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, $Rn", "", []> {
515 let Inst{5-4} = Rn{5-4};
516 let DecoderMethod = "DecodeVLDInstruction";
519 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
520 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
521 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
523 def VLD2q8 : VLD2Q<{0,0,?,?}, "8", VecListFourD>;
524 def VLD2q16 : VLD2Q<{0,1,?,?}, "16", VecListFourD>;
525 def VLD2q32 : VLD2Q<{1,0,?,?}, "32", VecListFourD>;
527 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
528 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
529 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
531 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
532 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
533 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
535 // ...with address register writeback:
536 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
537 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
538 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
539 "vld2", Dt, "$Vd, $Rn$Rm",
540 "$Rn.addr = $wb", []> {
541 let Inst{5-4} = Rn{5-4};
542 let DecoderMethod = "DecodeVLDInstruction";
544 class VLD2QWB<bits<4> op7_4, string Dt, RegisterOperand VdTy>
545 : NLdSt<0, 0b10, 0b0011, op7_4,
546 (outs VdTy:$Vd, GPR:$wb),
547 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
548 "vld2", Dt, "$Vd, $Rn$Rm",
549 "$Rn.addr = $wb", []> {
550 let Inst{5-4} = Rn{5-4};
551 let DecoderMethod = "DecodeVLDInstruction";
554 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
555 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
556 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
558 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8", VecListFourD>;
559 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16", VecListFourD>;
560 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32", VecListFourD>;
562 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
563 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
564 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
566 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
567 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
568 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
570 // ...with double-spaced registers
571 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
572 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
573 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
574 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
575 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
576 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
578 // VLD3 : Vector Load (multiple 3-element structures)
579 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
580 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
581 (ins addrmode6:$Rn), IIC_VLD3,
582 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
585 let DecoderMethod = "DecodeVLDInstruction";
588 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
589 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
590 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
592 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
593 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
594 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
596 // ...with address register writeback:
597 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
598 : NLdSt<0, 0b10, op11_8, op7_4,
599 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
600 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
601 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
602 "$Rn.addr = $wb", []> {
604 let DecoderMethod = "DecodeVLDInstruction";
607 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
608 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
609 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
611 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
612 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
613 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
615 // ...with double-spaced registers:
616 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
617 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
618 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
619 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
620 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
621 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
623 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
624 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
625 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
627 // ...alternate versions to be allocated odd register numbers:
628 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
629 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
630 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
632 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
633 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
634 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
636 // VLD4 : Vector Load (multiple 4-element structures)
637 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
638 : NLdSt<0, 0b10, op11_8, op7_4,
639 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
640 (ins addrmode6:$Rn), IIC_VLD4,
641 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
643 let Inst{5-4} = Rn{5-4};
644 let DecoderMethod = "DecodeVLDInstruction";
647 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
648 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
649 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
651 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
652 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
653 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
655 // ...with address register writeback:
656 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
657 : NLdSt<0, 0b10, op11_8, op7_4,
658 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
659 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
660 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
661 "$Rn.addr = $wb", []> {
662 let Inst{5-4} = Rn{5-4};
663 let DecoderMethod = "DecodeVLDInstruction";
666 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
667 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
668 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
670 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
671 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
672 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
674 // ...with double-spaced registers:
675 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
676 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
677 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
678 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
679 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
680 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
682 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
683 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
684 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
686 // ...alternate versions to be allocated odd register numbers:
687 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
688 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
689 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
691 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
692 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
693 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
695 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
697 // Classes for VLD*LN pseudo-instructions with multi-register operands.
698 // These are expanded to real instructions after register allocation.
699 class VLDQLNPseudo<InstrItinClass itin>
700 : PseudoNLdSt<(outs QPR:$dst),
701 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
702 itin, "$src = $dst">;
703 class VLDQLNWBPseudo<InstrItinClass itin>
704 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
705 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
706 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
707 class VLDQQLNPseudo<InstrItinClass itin>
708 : PseudoNLdSt<(outs QQPR:$dst),
709 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
710 itin, "$src = $dst">;
711 class VLDQQLNWBPseudo<InstrItinClass itin>
712 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
713 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
714 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
715 class VLDQQQQLNPseudo<InstrItinClass itin>
716 : PseudoNLdSt<(outs QQQQPR:$dst),
717 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
718 itin, "$src = $dst">;
719 class VLDQQQQLNWBPseudo<InstrItinClass itin>
720 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
721 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
722 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
724 // VLD1LN : Vector Load (single element to one lane)
725 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
727 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
728 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
729 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
731 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
732 (i32 (LoadOp addrmode6:$Rn)),
735 let DecoderMethod = "DecodeVLD1LN";
737 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
739 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
740 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
741 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
743 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
744 (i32 (LoadOp addrmode6oneL32:$Rn)),
747 let DecoderMethod = "DecodeVLD1LN";
749 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
750 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
751 (i32 (LoadOp addrmode6:$addr)),
755 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
756 let Inst{7-5} = lane{2-0};
758 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
759 let Inst{7-6} = lane{1-0};
762 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
763 let Inst{7} = lane{0};
768 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
769 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
770 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
772 def : Pat<(vector_insert (v2f32 DPR:$src),
773 (f32 (load addrmode6:$addr)), imm:$lane),
774 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
775 def : Pat<(vector_insert (v4f32 QPR:$src),
776 (f32 (load addrmode6:$addr)), imm:$lane),
777 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
779 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
781 // ...with address register writeback:
782 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
783 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
784 (ins addrmode6:$Rn, am6offset:$Rm,
785 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
786 "\\{$Vd[$lane]\\}, $Rn$Rm",
787 "$src = $Vd, $Rn.addr = $wb", []> {
788 let DecoderMethod = "DecodeVLD1LN";
791 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
792 let Inst{7-5} = lane{2-0};
794 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
795 let Inst{7-6} = lane{1-0};
798 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
799 let Inst{7} = lane{0};
804 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
805 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
806 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
808 // VLD2LN : Vector Load (single 2-element structure to one lane)
809 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
810 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
811 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
812 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
813 "$src1 = $Vd, $src2 = $dst2", []> {
816 let DecoderMethod = "DecodeVLD2LN";
819 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
820 let Inst{7-5} = lane{2-0};
822 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
823 let Inst{7-6} = lane{1-0};
825 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
826 let Inst{7} = lane{0};
829 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
830 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
831 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
833 // ...with double-spaced registers:
834 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
835 let Inst{7-6} = lane{1-0};
837 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
838 let Inst{7} = lane{0};
841 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
842 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
844 // ...with address register writeback:
845 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
846 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
847 (ins addrmode6:$Rn, am6offset:$Rm,
848 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
849 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
850 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
852 let DecoderMethod = "DecodeVLD2LN";
855 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
856 let Inst{7-5} = lane{2-0};
858 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
859 let Inst{7-6} = lane{1-0};
861 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
862 let Inst{7} = lane{0};
865 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
866 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
867 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
869 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
870 let Inst{7-6} = lane{1-0};
872 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
873 let Inst{7} = lane{0};
876 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
877 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
879 // VLD3LN : Vector Load (single 3-element structure to one lane)
880 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
881 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
882 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
883 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
884 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
885 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
887 let DecoderMethod = "DecodeVLD3LN";
890 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
891 let Inst{7-5} = lane{2-0};
893 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
894 let Inst{7-6} = lane{1-0};
896 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
897 let Inst{7} = lane{0};
900 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
901 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
902 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
904 // ...with double-spaced registers:
905 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
906 let Inst{7-6} = lane{1-0};
908 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
909 let Inst{7} = lane{0};
912 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
913 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
915 // ...with address register writeback:
916 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
917 : NLdStLn<1, 0b10, op11_8, op7_4,
918 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
919 (ins addrmode6:$Rn, am6offset:$Rm,
920 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
921 IIC_VLD3lnu, "vld3", Dt,
922 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
923 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
925 let DecoderMethod = "DecodeVLD3LN";
928 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
929 let Inst{7-5} = lane{2-0};
931 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
932 let Inst{7-6} = lane{1-0};
934 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
935 let Inst{7} = lane{0};
938 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
939 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
940 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
942 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
943 let Inst{7-6} = lane{1-0};
945 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
946 let Inst{7} = lane{0};
949 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
950 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
952 // VLD4LN : Vector Load (single 4-element structure to one lane)
953 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
954 : NLdStLn<1, 0b10, op11_8, op7_4,
955 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
956 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
957 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
958 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
959 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
962 let DecoderMethod = "DecodeVLD4LN";
965 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
966 let Inst{7-5} = lane{2-0};
968 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
969 let Inst{7-6} = lane{1-0};
971 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
972 let Inst{7} = lane{0};
976 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
977 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
978 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
980 // ...with double-spaced registers:
981 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
982 let Inst{7-6} = lane{1-0};
984 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
985 let Inst{7} = lane{0};
989 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
990 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
992 // ...with address register writeback:
993 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
994 : NLdStLn<1, 0b10, op11_8, op7_4,
995 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
996 (ins addrmode6:$Rn, am6offset:$Rm,
997 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
998 IIC_VLD4lnu, "vld4", Dt,
999 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
1000 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
1002 let Inst{4} = Rn{4};
1003 let DecoderMethod = "DecodeVLD4LN" ;
1006 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
1007 let Inst{7-5} = lane{2-0};
1009 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
1010 let Inst{7-6} = lane{1-0};
1012 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
1013 let Inst{7} = lane{0};
1014 let Inst{5} = Rn{5};
1017 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1018 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1019 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
1021 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
1022 let Inst{7-6} = lane{1-0};
1024 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
1025 let Inst{7} = lane{0};
1026 let Inst{5} = Rn{5};
1029 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1030 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1032 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1034 // VLD1DUP : Vector Load (single element to all lanes)
1035 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1036 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListOneDAllLanes:$Vd),
1037 (ins addrmode6dup:$Rn),
1038 IIC_VLD1dup, "vld1", Dt, "$Vd, $Rn", "",
1039 [(set VecListOneDAllLanes:$Vd,
1040 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1042 let Inst{4} = Rn{4};
1043 let DecoderMethod = "DecodeVLD1DupInstruction";
1045 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1046 let Pattern = [(set QPR:$dst,
1047 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1050 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1051 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1052 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1054 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1055 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1056 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1058 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1059 (VLD1DUPd32 addrmode6:$addr)>;
1060 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1061 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1063 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1065 class VLD1QDUP<bits<4> op7_4, string Dt>
1066 : NLdSt<1, 0b10, 0b1100, op7_4, (outs VecListTwoDAllLanes:$Vd),
1067 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1068 "vld1", Dt, "$Vd, $Rn", "", []> {
1070 let Inst{4} = Rn{4};
1071 let DecoderMethod = "DecodeVLD1DupInstruction";
1074 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1075 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1076 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1078 // ...with address register writeback:
1079 multiclass VLD1DUPWB<bits<4> op7_4, string Dt> {
1080 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1081 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1082 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1083 "vld1", Dt, "$Vd, $Rn!",
1084 "$Rn.addr = $wb", []> {
1085 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1086 let Inst{4} = Rn{4};
1087 let DecoderMethod = "DecodeVLD1DupInstruction";
1088 let AsmMatchConverter = "cvtVLDwbFixed";
1090 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1091 (outs VecListOneDAllLanes:$Vd, GPR:$wb),
1092 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1093 "vld1", Dt, "$Vd, $Rn, $Rm",
1094 "$Rn.addr = $wb", []> {
1095 let Inst{4} = Rn{4};
1096 let DecoderMethod = "DecodeVLD1DupInstruction";
1097 let AsmMatchConverter = "cvtVLDwbRegister";
1100 multiclass VLD1QDUPWB<bits<4> op7_4, string Dt> {
1101 def _fixed : NLdSt<1, 0b10, 0b1100, op7_4,
1102 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1103 (ins addrmode6dup:$Rn), IIC_VLD1dupu,
1104 "vld1", Dt, "$Vd, $Rn!",
1105 "$Rn.addr = $wb", []> {
1106 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1107 let Inst{4} = Rn{4};
1108 let DecoderMethod = "DecodeVLD1DupInstruction";
1109 let AsmMatchConverter = "cvtVLDwbFixed";
1111 def _register : NLdSt<1, 0b10, 0b1100, op7_4,
1112 (outs VecListTwoDAllLanes:$Vd, GPR:$wb),
1113 (ins addrmode6dup:$Rn, rGPR:$Rm), IIC_VLD1dupu,
1114 "vld1", Dt, "$Vd, $Rn, $Rm",
1115 "$Rn.addr = $wb", []> {
1116 let Inst{4} = Rn{4};
1117 let DecoderMethod = "DecodeVLD1DupInstruction";
1118 let AsmMatchConverter = "cvtVLDwbRegister";
1122 defm VLD1DUPd8wb : VLD1DUPWB<{0,0,0,0}, "8">;
1123 defm VLD1DUPd16wb : VLD1DUPWB<{0,1,0,?}, "16">;
1124 defm VLD1DUPd32wb : VLD1DUPWB<{1,0,0,?}, "32">;
1126 defm VLD1DUPq8wb : VLD1QDUPWB<{0,0,1,0}, "8">;
1127 defm VLD1DUPq16wb : VLD1QDUPWB<{0,1,1,?}, "16">;
1128 defm VLD1DUPq32wb : VLD1QDUPWB<{1,0,1,?}, "32">;
1130 def VLD1DUPq8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1131 def VLD1DUPq16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1132 def VLD1DUPq32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1dupu>;
1133 def VLD1DUPq8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1134 def VLD1DUPq16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1135 def VLD1DUPq32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1dupu>;
1137 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1138 class VLD2DUP<bits<4> op7_4, string Dt>
1139 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1140 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1141 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1143 let Inst{4} = Rn{4};
1144 let DecoderMethod = "DecodeVLD2DupInstruction";
1147 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1148 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1149 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1151 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1152 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1153 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1155 // ...with double-spaced registers (not used for codegen):
1156 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1157 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1158 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1160 // ...with address register writeback:
1161 class VLD2DUPWB<bits<4> op7_4, string Dt>
1162 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1163 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1164 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1165 let Inst{4} = Rn{4};
1166 let DecoderMethod = "DecodeVLD2DupInstruction";
1169 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1170 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1171 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1173 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1174 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1175 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1177 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1178 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1179 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1181 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1182 class VLD3DUP<bits<4> op7_4, string Dt>
1183 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1184 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1185 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1188 let DecoderMethod = "DecodeVLD3DupInstruction";
1191 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1192 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1193 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1195 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1196 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1197 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1199 // ...with double-spaced registers (not used for codegen):
1200 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1201 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1202 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1204 // ...with address register writeback:
1205 class VLD3DUPWB<bits<4> op7_4, string Dt>
1206 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1207 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1208 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1209 "$Rn.addr = $wb", []> {
1211 let DecoderMethod = "DecodeVLD3DupInstruction";
1214 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1215 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1216 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1218 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1219 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1220 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1222 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1223 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1224 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1226 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1227 class VLD4DUP<bits<4> op7_4, string Dt>
1228 : NLdSt<1, 0b10, 0b1111, op7_4,
1229 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1230 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1231 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1233 let Inst{4} = Rn{4};
1234 let DecoderMethod = "DecodeVLD4DupInstruction";
1237 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1238 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1239 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1241 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1242 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1243 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1245 // ...with double-spaced registers (not used for codegen):
1246 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1247 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1248 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1250 // ...with address register writeback:
1251 class VLD4DUPWB<bits<4> op7_4, string Dt>
1252 : NLdSt<1, 0b10, 0b1111, op7_4,
1253 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1254 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1255 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1256 "$Rn.addr = $wb", []> {
1257 let Inst{4} = Rn{4};
1258 let DecoderMethod = "DecodeVLD4DupInstruction";
1261 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1262 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1263 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1265 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1266 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1267 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1269 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1270 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1271 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1273 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1275 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1277 // Classes for VST* pseudo-instructions with multi-register operands.
1278 // These are expanded to real instructions after register allocation.
1279 class VSTQPseudo<InstrItinClass itin>
1280 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1281 class VSTQWBPseudo<InstrItinClass itin>
1282 : PseudoNLdSt<(outs GPR:$wb),
1283 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1284 "$addr.addr = $wb">;
1285 class VSTQWBfixedPseudo<InstrItinClass itin>
1286 : PseudoNLdSt<(outs GPR:$wb),
1287 (ins addrmode6:$addr, QPR:$src), itin,
1288 "$addr.addr = $wb">;
1289 class VSTQWBregisterPseudo<InstrItinClass itin>
1290 : PseudoNLdSt<(outs GPR:$wb),
1291 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1292 "$addr.addr = $wb">;
1293 class VSTQQPseudo<InstrItinClass itin>
1294 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1295 class VSTQQWBPseudo<InstrItinClass itin>
1296 : PseudoNLdSt<(outs GPR:$wb),
1297 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1298 "$addr.addr = $wb">;
1299 class VSTQQQQPseudo<InstrItinClass itin>
1300 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1301 class VSTQQQQWBPseudo<InstrItinClass itin>
1302 : PseudoNLdSt<(outs GPR:$wb),
1303 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1304 "$addr.addr = $wb">;
1306 // VST1 : Vector Store (multiple single elements)
1307 class VST1D<bits<4> op7_4, string Dt>
1308 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1309 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1311 let Inst{4} = Rn{4};
1312 let DecoderMethod = "DecodeVSTInstruction";
1314 class VST1Q<bits<4> op7_4, string Dt>
1315 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1316 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1318 let Inst{5-4} = Rn{5-4};
1319 let DecoderMethod = "DecodeVSTInstruction";
1322 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1323 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1324 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1325 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1327 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1328 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1329 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1330 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1332 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1333 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1334 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1335 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1337 // ...with address register writeback:
1338 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1339 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1340 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1341 "vst1", Dt, "$Vd, $Rn!",
1342 "$Rn.addr = $wb", []> {
1343 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1344 let Inst{4} = Rn{4};
1345 let DecoderMethod = "DecodeVSTInstruction";
1346 let AsmMatchConverter = "cvtVSTwbFixed";
1348 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1349 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1351 "vst1", Dt, "$Vd, $Rn, $Rm",
1352 "$Rn.addr = $wb", []> {
1353 let Inst{4} = Rn{4};
1354 let DecoderMethod = "DecodeVSTInstruction";
1355 let AsmMatchConverter = "cvtVSTwbRegister";
1358 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1359 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1360 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1361 "vst1", Dt, "$Vd, $Rn!",
1362 "$Rn.addr = $wb", []> {
1363 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1364 let Inst{5-4} = Rn{5-4};
1365 let DecoderMethod = "DecodeVSTInstruction";
1366 let AsmMatchConverter = "cvtVSTwbFixed";
1368 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1369 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1371 "vst1", Dt, "$Vd, $Rn, $Rm",
1372 "$Rn.addr = $wb", []> {
1373 let Inst{5-4} = Rn{5-4};
1374 let DecoderMethod = "DecodeVSTInstruction";
1375 let AsmMatchConverter = "cvtVSTwbRegister";
1379 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1380 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1381 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1382 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1384 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1385 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1386 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1387 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1389 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1390 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1391 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1392 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1393 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1394 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1395 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1396 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1398 // ...with 3 registers
1399 class VST1D3<bits<4> op7_4, string Dt>
1400 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1401 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1402 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1404 let Inst{4} = Rn{4};
1405 let DecoderMethod = "DecodeVSTInstruction";
1407 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1408 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1409 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1410 "vst1", Dt, "$Vd, $Rn!",
1411 "$Rn.addr = $wb", []> {
1412 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1413 let Inst{5-4} = Rn{5-4};
1414 let DecoderMethod = "DecodeVSTInstruction";
1415 let AsmMatchConverter = "cvtVSTwbFixed";
1417 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1418 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1420 "vst1", Dt, "$Vd, $Rn, $Rm",
1421 "$Rn.addr = $wb", []> {
1422 let Inst{5-4} = Rn{5-4};
1423 let DecoderMethod = "DecodeVSTInstruction";
1424 let AsmMatchConverter = "cvtVSTwbRegister";
1428 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1429 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1430 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1431 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1433 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1434 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1435 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1436 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1438 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1439 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1440 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1442 // ...with 4 registers
1443 class VST1D4<bits<4> op7_4, string Dt>
1444 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1445 (ins addrmode6:$Rn, VecListFourD:$Vd),
1446 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1449 let Inst{5-4} = Rn{5-4};
1450 let DecoderMethod = "DecodeVSTInstruction";
1452 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1453 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1454 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1455 "vst1", Dt, "$Vd, $Rn!",
1456 "$Rn.addr = $wb", []> {
1457 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1458 let Inst{5-4} = Rn{5-4};
1459 let DecoderMethod = "DecodeVSTInstruction";
1460 let AsmMatchConverter = "cvtVSTwbFixed";
1462 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1463 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1465 "vst1", Dt, "$Vd, $Rn, $Rm",
1466 "$Rn.addr = $wb", []> {
1467 let Inst{5-4} = Rn{5-4};
1468 let DecoderMethod = "DecodeVSTInstruction";
1469 let AsmMatchConverter = "cvtVSTwbRegister";
1473 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1474 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1475 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1476 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1478 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1479 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1480 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1481 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1483 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1484 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1485 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1487 // VST2 : Vector Store (multiple 2-element structures)
1488 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1489 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1490 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1491 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1493 let Inst{5-4} = Rn{5-4};
1494 let DecoderMethod = "DecodeVSTInstruction";
1496 class VST2Q<bits<4> op7_4, string Dt>
1497 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1498 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1499 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1502 let Inst{5-4} = Rn{5-4};
1503 let DecoderMethod = "DecodeVSTInstruction";
1506 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1507 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1508 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1510 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1511 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1512 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1514 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1515 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1516 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1518 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1519 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1520 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1522 // ...with address register writeback:
1523 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1524 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1525 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1526 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1527 "$Rn.addr = $wb", []> {
1528 let Inst{5-4} = Rn{5-4};
1529 let DecoderMethod = "DecodeVSTInstruction";
1531 class VST2QWB<bits<4> op7_4, string Dt>
1532 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1533 (ins addrmode6:$Rn, am6offset:$Rm,
1534 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1535 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1536 "$Rn.addr = $wb", []> {
1537 let Inst{5-4} = Rn{5-4};
1538 let DecoderMethod = "DecodeVSTInstruction";
1541 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1542 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1543 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1545 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1546 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1547 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1549 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1550 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1551 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1553 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1554 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1555 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1557 // ...with double-spaced registers
1558 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1559 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1560 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1561 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1562 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1563 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1565 // VST3 : Vector Store (multiple 3-element structures)
1566 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1567 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1568 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1569 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1571 let Inst{4} = Rn{4};
1572 let DecoderMethod = "DecodeVSTInstruction";
1575 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1576 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1577 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1579 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1580 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1581 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1583 // ...with address register writeback:
1584 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1585 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1586 (ins addrmode6:$Rn, am6offset:$Rm,
1587 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1588 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1589 "$Rn.addr = $wb", []> {
1590 let Inst{4} = Rn{4};
1591 let DecoderMethod = "DecodeVSTInstruction";
1594 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1595 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1596 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1598 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1599 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1600 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1602 // ...with double-spaced registers:
1603 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1604 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1605 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1606 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1607 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1608 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1610 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1611 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1612 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1614 // ...alternate versions to be allocated odd register numbers:
1615 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1616 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1617 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1619 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1620 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1621 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1623 // VST4 : Vector Store (multiple 4-element structures)
1624 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1625 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1626 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1627 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1630 let Inst{5-4} = Rn{5-4};
1631 let DecoderMethod = "DecodeVSTInstruction";
1634 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1635 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1636 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1638 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1639 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1640 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1642 // ...with address register writeback:
1643 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1644 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1645 (ins addrmode6:$Rn, am6offset:$Rm,
1646 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1647 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1648 "$Rn.addr = $wb", []> {
1649 let Inst{5-4} = Rn{5-4};
1650 let DecoderMethod = "DecodeVSTInstruction";
1653 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1654 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1655 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1657 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1658 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1659 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1661 // ...with double-spaced registers:
1662 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1663 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1664 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1665 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1666 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1667 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1669 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1670 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1671 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1673 // ...alternate versions to be allocated odd register numbers:
1674 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1675 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1676 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1678 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1679 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1680 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1682 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1684 // Classes for VST*LN pseudo-instructions with multi-register operands.
1685 // These are expanded to real instructions after register allocation.
1686 class VSTQLNPseudo<InstrItinClass itin>
1687 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1689 class VSTQLNWBPseudo<InstrItinClass itin>
1690 : PseudoNLdSt<(outs GPR:$wb),
1691 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1692 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1693 class VSTQQLNPseudo<InstrItinClass itin>
1694 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1696 class VSTQQLNWBPseudo<InstrItinClass itin>
1697 : PseudoNLdSt<(outs GPR:$wb),
1698 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1699 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1700 class VSTQQQQLNPseudo<InstrItinClass itin>
1701 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1703 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1704 : PseudoNLdSt<(outs GPR:$wb),
1705 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1706 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1708 // VST1LN : Vector Store (single element from one lane)
1709 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1710 PatFrag StoreOp, SDNode ExtractOp>
1711 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1712 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1713 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1714 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1716 let DecoderMethod = "DecodeVST1LN";
1718 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1719 PatFrag StoreOp, SDNode ExtractOp>
1720 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1721 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1722 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1723 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1725 let DecoderMethod = "DecodeVST1LN";
1727 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1728 : VSTQLNPseudo<IIC_VST1ln> {
1729 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1733 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1735 let Inst{7-5} = lane{2-0};
1737 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1739 let Inst{7-6} = lane{1-0};
1740 let Inst{4} = Rn{5};
1743 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1744 let Inst{7} = lane{0};
1745 let Inst{5-4} = Rn{5-4};
1748 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1749 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1750 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1752 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1753 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1754 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1755 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1757 // ...with address register writeback:
1758 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1759 PatFrag StoreOp, SDNode ExtractOp>
1760 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1761 (ins addrmode6:$Rn, am6offset:$Rm,
1762 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1763 "\\{$Vd[$lane]\\}, $Rn$Rm",
1765 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1766 addrmode6:$Rn, am6offset:$Rm))]> {
1767 let DecoderMethod = "DecodeVST1LN";
1769 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1770 : VSTQLNWBPseudo<IIC_VST1lnu> {
1771 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1772 addrmode6:$addr, am6offset:$offset))];
1775 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1777 let Inst{7-5} = lane{2-0};
1779 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1781 let Inst{7-6} = lane{1-0};
1782 let Inst{4} = Rn{5};
1784 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1786 let Inst{7} = lane{0};
1787 let Inst{5-4} = Rn{5-4};
1790 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1791 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1792 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1794 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1796 // VST2LN : Vector Store (single 2-element structure from one lane)
1797 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1798 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1799 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1800 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1803 let Inst{4} = Rn{4};
1804 let DecoderMethod = "DecodeVST2LN";
1807 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1808 let Inst{7-5} = lane{2-0};
1810 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1811 let Inst{7-6} = lane{1-0};
1813 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1814 let Inst{7} = lane{0};
1817 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1818 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1819 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1821 // ...with double-spaced registers:
1822 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1823 let Inst{7-6} = lane{1-0};
1824 let Inst{4} = Rn{4};
1826 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1827 let Inst{7} = lane{0};
1828 let Inst{4} = Rn{4};
1831 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1832 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1834 // ...with address register writeback:
1835 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1836 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1837 (ins addrmode6:$addr, am6offset:$offset,
1838 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1839 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1840 "$addr.addr = $wb", []> {
1841 let Inst{4} = Rn{4};
1842 let DecoderMethod = "DecodeVST2LN";
1845 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1846 let Inst{7-5} = lane{2-0};
1848 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1849 let Inst{7-6} = lane{1-0};
1851 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1852 let Inst{7} = lane{0};
1855 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1856 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1857 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1859 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1860 let Inst{7-6} = lane{1-0};
1862 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1863 let Inst{7} = lane{0};
1866 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1867 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1869 // VST3LN : Vector Store (single 3-element structure from one lane)
1870 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1871 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1872 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1873 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1874 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1876 let DecoderMethod = "DecodeVST3LN";
1879 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1880 let Inst{7-5} = lane{2-0};
1882 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1883 let Inst{7-6} = lane{1-0};
1885 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1886 let Inst{7} = lane{0};
1889 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1890 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1891 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1893 // ...with double-spaced registers:
1894 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1895 let Inst{7-6} = lane{1-0};
1897 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1898 let Inst{7} = lane{0};
1901 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1902 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1904 // ...with address register writeback:
1905 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1906 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1907 (ins addrmode6:$Rn, am6offset:$Rm,
1908 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1909 IIC_VST3lnu, "vst3", Dt,
1910 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1911 "$Rn.addr = $wb", []> {
1912 let DecoderMethod = "DecodeVST3LN";
1915 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1916 let Inst{7-5} = lane{2-0};
1918 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1919 let Inst{7-6} = lane{1-0};
1921 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1922 let Inst{7} = lane{0};
1925 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1926 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1927 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1929 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1930 let Inst{7-6} = lane{1-0};
1932 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1933 let Inst{7} = lane{0};
1936 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1937 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1939 // VST4LN : Vector Store (single 4-element structure from one lane)
1940 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1941 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1942 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1943 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1944 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1947 let Inst{4} = Rn{4};
1948 let DecoderMethod = "DecodeVST4LN";
1951 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1952 let Inst{7-5} = lane{2-0};
1954 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1955 let Inst{7-6} = lane{1-0};
1957 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1958 let Inst{7} = lane{0};
1959 let Inst{5} = Rn{5};
1962 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1963 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1964 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1966 // ...with double-spaced registers:
1967 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1968 let Inst{7-6} = lane{1-0};
1970 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1971 let Inst{7} = lane{0};
1972 let Inst{5} = Rn{5};
1975 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1976 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1978 // ...with address register writeback:
1979 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1980 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1981 (ins addrmode6:$Rn, am6offset:$Rm,
1982 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1983 IIC_VST4lnu, "vst4", Dt,
1984 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1985 "$Rn.addr = $wb", []> {
1986 let Inst{4} = Rn{4};
1987 let DecoderMethod = "DecodeVST4LN";
1990 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1991 let Inst{7-5} = lane{2-0};
1993 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1994 let Inst{7-6} = lane{1-0};
1996 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1997 let Inst{7} = lane{0};
1998 let Inst{5} = Rn{5};
2001 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2002 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2003 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
2005 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
2006 let Inst{7-6} = lane{1-0};
2008 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
2009 let Inst{7} = lane{0};
2010 let Inst{5} = Rn{5};
2013 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2014 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
2016 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
2019 //===----------------------------------------------------------------------===//
2020 // NEON pattern fragments
2021 //===----------------------------------------------------------------------===//
2023 // Extract D sub-registers of Q registers.
2024 def DSubReg_i8_reg : SDNodeXForm<imm, [{
2025 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2026 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
2028 def DSubReg_i16_reg : SDNodeXForm<imm, [{
2029 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2030 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
2032 def DSubReg_i32_reg : SDNodeXForm<imm, [{
2033 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2034 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
2036 def DSubReg_f64_reg : SDNodeXForm<imm, [{
2037 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
2038 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
2041 // Extract S sub-registers of Q/D registers.
2042 def SSubReg_f32_reg : SDNodeXForm<imm, [{
2043 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
2044 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
2047 // Translate lane numbers from Q registers to D subregs.
2048 def SubReg_i8_lane : SDNodeXForm<imm, [{
2049 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
2051 def SubReg_i16_lane : SDNodeXForm<imm, [{
2052 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
2054 def SubReg_i32_lane : SDNodeXForm<imm, [{
2055 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
2058 //===----------------------------------------------------------------------===//
2059 // Instruction Classes
2060 //===----------------------------------------------------------------------===//
2062 // Basic 2-register operations: double- and quad-register.
2063 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2064 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2065 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2066 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2067 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2068 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2069 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2070 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2071 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2072 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2073 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2074 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2076 // Basic 2-register intrinsics, both double- and quad-register.
2077 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2078 bits<2> op17_16, bits<5> op11_7, bit op4,
2079 InstrItinClass itin, string OpcodeStr, string Dt,
2080 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2081 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2082 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2083 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2084 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2085 bits<2> op17_16, bits<5> op11_7, bit op4,
2086 InstrItinClass itin, string OpcodeStr, string Dt,
2087 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2088 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2089 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2090 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2092 // Narrow 2-register operations.
2093 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2094 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2095 InstrItinClass itin, string OpcodeStr, string Dt,
2096 ValueType TyD, ValueType TyQ, SDNode OpNode>
2097 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2098 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2099 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2101 // Narrow 2-register intrinsics.
2102 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2103 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2104 InstrItinClass itin, string OpcodeStr, string Dt,
2105 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2106 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2107 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2108 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2110 // Long 2-register operations (currently only used for VMOVL).
2111 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2112 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2113 InstrItinClass itin, string OpcodeStr, string Dt,
2114 ValueType TyQ, ValueType TyD, SDNode OpNode>
2115 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2116 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2117 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2119 // Long 2-register intrinsics.
2120 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2121 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2122 InstrItinClass itin, string OpcodeStr, string Dt,
2123 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2124 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2125 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2126 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2128 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2129 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2130 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2131 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2132 OpcodeStr, Dt, "$Vd, $Vm",
2133 "$src1 = $Vd, $src2 = $Vm", []>;
2134 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2135 InstrItinClass itin, string OpcodeStr, string Dt>
2136 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2137 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2138 "$src1 = $Vd, $src2 = $Vm", []>;
2140 // Basic 3-register operations: double- and quad-register.
2141 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2142 InstrItinClass itin, string OpcodeStr, string Dt,
2143 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2144 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2145 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2146 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2147 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2148 let isCommutable = Commutable;
2150 // Same as N3VD but no data type.
2151 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2152 InstrItinClass itin, string OpcodeStr,
2153 ValueType ResTy, ValueType OpTy,
2154 SDNode OpNode, bit Commutable>
2155 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2156 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2157 OpcodeStr, "$Vd, $Vn, $Vm", "",
2158 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2159 let isCommutable = Commutable;
2162 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2163 InstrItinClass itin, string OpcodeStr, string Dt,
2164 ValueType Ty, SDNode ShOp>
2165 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2166 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2167 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2169 (Ty (ShOp (Ty DPR:$Vn),
2170 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2171 let isCommutable = 0;
2173 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2174 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2175 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2176 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2177 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2179 (Ty (ShOp (Ty DPR:$Vn),
2180 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2181 let isCommutable = 0;
2184 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2185 InstrItinClass itin, string OpcodeStr, string Dt,
2186 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2187 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2188 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2189 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2190 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2191 let isCommutable = Commutable;
2193 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2194 InstrItinClass itin, string OpcodeStr,
2195 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2196 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2197 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2198 OpcodeStr, "$Vd, $Vn, $Vm", "",
2199 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2200 let isCommutable = Commutable;
2202 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2203 InstrItinClass itin, string OpcodeStr, string Dt,
2204 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2205 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2206 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2207 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2208 [(set (ResTy QPR:$Vd),
2209 (ResTy (ShOp (ResTy QPR:$Vn),
2210 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2212 let isCommutable = 0;
2214 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2215 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2216 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2217 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2218 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2219 [(set (ResTy QPR:$Vd),
2220 (ResTy (ShOp (ResTy QPR:$Vn),
2221 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2223 let isCommutable = 0;
2226 // Basic 3-register intrinsics, both double- and quad-register.
2227 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2228 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2229 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2230 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2231 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2232 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2233 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2234 let isCommutable = Commutable;
2236 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2237 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2238 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2239 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2240 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2242 (Ty (IntOp (Ty DPR:$Vn),
2243 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2245 let isCommutable = 0;
2247 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2248 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2249 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2250 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2251 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2253 (Ty (IntOp (Ty DPR:$Vn),
2254 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2255 let isCommutable = 0;
2257 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2258 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2259 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2260 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2261 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2262 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2263 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2264 let isCommutable = 0;
2267 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2268 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2269 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2270 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2271 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2272 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2273 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2274 let isCommutable = Commutable;
2276 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2277 string OpcodeStr, string Dt,
2278 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2279 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2280 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2281 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2282 [(set (ResTy QPR:$Vd),
2283 (ResTy (IntOp (ResTy QPR:$Vn),
2284 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2286 let isCommutable = 0;
2288 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2289 string OpcodeStr, string Dt,
2290 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2291 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2292 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2293 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2294 [(set (ResTy QPR:$Vd),
2295 (ResTy (IntOp (ResTy QPR:$Vn),
2296 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2298 let isCommutable = 0;
2300 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2301 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2302 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2303 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2304 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2305 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2306 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2307 let isCommutable = 0;
2310 // Multiply-Add/Sub operations: double- and quad-register.
2311 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2312 InstrItinClass itin, string OpcodeStr, string Dt,
2313 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2314 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2315 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2316 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2317 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2318 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2320 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2321 string OpcodeStr, string Dt,
2322 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2323 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2325 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2327 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2329 (Ty (ShOp (Ty DPR:$src1),
2331 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2333 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2334 string OpcodeStr, string Dt,
2335 ValueType Ty, SDNode MulOp, SDNode ShOp>
2336 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2338 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2340 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2342 (Ty (ShOp (Ty DPR:$src1),
2344 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2347 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2348 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2349 SDPatternOperator MulOp, SDPatternOperator OpNode>
2350 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2351 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2352 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2353 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2354 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2355 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2356 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2357 SDPatternOperator MulOp, SDPatternOperator ShOp>
2358 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2360 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2362 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2363 [(set (ResTy QPR:$Vd),
2364 (ResTy (ShOp (ResTy QPR:$src1),
2365 (ResTy (MulOp QPR:$Vn,
2366 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2368 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2369 string OpcodeStr, string Dt,
2370 ValueType ResTy, ValueType OpTy,
2371 SDNode MulOp, SDNode ShOp>
2372 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2374 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2376 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2377 [(set (ResTy QPR:$Vd),
2378 (ResTy (ShOp (ResTy QPR:$src1),
2379 (ResTy (MulOp QPR:$Vn,
2380 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2383 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2384 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2385 InstrItinClass itin, string OpcodeStr, string Dt,
2386 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2387 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2388 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2389 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2390 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2391 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2392 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2393 InstrItinClass itin, string OpcodeStr, string Dt,
2394 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2395 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2396 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2397 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2398 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2399 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2401 // Neon 3-argument intrinsics, both double- and quad-register.
2402 // The destination register is also used as the first source operand register.
2403 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2404 InstrItinClass itin, string OpcodeStr, string Dt,
2405 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2406 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2407 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2408 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2409 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2410 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2411 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2412 InstrItinClass itin, string OpcodeStr, string Dt,
2413 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2414 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2415 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2416 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2417 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2418 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2420 // Long Multiply-Add/Sub operations.
2421 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2422 InstrItinClass itin, string OpcodeStr, string Dt,
2423 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2424 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2425 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2426 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2427 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2428 (TyQ (MulOp (TyD DPR:$Vn),
2429 (TyD DPR:$Vm)))))]>;
2430 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2431 InstrItinClass itin, string OpcodeStr, string Dt,
2432 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2433 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2434 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2436 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2438 (OpNode (TyQ QPR:$src1),
2439 (TyQ (MulOp (TyD DPR:$Vn),
2440 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2442 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2443 InstrItinClass itin, string OpcodeStr, string Dt,
2444 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2445 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2446 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2448 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2450 (OpNode (TyQ QPR:$src1),
2451 (TyQ (MulOp (TyD DPR:$Vn),
2452 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2455 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2456 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2457 InstrItinClass itin, string OpcodeStr, string Dt,
2458 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2460 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2461 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2462 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2463 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2464 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2465 (TyD DPR:$Vm)))))))]>;
2467 // Neon Long 3-argument intrinsic. The destination register is
2468 // a quad-register and is also used as the first source operand register.
2469 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2470 InstrItinClass itin, string OpcodeStr, string Dt,
2471 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2472 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2473 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2474 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2476 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2477 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2478 string OpcodeStr, string Dt,
2479 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2480 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2482 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2484 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2485 [(set (ResTy QPR:$Vd),
2486 (ResTy (IntOp (ResTy QPR:$src1),
2488 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2490 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2491 InstrItinClass itin, string OpcodeStr, string Dt,
2492 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2493 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2495 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2497 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2498 [(set (ResTy QPR:$Vd),
2499 (ResTy (IntOp (ResTy QPR:$src1),
2501 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2504 // Narrowing 3-register intrinsics.
2505 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2506 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2507 Intrinsic IntOp, bit Commutable>
2508 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2509 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2510 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2511 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2512 let isCommutable = Commutable;
2515 // Long 3-register operations.
2516 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2517 InstrItinClass itin, string OpcodeStr, string Dt,
2518 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2519 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2520 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2521 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2522 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2523 let isCommutable = Commutable;
2525 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2526 InstrItinClass itin, string OpcodeStr, string Dt,
2527 ValueType TyQ, ValueType TyD, SDNode OpNode>
2528 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2529 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2530 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2532 (TyQ (OpNode (TyD DPR:$Vn),
2533 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2534 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2535 InstrItinClass itin, string OpcodeStr, string Dt,
2536 ValueType TyQ, ValueType TyD, SDNode OpNode>
2537 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2538 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2539 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2541 (TyQ (OpNode (TyD DPR:$Vn),
2542 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2544 // Long 3-register operations with explicitly extended operands.
2545 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2546 InstrItinClass itin, string OpcodeStr, string Dt,
2547 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2549 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2550 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2551 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2552 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2553 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2554 let isCommutable = Commutable;
2557 // Long 3-register intrinsics with explicit extend (VABDL).
2558 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2559 InstrItinClass itin, string OpcodeStr, string Dt,
2560 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2562 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2563 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2564 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2565 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2566 (TyD DPR:$Vm))))))]> {
2567 let isCommutable = Commutable;
2570 // Long 3-register intrinsics.
2571 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2572 InstrItinClass itin, string OpcodeStr, string Dt,
2573 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2574 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2575 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2576 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2577 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2578 let isCommutable = Commutable;
2580 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2581 string OpcodeStr, string Dt,
2582 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2583 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2584 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2585 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2586 [(set (ResTy QPR:$Vd),
2587 (ResTy (IntOp (OpTy DPR:$Vn),
2588 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2590 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2591 InstrItinClass itin, string OpcodeStr, string Dt,
2592 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2593 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2594 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2595 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2596 [(set (ResTy QPR:$Vd),
2597 (ResTy (IntOp (OpTy DPR:$Vn),
2598 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2601 // Wide 3-register operations.
2602 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2603 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2604 SDNode OpNode, SDNode ExtOp, bit Commutable>
2605 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2606 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2607 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2608 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2609 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2610 let isCommutable = Commutable;
2613 // Pairwise long 2-register intrinsics, both double- and quad-register.
2614 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2615 bits<2> op17_16, bits<5> op11_7, bit op4,
2616 string OpcodeStr, string Dt,
2617 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2618 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2619 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2620 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2621 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2622 bits<2> op17_16, bits<5> op11_7, bit op4,
2623 string OpcodeStr, string Dt,
2624 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2625 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2626 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2627 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2629 // Pairwise long 2-register accumulate intrinsics,
2630 // both double- and quad-register.
2631 // The destination register is also used as the first source operand register.
2632 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2633 bits<2> op17_16, bits<5> op11_7, bit op4,
2634 string OpcodeStr, string Dt,
2635 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2636 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2637 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2638 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2639 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2640 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2641 bits<2> op17_16, bits<5> op11_7, bit op4,
2642 string OpcodeStr, string Dt,
2643 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2644 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2645 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2646 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2647 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2649 // Shift by immediate,
2650 // both double- and quad-register.
2651 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2652 Format f, InstrItinClass itin, Operand ImmTy,
2653 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2654 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2655 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2656 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2657 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2658 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2659 Format f, InstrItinClass itin, Operand ImmTy,
2660 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2661 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2662 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2663 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2664 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2666 // Long shift by immediate.
2667 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2668 string OpcodeStr, string Dt,
2669 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2670 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2671 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2672 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2673 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2674 (i32 imm:$SIMM))))]>;
2676 // Narrow shift by immediate.
2677 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2678 InstrItinClass itin, string OpcodeStr, string Dt,
2679 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2680 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2681 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2682 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2683 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2684 (i32 imm:$SIMM))))]>;
2686 // Shift right by immediate and accumulate,
2687 // both double- and quad-register.
2688 class N2VDShAdd<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, 0, op4, (outs DPR:$Vd),
2692 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2693 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2694 [(set DPR:$Vd, (Ty (add DPR:$src1,
2695 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2696 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2697 Operand ImmTy, string OpcodeStr, string Dt,
2698 ValueType Ty, SDNode ShOp>
2699 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2700 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2701 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2702 [(set QPR:$Vd, (Ty (add QPR:$src1,
2703 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2705 // Shift by immediate and insert,
2706 // both double- and quad-register.
2707 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2708 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2709 ValueType Ty,SDNode ShOp>
2710 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2711 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2712 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2713 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2714 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2715 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2716 ValueType Ty,SDNode ShOp>
2717 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2718 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2719 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2720 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2722 // Convert, with fractional bits immediate,
2723 // both double- and quad-register.
2724 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2725 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2727 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2728 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2729 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2730 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2731 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2732 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2734 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2735 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2736 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2737 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2739 //===----------------------------------------------------------------------===//
2741 //===----------------------------------------------------------------------===//
2743 // Abbreviations used in multiclass suffixes:
2744 // Q = quarter int (8 bit) elements
2745 // H = half int (16 bit) elements
2746 // S = single int (32 bit) elements
2747 // D = double int (64 bit) elements
2749 // Neon 2-register vector operations and intrinsics.
2751 // Neon 2-register comparisons.
2752 // source operand element sizes of 8, 16 and 32 bits:
2753 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2754 bits<5> op11_7, bit op4, string opc, string Dt,
2755 string asm, SDNode OpNode> {
2756 // 64-bit vector types.
2757 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2758 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2759 opc, !strconcat(Dt, "8"), asm, "",
2760 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2761 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2762 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2763 opc, !strconcat(Dt, "16"), asm, "",
2764 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2765 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2766 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2767 opc, !strconcat(Dt, "32"), asm, "",
2768 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2769 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2770 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2771 opc, "f32", asm, "",
2772 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2773 let Inst{10} = 1; // overwrite F = 1
2776 // 128-bit vector types.
2777 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2778 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2779 opc, !strconcat(Dt, "8"), asm, "",
2780 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2781 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2782 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2783 opc, !strconcat(Dt, "16"), asm, "",
2784 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2785 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2786 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2787 opc, !strconcat(Dt, "32"), asm, "",
2788 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2789 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2790 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2791 opc, "f32", asm, "",
2792 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2793 let Inst{10} = 1; // overwrite F = 1
2798 // Neon 2-register vector intrinsics,
2799 // element sizes of 8, 16 and 32 bits:
2800 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2801 bits<5> op11_7, bit op4,
2802 InstrItinClass itinD, InstrItinClass itinQ,
2803 string OpcodeStr, string Dt, Intrinsic IntOp> {
2804 // 64-bit vector types.
2805 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2806 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2807 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2808 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2809 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2810 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2812 // 128-bit vector types.
2813 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2814 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2815 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2816 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2817 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2818 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2822 // Neon Narrowing 2-register vector operations,
2823 // source operand element sizes of 16, 32 and 64 bits:
2824 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2825 bits<5> op11_7, bit op6, bit op4,
2826 InstrItinClass itin, string OpcodeStr, string Dt,
2828 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2829 itin, OpcodeStr, !strconcat(Dt, "16"),
2830 v8i8, v8i16, OpNode>;
2831 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2832 itin, OpcodeStr, !strconcat(Dt, "32"),
2833 v4i16, v4i32, OpNode>;
2834 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2835 itin, OpcodeStr, !strconcat(Dt, "64"),
2836 v2i32, v2i64, OpNode>;
2839 // Neon Narrowing 2-register vector intrinsics,
2840 // source operand element sizes of 16, 32 and 64 bits:
2841 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2842 bits<5> op11_7, bit op6, bit op4,
2843 InstrItinClass itin, string OpcodeStr, string Dt,
2845 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2846 itin, OpcodeStr, !strconcat(Dt, "16"),
2847 v8i8, v8i16, IntOp>;
2848 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2849 itin, OpcodeStr, !strconcat(Dt, "32"),
2850 v4i16, v4i32, IntOp>;
2851 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2852 itin, OpcodeStr, !strconcat(Dt, "64"),
2853 v2i32, v2i64, IntOp>;
2857 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2858 // source operand element sizes of 16, 32 and 64 bits:
2859 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2860 string OpcodeStr, string Dt, SDNode OpNode> {
2861 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2862 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2863 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2864 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2865 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2866 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2870 // Neon 3-register vector operations.
2872 // First with only element sizes of 8, 16 and 32 bits:
2873 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2874 InstrItinClass itinD16, InstrItinClass itinD32,
2875 InstrItinClass itinQ16, InstrItinClass itinQ32,
2876 string OpcodeStr, string Dt,
2877 SDNode OpNode, bit Commutable = 0> {
2878 // 64-bit vector types.
2879 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2880 OpcodeStr, !strconcat(Dt, "8"),
2881 v8i8, v8i8, OpNode, Commutable>;
2882 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2883 OpcodeStr, !strconcat(Dt, "16"),
2884 v4i16, v4i16, OpNode, Commutable>;
2885 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2886 OpcodeStr, !strconcat(Dt, "32"),
2887 v2i32, v2i32, OpNode, Commutable>;
2889 // 128-bit vector types.
2890 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2891 OpcodeStr, !strconcat(Dt, "8"),
2892 v16i8, v16i8, OpNode, Commutable>;
2893 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2894 OpcodeStr, !strconcat(Dt, "16"),
2895 v8i16, v8i16, OpNode, Commutable>;
2896 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2897 OpcodeStr, !strconcat(Dt, "32"),
2898 v4i32, v4i32, OpNode, Commutable>;
2901 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2902 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2904 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2906 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2907 v8i16, v4i16, ShOp>;
2908 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2909 v4i32, v2i32, ShOp>;
2912 // ....then also with element size 64 bits:
2913 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2914 InstrItinClass itinD, InstrItinClass itinQ,
2915 string OpcodeStr, string Dt,
2916 SDNode OpNode, bit Commutable = 0>
2917 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2918 OpcodeStr, Dt, OpNode, Commutable> {
2919 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2920 OpcodeStr, !strconcat(Dt, "64"),
2921 v1i64, v1i64, OpNode, Commutable>;
2922 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2923 OpcodeStr, !strconcat(Dt, "64"),
2924 v2i64, v2i64, OpNode, Commutable>;
2928 // Neon 3-register vector intrinsics.
2930 // First with only element sizes of 16 and 32 bits:
2931 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2932 InstrItinClass itinD16, InstrItinClass itinD32,
2933 InstrItinClass itinQ16, InstrItinClass itinQ32,
2934 string OpcodeStr, string Dt,
2935 Intrinsic IntOp, bit Commutable = 0> {
2936 // 64-bit vector types.
2937 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2938 OpcodeStr, !strconcat(Dt, "16"),
2939 v4i16, v4i16, IntOp, Commutable>;
2940 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2941 OpcodeStr, !strconcat(Dt, "32"),
2942 v2i32, v2i32, IntOp, Commutable>;
2944 // 128-bit vector types.
2945 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2946 OpcodeStr, !strconcat(Dt, "16"),
2947 v8i16, v8i16, IntOp, Commutable>;
2948 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2949 OpcodeStr, !strconcat(Dt, "32"),
2950 v4i32, v4i32, IntOp, Commutable>;
2952 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2953 InstrItinClass itinD16, InstrItinClass itinD32,
2954 InstrItinClass itinQ16, InstrItinClass itinQ32,
2955 string OpcodeStr, string Dt,
2957 // 64-bit vector types.
2958 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2959 OpcodeStr, !strconcat(Dt, "16"),
2960 v4i16, v4i16, IntOp>;
2961 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2962 OpcodeStr, !strconcat(Dt, "32"),
2963 v2i32, v2i32, IntOp>;
2965 // 128-bit vector types.
2966 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2967 OpcodeStr, !strconcat(Dt, "16"),
2968 v8i16, v8i16, IntOp>;
2969 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2970 OpcodeStr, !strconcat(Dt, "32"),
2971 v4i32, v4i32, IntOp>;
2974 multiclass N3VIntSL_HS<bits<4> op11_8,
2975 InstrItinClass itinD16, InstrItinClass itinD32,
2976 InstrItinClass itinQ16, InstrItinClass itinQ32,
2977 string OpcodeStr, string Dt, Intrinsic IntOp> {
2978 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2979 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2980 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2981 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2982 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2983 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2984 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2985 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2988 // ....then also with element size of 8 bits:
2989 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2990 InstrItinClass itinD16, InstrItinClass itinD32,
2991 InstrItinClass itinQ16, InstrItinClass itinQ32,
2992 string OpcodeStr, string Dt,
2993 Intrinsic IntOp, bit Commutable = 0>
2994 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2995 OpcodeStr, Dt, IntOp, Commutable> {
2996 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2997 OpcodeStr, !strconcat(Dt, "8"),
2998 v8i8, v8i8, IntOp, Commutable>;
2999 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3000 OpcodeStr, !strconcat(Dt, "8"),
3001 v16i8, v16i8, IntOp, Commutable>;
3003 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3004 InstrItinClass itinD16, InstrItinClass itinD32,
3005 InstrItinClass itinQ16, InstrItinClass itinQ32,
3006 string OpcodeStr, string Dt,
3008 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3009 OpcodeStr, Dt, IntOp> {
3010 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
3011 OpcodeStr, !strconcat(Dt, "8"),
3013 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
3014 OpcodeStr, !strconcat(Dt, "8"),
3015 v16i8, v16i8, IntOp>;
3019 // ....then also with element size of 64 bits:
3020 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3021 InstrItinClass itinD16, InstrItinClass itinD32,
3022 InstrItinClass itinQ16, InstrItinClass itinQ32,
3023 string OpcodeStr, string Dt,
3024 Intrinsic IntOp, bit Commutable = 0>
3025 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3026 OpcodeStr, Dt, IntOp, Commutable> {
3027 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
3028 OpcodeStr, !strconcat(Dt, "64"),
3029 v1i64, v1i64, IntOp, Commutable>;
3030 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3031 OpcodeStr, !strconcat(Dt, "64"),
3032 v2i64, v2i64, IntOp, Commutable>;
3034 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
3035 InstrItinClass itinD16, InstrItinClass itinD32,
3036 InstrItinClass itinQ16, InstrItinClass itinQ32,
3037 string OpcodeStr, string Dt,
3039 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
3040 OpcodeStr, Dt, IntOp> {
3041 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
3042 OpcodeStr, !strconcat(Dt, "64"),
3043 v1i64, v1i64, IntOp>;
3044 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
3045 OpcodeStr, !strconcat(Dt, "64"),
3046 v2i64, v2i64, IntOp>;
3049 // Neon Narrowing 3-register vector intrinsics,
3050 // source operand element sizes of 16, 32 and 64 bits:
3051 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3052 string OpcodeStr, string Dt,
3053 Intrinsic IntOp, bit Commutable = 0> {
3054 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
3055 OpcodeStr, !strconcat(Dt, "16"),
3056 v8i8, v8i16, IntOp, Commutable>;
3057 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
3058 OpcodeStr, !strconcat(Dt, "32"),
3059 v4i16, v4i32, IntOp, Commutable>;
3060 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
3061 OpcodeStr, !strconcat(Dt, "64"),
3062 v2i32, v2i64, IntOp, Commutable>;
3066 // Neon Long 3-register vector operations.
3068 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3069 InstrItinClass itin16, InstrItinClass itin32,
3070 string OpcodeStr, string Dt,
3071 SDNode OpNode, bit Commutable = 0> {
3072 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3073 OpcodeStr, !strconcat(Dt, "8"),
3074 v8i16, v8i8, OpNode, Commutable>;
3075 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3076 OpcodeStr, !strconcat(Dt, "16"),
3077 v4i32, v4i16, OpNode, Commutable>;
3078 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3079 OpcodeStr, !strconcat(Dt, "32"),
3080 v2i64, v2i32, OpNode, Commutable>;
3083 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3084 InstrItinClass itin, string OpcodeStr, string Dt,
3086 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3087 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3088 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3089 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3092 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3093 InstrItinClass itin16, InstrItinClass itin32,
3094 string OpcodeStr, string Dt,
3095 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3096 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3097 OpcodeStr, !strconcat(Dt, "8"),
3098 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3099 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3100 OpcodeStr, !strconcat(Dt, "16"),
3101 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3102 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3103 OpcodeStr, !strconcat(Dt, "32"),
3104 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3107 // Neon Long 3-register vector intrinsics.
3109 // First with only element sizes of 16 and 32 bits:
3110 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3111 InstrItinClass itin16, InstrItinClass itin32,
3112 string OpcodeStr, string Dt,
3113 Intrinsic IntOp, bit Commutable = 0> {
3114 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3115 OpcodeStr, !strconcat(Dt, "16"),
3116 v4i32, v4i16, IntOp, Commutable>;
3117 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3118 OpcodeStr, !strconcat(Dt, "32"),
3119 v2i64, v2i32, IntOp, Commutable>;
3122 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3123 InstrItinClass itin, string OpcodeStr, string Dt,
3125 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3126 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3127 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3128 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3131 // ....then also with element size of 8 bits:
3132 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3133 InstrItinClass itin16, InstrItinClass itin32,
3134 string OpcodeStr, string Dt,
3135 Intrinsic IntOp, bit Commutable = 0>
3136 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3137 IntOp, Commutable> {
3138 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3139 OpcodeStr, !strconcat(Dt, "8"),
3140 v8i16, v8i8, IntOp, Commutable>;
3143 // ....with explicit extend (VABDL).
3144 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3145 InstrItinClass itin, string OpcodeStr, string Dt,
3146 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3147 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3148 OpcodeStr, !strconcat(Dt, "8"),
3149 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3150 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3151 OpcodeStr, !strconcat(Dt, "16"),
3152 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3153 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3154 OpcodeStr, !strconcat(Dt, "32"),
3155 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3159 // Neon Wide 3-register vector intrinsics,
3160 // source operand element sizes of 8, 16 and 32 bits:
3161 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3162 string OpcodeStr, string Dt,
3163 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3164 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3165 OpcodeStr, !strconcat(Dt, "8"),
3166 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3167 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3168 OpcodeStr, !strconcat(Dt, "16"),
3169 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3170 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3171 OpcodeStr, !strconcat(Dt, "32"),
3172 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3176 // Neon Multiply-Op vector operations,
3177 // element sizes of 8, 16 and 32 bits:
3178 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3179 InstrItinClass itinD16, InstrItinClass itinD32,
3180 InstrItinClass itinQ16, InstrItinClass itinQ32,
3181 string OpcodeStr, string Dt, SDNode OpNode> {
3182 // 64-bit vector types.
3183 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3184 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3185 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3186 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3187 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3188 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3190 // 128-bit vector types.
3191 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3192 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3193 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3194 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3195 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3196 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3199 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3200 InstrItinClass itinD16, InstrItinClass itinD32,
3201 InstrItinClass itinQ16, InstrItinClass itinQ32,
3202 string OpcodeStr, string Dt, SDNode ShOp> {
3203 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3204 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3205 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3206 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3207 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3208 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3210 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3211 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3215 // Neon Intrinsic-Op vector operations,
3216 // element sizes of 8, 16 and 32 bits:
3217 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3218 InstrItinClass itinD, InstrItinClass itinQ,
3219 string OpcodeStr, string Dt, Intrinsic IntOp,
3221 // 64-bit vector types.
3222 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3223 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3224 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3225 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3226 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3227 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3229 // 128-bit vector types.
3230 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3231 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3232 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3233 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3234 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3235 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3238 // Neon 3-argument intrinsics,
3239 // element sizes of 8, 16 and 32 bits:
3240 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3241 InstrItinClass itinD, InstrItinClass itinQ,
3242 string OpcodeStr, string Dt, Intrinsic IntOp> {
3243 // 64-bit vector types.
3244 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3245 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3246 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3247 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3248 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3249 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3251 // 128-bit vector types.
3252 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3253 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3254 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3255 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3256 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3257 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3261 // Neon Long Multiply-Op vector operations,
3262 // element sizes of 8, 16 and 32 bits:
3263 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3264 InstrItinClass itin16, InstrItinClass itin32,
3265 string OpcodeStr, string Dt, SDNode MulOp,
3267 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3268 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3269 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3270 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3271 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3272 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3275 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3276 string Dt, SDNode MulOp, SDNode OpNode> {
3277 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3278 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3279 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3280 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3284 // Neon Long 3-argument intrinsics.
3286 // First with only element sizes of 16 and 32 bits:
3287 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3288 InstrItinClass itin16, InstrItinClass itin32,
3289 string OpcodeStr, string Dt, Intrinsic IntOp> {
3290 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3291 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3292 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3293 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3296 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3297 string OpcodeStr, string Dt, Intrinsic IntOp> {
3298 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3299 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3300 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3301 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3304 // ....then also with element size of 8 bits:
3305 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3306 InstrItinClass itin16, InstrItinClass itin32,
3307 string OpcodeStr, string Dt, Intrinsic IntOp>
3308 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3309 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3310 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3313 // ....with explicit extend (VABAL).
3314 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3315 InstrItinClass itin, string OpcodeStr, string Dt,
3316 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3317 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3318 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3319 IntOp, ExtOp, OpNode>;
3320 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3321 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3322 IntOp, ExtOp, OpNode>;
3323 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3324 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3325 IntOp, ExtOp, OpNode>;
3329 // Neon Pairwise long 2-register intrinsics,
3330 // element sizes of 8, 16 and 32 bits:
3331 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3332 bits<5> op11_7, bit op4,
3333 string OpcodeStr, string Dt, Intrinsic IntOp> {
3334 // 64-bit vector types.
3335 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3336 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3337 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3338 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3339 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3340 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3342 // 128-bit vector types.
3343 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3344 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3345 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3346 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3347 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3348 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3352 // Neon Pairwise long 2-register accumulate intrinsics,
3353 // element sizes of 8, 16 and 32 bits:
3354 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3355 bits<5> op11_7, bit op4,
3356 string OpcodeStr, string Dt, Intrinsic IntOp> {
3357 // 64-bit vector types.
3358 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3359 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3360 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3361 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3362 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3363 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3365 // 128-bit vector types.
3366 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3367 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3368 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3369 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3370 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3371 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3375 // Neon 2-register vector shift by immediate,
3376 // with f of either N2RegVShLFrm or N2RegVShRFrm
3377 // element sizes of 8, 16, 32 and 64 bits:
3378 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3379 InstrItinClass itin, string OpcodeStr, string Dt,
3381 // 64-bit vector types.
3382 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3383 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3384 let Inst{21-19} = 0b001; // imm6 = 001xxx
3386 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3387 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3388 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3390 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3391 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3392 let Inst{21} = 0b1; // imm6 = 1xxxxx
3394 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3395 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3398 // 128-bit vector types.
3399 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3400 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3401 let Inst{21-19} = 0b001; // imm6 = 001xxx
3403 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3404 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3405 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3407 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3408 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3409 let Inst{21} = 0b1; // imm6 = 1xxxxx
3411 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3412 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3415 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3416 InstrItinClass itin, string OpcodeStr, string Dt,
3418 // 64-bit vector types.
3419 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3420 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3421 let Inst{21-19} = 0b001; // imm6 = 001xxx
3423 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3424 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3425 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3427 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3428 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3429 let Inst{21} = 0b1; // imm6 = 1xxxxx
3431 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3432 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3435 // 128-bit vector types.
3436 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3437 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3438 let Inst{21-19} = 0b001; // imm6 = 001xxx
3440 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3441 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3442 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3444 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3445 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3446 let Inst{21} = 0b1; // imm6 = 1xxxxx
3448 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3449 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3453 // Neon Shift-Accumulate vector operations,
3454 // element sizes of 8, 16, 32 and 64 bits:
3455 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3456 string OpcodeStr, string Dt, SDNode ShOp> {
3457 // 64-bit vector types.
3458 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3459 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3460 let Inst{21-19} = 0b001; // imm6 = 001xxx
3462 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3463 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3464 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3466 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3467 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3468 let Inst{21} = 0b1; // imm6 = 1xxxxx
3470 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3471 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3474 // 128-bit vector types.
3475 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3476 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3477 let Inst{21-19} = 0b001; // imm6 = 001xxx
3479 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3480 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3481 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3483 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3484 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3485 let Inst{21} = 0b1; // imm6 = 1xxxxx
3487 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3488 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3492 // Neon Shift-Insert vector operations,
3493 // with f of either N2RegVShLFrm or N2RegVShRFrm
3494 // element sizes of 8, 16, 32 and 64 bits:
3495 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3497 // 64-bit vector types.
3498 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3499 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3500 let Inst{21-19} = 0b001; // imm6 = 001xxx
3502 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3503 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3504 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3506 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3507 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3508 let Inst{21} = 0b1; // imm6 = 1xxxxx
3510 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3511 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3514 // 128-bit vector types.
3515 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3516 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3517 let Inst{21-19} = 0b001; // imm6 = 001xxx
3519 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3520 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3521 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3523 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3524 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3525 let Inst{21} = 0b1; // imm6 = 1xxxxx
3527 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3528 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3531 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3533 // 64-bit vector types.
3534 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3535 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3536 let Inst{21-19} = 0b001; // imm6 = 001xxx
3538 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3539 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3540 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3542 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3543 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3544 let Inst{21} = 0b1; // imm6 = 1xxxxx
3546 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3547 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3550 // 128-bit vector types.
3551 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3552 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3553 let Inst{21-19} = 0b001; // imm6 = 001xxx
3555 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3556 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3557 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3559 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3560 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3561 let Inst{21} = 0b1; // imm6 = 1xxxxx
3563 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3564 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3568 // Neon Shift Long operations,
3569 // element sizes of 8, 16, 32 bits:
3570 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3571 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3572 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3573 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3574 let Inst{21-19} = 0b001; // imm6 = 001xxx
3576 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3577 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3578 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3580 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3581 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3582 let Inst{21} = 0b1; // imm6 = 1xxxxx
3586 // Neon Shift Narrow operations,
3587 // element sizes of 16, 32, 64 bits:
3588 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3589 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3591 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3592 OpcodeStr, !strconcat(Dt, "16"),
3593 v8i8, v8i16, shr_imm8, OpNode> {
3594 let Inst{21-19} = 0b001; // imm6 = 001xxx
3596 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3597 OpcodeStr, !strconcat(Dt, "32"),
3598 v4i16, v4i32, shr_imm16, OpNode> {
3599 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3601 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3602 OpcodeStr, !strconcat(Dt, "64"),
3603 v2i32, v2i64, shr_imm32, OpNode> {
3604 let Inst{21} = 0b1; // imm6 = 1xxxxx
3608 //===----------------------------------------------------------------------===//
3609 // Instruction Definitions.
3610 //===----------------------------------------------------------------------===//
3612 // Vector Add Operations.
3614 // VADD : Vector Add (integer and floating-point)
3615 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3617 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3618 v2f32, v2f32, fadd, 1>;
3619 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3620 v4f32, v4f32, fadd, 1>;
3621 // VADDL : Vector Add Long (Q = D + D)
3622 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3623 "vaddl", "s", add, sext, 1>;
3624 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3625 "vaddl", "u", add, zext, 1>;
3626 // VADDW : Vector Add Wide (Q = Q + D)
3627 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3628 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3629 // VHADD : Vector Halving Add
3630 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3631 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3632 "vhadd", "s", int_arm_neon_vhadds, 1>;
3633 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3634 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3635 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3636 // VRHADD : Vector Rounding Halving Add
3637 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3638 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3639 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3640 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3641 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3642 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3643 // VQADD : Vector Saturating Add
3644 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3645 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3646 "vqadd", "s", int_arm_neon_vqadds, 1>;
3647 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3648 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3649 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3650 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3651 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3652 int_arm_neon_vaddhn, 1>;
3653 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3654 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3655 int_arm_neon_vraddhn, 1>;
3657 // Vector Multiply Operations.
3659 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3660 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3661 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3662 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3663 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3664 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3665 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3666 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3667 v2f32, v2f32, fmul, 1>;
3668 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3669 v4f32, v4f32, fmul, 1>;
3670 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3671 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3672 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3675 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3676 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3677 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3678 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3679 (DSubReg_i16_reg imm:$lane))),
3680 (SubReg_i16_lane imm:$lane)))>;
3681 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3682 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3683 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3684 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3685 (DSubReg_i32_reg imm:$lane))),
3686 (SubReg_i32_lane imm:$lane)))>;
3687 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3688 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3689 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3690 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3691 (DSubReg_i32_reg imm:$lane))),
3692 (SubReg_i32_lane imm:$lane)))>;
3694 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3695 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3696 IIC_VMULi16Q, IIC_VMULi32Q,
3697 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3698 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3699 IIC_VMULi16Q, IIC_VMULi32Q,
3700 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3701 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3702 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3704 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3705 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3706 (DSubReg_i16_reg imm:$lane))),
3707 (SubReg_i16_lane imm:$lane)))>;
3708 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3709 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3711 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3712 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3713 (DSubReg_i32_reg imm:$lane))),
3714 (SubReg_i32_lane imm:$lane)))>;
3716 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3717 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3718 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3719 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3720 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3721 IIC_VMULi16Q, IIC_VMULi32Q,
3722 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3723 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3724 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3726 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3727 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3728 (DSubReg_i16_reg imm:$lane))),
3729 (SubReg_i16_lane imm:$lane)))>;
3730 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3731 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3733 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3734 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3735 (DSubReg_i32_reg imm:$lane))),
3736 (SubReg_i32_lane imm:$lane)))>;
3738 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3739 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3740 "vmull", "s", NEONvmulls, 1>;
3741 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3742 "vmull", "u", NEONvmullu, 1>;
3743 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3744 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3745 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3746 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3748 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3749 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3750 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3751 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3752 "vqdmull", "s", int_arm_neon_vqdmull>;
3754 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3756 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3757 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3758 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3759 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3760 v2f32, fmul_su, fadd_mlx>,
3761 Requires<[HasNEON, UseFPVMLx]>;
3762 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3763 v4f32, fmul_su, fadd_mlx>,
3764 Requires<[HasNEON, UseFPVMLx]>;
3765 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3766 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3767 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3768 v2f32, fmul_su, fadd_mlx>,
3769 Requires<[HasNEON, UseFPVMLx]>;
3770 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3771 v4f32, v2f32, fmul_su, fadd_mlx>,
3772 Requires<[HasNEON, UseFPVMLx]>;
3774 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3775 (mul (v8i16 QPR:$src2),
3776 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3777 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3778 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3779 (DSubReg_i16_reg imm:$lane))),
3780 (SubReg_i16_lane imm:$lane)))>;
3782 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3783 (mul (v4i32 QPR:$src2),
3784 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3785 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3786 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3787 (DSubReg_i32_reg imm:$lane))),
3788 (SubReg_i32_lane imm:$lane)))>;
3790 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3791 (fmul_su (v4f32 QPR:$src2),
3792 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3793 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3795 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3796 (DSubReg_i32_reg imm:$lane))),
3797 (SubReg_i32_lane imm:$lane)))>,
3798 Requires<[HasNEON, UseFPVMLx]>;
3800 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3801 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3802 "vmlal", "s", NEONvmulls, add>;
3803 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3804 "vmlal", "u", NEONvmullu, add>;
3806 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3807 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3809 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3810 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3811 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3812 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3814 // VMLS : Vector Multiply Subtract (integer and floating-point)
3815 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3816 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3817 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3818 v2f32, fmul_su, fsub_mlx>,
3819 Requires<[HasNEON, UseFPVMLx]>;
3820 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3821 v4f32, fmul_su, fsub_mlx>,
3822 Requires<[HasNEON, UseFPVMLx]>;
3823 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3824 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3825 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3826 v2f32, fmul_su, fsub_mlx>,
3827 Requires<[HasNEON, UseFPVMLx]>;
3828 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3829 v4f32, v2f32, fmul_su, fsub_mlx>,
3830 Requires<[HasNEON, UseFPVMLx]>;
3832 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3833 (mul (v8i16 QPR:$src2),
3834 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3835 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3836 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3837 (DSubReg_i16_reg imm:$lane))),
3838 (SubReg_i16_lane imm:$lane)))>;
3840 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3841 (mul (v4i32 QPR:$src2),
3842 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3843 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3844 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3845 (DSubReg_i32_reg imm:$lane))),
3846 (SubReg_i32_lane imm:$lane)))>;
3848 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3849 (fmul_su (v4f32 QPR:$src2),
3850 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3851 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3852 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3853 (DSubReg_i32_reg imm:$lane))),
3854 (SubReg_i32_lane imm:$lane)))>,
3855 Requires<[HasNEON, UseFPVMLx]>;
3857 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3858 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3859 "vmlsl", "s", NEONvmulls, sub>;
3860 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3861 "vmlsl", "u", NEONvmullu, sub>;
3863 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3864 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3866 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3867 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3868 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3869 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3871 // Vector Subtract Operations.
3873 // VSUB : Vector Subtract (integer and floating-point)
3874 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3875 "vsub", "i", sub, 0>;
3876 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3877 v2f32, v2f32, fsub, 0>;
3878 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3879 v4f32, v4f32, fsub, 0>;
3880 // VSUBL : Vector Subtract Long (Q = D - D)
3881 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3882 "vsubl", "s", sub, sext, 0>;
3883 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3884 "vsubl", "u", sub, zext, 0>;
3885 // VSUBW : Vector Subtract Wide (Q = Q - D)
3886 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3887 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3888 // VHSUB : Vector Halving Subtract
3889 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3890 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3891 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3892 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3893 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3894 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3895 // VQSUB : Vector Saturing Subtract
3896 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3897 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3898 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3899 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3900 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3901 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3902 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3903 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3904 int_arm_neon_vsubhn, 0>;
3905 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3906 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3907 int_arm_neon_vrsubhn, 0>;
3909 // Vector Comparisons.
3911 // VCEQ : Vector Compare Equal
3912 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3913 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3914 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3916 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3919 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3920 "$Vd, $Vm, #0", NEONvceqz>;
3922 // VCGE : Vector Compare Greater Than or Equal
3923 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3924 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3925 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3926 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3927 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3929 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3932 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3933 "$Vd, $Vm, #0", NEONvcgez>;
3934 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3935 "$Vd, $Vm, #0", NEONvclez>;
3937 // VCGT : Vector Compare Greater Than
3938 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3939 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3940 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3941 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3942 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3944 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3947 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3948 "$Vd, $Vm, #0", NEONvcgtz>;
3949 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3950 "$Vd, $Vm, #0", NEONvcltz>;
3952 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3953 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3954 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3955 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3956 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3957 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3958 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3959 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3960 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3961 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3962 // VTST : Vector Test Bits
3963 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3964 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3966 // Vector Bitwise Operations.
3968 def vnotd : PatFrag<(ops node:$in),
3969 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3970 def vnotq : PatFrag<(ops node:$in),
3971 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3974 // VAND : Vector Bitwise AND
3975 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3976 v2i32, v2i32, and, 1>;
3977 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3978 v4i32, v4i32, and, 1>;
3980 // VEOR : Vector Bitwise Exclusive OR
3981 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3982 v2i32, v2i32, xor, 1>;
3983 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3984 v4i32, v4i32, xor, 1>;
3986 // VORR : Vector Bitwise OR
3987 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3988 v2i32, v2i32, or, 1>;
3989 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3990 v4i32, v4i32, or, 1>;
3992 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3993 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3995 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3997 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3998 let Inst{9} = SIMM{9};
4001 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
4002 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4004 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4006 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
4007 let Inst{10-9} = SIMM{10-9};
4010 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
4011 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4013 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
4015 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4016 let Inst{9} = SIMM{9};
4019 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
4020 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4022 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
4024 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
4025 let Inst{10-9} = SIMM{10-9};
4029 // VBIC : Vector Bitwise Bit Clear (AND NOT)
4030 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4031 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4032 "vbic", "$Vd, $Vn, $Vm", "",
4033 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
4034 (vnotd DPR:$Vm))))]>;
4035 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4036 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4037 "vbic", "$Vd, $Vn, $Vm", "",
4038 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
4039 (vnotq QPR:$Vm))))]>;
4041 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
4042 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
4044 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4046 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4047 let Inst{9} = SIMM{9};
4050 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
4051 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
4053 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4055 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
4056 let Inst{10-9} = SIMM{10-9};
4059 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
4060 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
4062 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4064 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4065 let Inst{9} = SIMM{9};
4068 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4069 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4071 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4073 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4074 let Inst{10-9} = SIMM{10-9};
4077 // VORN : Vector Bitwise OR NOT
4078 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4079 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4080 "vorn", "$Vd, $Vn, $Vm", "",
4081 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4082 (vnotd DPR:$Vm))))]>;
4083 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4084 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4085 "vorn", "$Vd, $Vn, $Vm", "",
4086 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4087 (vnotq QPR:$Vm))))]>;
4089 // VMVN : Vector Bitwise NOT (Immediate)
4091 let isReMaterializable = 1 in {
4093 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4094 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4095 "vmvn", "i16", "$Vd, $SIMM", "",
4096 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4097 let Inst{9} = SIMM{9};
4100 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4101 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4102 "vmvn", "i16", "$Vd, $SIMM", "",
4103 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4104 let Inst{9} = SIMM{9};
4107 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4108 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4109 "vmvn", "i32", "$Vd, $SIMM", "",
4110 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4111 let Inst{11-8} = SIMM{11-8};
4114 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4115 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4116 "vmvn", "i32", "$Vd, $SIMM", "",
4117 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4118 let Inst{11-8} = SIMM{11-8};
4122 // VMVN : Vector Bitwise NOT
4123 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4124 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4125 "vmvn", "$Vd, $Vm", "",
4126 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4127 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4128 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4129 "vmvn", "$Vd, $Vm", "",
4130 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4131 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4132 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4134 // VBSL : Vector Bitwise Select
4135 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4136 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4137 N3RegFrm, IIC_VCNTiD,
4138 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4140 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4142 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4143 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4144 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4146 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4147 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4148 N3RegFrm, IIC_VCNTiQ,
4149 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4151 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4153 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4154 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4155 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4157 // VBIF : Vector Bitwise Insert if False
4158 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4159 // FIXME: This instruction's encoding MAY NOT BE correct.
4160 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4161 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4162 N3RegFrm, IIC_VBINiD,
4163 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4165 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4166 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4167 N3RegFrm, IIC_VBINiQ,
4168 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4171 // VBIT : Vector Bitwise Insert if True
4172 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4173 // FIXME: This instruction's encoding MAY NOT BE correct.
4174 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4175 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4176 N3RegFrm, IIC_VBINiD,
4177 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4179 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4180 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4181 N3RegFrm, IIC_VBINiQ,
4182 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4185 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4186 // for equivalent operations with different register constraints; it just
4189 // Vector Absolute Differences.
4191 // VABD : Vector Absolute Difference
4192 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4193 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4194 "vabd", "s", int_arm_neon_vabds, 1>;
4195 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4196 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4197 "vabd", "u", int_arm_neon_vabdu, 1>;
4198 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4199 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4200 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4201 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4203 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4204 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4205 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4206 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4207 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4209 // VABA : Vector Absolute Difference and Accumulate
4210 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4211 "vaba", "s", int_arm_neon_vabds, add>;
4212 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4213 "vaba", "u", int_arm_neon_vabdu, add>;
4215 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4216 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4217 "vabal", "s", int_arm_neon_vabds, zext, add>;
4218 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4219 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4221 // Vector Maximum and Minimum.
4223 // VMAX : Vector Maximum
4224 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4225 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4226 "vmax", "s", int_arm_neon_vmaxs, 1>;
4227 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4228 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4229 "vmax", "u", int_arm_neon_vmaxu, 1>;
4230 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4232 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4233 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4235 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4237 // VMIN : Vector Minimum
4238 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4239 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4240 "vmin", "s", int_arm_neon_vmins, 1>;
4241 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4242 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4243 "vmin", "u", int_arm_neon_vminu, 1>;
4244 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4246 v2f32, v2f32, int_arm_neon_vmins, 1>;
4247 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4249 v4f32, v4f32, int_arm_neon_vmins, 1>;
4251 // Vector Pairwise Operations.
4253 // VPADD : Vector Pairwise Add
4254 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4256 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4257 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4259 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4260 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4262 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4263 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4264 IIC_VPBIND, "vpadd", "f32",
4265 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4267 // VPADDL : Vector Pairwise Add Long
4268 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4269 int_arm_neon_vpaddls>;
4270 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4271 int_arm_neon_vpaddlu>;
4273 // VPADAL : Vector Pairwise Add and Accumulate Long
4274 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4275 int_arm_neon_vpadals>;
4276 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4277 int_arm_neon_vpadalu>;
4279 // VPMAX : Vector Pairwise Maximum
4280 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4281 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4282 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4283 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4284 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4285 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4286 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4287 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4288 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4289 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4290 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4291 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4292 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4293 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4295 // VPMIN : Vector Pairwise Minimum
4296 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4297 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4298 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4299 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4300 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4301 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4302 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4303 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4304 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4305 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4306 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4307 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4308 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4309 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4311 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4313 // VRECPE : Vector Reciprocal Estimate
4314 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4315 IIC_VUNAD, "vrecpe", "u32",
4316 v2i32, v2i32, int_arm_neon_vrecpe>;
4317 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4318 IIC_VUNAQ, "vrecpe", "u32",
4319 v4i32, v4i32, int_arm_neon_vrecpe>;
4320 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4321 IIC_VUNAD, "vrecpe", "f32",
4322 v2f32, v2f32, int_arm_neon_vrecpe>;
4323 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4324 IIC_VUNAQ, "vrecpe", "f32",
4325 v4f32, v4f32, int_arm_neon_vrecpe>;
4327 // VRECPS : Vector Reciprocal Step
4328 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4329 IIC_VRECSD, "vrecps", "f32",
4330 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4331 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4332 IIC_VRECSQ, "vrecps", "f32",
4333 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4335 // VRSQRTE : Vector Reciprocal Square Root Estimate
4336 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4337 IIC_VUNAD, "vrsqrte", "u32",
4338 v2i32, v2i32, int_arm_neon_vrsqrte>;
4339 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4340 IIC_VUNAQ, "vrsqrte", "u32",
4341 v4i32, v4i32, int_arm_neon_vrsqrte>;
4342 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4343 IIC_VUNAD, "vrsqrte", "f32",
4344 v2f32, v2f32, int_arm_neon_vrsqrte>;
4345 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4346 IIC_VUNAQ, "vrsqrte", "f32",
4347 v4f32, v4f32, int_arm_neon_vrsqrte>;
4349 // VRSQRTS : Vector Reciprocal Square Root Step
4350 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4351 IIC_VRECSD, "vrsqrts", "f32",
4352 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4353 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4354 IIC_VRECSQ, "vrsqrts", "f32",
4355 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4359 // VSHL : Vector Shift
4360 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4361 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4362 "vshl", "s", int_arm_neon_vshifts>;
4363 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4364 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4365 "vshl", "u", int_arm_neon_vshiftu>;
4367 // VSHL : Vector Shift Left (Immediate)
4368 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4370 // VSHR : Vector Shift Right (Immediate)
4371 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4372 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4374 // VSHLL : Vector Shift Left Long
4375 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4376 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4378 // VSHLL : Vector Shift Left Long (with maximum shift count)
4379 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4380 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4381 ValueType OpTy, SDNode OpNode>
4382 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4383 ResTy, OpTy, OpNode> {
4384 let Inst{21-16} = op21_16;
4385 let DecoderMethod = "DecodeVSHLMaxInstruction";
4387 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4388 v8i16, v8i8, NEONvshlli>;
4389 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4390 v4i32, v4i16, NEONvshlli>;
4391 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4392 v2i64, v2i32, NEONvshlli>;
4394 // VSHRN : Vector Shift Right and Narrow
4395 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4398 // VRSHL : Vector Rounding Shift
4399 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4400 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4401 "vrshl", "s", int_arm_neon_vrshifts>;
4402 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4403 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4404 "vrshl", "u", int_arm_neon_vrshiftu>;
4405 // VRSHR : Vector Rounding Shift Right
4406 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4407 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4409 // VRSHRN : Vector Rounding Shift Right and Narrow
4410 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4413 // VQSHL : Vector Saturating Shift
4414 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4415 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4416 "vqshl", "s", int_arm_neon_vqshifts>;
4417 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4418 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4419 "vqshl", "u", int_arm_neon_vqshiftu>;
4420 // VQSHL : Vector Saturating Shift Left (Immediate)
4421 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4422 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4424 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4425 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4427 // VQSHRN : Vector Saturating Shift Right and Narrow
4428 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4430 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4433 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4434 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4437 // VQRSHL : Vector Saturating Rounding Shift
4438 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4439 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4440 "vqrshl", "s", int_arm_neon_vqrshifts>;
4441 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4442 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4443 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4445 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4446 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4448 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4451 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4452 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4455 // VSRA : Vector Shift Right and Accumulate
4456 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4457 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4458 // VRSRA : Vector Rounding Shift Right and Accumulate
4459 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4460 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4462 // VSLI : Vector Shift Left and Insert
4463 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4465 // VSRI : Vector Shift Right and Insert
4466 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4468 // Vector Absolute and Saturating Absolute.
4470 // VABS : Vector Absolute Value
4471 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4472 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4474 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4475 IIC_VUNAD, "vabs", "f32",
4476 v2f32, v2f32, int_arm_neon_vabs>;
4477 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4478 IIC_VUNAQ, "vabs", "f32",
4479 v4f32, v4f32, int_arm_neon_vabs>;
4481 // VQABS : Vector Saturating Absolute Value
4482 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4483 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4484 int_arm_neon_vqabs>;
4488 def vnegd : PatFrag<(ops node:$in),
4489 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4490 def vnegq : PatFrag<(ops node:$in),
4491 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4493 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4494 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4495 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4496 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4497 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4498 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4499 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4500 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4502 // VNEG : Vector Negate (integer)
4503 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4504 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4505 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4506 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4507 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4508 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4510 // VNEG : Vector Negate (floating-point)
4511 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4512 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4513 "vneg", "f32", "$Vd, $Vm", "",
4514 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4515 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4516 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4517 "vneg", "f32", "$Vd, $Vm", "",
4518 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4520 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4521 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4522 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4523 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4524 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4525 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4527 // VQNEG : Vector Saturating Negate
4528 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4529 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4530 int_arm_neon_vqneg>;
4532 // Vector Bit Counting Operations.
4534 // VCLS : Vector Count Leading Sign Bits
4535 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4536 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4538 // VCLZ : Vector Count Leading Zeros
4539 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4540 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4542 // VCNT : Vector Count One Bits
4543 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4544 IIC_VCNTiD, "vcnt", "8",
4545 v8i8, v8i8, int_arm_neon_vcnt>;
4546 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4547 IIC_VCNTiQ, "vcnt", "8",
4548 v16i8, v16i8, int_arm_neon_vcnt>;
4551 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4552 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4553 "vswp", "$Vd, $Vm", "", []>;
4554 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4555 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4556 "vswp", "$Vd, $Vm", "", []>;
4558 // Vector Move Operations.
4560 // VMOV : Vector Move (Register)
4561 def : InstAlias<"vmov${p} $Vd, $Vm",
4562 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4563 def : InstAlias<"vmov${p} $Vd, $Vm",
4564 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4565 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4566 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4567 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4568 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4570 // VMOV : Vector Move (Immediate)
4572 let isReMaterializable = 1 in {
4573 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4574 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4575 "vmov", "i8", "$Vd, $SIMM", "",
4576 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4577 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4578 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4579 "vmov", "i8", "$Vd, $SIMM", "",
4580 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4582 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4583 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4584 "vmov", "i16", "$Vd, $SIMM", "",
4585 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4586 let Inst{9} = SIMM{9};
4589 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4590 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4591 "vmov", "i16", "$Vd, $SIMM", "",
4592 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4593 let Inst{9} = SIMM{9};
4596 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4597 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4598 "vmov", "i32", "$Vd, $SIMM", "",
4599 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4600 let Inst{11-8} = SIMM{11-8};
4603 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4604 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4605 "vmov", "i32", "$Vd, $SIMM", "",
4606 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4607 let Inst{11-8} = SIMM{11-8};
4610 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4611 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4612 "vmov", "i64", "$Vd, $SIMM", "",
4613 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4614 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4615 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4616 "vmov", "i64", "$Vd, $SIMM", "",
4617 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4619 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4620 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4621 "vmov", "f32", "$Vd, $SIMM", "",
4622 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4623 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4624 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4625 "vmov", "f32", "$Vd, $SIMM", "",
4626 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4627 } // isReMaterializable
4629 // VMOV : Vector Get Lane (move scalar to ARM core register)
4631 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4632 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4633 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4634 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4636 let Inst{21} = lane{2};
4637 let Inst{6-5} = lane{1-0};
4639 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4640 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4641 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4642 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4644 let Inst{21} = lane{1};
4645 let Inst{6} = lane{0};
4647 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4648 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4649 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4650 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4652 let Inst{21} = lane{2};
4653 let Inst{6-5} = lane{1-0};
4655 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4656 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4657 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4658 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4660 let Inst{21} = lane{1};
4661 let Inst{6} = lane{0};
4663 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4664 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4665 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4666 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4668 let Inst{21} = lane{0};
4670 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4671 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4672 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4673 (DSubReg_i8_reg imm:$lane))),
4674 (SubReg_i8_lane imm:$lane))>;
4675 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4676 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4677 (DSubReg_i16_reg imm:$lane))),
4678 (SubReg_i16_lane imm:$lane))>;
4679 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4680 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4681 (DSubReg_i8_reg imm:$lane))),
4682 (SubReg_i8_lane imm:$lane))>;
4683 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4684 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4685 (DSubReg_i16_reg imm:$lane))),
4686 (SubReg_i16_lane imm:$lane))>;
4687 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4688 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4689 (DSubReg_i32_reg imm:$lane))),
4690 (SubReg_i32_lane imm:$lane))>;
4691 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4692 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4693 (SSubReg_f32_reg imm:$src2))>;
4694 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4695 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4696 (SSubReg_f32_reg imm:$src2))>;
4697 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4698 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4699 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4700 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4703 // VMOV : Vector Set Lane (move ARM core register to scalar)
4705 let Constraints = "$src1 = $V" in {
4706 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4707 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4708 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4709 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4710 GPR:$R, imm:$lane))]> {
4711 let Inst{21} = lane{2};
4712 let Inst{6-5} = lane{1-0};
4714 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4715 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4716 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4717 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4718 GPR:$R, imm:$lane))]> {
4719 let Inst{21} = lane{1};
4720 let Inst{6} = lane{0};
4722 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4723 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4724 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4725 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4726 GPR:$R, imm:$lane))]> {
4727 let Inst{21} = lane{0};
4730 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4731 (v16i8 (INSERT_SUBREG QPR:$src1,
4732 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4733 (DSubReg_i8_reg imm:$lane))),
4734 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4735 (DSubReg_i8_reg imm:$lane)))>;
4736 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4737 (v8i16 (INSERT_SUBREG QPR:$src1,
4738 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4739 (DSubReg_i16_reg imm:$lane))),
4740 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4741 (DSubReg_i16_reg imm:$lane)))>;
4742 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4743 (v4i32 (INSERT_SUBREG QPR:$src1,
4744 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4745 (DSubReg_i32_reg imm:$lane))),
4746 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4747 (DSubReg_i32_reg imm:$lane)))>;
4749 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4750 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4751 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4752 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4753 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4754 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4756 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4757 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4758 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4759 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4761 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4762 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4763 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4764 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4765 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4766 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4768 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4769 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4770 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4771 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4772 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4773 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4775 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4776 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4777 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4779 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4780 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4781 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4783 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4784 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4785 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4788 // VDUP : Vector Duplicate (from ARM core register to all elements)
4790 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4791 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4792 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4793 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4794 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4795 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4796 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4797 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4799 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4800 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4801 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4802 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4803 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4804 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4806 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4807 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4809 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4811 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4812 ValueType Ty, Operand IdxTy>
4813 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4814 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4815 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4817 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4818 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4819 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4820 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4821 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4822 VectorIndex32:$lane)))]>;
4824 // Inst{19-16} is partially specified depending on the element size.
4826 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4828 let Inst{19-17} = lane{2-0};
4830 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4832 let Inst{19-18} = lane{1-0};
4834 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4836 let Inst{19} = lane{0};
4838 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4840 let Inst{19-17} = lane{2-0};
4842 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4844 let Inst{19-18} = lane{1-0};
4846 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4848 let Inst{19} = lane{0};
4851 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4852 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4854 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4855 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4857 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4858 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4859 (DSubReg_i8_reg imm:$lane))),
4860 (SubReg_i8_lane imm:$lane)))>;
4861 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4862 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4863 (DSubReg_i16_reg imm:$lane))),
4864 (SubReg_i16_lane imm:$lane)))>;
4865 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4866 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4867 (DSubReg_i32_reg imm:$lane))),
4868 (SubReg_i32_lane imm:$lane)))>;
4869 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4870 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4871 (DSubReg_i32_reg imm:$lane))),
4872 (SubReg_i32_lane imm:$lane)))>;
4874 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4875 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4876 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4877 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4879 // VMOVN : Vector Narrowing Move
4880 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4881 "vmovn", "i", trunc>;
4882 // VQMOVN : Vector Saturating Narrowing Move
4883 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4884 "vqmovn", "s", int_arm_neon_vqmovns>;
4885 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4886 "vqmovn", "u", int_arm_neon_vqmovnu>;
4887 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4888 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4889 // VMOVL : Vector Lengthening Move
4890 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4891 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4893 // Vector Conversions.
4895 // VCVT : Vector Convert Between Floating-Point and Integers
4896 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4897 v2i32, v2f32, fp_to_sint>;
4898 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4899 v2i32, v2f32, fp_to_uint>;
4900 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4901 v2f32, v2i32, sint_to_fp>;
4902 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4903 v2f32, v2i32, uint_to_fp>;
4905 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4906 v4i32, v4f32, fp_to_sint>;
4907 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4908 v4i32, v4f32, fp_to_uint>;
4909 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4910 v4f32, v4i32, sint_to_fp>;
4911 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4912 v4f32, v4i32, uint_to_fp>;
4914 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4915 let DecoderMethod = "DecodeVCVTD" in {
4916 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4917 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4918 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4919 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4920 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4921 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4922 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4923 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4926 let DecoderMethod = "DecodeVCVTQ" in {
4927 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4928 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4929 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4930 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4931 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4932 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4933 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4934 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4937 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4938 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4939 IIC_VUNAQ, "vcvt", "f16.f32",
4940 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4941 Requires<[HasNEON, HasFP16]>;
4942 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4943 IIC_VUNAQ, "vcvt", "f32.f16",
4944 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4945 Requires<[HasNEON, HasFP16]>;
4949 // VREV64 : Vector Reverse elements within 64-bit doublewords
4951 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4952 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4953 (ins DPR:$Vm), IIC_VMOVD,
4954 OpcodeStr, Dt, "$Vd, $Vm", "",
4955 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4956 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4957 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4958 (ins QPR:$Vm), IIC_VMOVQ,
4959 OpcodeStr, Dt, "$Vd, $Vm", "",
4960 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4962 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4963 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4964 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4965 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4967 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4968 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4969 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4970 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4972 // VREV32 : Vector Reverse elements within 32-bit words
4974 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4975 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4976 (ins DPR:$Vm), IIC_VMOVD,
4977 OpcodeStr, Dt, "$Vd, $Vm", "",
4978 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4979 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4980 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4981 (ins QPR:$Vm), IIC_VMOVQ,
4982 OpcodeStr, Dt, "$Vd, $Vm", "",
4983 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4985 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4986 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4988 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4989 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4991 // VREV16 : Vector Reverse elements within 16-bit halfwords
4993 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4994 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4995 (ins DPR:$Vm), IIC_VMOVD,
4996 OpcodeStr, Dt, "$Vd, $Vm", "",
4997 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4998 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4999 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
5000 (ins QPR:$Vm), IIC_VMOVQ,
5001 OpcodeStr, Dt, "$Vd, $Vm", "",
5002 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
5004 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
5005 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
5007 // Other Vector Shuffles.
5009 // Aligned extractions: really just dropping registers
5011 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
5012 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
5013 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
5015 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
5017 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
5019 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
5021 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
5023 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
5026 // VEXT : Vector Extract
5028 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
5029 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
5030 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
5031 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5032 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
5033 (Ty DPR:$Vm), imm:$index)))]> {
5035 let Inst{11-8} = index{3-0};
5038 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
5039 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
5040 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
5041 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
5042 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
5043 (Ty QPR:$Vm), imm:$index)))]> {
5045 let Inst{11-8} = index{3-0};
5048 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
5049 let Inst{11-8} = index{3-0};
5051 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
5052 let Inst{11-9} = index{2-0};
5055 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
5056 let Inst{11-10} = index{1-0};
5057 let Inst{9-8} = 0b00;
5059 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
5062 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5064 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
5065 let Inst{11-8} = index{3-0};
5067 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
5068 let Inst{11-9} = index{2-0};
5071 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
5072 let Inst{11-10} = index{1-0};
5073 let Inst{9-8} = 0b00;
5075 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5078 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5080 // VTRN : Vector Transpose
5082 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5083 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5084 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5086 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5087 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5088 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5090 // VUZP : Vector Unzip (Deinterleave)
5092 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5093 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5094 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5096 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5097 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5098 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5100 // VZIP : Vector Zip (Interleave)
5102 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5103 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5104 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5106 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5107 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5108 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5110 // Vector Table Lookup and Table Extension.
5112 // VTBL : Vector Table Lookup
5113 let DecoderMethod = "DecodeTBLInstruction" in {
5115 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5116 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5117 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5118 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5119 let hasExtraSrcRegAllocReq = 1 in {
5121 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5122 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5123 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
5125 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5126 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5127 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
5129 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5130 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
5132 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
5133 } // hasExtraSrcRegAllocReq = 1
5136 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5138 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5140 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5142 // VTBX : Vector Table Extension
5144 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5145 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5146 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5147 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5148 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5149 let hasExtraSrcRegAllocReq = 1 in {
5151 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5152 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5153 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5155 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5156 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5157 NVTBLFrm, IIC_VTBX3,
5158 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5161 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5162 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5163 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5165 } // hasExtraSrcRegAllocReq = 1
5168 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5169 IIC_VTBX2, "$orig = $dst", []>;
5171 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5172 IIC_VTBX3, "$orig = $dst", []>;
5174 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5175 IIC_VTBX4, "$orig = $dst", []>;
5176 } // DecoderMethod = "DecodeTBLInstruction"
5178 //===----------------------------------------------------------------------===//
5179 // NEON instructions for single-precision FP math
5180 //===----------------------------------------------------------------------===//
5182 class N2VSPat<SDNode OpNode, NeonI Inst>
5183 : NEONFPPat<(f32 (OpNode SPR:$a)),
5185 (v2f32 (COPY_TO_REGCLASS (Inst
5187 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5188 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5190 class N3VSPat<SDNode OpNode, NeonI Inst>
5191 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5193 (v2f32 (COPY_TO_REGCLASS (Inst
5195 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5198 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5199 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5201 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5202 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5204 (v2f32 (COPY_TO_REGCLASS (Inst
5206 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5209 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5212 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5213 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5215 def : N3VSPat<fadd, VADDfd>;
5216 def : N3VSPat<fsub, VSUBfd>;
5217 def : N3VSPat<fmul, VMULfd>;
5218 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5219 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5220 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5221 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5222 def : N2VSPat<fabs, VABSfd>;
5223 def : N2VSPat<fneg, VNEGfd>;
5224 def : N3VSPat<NEONfmax, VMAXfd>;
5225 def : N3VSPat<NEONfmin, VMINfd>;
5226 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5227 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5228 def : N2VSPat<arm_sitof, VCVTs2fd>;
5229 def : N2VSPat<arm_uitof, VCVTu2fd>;
5231 //===----------------------------------------------------------------------===//
5232 // Non-Instruction Patterns
5233 //===----------------------------------------------------------------------===//
5236 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5237 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5238 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5239 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5240 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5241 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5242 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5243 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5244 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5245 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5246 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5247 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5248 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5249 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5250 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5251 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5252 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5253 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5254 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5255 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5256 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5257 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5258 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5259 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5260 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5261 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5262 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5263 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5264 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5265 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5267 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5268 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5269 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5270 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5271 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5272 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5273 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5274 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5275 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5276 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5277 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5278 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5279 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5280 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5281 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5282 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5283 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5284 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5285 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5286 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5287 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5288 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5289 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5290 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5291 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5292 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5293 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5294 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5295 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5296 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5299 //===----------------------------------------------------------------------===//
5300 // Assembler aliases
5303 // VAND/VEOR/VORR accept but do not require a type suffix.
5304 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5305 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5306 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5307 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5308 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5309 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5310 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5311 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5312 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5313 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5314 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5315 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5317 // VLD1 requires a size suffix, but also accepts type specific variants.
5318 // Load one D register.
5319 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5320 (VLD1d8 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5321 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5322 (VLD1d16 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5323 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5324 (VLD1d32 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5325 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5326 (VLD1d64 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5327 // with writeback, fixed stride
5328 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5329 (VLD1d8wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5330 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5331 (VLD1d16wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5332 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5333 (VLD1d32wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5334 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5335 (VLD1d64wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5336 // with writeback, register stride
5337 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5338 (VLD1d8wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5339 rGPR:$Rm, pred:$p)>;
5340 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5341 (VLD1d16wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5342 rGPR:$Rm, pred:$p)>;
5343 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5344 (VLD1d32wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5345 rGPR:$Rm, pred:$p)>;
5346 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5347 (VLD1d64wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5348 rGPR:$Rm, pred:$p)>;
5350 // Load two D registers.
5351 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5352 (VLD1q8 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5353 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5354 (VLD1q16 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5355 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5356 (VLD1q32 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5357 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5358 (VLD1q64 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5359 // with writeback, fixed stride
5360 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5361 (VLD1q8wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5362 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5363 (VLD1q16wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5364 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5365 (VLD1q32wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5366 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5367 (VLD1q64wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5368 // with writeback, register stride
5369 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5370 (VLD1q8wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5371 rGPR:$Rm, pred:$p)>;
5372 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5373 (VLD1q16wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5374 rGPR:$Rm, pred:$p)>;
5375 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5376 (VLD1q32wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5377 rGPR:$Rm, pred:$p)>;
5378 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5379 (VLD1q64wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5380 rGPR:$Rm, pred:$p)>;
5382 // Load three D registers.
5383 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5384 (VLD1d8T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5385 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5386 (VLD1d16T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5387 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5388 (VLD1d32T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5389 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5390 (VLD1d64T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5391 // with writeback, fixed stride
5392 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5393 (VLD1d8Twb_fixed VecListThreeD:$Vd, zero_reg,
5394 addrmode6:$Rn, pred:$p)>;
5395 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5396 (VLD1d16Twb_fixed VecListThreeD:$Vd, zero_reg,
5397 addrmode6:$Rn, pred:$p)>;
5398 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5399 (VLD1d32Twb_fixed VecListThreeD:$Vd, zero_reg,
5400 addrmode6:$Rn, pred:$p)>;
5401 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5402 (VLD1d64Twb_fixed VecListThreeD:$Vd, zero_reg,
5403 addrmode6:$Rn, pred:$p)>;
5404 // with writeback, register stride
5405 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5406 (VLD1d8Twb_register VecListThreeD:$Vd, zero_reg,
5407 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5408 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5409 (VLD1d16Twb_register VecListThreeD:$Vd, zero_reg,
5410 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5411 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5412 (VLD1d32Twb_register VecListThreeD:$Vd, zero_reg,
5413 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5414 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5415 (VLD1d64Twb_register VecListThreeD:$Vd, zero_reg,
5416 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5419 // Load four D registers.
5420 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5421 (VLD1d8Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5422 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5423 (VLD1d16Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5424 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5425 (VLD1d32Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5426 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5427 (VLD1d64Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5428 // with writeback, fixed stride
5429 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5430 (VLD1d8Qwb_fixed VecListFourD:$Vd, zero_reg,
5431 addrmode6:$Rn, pred:$p)>;
5432 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5433 (VLD1d16Qwb_fixed VecListFourD:$Vd, zero_reg,
5434 addrmode6:$Rn, pred:$p)>;
5435 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5436 (VLD1d32Qwb_fixed VecListFourD:$Vd, zero_reg,
5437 addrmode6:$Rn, pred:$p)>;
5438 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5439 (VLD1d64Qwb_fixed VecListFourD:$Vd, zero_reg,
5440 addrmode6:$Rn, pred:$p)>;
5441 // with writeback, register stride
5442 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5443 (VLD1d8Qwb_register VecListFourD:$Vd, zero_reg,
5444 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5445 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5446 (VLD1d16Qwb_register VecListFourD:$Vd, zero_reg,
5447 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5448 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5449 (VLD1d32Qwb_register VecListFourD:$Vd, zero_reg,
5450 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5451 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5452 (VLD1d64Qwb_register VecListFourD:$Vd, zero_reg,
5453 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5455 // VST1 requires a size suffix, but also accepts type specific variants.
5456 // Store one D register.
5457 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5458 (VST1d8 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5459 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5460 (VST1d16 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5461 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5462 (VST1d32 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5463 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5464 (VST1d64 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5465 // with writeback, fixed stride
5466 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5467 (VST1d8wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5468 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5469 (VST1d16wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5470 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5471 (VST1d32wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5472 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5473 (VST1d64wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5474 // with writeback, register stride
5475 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5476 (VST1d8wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5477 VecListOneD:$Vd, pred:$p)>;
5478 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5479 (VST1d16wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5480 VecListOneD:$Vd, pred:$p)>;
5481 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5482 (VST1d32wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5483 VecListOneD:$Vd, pred:$p)>;
5484 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5485 (VST1d64wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5486 VecListOneD:$Vd, pred:$p)>;
5488 // Store two D registers.
5489 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5490 (VST1q8 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5491 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5492 (VST1q16 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5493 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5494 (VST1q32 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5495 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5496 (VST1q64 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5497 // with writeback, fixed stride
5498 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5499 (VST1q8wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5500 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5501 (VST1q16wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5502 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5503 (VST1q32wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5504 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5505 (VST1q64wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5506 // with writeback, register stride
5507 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5508 (VST1q8wb_register zero_reg, addrmode6:$Rn,
5509 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5510 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5511 (VST1q16wb_register zero_reg, addrmode6:$Rn,
5512 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5513 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5514 (VST1q32wb_register zero_reg, addrmode6:$Rn,
5515 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5516 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5517 (VST1q64wb_register zero_reg, addrmode6:$Rn,
5518 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5520 // Load three D registers.
5521 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5522 (VST1d8T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5523 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5524 (VST1d16T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5525 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5526 (VST1d32T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5527 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5528 (VST1d64T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5529 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5530 (VST1d8Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5531 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5532 (VST1d16Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5533 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5534 (VST1d32Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5535 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5536 (VST1d64Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5537 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5538 (VST1d8Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5539 VecListThreeD:$Vd, pred:$p)>;
5540 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5541 (VST1d16Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5542 VecListThreeD:$Vd, pred:$p)>;
5543 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5544 (VST1d32Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5545 VecListThreeD:$Vd, pred:$p)>;
5546 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5547 (VST1d64Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5548 VecListThreeD:$Vd, pred:$p)>;
5550 // Load four D registers.
5551 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5552 (VST1d8Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5553 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5554 (VST1d16Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5555 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5556 (VST1d32Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5557 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5558 (VST1d64Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5559 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5560 (VST1d8Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5561 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5562 (VST1d16Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5563 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5564 (VST1d32Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5565 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5566 (VST1d64Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5567 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5568 (VST1d8Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5569 VecListFourD:$Vd, pred:$p)>;
5570 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5571 (VST1d16Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5572 VecListFourD:$Vd, pred:$p)>;
5573 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5574 (VST1d32Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5575 VecListFourD:$Vd, pred:$p)>;
5576 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5577 (VST1d64Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5578 VecListFourD:$Vd, pred:$p)>;
5581 // VTRN instructions data type suffix aliases for more-specific types.
5582 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Dd, $Dm",
5583 (VTRNd8 DPR:$Dd, DPR:$Dm, pred:$p)>;
5584 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5585 (VTRNd16 DPR:$Dd, DPR:$Dm, pred:$p)>;
5586 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5587 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5589 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Qd, $Qm",
5590 (VTRNq8 QPR:$Qd, QPR:$Qm, pred:$p)>;
5591 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5592 (VTRNq16 QPR:$Qd, QPR:$Qm, pred:$p)>;
5593 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5594 (VTRNq32 QPR:$Qd, QPR:$Qm, pred:$p)>;
5596 // FIXME: Proof of concept pseudos. We want to parameterize these for all
5597 // the suffices we have to support.
5598 defm VLD1LNdAsm : NEONDT8AsmPseudoInst<"vld1${p}", "$list, $addr",
5599 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5600 defm VLD1LNdAsm : NEONDT16AsmPseudoInst<"vld1${p}", "$list, $addr",
5601 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;
5602 defm VLD1LNdAsm : NEONDT32AsmPseudoInst<"vld1${p}", "$list, $addr",
5603 (ins VecListOneDByteIndexed:$list, addrmode6:$addr, pred:$p)>;