two-operand aliases.
5494 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5495 (VSHLsv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5496 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5497 (VSHLsv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5498 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5499 (VSHLsv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5500 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5501 (VSHLsv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5502 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5503 (VSHLuv8i8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5504 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5505 (VSHLuv4i16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5506 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5507 (VSHLuv2i32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5508 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5509 (VSHLuv1i64 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, pred:$p)>;
5511 def : NEONInstAlias<"vshl${p}.s8 $Vdn, $Vm",
5512 (VSHLsv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5513 def : NEONInstAlias<"vshl${p}.s16 $Vdn, $Vm",
5514 (VSHLsv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5515 def : NEONInstAlias<"vshl${p}.s32 $Vdn, $Vm",
5516 (VSHLsv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5517 def : NEONInstAlias<"vshl${p}.s64 $Vdn, $Vm",
5518 (VSHLsv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5519 def : NEONInstAlias<"vshl${p}.u8 $Vdn, $Vm",
5520 (VSHLuv16i8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5521 def : NEONInstAlias<"vshl${p}.u16 $Vdn, $Vm",
5522 (VSHLuv8i16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5523 def : NEONInstAlias<"vshl${p}.u32 $Vdn, $Vm",
5524 (VSHLuv4i32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5525 def : NEONInstAlias<"vshl${p}.u64 $Vdn, $Vm",
5526 (VSHLuv2i64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, pred:$p)>;
5528 // VSHL (immediate) two-operand aliases.
5529 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5530 (VSHRsv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5531 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5532 (VSHRsv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5533 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5534 (VSHRsv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5535 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5536 (VSHRsv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5538 def : NEONInstAlias<"vshr${p}.s8 $Vdn, $imm",
5539 (VSHRsv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5540 def : NEONInstAlias<"vshr${p}.s16 $Vdn, $imm",
5541 (VSHRsv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5542 def : NEONInstAlias<"vshr${p}.s32 $Vdn, $imm",
5543 (VSHRsv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5544 def : NEONInstAlias<"vshr${p}.s64 $Vdn, $imm",
5545 (VSHRsv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5547 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5548 (VSHRuv8i8 DPR:$Vdn, DPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5549 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5550 (VSHRuv4i16 DPR:$Vdn, DPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5551 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5552 (VSHRuv2i32 DPR:$Vdn, DPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5553 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5554 (VSHRuv1i64 DPR:$Vdn, DPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5556 def : NEONInstAlias<"vshr${p}.u8 $Vdn, $imm",
5557 (VSHRuv16i8 QPR:$Vdn, QPR:$Vdn, shr_imm8:$imm, pred:$p)>;
5558 def : NEONInstAlias<"vshr${p}.u16 $Vdn, $imm",
5559 (VSHRuv8i16 QPR:$Vdn, QPR:$Vdn, shr_imm16:$imm, pred:$p)>;
5560 def : NEONInstAlias<"vshr${p}.u32 $Vdn, $imm",
5561 (VSHRuv4i32 QPR:$Vdn, QPR:$Vdn, shr_imm32:$imm, pred:$p)>;
5562 def : NEONInstAlias<"vshr${p}.u64 $Vdn, $imm",
5563 (VSHRuv2i64 QPR:$Vdn, QPR:$Vdn, shr_imm64:$imm, pred:$p)>;
5565 // VLD1 single-lane pseudo-instructions. These need special handling for
5566 // the lane index that an InstAlias can't handle, so we use these instead.
5567 defm VLD1LNdAsm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr",
5568 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5569 defm VLD1LNdAsm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr",
5570 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5571 defm VLD1LNdAsm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr",
5572 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5574 defm VLD1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr!",
5575 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5576 defm VLD1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr!",
5577 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5578 defm VLD1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr!",
5579 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5580 defm VLD1LNdWB_register_Asm :
5581 NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5582 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5583 rGPR:$Rm, pred:$p)>;
5584 defm VLD1LNdWB_register_Asm :
5585 NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5586 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5587 rGPR:$Rm, pred:$p)>;
5588 defm VLD1LNdWB_register_Asm :
5589 NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr, $Rm",
5590 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5591 rGPR:$Rm, pred:$p)>;
5594 // VST1 single-lane pseudo-instructions. These need special handling for
5595 // the lane index that an InstAlias can't handle, so we use these instead.
5596 defm VST1LNdAsm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr",
5597 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5598 defm VST1LNdAsm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr",
5599 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5600 defm VST1LNdAsm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr",
5601 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5603 defm VST1LNdWB_fixed_Asm : NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr!",
5604 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5605 defm VST1LNdWB_fixed_Asm : NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr!",
5606 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5607 defm VST1LNdWB_fixed_Asm : NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr!",
5608 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5609 defm VST1LNdWB_register_Asm :
5610 NEONDT8AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5611 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5612 rGPR:$Rm, pred:$p)>;
5613 defm VST1LNdWB_register_Asm :
5614 NEONDT16AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5615 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5616 rGPR:$Rm, pred:$p)>;
5617 defm VST1LNdWB_register_Asm :
5618 NEONDT32AsmPseudoInst<"vst1${p}", "$list, $addr, $Rm",
5619 (ins VecListOneDByteIndexed:$list, addrmode6:$addr,
5620 rGPR:$Rm, pred:$p)>;
5622 // VMOV takes an optional datatype suffix
5623 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5624 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
5625 defm : VFPDTAnyInstAlias<"vmov${p}", "$Vd, $Vm",
5626 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
5628 // VCLT (register) is an assembler alias for VCGT w/ the operands reversed.
5629 // D-register versions.
5630 def : NEONInstAlias<"vclt${p}.s8 $Dd, $Dn, $Dm",
5631 (VCGTsv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5632 def : NEONInstAlias<"vclt${p}.s16 $Dd, $Dn, $Dm",
5633 (VCGTsv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5634 def : NEONInstAlias<"vclt${p}.s32 $Dd, $Dn, $Dm",
5635 (VCGTsv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5636 def : NEONInstAlias<"vclt${p}.u8 $Dd, $Dn, $Dm",
5637 (VCGTuv8i8 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5638 def : NEONInstAlias<"vclt${p}.u16 $Dd, $Dn, $Dm",
5639 (VCGTuv4i16 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5640 def : NEONInstAlias<"vclt${p}.u32 $Dd, $Dn, $Dm",
5641 (VCGTuv2i32 DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5642 def : NEONInstAlias<"vclt${p}.f32 $Dd, $Dn, $Dm",
5643 (VCGTfd DPR:$Dd, DPR:$Dm, DPR:$Dn, pred:$p)>;
5644 // Q-register versions.
5645 def : NEONInstAlias<"vclt${p}.s8 $Qd, $Qn, $Qm",
5646 (VCGTsv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5647 def : NEONInstAlias<"vclt${p}.s16 $Qd, $Qn, $Qm",
5648 (VCGTsv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5649 def : NEONInstAlias<"vclt${p}.s32 $Qd, $Qn, $Qm",
5650 (VCGTsv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5651 def : NEONInstAlias<"vclt${p}.u8 $Qd, $Qn, $Qm",
5652 (VCGTuv16i8 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5653 def : NEONInstAlias<"vclt${p}.u16 $Qd, $Qn, $Qm",
5654 (VCGTuv8i16 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5655 def : NEONInstAlias<"vclt${p}.u32 $Qd, $Qn, $Qm",
5656 (VCGTuv4i32 QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5657 def : NEONInstAlias<"vclt${p}.f32 $Qd, $Qn, $Qm",
5658 (VCGTfq QPR:$Qd, QPR:$Qm, QPR:$Qn, pred:$p)>;
5660 // Two-operand variants for VEXT
5661 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5662 (VEXTd8 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_7:$imm, pred:$p)>;
5663 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5664 (VEXTd16 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_3:$imm, pred:$p)>;
5665 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5666 (VEXTd32 DPR:$Vdn, DPR:$Vdn, DPR:$Vm, imm0_1:$imm, pred:$p)>;
5668 def : NEONInstAlias<"vext${p}.8 $Vdn, $Vm, $imm",
5669 (VEXTq8 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_15:$imm, pred:$p)>;
5670 def : NEONInstAlias<"vext${p}.16 $Vdn, $Vm, $imm",
5671 (VEXTq16 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_7:$imm, pred:$p)>;
5672 def : NEONInstAlias<"vext${p}.32 $Vdn, $Vm, $imm",
5673 (VEXTq32 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_3:$imm, pred:$p)>;
5674 def : NEONInstAlias<"vext${p}.64 $Vdn, $Vm, $imm",
5675 (VEXTq64 QPR:$Vdn, QPR:$Vdn, QPR:$Vm, imm0_1:$imm, pred:$p)>;