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 //===----------------------------------------------------------------------===//
124 // NEON-specific DAG Nodes.
125 //===----------------------------------------------------------------------===//
127 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
128 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
130 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
131 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
132 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
133 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
134 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
135 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
136 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
137 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
138 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
139 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
140 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
142 // Types for vector shift by immediates. The "SHX" version is for long and
143 // narrow operations where the source and destination vectors have different
144 // types. The "SHINS" version is for shift and insert operations.
145 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
147 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
149 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
150 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
152 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
153 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
154 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
155 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
156 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
157 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
158 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
160 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
161 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
162 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
164 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
165 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
166 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
167 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
168 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
169 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
171 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
172 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
173 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
175 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
176 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
178 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
180 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
181 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
183 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
184 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
185 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
186 def NEONvmovFPImm : SDNode<"ARMISD::VMOVFPIMM", SDTARMVMOVIMM>;
188 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
190 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
191 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
193 def NEONvbsl : SDNode<"ARMISD::VBSL",
194 SDTypeProfile<1, 3, [SDTCisVec<0>,
197 SDTCisSameAs<0, 3>]>>;
199 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
201 // VDUPLANE can produce a quad-register result from a double-register source,
202 // so the result is not constrained to match the source.
203 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
204 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
207 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
208 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
209 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
211 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
212 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
213 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
214 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
216 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
218 SDTCisSameAs<0, 3>]>;
219 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
220 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
221 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
223 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
224 SDTCisSameAs<1, 2>]>;
225 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
226 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
228 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
229 SDTCisSameAs<0, 2>]>;
230 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
231 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
233 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
234 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
235 unsigned EltBits = 0;
236 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
237 return (EltBits == 32 && EltVal == 0);
240 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
241 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
242 unsigned EltBits = 0;
243 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
244 return (EltBits == 8 && EltVal == 0xff);
247 //===----------------------------------------------------------------------===//
248 // NEON load / store instructions
249 //===----------------------------------------------------------------------===//
251 // Use VLDM to load a Q register as a D register pair.
252 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
254 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
256 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
258 // Use VSTM to store a Q register as a D register pair.
259 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
261 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
263 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
265 // Classes for VLD* pseudo-instructions with multi-register operands.
266 // These are expanded to real instructions after register allocation.
267 class VLDQPseudo<InstrItinClass itin>
268 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
269 class VLDQWBPseudo<InstrItinClass itin>
270 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
271 (ins addrmode6:$addr, am6offset:$offset), itin,
273 class VLDQWBfixedPseudo<InstrItinClass itin>
274 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
275 (ins addrmode6:$addr), itin,
277 class VLDQWBregisterPseudo<InstrItinClass itin>
278 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
279 (ins addrmode6:$addr, rGPR:$offset), itin,
281 class VLDQQPseudo<InstrItinClass itin>
282 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
283 class VLDQQWBPseudo<InstrItinClass itin>
284 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
285 (ins addrmode6:$addr, am6offset:$offset), itin,
287 class VLDQQQQPseudo<InstrItinClass itin>
288 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
290 class VLDQQQQWBPseudo<InstrItinClass itin>
291 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
292 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
293 "$addr.addr = $wb, $src = $dst">;
295 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
297 // VLD1 : Vector Load (multiple single elements)
298 class VLD1D<bits<4> op7_4, string Dt>
299 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
300 (ins addrmode6:$Rn), IIC_VLD1,
301 "vld1", Dt, "$Vd, $Rn", "", []> {
304 let DecoderMethod = "DecodeVLDInstruction";
306 class VLD1Q<bits<4> op7_4, string Dt>
307 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
308 (ins addrmode6:$Rn), IIC_VLD1x2,
309 "vld1", Dt, "$Vd, $Rn", "", []> {
311 let Inst{5-4} = Rn{5-4};
312 let DecoderMethod = "DecodeVLDInstruction";
315 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
316 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
317 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
318 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
320 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
321 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
322 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
323 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
325 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
326 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
327 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
328 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
330 // ...with address register writeback:
331 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
332 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
333 (ins addrmode6:$Rn), IIC_VLD1u,
334 "vld1", Dt, "$Vd, $Rn!",
335 "$Rn.addr = $wb", []> {
336 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
338 let DecoderMethod = "DecodeVLDInstruction";
339 let AsmMatchConverter = "cvtVLDwbFixed";
341 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
342 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
343 "vld1", Dt, "$Vd, $Rn, $Rm",
344 "$Rn.addr = $wb", []> {
346 let DecoderMethod = "DecodeVLDInstruction";
347 let AsmMatchConverter = "cvtVLDwbRegister";
350 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
351 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
352 (ins addrmode6:$Rn), IIC_VLD1x2u,
353 "vld1", Dt, "$Vd, $Rn!",
354 "$Rn.addr = $wb", []> {
355 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
356 let Inst{5-4} = Rn{5-4};
357 let DecoderMethod = "DecodeVLDInstruction";
358 let AsmMatchConverter = "cvtVLDwbFixed";
360 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
361 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
362 "vld1", Dt, "$Vd, $Rn, $Rm",
363 "$Rn.addr = $wb", []> {
364 let Inst{5-4} = Rn{5-4};
365 let DecoderMethod = "DecodeVLDInstruction";
366 let AsmMatchConverter = "cvtVLDwbRegister";
370 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
371 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
372 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
373 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
374 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
375 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
376 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
377 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
379 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
380 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
381 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
382 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
383 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
384 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
385 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
386 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
388 // ...with 3 registers
389 class VLD1D3<bits<4> op7_4, string Dt>
390 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
391 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
392 "$Vd, $Rn", "", []> {
395 let DecoderMethod = "DecodeVLDInstruction";
397 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
398 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
399 (ins addrmode6:$Rn), IIC_VLD1x2u,
400 "vld1", Dt, "$Vd, $Rn!",
401 "$Rn.addr = $wb", []> {
402 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
404 let DecoderMethod = "DecodeVLDInstruction";
405 let AsmMatchConverter = "cvtVLDwbFixed";
407 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
408 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
409 "vld1", Dt, "$Vd, $Rn, $Rm",
410 "$Rn.addr = $wb", []> {
412 let DecoderMethod = "DecodeVLDInstruction";
413 let AsmMatchConverter = "cvtVLDwbRegister";
417 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
418 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
419 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
420 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
422 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
423 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
424 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
425 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
427 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
429 // ...with 4 registers
430 class VLD1D4<bits<4> op7_4, string Dt>
431 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
432 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
433 "$Vd, $Rn", "", []> {
435 let Inst{5-4} = Rn{5-4};
436 let DecoderMethod = "DecodeVLDInstruction";
438 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
439 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
440 (ins addrmode6:$Rn), IIC_VLD1x2u,
441 "vld1", Dt, "$Vd, $Rn!",
442 "$Rn.addr = $wb", []> {
443 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
444 let Inst{5-4} = Rn{5-4};
445 let DecoderMethod = "DecodeVLDInstruction";
446 let AsmMatchConverter = "cvtVLDwbFixed";
448 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
449 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
450 "vld1", Dt, "$Vd, $Rn, $Rm",
451 "$Rn.addr = $wb", []> {
452 let Inst{5-4} = Rn{5-4};
453 let DecoderMethod = "DecodeVLDInstruction";
454 let AsmMatchConverter = "cvtVLDwbRegister";
458 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
459 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
460 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
461 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
463 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
464 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
465 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
466 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
468 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
470 // VLD2 : Vector Load (multiple 2-element structures)
471 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
472 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
473 (ins addrmode6:$Rn), IIC_VLD2,
474 "vld2", Dt, "$Vd, $Rn", "", []> {
476 let Inst{5-4} = Rn{5-4};
477 let DecoderMethod = "DecodeVLDInstruction";
479 class VLD2Q<bits<4> op7_4, string Dt, RegisterOperand VdTy>
480 : NLdSt<0, 0b10, 0b0011, op7_4,
482 (ins addrmode6:$Rn), IIC_VLD2x2,
483 "vld2", Dt, "$Vd, $Rn", "", []> {
485 let Inst{5-4} = Rn{5-4};
486 let DecoderMethod = "DecodeVLDInstruction";
489 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
490 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
491 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
493 def VLD2q8 : VLD2Q<{0,0,?,?}, "8", VecListFourD>;
494 def VLD2q16 : VLD2Q<{0,1,?,?}, "16", VecListFourD>;
495 def VLD2q32 : VLD2Q<{1,0,?,?}, "32", VecListFourD>;
497 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
498 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
499 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
501 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
502 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
503 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
505 // ...with address register writeback:
506 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
507 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
508 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
509 "vld2", Dt, "$Vd, $Rn$Rm",
510 "$Rn.addr = $wb", []> {
511 let Inst{5-4} = Rn{5-4};
512 let DecoderMethod = "DecodeVLDInstruction";
514 class VLD2QWB<bits<4> op7_4, string Dt, RegisterOperand VdTy>
515 : NLdSt<0, 0b10, 0b0011, op7_4,
516 (outs VdTy:$Vd, GPR:$wb),
517 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
518 "vld2", Dt, "$Vd, $Rn$Rm",
519 "$Rn.addr = $wb", []> {
520 let Inst{5-4} = Rn{5-4};
521 let DecoderMethod = "DecodeVLDInstruction";
524 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
525 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
526 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
528 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8", VecListFourD>;
529 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16", VecListFourD>;
530 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32", VecListFourD>;
532 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
533 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
534 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
536 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
537 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
538 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
540 // ...with double-spaced registers
541 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
542 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
543 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
544 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
545 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
546 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
548 // VLD3 : Vector Load (multiple 3-element structures)
549 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
550 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
551 (ins addrmode6:$Rn), IIC_VLD3,
552 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
555 let DecoderMethod = "DecodeVLDInstruction";
558 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
559 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
560 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
562 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
563 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
564 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
566 // ...with address register writeback:
567 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
568 : NLdSt<0, 0b10, op11_8, op7_4,
569 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
570 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
571 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
572 "$Rn.addr = $wb", []> {
574 let DecoderMethod = "DecodeVLDInstruction";
577 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
578 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
579 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
581 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
582 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
583 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
585 // ...with double-spaced registers:
586 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
587 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
588 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
589 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
590 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
591 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
593 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
594 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
595 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
597 // ...alternate versions to be allocated odd register numbers:
598 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
599 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
600 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
602 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
603 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
604 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
606 // VLD4 : Vector Load (multiple 4-element structures)
607 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
608 : NLdSt<0, 0b10, op11_8, op7_4,
609 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
610 (ins addrmode6:$Rn), IIC_VLD4,
611 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
613 let Inst{5-4} = Rn{5-4};
614 let DecoderMethod = "DecodeVLDInstruction";
617 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
618 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
619 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
621 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
622 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
623 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
625 // ...with address register writeback:
626 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
627 : NLdSt<0, 0b10, op11_8, op7_4,
628 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
629 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
630 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
631 "$Rn.addr = $wb", []> {
632 let Inst{5-4} = Rn{5-4};
633 let DecoderMethod = "DecodeVLDInstruction";
636 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
637 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
638 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
640 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
641 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
642 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
644 // ...with double-spaced registers:
645 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
646 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
647 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
648 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
649 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
650 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
652 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
653 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
654 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
656 // ...alternate versions to be allocated odd register numbers:
657 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
658 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
659 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
661 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
662 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
663 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
665 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
667 // Classes for VLD*LN pseudo-instructions with multi-register operands.
668 // These are expanded to real instructions after register allocation.
669 class VLDQLNPseudo<InstrItinClass itin>
670 : PseudoNLdSt<(outs QPR:$dst),
671 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
672 itin, "$src = $dst">;
673 class VLDQLNWBPseudo<InstrItinClass itin>
674 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
675 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
676 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
677 class VLDQQLNPseudo<InstrItinClass itin>
678 : PseudoNLdSt<(outs QQPR:$dst),
679 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
680 itin, "$src = $dst">;
681 class VLDQQLNWBPseudo<InstrItinClass itin>
682 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
683 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
684 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
685 class VLDQQQQLNPseudo<InstrItinClass itin>
686 : PseudoNLdSt<(outs QQQQPR:$dst),
687 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
688 itin, "$src = $dst">;
689 class VLDQQQQLNWBPseudo<InstrItinClass itin>
690 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
691 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
692 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
694 // VLD1LN : Vector Load (single element to one lane)
695 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
697 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
698 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
699 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
701 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
702 (i32 (LoadOp addrmode6:$Rn)),
705 let DecoderMethod = "DecodeVLD1LN";
707 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
709 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
710 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
711 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
713 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
714 (i32 (LoadOp addrmode6oneL32:$Rn)),
717 let DecoderMethod = "DecodeVLD1LN";
719 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
720 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
721 (i32 (LoadOp addrmode6:$addr)),
725 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
726 let Inst{7-5} = lane{2-0};
728 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
729 let Inst{7-6} = lane{1-0};
732 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
733 let Inst{7} = lane{0};
738 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
739 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
740 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
742 def : Pat<(vector_insert (v2f32 DPR:$src),
743 (f32 (load addrmode6:$addr)), imm:$lane),
744 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
745 def : Pat<(vector_insert (v4f32 QPR:$src),
746 (f32 (load addrmode6:$addr)), imm:$lane),
747 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
749 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
751 // ...with address register writeback:
752 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
753 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
754 (ins addrmode6:$Rn, am6offset:$Rm,
755 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
756 "\\{$Vd[$lane]\\}, $Rn$Rm",
757 "$src = $Vd, $Rn.addr = $wb", []> {
758 let DecoderMethod = "DecodeVLD1LN";
761 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
762 let Inst{7-5} = lane{2-0};
764 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
765 let Inst{7-6} = lane{1-0};
768 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
769 let Inst{7} = lane{0};
774 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
775 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
776 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
778 // VLD2LN : Vector Load (single 2-element structure to one lane)
779 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
780 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
781 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
782 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
783 "$src1 = $Vd, $src2 = $dst2", []> {
786 let DecoderMethod = "DecodeVLD2LN";
789 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
790 let Inst{7-5} = lane{2-0};
792 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
793 let Inst{7-6} = lane{1-0};
795 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
796 let Inst{7} = lane{0};
799 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
800 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
801 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
803 // ...with double-spaced registers:
804 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
805 let Inst{7-6} = lane{1-0};
807 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
808 let Inst{7} = lane{0};
811 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
812 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
814 // ...with address register writeback:
815 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
816 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
817 (ins addrmode6:$Rn, am6offset:$Rm,
818 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
819 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
820 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
822 let DecoderMethod = "DecodeVLD2LN";
825 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
826 let Inst{7-5} = lane{2-0};
828 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
829 let Inst{7-6} = lane{1-0};
831 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
832 let Inst{7} = lane{0};
835 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
836 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
837 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
839 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
840 let Inst{7-6} = lane{1-0};
842 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
843 let Inst{7} = lane{0};
846 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
847 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
849 // VLD3LN : Vector Load (single 3-element structure to one lane)
850 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
851 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
852 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
853 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
854 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
855 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
857 let DecoderMethod = "DecodeVLD3LN";
860 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
861 let Inst{7-5} = lane{2-0};
863 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
864 let Inst{7-6} = lane{1-0};
866 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
867 let Inst{7} = lane{0};
870 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
871 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
872 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
874 // ...with double-spaced registers:
875 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
876 let Inst{7-6} = lane{1-0};
878 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
879 let Inst{7} = lane{0};
882 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
883 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
885 // ...with address register writeback:
886 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
887 : NLdStLn<1, 0b10, op11_8, op7_4,
888 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
889 (ins addrmode6:$Rn, am6offset:$Rm,
890 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
891 IIC_VLD3lnu, "vld3", Dt,
892 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
893 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
895 let DecoderMethod = "DecodeVLD3LN";
898 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
899 let Inst{7-5} = lane{2-0};
901 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
902 let Inst{7-6} = lane{1-0};
904 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
905 let Inst{7} = lane{0};
908 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
909 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
910 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
912 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
913 let Inst{7-6} = lane{1-0};
915 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
916 let Inst{7} = lane{0};
919 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
920 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
922 // VLD4LN : Vector Load (single 4-element structure to one lane)
923 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
924 : NLdStLn<1, 0b10, op11_8, op7_4,
925 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
926 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
927 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
928 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
929 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
932 let DecoderMethod = "DecodeVLD4LN";
935 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
936 let Inst{7-5} = lane{2-0};
938 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
939 let Inst{7-6} = lane{1-0};
941 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
942 let Inst{7} = lane{0};
946 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
947 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
948 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
950 // ...with double-spaced registers:
951 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
952 let Inst{7-6} = lane{1-0};
954 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
955 let Inst{7} = lane{0};
959 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
960 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
962 // ...with address register writeback:
963 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
964 : NLdStLn<1, 0b10, op11_8, op7_4,
965 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
966 (ins addrmode6:$Rn, am6offset:$Rm,
967 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
968 IIC_VLD4lnu, "vld4", Dt,
969 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
970 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
973 let DecoderMethod = "DecodeVLD4LN" ;
976 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
977 let Inst{7-5} = lane{2-0};
979 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
980 let Inst{7-6} = lane{1-0};
982 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
983 let Inst{7} = lane{0};
987 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
988 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
989 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
991 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
992 let Inst{7-6} = lane{1-0};
994 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
995 let Inst{7} = lane{0};
999 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1000 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
1002 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1004 // VLD1DUP : Vector Load (single element to all lanes)
1005 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
1006 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
1007 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
1008 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1010 let Inst{4} = Rn{4};
1011 let DecoderMethod = "DecodeVLD1DupInstruction";
1013 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1014 let Pattern = [(set QPR:$dst,
1015 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1018 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1019 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1020 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1022 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1023 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1024 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1026 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1027 (VLD1DUPd32 addrmode6:$addr)>;
1028 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1029 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1031 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1033 class VLD1QDUP<bits<4> op7_4, string Dt>
1034 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
1035 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1036 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1038 let Inst{4} = Rn{4};
1039 let DecoderMethod = "DecodeVLD1DupInstruction";
1042 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1043 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1044 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1046 // ...with address register writeback:
1047 class VLD1DUPWB<bits<4> op7_4, string Dt>
1048 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
1049 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1050 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1051 let Inst{4} = Rn{4};
1052 let DecoderMethod = "DecodeVLD1DupInstruction";
1054 class VLD1QDUPWB<bits<4> op7_4, string Dt>
1055 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1056 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1057 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1058 let Inst{4} = Rn{4};
1059 let DecoderMethod = "DecodeVLD1DupInstruction";
1062 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
1063 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
1064 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
1066 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
1067 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
1068 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
1070 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1071 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1072 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1074 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1075 class VLD2DUP<bits<4> op7_4, string Dt>
1076 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1077 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1078 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1080 let Inst{4} = Rn{4};
1081 let DecoderMethod = "DecodeVLD2DupInstruction";
1084 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1085 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1086 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1088 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1089 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1090 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1092 // ...with double-spaced registers (not used for codegen):
1093 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1094 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1095 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1097 // ...with address register writeback:
1098 class VLD2DUPWB<bits<4> op7_4, string Dt>
1099 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1100 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1101 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1102 let Inst{4} = Rn{4};
1103 let DecoderMethod = "DecodeVLD2DupInstruction";
1106 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1107 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1108 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1110 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1111 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1112 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1114 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1115 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1116 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1118 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1119 class VLD3DUP<bits<4> op7_4, string Dt>
1120 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1121 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1122 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1125 let DecoderMethod = "DecodeVLD3DupInstruction";
1128 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1129 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1130 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1132 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1133 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1134 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1136 // ...with double-spaced registers (not used for codegen):
1137 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1138 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1139 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1141 // ...with address register writeback:
1142 class VLD3DUPWB<bits<4> op7_4, string Dt>
1143 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1144 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1145 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1146 "$Rn.addr = $wb", []> {
1148 let DecoderMethod = "DecodeVLD3DupInstruction";
1151 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1152 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1153 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1155 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1156 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1157 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1159 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1160 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1161 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1163 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1164 class VLD4DUP<bits<4> op7_4, string Dt>
1165 : NLdSt<1, 0b10, 0b1111, op7_4,
1166 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1167 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1168 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1170 let Inst{4} = Rn{4};
1171 let DecoderMethod = "DecodeVLD4DupInstruction";
1174 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1175 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1176 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1178 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1179 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1180 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1182 // ...with double-spaced registers (not used for codegen):
1183 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1184 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1185 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1187 // ...with address register writeback:
1188 class VLD4DUPWB<bits<4> op7_4, string Dt>
1189 : NLdSt<1, 0b10, 0b1111, op7_4,
1190 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1191 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1192 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1193 "$Rn.addr = $wb", []> {
1194 let Inst{4} = Rn{4};
1195 let DecoderMethod = "DecodeVLD4DupInstruction";
1198 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1199 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1200 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1202 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1203 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1204 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1206 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1207 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1208 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1210 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1212 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1214 // Classes for VST* pseudo-instructions with multi-register operands.
1215 // These are expanded to real instructions after register allocation.
1216 class VSTQPseudo<InstrItinClass itin>
1217 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1218 class VSTQWBPseudo<InstrItinClass itin>
1219 : PseudoNLdSt<(outs GPR:$wb),
1220 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1221 "$addr.addr = $wb">;
1222 class VSTQWBfixedPseudo<InstrItinClass itin>
1223 : PseudoNLdSt<(outs GPR:$wb),
1224 (ins addrmode6:$addr, QPR:$src), itin,
1225 "$addr.addr = $wb">;
1226 class VSTQWBregisterPseudo<InstrItinClass itin>
1227 : PseudoNLdSt<(outs GPR:$wb),
1228 (ins addrmode6:$addr, rGPR:$offset, QPR:$src), itin,
1229 "$addr.addr = $wb">;
1230 class VSTQQPseudo<InstrItinClass itin>
1231 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1232 class VSTQQWBPseudo<InstrItinClass itin>
1233 : PseudoNLdSt<(outs GPR:$wb),
1234 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1235 "$addr.addr = $wb">;
1236 class VSTQQQQPseudo<InstrItinClass itin>
1237 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1238 class VSTQQQQWBPseudo<InstrItinClass itin>
1239 : PseudoNLdSt<(outs GPR:$wb),
1240 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1241 "$addr.addr = $wb">;
1243 // VST1 : Vector Store (multiple single elements)
1244 class VST1D<bits<4> op7_4, string Dt>
1245 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1246 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1248 let Inst{4} = Rn{4};
1249 let DecoderMethod = "DecodeVSTInstruction";
1251 class VST1Q<bits<4> op7_4, string Dt>
1252 : NLdSt<0,0b00,0b1010,op7_4, (outs), (ins addrmode6:$Rn, VecListTwoD:$Vd),
1253 IIC_VST1x2, "vst1", Dt, "$Vd, $Rn", "", []> {
1255 let Inst{5-4} = Rn{5-4};
1256 let DecoderMethod = "DecodeVSTInstruction";
1259 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1260 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1261 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1262 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1264 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1265 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1266 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1267 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1269 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1270 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1271 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1272 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1274 // ...with address register writeback:
1275 multiclass VST1DWB<bits<4> op7_4, string Dt> {
1276 def _fixed : NLdSt<0,0b00, 0b0111,op7_4, (outs GPR:$wb),
1277 (ins addrmode6:$Rn, VecListOneD:$Vd), IIC_VLD1u,
1278 "vst1", Dt, "$Vd, $Rn!",
1279 "$Rn.addr = $wb", []> {
1280 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1281 let Inst{4} = Rn{4};
1282 let DecoderMethod = "DecodeVSTInstruction";
1283 let AsmMatchConverter = "cvtVSTwbFixed";
1285 def _register : NLdSt<0,0b00,0b0111,op7_4, (outs GPR:$wb),
1286 (ins addrmode6:$Rn, rGPR:$Rm, VecListOneD:$Vd),
1288 "vst1", Dt, "$Vd, $Rn, $Rm",
1289 "$Rn.addr = $wb", []> {
1290 let Inst{4} = Rn{4};
1291 let DecoderMethod = "DecodeVSTInstruction";
1292 let AsmMatchConverter = "cvtVSTwbRegister";
1295 multiclass VST1QWB<bits<4> op7_4, string Dt> {
1296 def _fixed : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1297 (ins addrmode6:$Rn, VecListTwoD:$Vd), IIC_VLD1x2u,
1298 "vst1", Dt, "$Vd, $Rn!",
1299 "$Rn.addr = $wb", []> {
1300 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1301 let Inst{5-4} = Rn{5-4};
1302 let DecoderMethod = "DecodeVSTInstruction";
1303 let AsmMatchConverter = "cvtVSTwbFixed";
1305 def _register : NLdSt<0,0b00,0b1010,op7_4, (outs GPR:$wb),
1306 (ins addrmode6:$Rn, rGPR:$Rm, VecListTwoD:$Vd),
1308 "vst1", Dt, "$Vd, $Rn, $Rm",
1309 "$Rn.addr = $wb", []> {
1310 let Inst{5-4} = Rn{5-4};
1311 let DecoderMethod = "DecodeVSTInstruction";
1312 let AsmMatchConverter = "cvtVSTwbRegister";
1316 defm VST1d8wb : VST1DWB<{0,0,0,?}, "8">;
1317 defm VST1d16wb : VST1DWB<{0,1,0,?}, "16">;
1318 defm VST1d32wb : VST1DWB<{1,0,0,?}, "32">;
1319 defm VST1d64wb : VST1DWB<{1,1,0,?}, "64">;
1321 defm VST1q8wb : VST1QWB<{0,0,?,?}, "8">;
1322 defm VST1q16wb : VST1QWB<{0,1,?,?}, "16">;
1323 defm VST1q32wb : VST1QWB<{1,0,?,?}, "32">;
1324 defm VST1q64wb : VST1QWB<{1,1,?,?}, "64">;
1326 def VST1q8PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1327 def VST1q16PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1328 def VST1q32PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1329 def VST1q64PseudoWB_fixed : VSTQWBfixedPseudo<IIC_VST1x2u>;
1330 def VST1q8PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1331 def VST1q16PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1332 def VST1q32PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1333 def VST1q64PseudoWB_register : VSTQWBregisterPseudo<IIC_VST1x2u>;
1335 // ...with 3 registers
1336 class VST1D3<bits<4> op7_4, string Dt>
1337 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1338 (ins addrmode6:$Rn, VecListThreeD:$Vd),
1339 IIC_VST1x3, "vst1", Dt, "$Vd, $Rn", "", []> {
1341 let Inst{4} = Rn{4};
1342 let DecoderMethod = "DecodeVSTInstruction";
1344 multiclass VST1D3WB<bits<4> op7_4, string Dt> {
1345 def _fixed : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1346 (ins addrmode6:$Rn, VecListThreeD:$Vd), IIC_VLD1x3u,
1347 "vst1", Dt, "$Vd, $Rn!",
1348 "$Rn.addr = $wb", []> {
1349 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1350 let Inst{5-4} = Rn{5-4};
1351 let DecoderMethod = "DecodeVSTInstruction";
1352 let AsmMatchConverter = "cvtVSTwbFixed";
1354 def _register : NLdSt<0,0b00,0b0110,op7_4, (outs GPR:$wb),
1355 (ins addrmode6:$Rn, rGPR:$Rm, VecListThreeD:$Vd),
1357 "vst1", Dt, "$Vd, $Rn, $Rm",
1358 "$Rn.addr = $wb", []> {
1359 let Inst{5-4} = Rn{5-4};
1360 let DecoderMethod = "DecodeVSTInstruction";
1361 let AsmMatchConverter = "cvtVSTwbRegister";
1365 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1366 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1367 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1368 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1370 defm VST1d8Twb : VST1D3WB<{0,0,0,?}, "8">;
1371 defm VST1d16Twb : VST1D3WB<{0,1,0,?}, "16">;
1372 defm VST1d32Twb : VST1D3WB<{1,0,0,?}, "32">;
1373 defm VST1d64Twb : VST1D3WB<{1,1,0,?}, "64">;
1375 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1376 def VST1d64TPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x3u>;
1377 def VST1d64TPseudoWB_register : VSTQQWBPseudo<IIC_VST1x3u>;
1379 // ...with 4 registers
1380 class VST1D4<bits<4> op7_4, string Dt>
1381 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1382 (ins addrmode6:$Rn, VecListFourD:$Vd),
1383 IIC_VST1x4, "vst1", Dt, "$Vd, $Rn", "",
1386 let Inst{5-4} = Rn{5-4};
1387 let DecoderMethod = "DecodeVSTInstruction";
1389 multiclass VST1D4WB<bits<4> op7_4, string Dt> {
1390 def _fixed : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1391 (ins addrmode6:$Rn, VecListFourD:$Vd), IIC_VLD1x4u,
1392 "vst1", Dt, "$Vd, $Rn!",
1393 "$Rn.addr = $wb", []> {
1394 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
1395 let Inst{5-4} = Rn{5-4};
1396 let DecoderMethod = "DecodeVSTInstruction";
1397 let AsmMatchConverter = "cvtVSTwbFixed";
1399 def _register : NLdSt<0,0b00,0b0010,op7_4, (outs GPR:$wb),
1400 (ins addrmode6:$Rn, rGPR:$Rm, VecListFourD:$Vd),
1402 "vst1", Dt, "$Vd, $Rn, $Rm",
1403 "$Rn.addr = $wb", []> {
1404 let Inst{5-4} = Rn{5-4};
1405 let DecoderMethod = "DecodeVSTInstruction";
1406 let AsmMatchConverter = "cvtVSTwbRegister";
1410 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1411 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1412 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1413 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1415 defm VST1d8Qwb : VST1D4WB<{0,0,?,?}, "8">;
1416 defm VST1d16Qwb : VST1D4WB<{0,1,?,?}, "16">;
1417 defm VST1d32Qwb : VST1D4WB<{1,0,?,?}, "32">;
1418 defm VST1d64Qwb : VST1D4WB<{1,1,?,?}, "64">;
1420 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1421 def VST1d64QPseudoWB_fixed : VSTQQWBPseudo<IIC_VST1x4u>;
1422 def VST1d64QPseudoWB_register : VSTQQWBPseudo<IIC_VST1x4u>;
1424 // VST2 : Vector Store (multiple 2-element structures)
1425 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1426 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1427 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1428 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1430 let Inst{5-4} = Rn{5-4};
1431 let DecoderMethod = "DecodeVSTInstruction";
1433 class VST2Q<bits<4> op7_4, string Dt>
1434 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1435 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1436 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1439 let Inst{5-4} = Rn{5-4};
1440 let DecoderMethod = "DecodeVSTInstruction";
1443 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1444 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1445 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1447 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1448 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1449 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1451 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1452 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1453 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1455 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1456 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1457 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1459 // ...with address register writeback:
1460 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1461 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1462 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1463 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1464 "$Rn.addr = $wb", []> {
1465 let Inst{5-4} = Rn{5-4};
1466 let DecoderMethod = "DecodeVSTInstruction";
1468 class VST2QWB<bits<4> op7_4, string Dt>
1469 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1470 (ins addrmode6:$Rn, am6offset:$Rm,
1471 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1472 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1473 "$Rn.addr = $wb", []> {
1474 let Inst{5-4} = Rn{5-4};
1475 let DecoderMethod = "DecodeVSTInstruction";
1478 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1479 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1480 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1482 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1483 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1484 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1486 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1487 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1488 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1490 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1491 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1492 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1494 // ...with double-spaced registers
1495 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1496 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1497 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1498 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1499 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1500 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1502 // VST3 : Vector Store (multiple 3-element structures)
1503 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1504 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1505 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1506 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1508 let Inst{4} = Rn{4};
1509 let DecoderMethod = "DecodeVSTInstruction";
1512 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1513 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1514 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1516 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1517 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1518 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1520 // ...with address register writeback:
1521 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1522 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1523 (ins addrmode6:$Rn, am6offset:$Rm,
1524 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1525 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1526 "$Rn.addr = $wb", []> {
1527 let Inst{4} = Rn{4};
1528 let DecoderMethod = "DecodeVSTInstruction";
1531 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1532 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1533 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1535 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1536 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1537 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1539 // ...with double-spaced registers:
1540 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1541 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1542 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1543 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1544 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1545 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1547 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1548 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1549 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1551 // ...alternate versions to be allocated odd register numbers:
1552 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1553 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1554 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1556 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1557 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1558 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1560 // VST4 : Vector Store (multiple 4-element structures)
1561 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1562 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1563 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1564 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1567 let Inst{5-4} = Rn{5-4};
1568 let DecoderMethod = "DecodeVSTInstruction";
1571 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1572 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1573 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1575 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1576 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1577 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1579 // ...with address register writeback:
1580 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1581 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1582 (ins addrmode6:$Rn, am6offset:$Rm,
1583 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1584 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1585 "$Rn.addr = $wb", []> {
1586 let Inst{5-4} = Rn{5-4};
1587 let DecoderMethod = "DecodeVSTInstruction";
1590 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1591 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1592 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1594 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1595 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1596 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1598 // ...with double-spaced registers:
1599 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1600 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1601 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1602 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1603 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1604 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1606 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1607 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1608 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1610 // ...alternate versions to be allocated odd register numbers:
1611 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1612 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1613 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1615 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1616 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1617 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1619 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1621 // Classes for VST*LN pseudo-instructions with multi-register operands.
1622 // These are expanded to real instructions after register allocation.
1623 class VSTQLNPseudo<InstrItinClass itin>
1624 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1626 class VSTQLNWBPseudo<InstrItinClass itin>
1627 : PseudoNLdSt<(outs GPR:$wb),
1628 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1629 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1630 class VSTQQLNPseudo<InstrItinClass itin>
1631 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1633 class VSTQQLNWBPseudo<InstrItinClass itin>
1634 : PseudoNLdSt<(outs GPR:$wb),
1635 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1636 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1637 class VSTQQQQLNPseudo<InstrItinClass itin>
1638 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1640 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1641 : PseudoNLdSt<(outs GPR:$wb),
1642 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1643 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1645 // VST1LN : Vector Store (single element from one lane)
1646 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1647 PatFrag StoreOp, SDNode ExtractOp>
1648 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1649 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1650 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1651 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1653 let DecoderMethod = "DecodeVST1LN";
1655 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1656 PatFrag StoreOp, SDNode ExtractOp>
1657 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1658 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1659 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1660 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1662 let DecoderMethod = "DecodeVST1LN";
1664 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1665 : VSTQLNPseudo<IIC_VST1ln> {
1666 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1670 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1672 let Inst{7-5} = lane{2-0};
1674 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1676 let Inst{7-6} = lane{1-0};
1677 let Inst{4} = Rn{5};
1680 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1681 let Inst{7} = lane{0};
1682 let Inst{5-4} = Rn{5-4};
1685 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1686 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1687 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1689 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1690 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1691 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1692 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1694 // ...with address register writeback:
1695 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1696 PatFrag StoreOp, SDNode ExtractOp>
1697 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1698 (ins addrmode6:$Rn, am6offset:$Rm,
1699 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1700 "\\{$Vd[$lane]\\}, $Rn$Rm",
1702 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1703 addrmode6:$Rn, am6offset:$Rm))]> {
1704 let DecoderMethod = "DecodeVST1LN";
1706 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1707 : VSTQLNWBPseudo<IIC_VST1lnu> {
1708 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1709 addrmode6:$addr, am6offset:$offset))];
1712 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1714 let Inst{7-5} = lane{2-0};
1716 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1718 let Inst{7-6} = lane{1-0};
1719 let Inst{4} = Rn{5};
1721 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1723 let Inst{7} = lane{0};
1724 let Inst{5-4} = Rn{5-4};
1727 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1728 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1729 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1731 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1733 // VST2LN : Vector Store (single 2-element structure from one lane)
1734 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1735 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1736 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1737 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1740 let Inst{4} = Rn{4};
1741 let DecoderMethod = "DecodeVST2LN";
1744 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1745 let Inst{7-5} = lane{2-0};
1747 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1748 let Inst{7-6} = lane{1-0};
1750 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1751 let Inst{7} = lane{0};
1754 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1755 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1756 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1758 // ...with double-spaced registers:
1759 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1760 let Inst{7-6} = lane{1-0};
1761 let Inst{4} = Rn{4};
1763 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1764 let Inst{7} = lane{0};
1765 let Inst{4} = Rn{4};
1768 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1769 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1771 // ...with address register writeback:
1772 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1773 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1774 (ins addrmode6:$addr, am6offset:$offset,
1775 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1776 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1777 "$addr.addr = $wb", []> {
1778 let Inst{4} = Rn{4};
1779 let DecoderMethod = "DecodeVST2LN";
1782 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1783 let Inst{7-5} = lane{2-0};
1785 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1786 let Inst{7-6} = lane{1-0};
1788 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1789 let Inst{7} = lane{0};
1792 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1793 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1794 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1796 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1797 let Inst{7-6} = lane{1-0};
1799 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1800 let Inst{7} = lane{0};
1803 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1804 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1806 // VST3LN : Vector Store (single 3-element structure from one lane)
1807 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1808 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1809 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1810 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1811 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1813 let DecoderMethod = "DecodeVST3LN";
1816 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1817 let Inst{7-5} = lane{2-0};
1819 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1820 let Inst{7-6} = lane{1-0};
1822 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1823 let Inst{7} = lane{0};
1826 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1827 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1828 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1830 // ...with double-spaced registers:
1831 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1832 let Inst{7-6} = lane{1-0};
1834 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1835 let Inst{7} = lane{0};
1838 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1839 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1841 // ...with address register writeback:
1842 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1843 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1844 (ins addrmode6:$Rn, am6offset:$Rm,
1845 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1846 IIC_VST3lnu, "vst3", Dt,
1847 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1848 "$Rn.addr = $wb", []> {
1849 let DecoderMethod = "DecodeVST3LN";
1852 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1853 let Inst{7-5} = lane{2-0};
1855 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1856 let Inst{7-6} = lane{1-0};
1858 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1859 let Inst{7} = lane{0};
1862 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1863 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1864 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1866 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1867 let Inst{7-6} = lane{1-0};
1869 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1870 let Inst{7} = lane{0};
1873 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1874 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1876 // VST4LN : Vector Store (single 4-element structure from one lane)
1877 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1878 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1879 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1880 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1881 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1884 let Inst{4} = Rn{4};
1885 let DecoderMethod = "DecodeVST4LN";
1888 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1889 let Inst{7-5} = lane{2-0};
1891 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1892 let Inst{7-6} = lane{1-0};
1894 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1895 let Inst{7} = lane{0};
1896 let Inst{5} = Rn{5};
1899 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1900 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1901 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1903 // ...with double-spaced registers:
1904 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1905 let Inst{7-6} = lane{1-0};
1907 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1908 let Inst{7} = lane{0};
1909 let Inst{5} = Rn{5};
1912 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1913 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1915 // ...with address register writeback:
1916 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1917 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1918 (ins addrmode6:$Rn, am6offset:$Rm,
1919 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1920 IIC_VST4lnu, "vst4", Dt,
1921 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1922 "$Rn.addr = $wb", []> {
1923 let Inst{4} = Rn{4};
1924 let DecoderMethod = "DecodeVST4LN";
1927 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1928 let Inst{7-5} = lane{2-0};
1930 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1931 let Inst{7-6} = lane{1-0};
1933 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1934 let Inst{7} = lane{0};
1935 let Inst{5} = Rn{5};
1938 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1939 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1940 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1942 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1943 let Inst{7-6} = lane{1-0};
1945 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1946 let Inst{7} = lane{0};
1947 let Inst{5} = Rn{5};
1950 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1951 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1953 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1956 //===----------------------------------------------------------------------===//
1957 // NEON pattern fragments
1958 //===----------------------------------------------------------------------===//
1960 // Extract D sub-registers of Q registers.
1961 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1962 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1963 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1965 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1966 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1967 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1969 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1970 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1971 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1973 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1974 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1975 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1978 // Extract S sub-registers of Q/D registers.
1979 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1980 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1981 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1984 // Translate lane numbers from Q registers to D subregs.
1985 def SubReg_i8_lane : SDNodeXForm<imm, [{
1986 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1988 def SubReg_i16_lane : SDNodeXForm<imm, [{
1989 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1991 def SubReg_i32_lane : SDNodeXForm<imm, [{
1992 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1995 //===----------------------------------------------------------------------===//
1996 // Instruction Classes
1997 //===----------------------------------------------------------------------===//
1999 // Basic 2-register operations: double- and quad-register.
2000 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2001 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2002 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2003 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2004 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
2005 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
2006 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2007 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
2008 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
2009 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2010 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
2011 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
2013 // Basic 2-register intrinsics, both double- and quad-register.
2014 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2015 bits<2> op17_16, bits<5> op11_7, bit op4,
2016 InstrItinClass itin, string OpcodeStr, string Dt,
2017 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2018 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2019 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2020 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2021 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2022 bits<2> op17_16, bits<5> op11_7, bit op4,
2023 InstrItinClass itin, string OpcodeStr, string Dt,
2024 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2025 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2026 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2027 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2029 // Narrow 2-register operations.
2030 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2031 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2032 InstrItinClass itin, string OpcodeStr, string Dt,
2033 ValueType TyD, ValueType TyQ, SDNode OpNode>
2034 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2035 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2036 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
2038 // Narrow 2-register intrinsics.
2039 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2040 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2041 InstrItinClass itin, string OpcodeStr, string Dt,
2042 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
2043 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
2044 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2045 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
2047 // Long 2-register operations (currently only used for VMOVL).
2048 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2049 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2050 InstrItinClass itin, string OpcodeStr, string Dt,
2051 ValueType TyQ, ValueType TyD, SDNode OpNode>
2052 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2053 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2054 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
2056 // Long 2-register intrinsics.
2057 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2058 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
2059 InstrItinClass itin, string OpcodeStr, string Dt,
2060 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2061 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
2062 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
2063 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
2065 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
2066 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
2067 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
2068 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
2069 OpcodeStr, Dt, "$Vd, $Vm",
2070 "$src1 = $Vd, $src2 = $Vm", []>;
2071 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2072 InstrItinClass itin, string OpcodeStr, string Dt>
2073 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2074 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2075 "$src1 = $Vd, $src2 = $Vm", []>;
2077 // Basic 3-register operations: double- and quad-register.
2078 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2079 InstrItinClass itin, string OpcodeStr, string Dt,
2080 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2081 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2082 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2083 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2084 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2085 let isCommutable = Commutable;
2087 // Same as N3VD but no data type.
2088 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2089 InstrItinClass itin, string OpcodeStr,
2090 ValueType ResTy, ValueType OpTy,
2091 SDNode OpNode, bit Commutable>
2092 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2093 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2094 OpcodeStr, "$Vd, $Vn, $Vm", "",
2095 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2096 let isCommutable = Commutable;
2099 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2100 InstrItinClass itin, string OpcodeStr, string Dt,
2101 ValueType Ty, SDNode ShOp>
2102 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2103 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2104 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2106 (Ty (ShOp (Ty DPR:$Vn),
2107 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2108 let isCommutable = 0;
2110 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2111 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2112 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2113 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2114 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2116 (Ty (ShOp (Ty DPR:$Vn),
2117 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2118 let isCommutable = 0;
2121 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2122 InstrItinClass itin, string OpcodeStr, string Dt,
2123 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2124 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2125 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2126 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2127 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2128 let isCommutable = Commutable;
2130 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2131 InstrItinClass itin, string OpcodeStr,
2132 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2133 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2134 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2135 OpcodeStr, "$Vd, $Vn, $Vm", "",
2136 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2137 let isCommutable = Commutable;
2139 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2140 InstrItinClass itin, string OpcodeStr, string Dt,
2141 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2142 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2143 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2144 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2145 [(set (ResTy QPR:$Vd),
2146 (ResTy (ShOp (ResTy QPR:$Vn),
2147 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2149 let isCommutable = 0;
2151 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2152 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2153 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2154 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2155 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2156 [(set (ResTy QPR:$Vd),
2157 (ResTy (ShOp (ResTy QPR:$Vn),
2158 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2160 let isCommutable = 0;
2163 // Basic 3-register intrinsics, both double- and quad-register.
2164 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2165 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2166 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2167 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2168 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2169 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2170 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2171 let isCommutable = Commutable;
2173 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2174 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2175 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2176 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2177 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2179 (Ty (IntOp (Ty DPR:$Vn),
2180 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2182 let isCommutable = 0;
2184 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2185 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2186 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2187 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2188 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2190 (Ty (IntOp (Ty DPR:$Vn),
2191 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2192 let isCommutable = 0;
2194 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2195 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2196 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2197 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2198 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2199 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2200 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2201 let isCommutable = 0;
2204 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2205 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2206 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2207 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2208 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2209 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2210 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2211 let isCommutable = Commutable;
2213 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2214 string OpcodeStr, string Dt,
2215 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2216 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2217 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2218 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2219 [(set (ResTy QPR:$Vd),
2220 (ResTy (IntOp (ResTy QPR:$Vn),
2221 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2223 let isCommutable = 0;
2225 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2226 string OpcodeStr, string Dt,
2227 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2228 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2229 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2230 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2231 [(set (ResTy QPR:$Vd),
2232 (ResTy (IntOp (ResTy QPR:$Vn),
2233 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2235 let isCommutable = 0;
2237 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2238 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2239 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2240 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2241 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2242 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2243 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2244 let isCommutable = 0;
2247 // Multiply-Add/Sub operations: double- and quad-register.
2248 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2249 InstrItinClass itin, string OpcodeStr, string Dt,
2250 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2251 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2252 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2253 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2254 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2255 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2257 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2258 string OpcodeStr, string Dt,
2259 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2260 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2262 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2264 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2266 (Ty (ShOp (Ty DPR:$src1),
2268 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2270 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2271 string OpcodeStr, string Dt,
2272 ValueType Ty, SDNode MulOp, SDNode ShOp>
2273 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2275 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2277 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2279 (Ty (ShOp (Ty DPR:$src1),
2281 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2284 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2285 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2286 SDPatternOperator MulOp, SDPatternOperator OpNode>
2287 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2288 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2289 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2290 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2291 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2292 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2293 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2294 SDPatternOperator MulOp, SDPatternOperator ShOp>
2295 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2297 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2299 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2300 [(set (ResTy QPR:$Vd),
2301 (ResTy (ShOp (ResTy QPR:$src1),
2302 (ResTy (MulOp QPR:$Vn,
2303 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2305 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2306 string OpcodeStr, string Dt,
2307 ValueType ResTy, ValueType OpTy,
2308 SDNode MulOp, SDNode ShOp>
2309 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2311 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2313 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2314 [(set (ResTy QPR:$Vd),
2315 (ResTy (ShOp (ResTy QPR:$src1),
2316 (ResTy (MulOp QPR:$Vn,
2317 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2320 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2321 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2322 InstrItinClass itin, string OpcodeStr, string Dt,
2323 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2324 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2325 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2326 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2327 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2328 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2329 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2330 InstrItinClass itin, string OpcodeStr, string Dt,
2331 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2332 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2333 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2334 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2335 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2336 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2338 // Neon 3-argument intrinsics, both double- and quad-register.
2339 // The destination register is also used as the first source operand register.
2340 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2341 InstrItinClass itin, string OpcodeStr, string Dt,
2342 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2343 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2344 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2345 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2346 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2347 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2348 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2349 InstrItinClass itin, string OpcodeStr, string Dt,
2350 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2351 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2352 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2353 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2354 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2355 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2357 // Long Multiply-Add/Sub operations.
2358 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2359 InstrItinClass itin, string OpcodeStr, string Dt,
2360 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2361 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2362 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2363 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2364 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2365 (TyQ (MulOp (TyD DPR:$Vn),
2366 (TyD DPR:$Vm)))))]>;
2367 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2368 InstrItinClass itin, string OpcodeStr, string Dt,
2369 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2370 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2371 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2373 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2375 (OpNode (TyQ QPR:$src1),
2376 (TyQ (MulOp (TyD DPR:$Vn),
2377 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2379 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2380 InstrItinClass itin, string OpcodeStr, string Dt,
2381 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2382 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2383 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2385 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2387 (OpNode (TyQ QPR:$src1),
2388 (TyQ (MulOp (TyD DPR:$Vn),
2389 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2392 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2393 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2394 InstrItinClass itin, string OpcodeStr, string Dt,
2395 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2397 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2398 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2399 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2400 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2401 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2402 (TyD DPR:$Vm)))))))]>;
2404 // Neon Long 3-argument intrinsic. The destination register is
2405 // a quad-register and is also used as the first source operand register.
2406 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2407 InstrItinClass itin, string OpcodeStr, string Dt,
2408 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2409 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2410 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2411 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2413 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2414 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2415 string OpcodeStr, string Dt,
2416 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2417 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2419 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2421 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2422 [(set (ResTy QPR:$Vd),
2423 (ResTy (IntOp (ResTy QPR:$src1),
2425 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2427 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2428 InstrItinClass itin, string OpcodeStr, string Dt,
2429 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2430 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2432 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2434 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2435 [(set (ResTy QPR:$Vd),
2436 (ResTy (IntOp (ResTy QPR:$src1),
2438 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2441 // Narrowing 3-register intrinsics.
2442 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2443 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2444 Intrinsic IntOp, bit Commutable>
2445 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2446 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2447 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2448 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2449 let isCommutable = Commutable;
2452 // Long 3-register operations.
2453 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2454 InstrItinClass itin, string OpcodeStr, string Dt,
2455 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2456 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2457 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2458 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2459 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2460 let isCommutable = Commutable;
2462 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2463 InstrItinClass itin, string OpcodeStr, string Dt,
2464 ValueType TyQ, ValueType TyD, SDNode OpNode>
2465 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2466 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2467 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2469 (TyQ (OpNode (TyD DPR:$Vn),
2470 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2471 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2472 InstrItinClass itin, string OpcodeStr, string Dt,
2473 ValueType TyQ, ValueType TyD, SDNode OpNode>
2474 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2475 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2476 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2478 (TyQ (OpNode (TyD DPR:$Vn),
2479 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2481 // Long 3-register operations with explicitly extended operands.
2482 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2483 InstrItinClass itin, string OpcodeStr, string Dt,
2484 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2486 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2487 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2488 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2489 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2490 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2491 let isCommutable = Commutable;
2494 // Long 3-register intrinsics with explicit extend (VABDL).
2495 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2496 InstrItinClass itin, string OpcodeStr, string Dt,
2497 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2499 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2500 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2501 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2502 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2503 (TyD DPR:$Vm))))))]> {
2504 let isCommutable = Commutable;
2507 // Long 3-register intrinsics.
2508 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2509 InstrItinClass itin, string OpcodeStr, string Dt,
2510 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2511 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2512 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2513 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2514 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2515 let isCommutable = Commutable;
2517 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2518 string OpcodeStr, string Dt,
2519 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2520 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2521 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2522 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2523 [(set (ResTy QPR:$Vd),
2524 (ResTy (IntOp (OpTy DPR:$Vn),
2525 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2527 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2528 InstrItinClass itin, string OpcodeStr, string Dt,
2529 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2530 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2531 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2532 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2533 [(set (ResTy QPR:$Vd),
2534 (ResTy (IntOp (OpTy DPR:$Vn),
2535 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2538 // Wide 3-register operations.
2539 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2540 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2541 SDNode OpNode, SDNode ExtOp, bit Commutable>
2542 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2543 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2544 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2545 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2546 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2547 let isCommutable = Commutable;
2550 // Pairwise long 2-register intrinsics, both double- and quad-register.
2551 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2552 bits<2> op17_16, bits<5> op11_7, bit op4,
2553 string OpcodeStr, string Dt,
2554 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2555 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2556 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2557 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2558 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2559 bits<2> op17_16, bits<5> op11_7, bit op4,
2560 string OpcodeStr, string Dt,
2561 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2562 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2563 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2564 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2566 // Pairwise long 2-register accumulate intrinsics,
2567 // both double- and quad-register.
2568 // The destination register is also used as the first source operand register.
2569 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2570 bits<2> op17_16, bits<5> op11_7, bit op4,
2571 string OpcodeStr, string Dt,
2572 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2573 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2574 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2575 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2576 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2577 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2578 bits<2> op17_16, bits<5> op11_7, bit op4,
2579 string OpcodeStr, string Dt,
2580 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2581 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2582 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2583 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2584 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2586 // Shift by immediate,
2587 // both double- and quad-register.
2588 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2589 Format f, InstrItinClass itin, Operand ImmTy,
2590 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2591 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2592 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2593 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2594 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2595 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2596 Format f, InstrItinClass itin, Operand ImmTy,
2597 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2598 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2599 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2600 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2601 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2603 // Long shift by immediate.
2604 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2605 string OpcodeStr, string Dt,
2606 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2607 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2608 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2609 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2610 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2611 (i32 imm:$SIMM))))]>;
2613 // Narrow shift by immediate.
2614 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2615 InstrItinClass itin, string OpcodeStr, string Dt,
2616 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2617 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2618 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2619 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2620 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2621 (i32 imm:$SIMM))))]>;
2623 // Shift right by immediate and accumulate,
2624 // both double- and quad-register.
2625 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2626 Operand ImmTy, string OpcodeStr, string Dt,
2627 ValueType Ty, SDNode ShOp>
2628 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2629 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2630 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2631 [(set DPR:$Vd, (Ty (add DPR:$src1,
2632 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2633 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2634 Operand ImmTy, string OpcodeStr, string Dt,
2635 ValueType Ty, SDNode ShOp>
2636 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2637 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2638 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2639 [(set QPR:$Vd, (Ty (add QPR:$src1,
2640 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2642 // Shift by immediate and insert,
2643 // both double- and quad-register.
2644 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2645 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2646 ValueType Ty,SDNode ShOp>
2647 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2648 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2649 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2650 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2651 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2652 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2653 ValueType Ty,SDNode ShOp>
2654 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2655 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2656 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2657 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2659 // Convert, with fractional bits immediate,
2660 // both double- and quad-register.
2661 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2662 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2664 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2665 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2666 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2667 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2668 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2669 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2671 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2672 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2673 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2674 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2676 //===----------------------------------------------------------------------===//
2678 //===----------------------------------------------------------------------===//
2680 // Abbreviations used in multiclass suffixes:
2681 // Q = quarter int (8 bit) elements
2682 // H = half int (16 bit) elements
2683 // S = single int (32 bit) elements
2684 // D = double int (64 bit) elements
2686 // Neon 2-register vector operations and intrinsics.
2688 // Neon 2-register comparisons.
2689 // source operand element sizes of 8, 16 and 32 bits:
2690 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2691 bits<5> op11_7, bit op4, string opc, string Dt,
2692 string asm, SDNode OpNode> {
2693 // 64-bit vector types.
2694 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2695 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2696 opc, !strconcat(Dt, "8"), asm, "",
2697 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2698 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2699 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2700 opc, !strconcat(Dt, "16"), asm, "",
2701 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2702 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2703 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2704 opc, !strconcat(Dt, "32"), asm, "",
2705 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2706 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2707 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2708 opc, "f32", asm, "",
2709 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2710 let Inst{10} = 1; // overwrite F = 1
2713 // 128-bit vector types.
2714 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2715 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2716 opc, !strconcat(Dt, "8"), asm, "",
2717 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2718 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2719 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2720 opc, !strconcat(Dt, "16"), asm, "",
2721 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2722 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2723 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2724 opc, !strconcat(Dt, "32"), asm, "",
2725 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2726 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2727 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2728 opc, "f32", asm, "",
2729 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2730 let Inst{10} = 1; // overwrite F = 1
2735 // Neon 2-register vector intrinsics,
2736 // element sizes of 8, 16 and 32 bits:
2737 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2738 bits<5> op11_7, bit op4,
2739 InstrItinClass itinD, InstrItinClass itinQ,
2740 string OpcodeStr, string Dt, Intrinsic IntOp> {
2741 // 64-bit vector types.
2742 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2743 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2744 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2745 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2746 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2747 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2749 // 128-bit vector types.
2750 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2751 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2752 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2753 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2754 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2755 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2759 // Neon Narrowing 2-register vector operations,
2760 // source operand element sizes of 16, 32 and 64 bits:
2761 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2762 bits<5> op11_7, bit op6, bit op4,
2763 InstrItinClass itin, string OpcodeStr, string Dt,
2765 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2766 itin, OpcodeStr, !strconcat(Dt, "16"),
2767 v8i8, v8i16, OpNode>;
2768 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2769 itin, OpcodeStr, !strconcat(Dt, "32"),
2770 v4i16, v4i32, OpNode>;
2771 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2772 itin, OpcodeStr, !strconcat(Dt, "64"),
2773 v2i32, v2i64, OpNode>;
2776 // Neon Narrowing 2-register vector intrinsics,
2777 // source operand element sizes of 16, 32 and 64 bits:
2778 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2779 bits<5> op11_7, bit op6, bit op4,
2780 InstrItinClass itin, string OpcodeStr, string Dt,
2782 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2783 itin, OpcodeStr, !strconcat(Dt, "16"),
2784 v8i8, v8i16, IntOp>;
2785 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2786 itin, OpcodeStr, !strconcat(Dt, "32"),
2787 v4i16, v4i32, IntOp>;
2788 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2789 itin, OpcodeStr, !strconcat(Dt, "64"),
2790 v2i32, v2i64, IntOp>;
2794 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2795 // source operand element sizes of 16, 32 and 64 bits:
2796 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2797 string OpcodeStr, string Dt, SDNode OpNode> {
2798 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2799 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2800 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2801 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2802 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2803 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2807 // Neon 3-register vector operations.
2809 // First with only element sizes of 8, 16 and 32 bits:
2810 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2811 InstrItinClass itinD16, InstrItinClass itinD32,
2812 InstrItinClass itinQ16, InstrItinClass itinQ32,
2813 string OpcodeStr, string Dt,
2814 SDNode OpNode, bit Commutable = 0> {
2815 // 64-bit vector types.
2816 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2817 OpcodeStr, !strconcat(Dt, "8"),
2818 v8i8, v8i8, OpNode, Commutable>;
2819 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2820 OpcodeStr, !strconcat(Dt, "16"),
2821 v4i16, v4i16, OpNode, Commutable>;
2822 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2823 OpcodeStr, !strconcat(Dt, "32"),
2824 v2i32, v2i32, OpNode, Commutable>;
2826 // 128-bit vector types.
2827 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2828 OpcodeStr, !strconcat(Dt, "8"),
2829 v16i8, v16i8, OpNode, Commutable>;
2830 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2831 OpcodeStr, !strconcat(Dt, "16"),
2832 v8i16, v8i16, OpNode, Commutable>;
2833 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2834 OpcodeStr, !strconcat(Dt, "32"),
2835 v4i32, v4i32, OpNode, Commutable>;
2838 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2839 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2841 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2843 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2844 v8i16, v4i16, ShOp>;
2845 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2846 v4i32, v2i32, ShOp>;
2849 // ....then also with element size 64 bits:
2850 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2851 InstrItinClass itinD, InstrItinClass itinQ,
2852 string OpcodeStr, string Dt,
2853 SDNode OpNode, bit Commutable = 0>
2854 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2855 OpcodeStr, Dt, OpNode, Commutable> {
2856 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2857 OpcodeStr, !strconcat(Dt, "64"),
2858 v1i64, v1i64, OpNode, Commutable>;
2859 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2860 OpcodeStr, !strconcat(Dt, "64"),
2861 v2i64, v2i64, OpNode, Commutable>;
2865 // Neon 3-register vector intrinsics.
2867 // First with only element sizes of 16 and 32 bits:
2868 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2869 InstrItinClass itinD16, InstrItinClass itinD32,
2870 InstrItinClass itinQ16, InstrItinClass itinQ32,
2871 string OpcodeStr, string Dt,
2872 Intrinsic IntOp, bit Commutable = 0> {
2873 // 64-bit vector types.
2874 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2875 OpcodeStr, !strconcat(Dt, "16"),
2876 v4i16, v4i16, IntOp, Commutable>;
2877 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2878 OpcodeStr, !strconcat(Dt, "32"),
2879 v2i32, v2i32, IntOp, Commutable>;
2881 // 128-bit vector types.
2882 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2883 OpcodeStr, !strconcat(Dt, "16"),
2884 v8i16, v8i16, IntOp, Commutable>;
2885 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2886 OpcodeStr, !strconcat(Dt, "32"),
2887 v4i32, v4i32, IntOp, Commutable>;
2889 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2890 InstrItinClass itinD16, InstrItinClass itinD32,
2891 InstrItinClass itinQ16, InstrItinClass itinQ32,
2892 string OpcodeStr, string Dt,
2894 // 64-bit vector types.
2895 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2896 OpcodeStr, !strconcat(Dt, "16"),
2897 v4i16, v4i16, IntOp>;
2898 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2899 OpcodeStr, !strconcat(Dt, "32"),
2900 v2i32, v2i32, IntOp>;
2902 // 128-bit vector types.
2903 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2904 OpcodeStr, !strconcat(Dt, "16"),
2905 v8i16, v8i16, IntOp>;
2906 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2907 OpcodeStr, !strconcat(Dt, "32"),
2908 v4i32, v4i32, IntOp>;
2911 multiclass N3VIntSL_HS<bits<4> op11_8,
2912 InstrItinClass itinD16, InstrItinClass itinD32,
2913 InstrItinClass itinQ16, InstrItinClass itinQ32,
2914 string OpcodeStr, string Dt, Intrinsic IntOp> {
2915 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2916 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2917 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2918 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2919 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2920 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2921 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2922 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2925 // ....then also with element size of 8 bits:
2926 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2927 InstrItinClass itinD16, InstrItinClass itinD32,
2928 InstrItinClass itinQ16, InstrItinClass itinQ32,
2929 string OpcodeStr, string Dt,
2930 Intrinsic IntOp, bit Commutable = 0>
2931 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2932 OpcodeStr, Dt, IntOp, Commutable> {
2933 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2934 OpcodeStr, !strconcat(Dt, "8"),
2935 v8i8, v8i8, IntOp, Commutable>;
2936 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2937 OpcodeStr, !strconcat(Dt, "8"),
2938 v16i8, v16i8, IntOp, Commutable>;
2940 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2941 InstrItinClass itinD16, InstrItinClass itinD32,
2942 InstrItinClass itinQ16, InstrItinClass itinQ32,
2943 string OpcodeStr, string Dt,
2945 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2946 OpcodeStr, Dt, IntOp> {
2947 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2948 OpcodeStr, !strconcat(Dt, "8"),
2950 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2951 OpcodeStr, !strconcat(Dt, "8"),
2952 v16i8, v16i8, IntOp>;
2956 // ....then also with element size of 64 bits:
2957 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2958 InstrItinClass itinD16, InstrItinClass itinD32,
2959 InstrItinClass itinQ16, InstrItinClass itinQ32,
2960 string OpcodeStr, string Dt,
2961 Intrinsic IntOp, bit Commutable = 0>
2962 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2963 OpcodeStr, Dt, IntOp, Commutable> {
2964 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2965 OpcodeStr, !strconcat(Dt, "64"),
2966 v1i64, v1i64, IntOp, Commutable>;
2967 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2968 OpcodeStr, !strconcat(Dt, "64"),
2969 v2i64, v2i64, IntOp, Commutable>;
2971 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2972 InstrItinClass itinD16, InstrItinClass itinD32,
2973 InstrItinClass itinQ16, InstrItinClass itinQ32,
2974 string OpcodeStr, string Dt,
2976 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2977 OpcodeStr, Dt, IntOp> {
2978 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2979 OpcodeStr, !strconcat(Dt, "64"),
2980 v1i64, v1i64, IntOp>;
2981 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2982 OpcodeStr, !strconcat(Dt, "64"),
2983 v2i64, v2i64, IntOp>;
2986 // Neon Narrowing 3-register vector intrinsics,
2987 // source operand element sizes of 16, 32 and 64 bits:
2988 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2989 string OpcodeStr, string Dt,
2990 Intrinsic IntOp, bit Commutable = 0> {
2991 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2992 OpcodeStr, !strconcat(Dt, "16"),
2993 v8i8, v8i16, IntOp, Commutable>;
2994 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2995 OpcodeStr, !strconcat(Dt, "32"),
2996 v4i16, v4i32, IntOp, Commutable>;
2997 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2998 OpcodeStr, !strconcat(Dt, "64"),
2999 v2i32, v2i64, IntOp, Commutable>;
3003 // Neon Long 3-register vector operations.
3005 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3006 InstrItinClass itin16, InstrItinClass itin32,
3007 string OpcodeStr, string Dt,
3008 SDNode OpNode, bit Commutable = 0> {
3009 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
3010 OpcodeStr, !strconcat(Dt, "8"),
3011 v8i16, v8i8, OpNode, Commutable>;
3012 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
3013 OpcodeStr, !strconcat(Dt, "16"),
3014 v4i32, v4i16, OpNode, Commutable>;
3015 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
3016 OpcodeStr, !strconcat(Dt, "32"),
3017 v2i64, v2i32, OpNode, Commutable>;
3020 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
3021 InstrItinClass itin, string OpcodeStr, string Dt,
3023 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
3024 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
3025 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
3026 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
3029 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3030 InstrItinClass itin16, InstrItinClass itin32,
3031 string OpcodeStr, string Dt,
3032 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3033 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
3034 OpcodeStr, !strconcat(Dt, "8"),
3035 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3036 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
3037 OpcodeStr, !strconcat(Dt, "16"),
3038 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3039 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
3040 OpcodeStr, !strconcat(Dt, "32"),
3041 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3044 // Neon Long 3-register vector intrinsics.
3046 // First with only element sizes of 16 and 32 bits:
3047 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3048 InstrItinClass itin16, InstrItinClass itin32,
3049 string OpcodeStr, string Dt,
3050 Intrinsic IntOp, bit Commutable = 0> {
3051 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
3052 OpcodeStr, !strconcat(Dt, "16"),
3053 v4i32, v4i16, IntOp, Commutable>;
3054 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
3055 OpcodeStr, !strconcat(Dt, "32"),
3056 v2i64, v2i32, IntOp, Commutable>;
3059 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
3060 InstrItinClass itin, string OpcodeStr, string Dt,
3062 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
3063 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3064 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
3065 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3068 // ....then also with element size of 8 bits:
3069 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3070 InstrItinClass itin16, InstrItinClass itin32,
3071 string OpcodeStr, string Dt,
3072 Intrinsic IntOp, bit Commutable = 0>
3073 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3074 IntOp, Commutable> {
3075 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3076 OpcodeStr, !strconcat(Dt, "8"),
3077 v8i16, v8i8, IntOp, Commutable>;
3080 // ....with explicit extend (VABDL).
3081 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3082 InstrItinClass itin, string OpcodeStr, string Dt,
3083 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3084 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3085 OpcodeStr, !strconcat(Dt, "8"),
3086 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3087 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3088 OpcodeStr, !strconcat(Dt, "16"),
3089 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3090 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3091 OpcodeStr, !strconcat(Dt, "32"),
3092 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3096 // Neon Wide 3-register vector intrinsics,
3097 // source operand element sizes of 8, 16 and 32 bits:
3098 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3099 string OpcodeStr, string Dt,
3100 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3101 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3102 OpcodeStr, !strconcat(Dt, "8"),
3103 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3104 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3105 OpcodeStr, !strconcat(Dt, "16"),
3106 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3107 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3108 OpcodeStr, !strconcat(Dt, "32"),
3109 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3113 // Neon Multiply-Op vector operations,
3114 // element sizes of 8, 16 and 32 bits:
3115 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3116 InstrItinClass itinD16, InstrItinClass itinD32,
3117 InstrItinClass itinQ16, InstrItinClass itinQ32,
3118 string OpcodeStr, string Dt, SDNode OpNode> {
3119 // 64-bit vector types.
3120 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3121 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3122 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3123 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3124 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3125 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3127 // 128-bit vector types.
3128 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3129 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3130 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3131 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3132 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3133 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3136 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3137 InstrItinClass itinD16, InstrItinClass itinD32,
3138 InstrItinClass itinQ16, InstrItinClass itinQ32,
3139 string OpcodeStr, string Dt, SDNode ShOp> {
3140 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3141 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3142 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3143 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3144 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3145 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3147 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3148 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3152 // Neon Intrinsic-Op vector operations,
3153 // element sizes of 8, 16 and 32 bits:
3154 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3155 InstrItinClass itinD, InstrItinClass itinQ,
3156 string OpcodeStr, string Dt, Intrinsic IntOp,
3158 // 64-bit vector types.
3159 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3160 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3161 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3162 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3163 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3164 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3166 // 128-bit vector types.
3167 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3168 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3169 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3170 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3171 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3172 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3175 // Neon 3-argument intrinsics,
3176 // element sizes of 8, 16 and 32 bits:
3177 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3178 InstrItinClass itinD, InstrItinClass itinQ,
3179 string OpcodeStr, string Dt, Intrinsic IntOp> {
3180 // 64-bit vector types.
3181 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3182 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3183 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3184 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3185 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3186 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3188 // 128-bit vector types.
3189 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3190 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3191 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3192 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3193 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3194 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3198 // Neon Long Multiply-Op vector operations,
3199 // element sizes of 8, 16 and 32 bits:
3200 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3201 InstrItinClass itin16, InstrItinClass itin32,
3202 string OpcodeStr, string Dt, SDNode MulOp,
3204 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3205 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3206 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3207 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3208 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3209 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3212 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3213 string Dt, SDNode MulOp, SDNode OpNode> {
3214 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3215 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3216 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3217 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3221 // Neon Long 3-argument intrinsics.
3223 // First with only element sizes of 16 and 32 bits:
3224 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3225 InstrItinClass itin16, InstrItinClass itin32,
3226 string OpcodeStr, string Dt, Intrinsic IntOp> {
3227 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3228 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3229 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3230 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3233 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3234 string OpcodeStr, string Dt, Intrinsic IntOp> {
3235 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3236 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3237 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3238 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3241 // ....then also with element size of 8 bits:
3242 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3243 InstrItinClass itin16, InstrItinClass itin32,
3244 string OpcodeStr, string Dt, Intrinsic IntOp>
3245 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3246 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3247 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3250 // ....with explicit extend (VABAL).
3251 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3252 InstrItinClass itin, string OpcodeStr, string Dt,
3253 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3254 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3255 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3256 IntOp, ExtOp, OpNode>;
3257 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3258 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3259 IntOp, ExtOp, OpNode>;
3260 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3261 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3262 IntOp, ExtOp, OpNode>;
3266 // Neon Pairwise long 2-register intrinsics,
3267 // element sizes of 8, 16 and 32 bits:
3268 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3269 bits<5> op11_7, bit op4,
3270 string OpcodeStr, string Dt, Intrinsic IntOp> {
3271 // 64-bit vector types.
3272 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3273 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3274 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3275 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3276 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3277 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3279 // 128-bit vector types.
3280 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3281 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3282 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3283 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3284 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3285 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3289 // Neon Pairwise long 2-register accumulate intrinsics,
3290 // element sizes of 8, 16 and 32 bits:
3291 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3292 bits<5> op11_7, bit op4,
3293 string OpcodeStr, string Dt, Intrinsic IntOp> {
3294 // 64-bit vector types.
3295 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3296 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3297 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3298 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3299 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3300 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3302 // 128-bit vector types.
3303 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3304 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3305 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3306 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3307 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3308 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3312 // Neon 2-register vector shift by immediate,
3313 // with f of either N2RegVShLFrm or N2RegVShRFrm
3314 // element sizes of 8, 16, 32 and 64 bits:
3315 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3316 InstrItinClass itin, string OpcodeStr, string Dt,
3318 // 64-bit vector types.
3319 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3320 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3321 let Inst{21-19} = 0b001; // imm6 = 001xxx
3323 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3324 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3325 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3327 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3328 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3329 let Inst{21} = 0b1; // imm6 = 1xxxxx
3331 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3332 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3335 // 128-bit vector types.
3336 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3337 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3338 let Inst{21-19} = 0b001; // imm6 = 001xxx
3340 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3341 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3342 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3344 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3345 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3346 let Inst{21} = 0b1; // imm6 = 1xxxxx
3348 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3349 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3352 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3353 InstrItinClass itin, string OpcodeStr, string Dt,
3355 // 64-bit vector types.
3356 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3357 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3358 let Inst{21-19} = 0b001; // imm6 = 001xxx
3360 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3361 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3362 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3364 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3365 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3366 let Inst{21} = 0b1; // imm6 = 1xxxxx
3368 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3369 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3372 // 128-bit vector types.
3373 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3374 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3375 let Inst{21-19} = 0b001; // imm6 = 001xxx
3377 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3378 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3379 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3381 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3382 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3383 let Inst{21} = 0b1; // imm6 = 1xxxxx
3385 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3386 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3390 // Neon Shift-Accumulate vector operations,
3391 // element sizes of 8, 16, 32 and 64 bits:
3392 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3393 string OpcodeStr, string Dt, SDNode ShOp> {
3394 // 64-bit vector types.
3395 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3396 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3397 let Inst{21-19} = 0b001; // imm6 = 001xxx
3399 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3400 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3401 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3403 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3404 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3405 let Inst{21} = 0b1; // imm6 = 1xxxxx
3407 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3408 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3411 // 128-bit vector types.
3412 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3413 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3414 let Inst{21-19} = 0b001; // imm6 = 001xxx
3416 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3417 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3418 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3420 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3421 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3422 let Inst{21} = 0b1; // imm6 = 1xxxxx
3424 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3425 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3429 // Neon Shift-Insert vector operations,
3430 // with f of either N2RegVShLFrm or N2RegVShRFrm
3431 // element sizes of 8, 16, 32 and 64 bits:
3432 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3434 // 64-bit vector types.
3435 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3436 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3437 let Inst{21-19} = 0b001; // imm6 = 001xxx
3439 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3440 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3441 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3443 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3444 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3445 let Inst{21} = 0b1; // imm6 = 1xxxxx
3447 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3448 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3451 // 128-bit vector types.
3452 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3453 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3454 let Inst{21-19} = 0b001; // imm6 = 001xxx
3456 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3457 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3458 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3460 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3461 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3462 let Inst{21} = 0b1; // imm6 = 1xxxxx
3464 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3465 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3468 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3470 // 64-bit vector types.
3471 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3472 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3473 let Inst{21-19} = 0b001; // imm6 = 001xxx
3475 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3476 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3477 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3479 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3480 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3481 let Inst{21} = 0b1; // imm6 = 1xxxxx
3483 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3484 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3487 // 128-bit vector types.
3488 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3489 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3490 let Inst{21-19} = 0b001; // imm6 = 001xxx
3492 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3493 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3494 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3496 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3497 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3498 let Inst{21} = 0b1; // imm6 = 1xxxxx
3500 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3501 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3505 // Neon Shift Long operations,
3506 // element sizes of 8, 16, 32 bits:
3507 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3508 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3509 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3510 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3511 let Inst{21-19} = 0b001; // imm6 = 001xxx
3513 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3514 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3515 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3517 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3518 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3519 let Inst{21} = 0b1; // imm6 = 1xxxxx
3523 // Neon Shift Narrow operations,
3524 // element sizes of 16, 32, 64 bits:
3525 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3526 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3528 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3529 OpcodeStr, !strconcat(Dt, "16"),
3530 v8i8, v8i16, shr_imm8, OpNode> {
3531 let Inst{21-19} = 0b001; // imm6 = 001xxx
3533 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3534 OpcodeStr, !strconcat(Dt, "32"),
3535 v4i16, v4i32, shr_imm16, OpNode> {
3536 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3538 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3539 OpcodeStr, !strconcat(Dt, "64"),
3540 v2i32, v2i64, shr_imm32, OpNode> {
3541 let Inst{21} = 0b1; // imm6 = 1xxxxx
3545 //===----------------------------------------------------------------------===//
3546 // Instruction Definitions.
3547 //===----------------------------------------------------------------------===//
3549 // Vector Add Operations.
3551 // VADD : Vector Add (integer and floating-point)
3552 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3554 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3555 v2f32, v2f32, fadd, 1>;
3556 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3557 v4f32, v4f32, fadd, 1>;
3558 // VADDL : Vector Add Long (Q = D + D)
3559 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3560 "vaddl", "s", add, sext, 1>;
3561 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3562 "vaddl", "u", add, zext, 1>;
3563 // VADDW : Vector Add Wide (Q = Q + D)
3564 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3565 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3566 // VHADD : Vector Halving Add
3567 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3568 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3569 "vhadd", "s", int_arm_neon_vhadds, 1>;
3570 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3571 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3572 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3573 // VRHADD : Vector Rounding Halving Add
3574 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3575 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3576 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3577 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3578 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3579 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3580 // VQADD : Vector Saturating Add
3581 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3582 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3583 "vqadd", "s", int_arm_neon_vqadds, 1>;
3584 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3585 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3586 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3587 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3588 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3589 int_arm_neon_vaddhn, 1>;
3590 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3591 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3592 int_arm_neon_vraddhn, 1>;
3594 // Vector Multiply Operations.
3596 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3597 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3598 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3599 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3600 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3601 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3602 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3603 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3604 v2f32, v2f32, fmul, 1>;
3605 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3606 v4f32, v4f32, fmul, 1>;
3607 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3608 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3609 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3612 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3613 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3614 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3615 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3616 (DSubReg_i16_reg imm:$lane))),
3617 (SubReg_i16_lane imm:$lane)))>;
3618 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3619 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3620 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3621 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3622 (DSubReg_i32_reg imm:$lane))),
3623 (SubReg_i32_lane imm:$lane)))>;
3624 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3625 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3626 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3627 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3628 (DSubReg_i32_reg imm:$lane))),
3629 (SubReg_i32_lane imm:$lane)))>;
3631 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3632 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3633 IIC_VMULi16Q, IIC_VMULi32Q,
3634 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3635 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3636 IIC_VMULi16Q, IIC_VMULi32Q,
3637 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3638 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3639 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3641 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3642 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3643 (DSubReg_i16_reg imm:$lane))),
3644 (SubReg_i16_lane imm:$lane)))>;
3645 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3646 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3648 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3649 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3650 (DSubReg_i32_reg imm:$lane))),
3651 (SubReg_i32_lane imm:$lane)))>;
3653 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3654 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3655 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3656 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3657 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3658 IIC_VMULi16Q, IIC_VMULi32Q,
3659 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3660 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3661 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3663 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3664 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3665 (DSubReg_i16_reg imm:$lane))),
3666 (SubReg_i16_lane imm:$lane)))>;
3667 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3668 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3670 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3671 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3672 (DSubReg_i32_reg imm:$lane))),
3673 (SubReg_i32_lane imm:$lane)))>;
3675 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3676 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3677 "vmull", "s", NEONvmulls, 1>;
3678 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3679 "vmull", "u", NEONvmullu, 1>;
3680 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3681 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3682 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3683 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3685 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3686 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3687 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3688 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3689 "vqdmull", "s", int_arm_neon_vqdmull>;
3691 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3693 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3694 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3695 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3696 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3697 v2f32, fmul_su, fadd_mlx>,
3698 Requires<[HasNEON, UseFPVMLx]>;
3699 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3700 v4f32, fmul_su, fadd_mlx>,
3701 Requires<[HasNEON, UseFPVMLx]>;
3702 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3703 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3704 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3705 v2f32, fmul_su, fadd_mlx>,
3706 Requires<[HasNEON, UseFPVMLx]>;
3707 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3708 v4f32, v2f32, fmul_su, fadd_mlx>,
3709 Requires<[HasNEON, UseFPVMLx]>;
3711 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3712 (mul (v8i16 QPR:$src2),
3713 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3714 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3715 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3716 (DSubReg_i16_reg imm:$lane))),
3717 (SubReg_i16_lane imm:$lane)))>;
3719 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3720 (mul (v4i32 QPR:$src2),
3721 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3722 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3723 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3724 (DSubReg_i32_reg imm:$lane))),
3725 (SubReg_i32_lane imm:$lane)))>;
3727 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3728 (fmul_su (v4f32 QPR:$src2),
3729 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3730 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3732 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3733 (DSubReg_i32_reg imm:$lane))),
3734 (SubReg_i32_lane imm:$lane)))>,
3735 Requires<[HasNEON, UseFPVMLx]>;
3737 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3738 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3739 "vmlal", "s", NEONvmulls, add>;
3740 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3741 "vmlal", "u", NEONvmullu, add>;
3743 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3744 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3746 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3747 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3748 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3749 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3751 // VMLS : Vector Multiply Subtract (integer and floating-point)
3752 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3753 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3754 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3755 v2f32, fmul_su, fsub_mlx>,
3756 Requires<[HasNEON, UseFPVMLx]>;
3757 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3758 v4f32, fmul_su, fsub_mlx>,
3759 Requires<[HasNEON, UseFPVMLx]>;
3760 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3761 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3762 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3763 v2f32, fmul_su, fsub_mlx>,
3764 Requires<[HasNEON, UseFPVMLx]>;
3765 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3766 v4f32, v2f32, fmul_su, fsub_mlx>,
3767 Requires<[HasNEON, UseFPVMLx]>;
3769 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3770 (mul (v8i16 QPR:$src2),
3771 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3772 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3773 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3774 (DSubReg_i16_reg imm:$lane))),
3775 (SubReg_i16_lane imm:$lane)))>;
3777 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3778 (mul (v4i32 QPR:$src2),
3779 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3780 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3781 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3782 (DSubReg_i32_reg imm:$lane))),
3783 (SubReg_i32_lane imm:$lane)))>;
3785 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3786 (fmul_su (v4f32 QPR:$src2),
3787 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3788 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3789 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3790 (DSubReg_i32_reg imm:$lane))),
3791 (SubReg_i32_lane imm:$lane)))>,
3792 Requires<[HasNEON, UseFPVMLx]>;
3794 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3795 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3796 "vmlsl", "s", NEONvmulls, sub>;
3797 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3798 "vmlsl", "u", NEONvmullu, sub>;
3800 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3801 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3803 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3804 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3805 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3806 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3808 // Vector Subtract Operations.
3810 // VSUB : Vector Subtract (integer and floating-point)
3811 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3812 "vsub", "i", sub, 0>;
3813 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3814 v2f32, v2f32, fsub, 0>;
3815 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3816 v4f32, v4f32, fsub, 0>;
3817 // VSUBL : Vector Subtract Long (Q = D - D)
3818 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3819 "vsubl", "s", sub, sext, 0>;
3820 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3821 "vsubl", "u", sub, zext, 0>;
3822 // VSUBW : Vector Subtract Wide (Q = Q - D)
3823 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3824 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3825 // VHSUB : Vector Halving Subtract
3826 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3827 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3828 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3829 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3830 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3831 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3832 // VQSUB : Vector Saturing Subtract
3833 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3834 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3835 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3836 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3837 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3838 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3839 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3840 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3841 int_arm_neon_vsubhn, 0>;
3842 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3843 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3844 int_arm_neon_vrsubhn, 0>;
3846 // Vector Comparisons.
3848 // VCEQ : Vector Compare Equal
3849 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3850 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3851 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3853 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3856 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3857 "$Vd, $Vm, #0", NEONvceqz>;
3859 // VCGE : Vector Compare Greater Than or Equal
3860 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3861 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3862 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3863 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3864 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3866 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3869 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3870 "$Vd, $Vm, #0", NEONvcgez>;
3871 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3872 "$Vd, $Vm, #0", NEONvclez>;
3874 // VCGT : Vector Compare Greater Than
3875 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3876 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3877 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3878 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3879 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3881 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3884 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3885 "$Vd, $Vm, #0", NEONvcgtz>;
3886 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3887 "$Vd, $Vm, #0", NEONvcltz>;
3889 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3890 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3891 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3892 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3893 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3894 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3895 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3896 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3897 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3898 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3899 // VTST : Vector Test Bits
3900 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3901 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3903 // Vector Bitwise Operations.
3905 def vnotd : PatFrag<(ops node:$in),
3906 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3907 def vnotq : PatFrag<(ops node:$in),
3908 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3911 // VAND : Vector Bitwise AND
3912 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3913 v2i32, v2i32, and, 1>;
3914 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3915 v4i32, v4i32, and, 1>;
3917 // VEOR : Vector Bitwise Exclusive OR
3918 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3919 v2i32, v2i32, xor, 1>;
3920 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3921 v4i32, v4i32, xor, 1>;
3923 // VORR : Vector Bitwise OR
3924 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3925 v2i32, v2i32, or, 1>;
3926 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3927 v4i32, v4i32, or, 1>;
3929 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3930 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3932 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3934 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3935 let Inst{9} = SIMM{9};
3938 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3939 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3941 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3943 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3944 let Inst{10-9} = SIMM{10-9};
3947 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3948 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3950 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3952 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3953 let Inst{9} = SIMM{9};
3956 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3957 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3959 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3961 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3962 let Inst{10-9} = SIMM{10-9};
3966 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3967 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3968 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3969 "vbic", "$Vd, $Vn, $Vm", "",
3970 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3971 (vnotd DPR:$Vm))))]>;
3972 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3973 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3974 "vbic", "$Vd, $Vn, $Vm", "",
3975 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3976 (vnotq QPR:$Vm))))]>;
3978 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3979 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3981 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3983 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3984 let Inst{9} = SIMM{9};
3987 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3988 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3990 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3992 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3993 let Inst{10-9} = SIMM{10-9};
3996 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3997 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3999 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
4001 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4002 let Inst{9} = SIMM{9};
4005 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
4006 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
4008 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
4010 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
4011 let Inst{10-9} = SIMM{10-9};
4014 // VORN : Vector Bitwise OR NOT
4015 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
4016 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
4017 "vorn", "$Vd, $Vn, $Vm", "",
4018 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
4019 (vnotd DPR:$Vm))))]>;
4020 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
4021 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
4022 "vorn", "$Vd, $Vn, $Vm", "",
4023 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
4024 (vnotq QPR:$Vm))))]>;
4026 // VMVN : Vector Bitwise NOT (Immediate)
4028 let isReMaterializable = 1 in {
4030 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
4031 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4032 "vmvn", "i16", "$Vd, $SIMM", "",
4033 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
4034 let Inst{9} = SIMM{9};
4037 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
4038 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4039 "vmvn", "i16", "$Vd, $SIMM", "",
4040 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
4041 let Inst{9} = SIMM{9};
4044 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
4045 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4046 "vmvn", "i32", "$Vd, $SIMM", "",
4047 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
4048 let Inst{11-8} = SIMM{11-8};
4051 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
4052 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4053 "vmvn", "i32", "$Vd, $SIMM", "",
4054 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
4055 let Inst{11-8} = SIMM{11-8};
4059 // VMVN : Vector Bitwise NOT
4060 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
4061 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
4062 "vmvn", "$Vd, $Vm", "",
4063 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
4064 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
4065 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
4066 "vmvn", "$Vd, $Vm", "",
4067 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
4068 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
4069 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4071 // VBSL : Vector Bitwise Select
4072 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4073 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4074 N3RegFrm, IIC_VCNTiD,
4075 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4077 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4079 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4080 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4081 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4083 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4084 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4085 N3RegFrm, IIC_VCNTiQ,
4086 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4088 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4090 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4091 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4092 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4094 // VBIF : Vector Bitwise Insert if False
4095 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4096 // FIXME: This instruction's encoding MAY NOT BE correct.
4097 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4098 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4099 N3RegFrm, IIC_VBINiD,
4100 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4102 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4103 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4104 N3RegFrm, IIC_VBINiQ,
4105 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4108 // VBIT : Vector Bitwise Insert if True
4109 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4110 // FIXME: This instruction's encoding MAY NOT BE correct.
4111 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4112 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4113 N3RegFrm, IIC_VBINiD,
4114 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4116 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4117 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4118 N3RegFrm, IIC_VBINiQ,
4119 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4122 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4123 // for equivalent operations with different register constraints; it just
4126 // Vector Absolute Differences.
4128 // VABD : Vector Absolute Difference
4129 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4130 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4131 "vabd", "s", int_arm_neon_vabds, 1>;
4132 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4133 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4134 "vabd", "u", int_arm_neon_vabdu, 1>;
4135 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4136 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4137 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4138 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4140 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4141 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4142 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4143 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4144 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4146 // VABA : Vector Absolute Difference and Accumulate
4147 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4148 "vaba", "s", int_arm_neon_vabds, add>;
4149 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4150 "vaba", "u", int_arm_neon_vabdu, add>;
4152 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4153 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4154 "vabal", "s", int_arm_neon_vabds, zext, add>;
4155 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4156 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4158 // Vector Maximum and Minimum.
4160 // VMAX : Vector Maximum
4161 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4162 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4163 "vmax", "s", int_arm_neon_vmaxs, 1>;
4164 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4165 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4166 "vmax", "u", int_arm_neon_vmaxu, 1>;
4167 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4169 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4170 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4172 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4174 // VMIN : Vector Minimum
4175 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4176 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4177 "vmin", "s", int_arm_neon_vmins, 1>;
4178 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4179 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4180 "vmin", "u", int_arm_neon_vminu, 1>;
4181 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4183 v2f32, v2f32, int_arm_neon_vmins, 1>;
4184 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4186 v4f32, v4f32, int_arm_neon_vmins, 1>;
4188 // Vector Pairwise Operations.
4190 // VPADD : Vector Pairwise Add
4191 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4193 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4194 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4196 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4197 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4199 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4200 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4201 IIC_VPBIND, "vpadd", "f32",
4202 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4204 // VPADDL : Vector Pairwise Add Long
4205 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4206 int_arm_neon_vpaddls>;
4207 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4208 int_arm_neon_vpaddlu>;
4210 // VPADAL : Vector Pairwise Add and Accumulate Long
4211 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4212 int_arm_neon_vpadals>;
4213 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4214 int_arm_neon_vpadalu>;
4216 // VPMAX : Vector Pairwise Maximum
4217 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4218 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4219 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4220 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4221 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4222 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4223 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4224 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4225 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4226 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4227 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4228 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4229 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4230 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4232 // VPMIN : Vector Pairwise Minimum
4233 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4234 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4235 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4236 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4237 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4238 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4239 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4240 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4241 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4242 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4243 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4244 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4245 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4246 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4248 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4250 // VRECPE : Vector Reciprocal Estimate
4251 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4252 IIC_VUNAD, "vrecpe", "u32",
4253 v2i32, v2i32, int_arm_neon_vrecpe>;
4254 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4255 IIC_VUNAQ, "vrecpe", "u32",
4256 v4i32, v4i32, int_arm_neon_vrecpe>;
4257 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4258 IIC_VUNAD, "vrecpe", "f32",
4259 v2f32, v2f32, int_arm_neon_vrecpe>;
4260 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4261 IIC_VUNAQ, "vrecpe", "f32",
4262 v4f32, v4f32, int_arm_neon_vrecpe>;
4264 // VRECPS : Vector Reciprocal Step
4265 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4266 IIC_VRECSD, "vrecps", "f32",
4267 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4268 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4269 IIC_VRECSQ, "vrecps", "f32",
4270 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4272 // VRSQRTE : Vector Reciprocal Square Root Estimate
4273 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4274 IIC_VUNAD, "vrsqrte", "u32",
4275 v2i32, v2i32, int_arm_neon_vrsqrte>;
4276 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4277 IIC_VUNAQ, "vrsqrte", "u32",
4278 v4i32, v4i32, int_arm_neon_vrsqrte>;
4279 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4280 IIC_VUNAD, "vrsqrte", "f32",
4281 v2f32, v2f32, int_arm_neon_vrsqrte>;
4282 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4283 IIC_VUNAQ, "vrsqrte", "f32",
4284 v4f32, v4f32, int_arm_neon_vrsqrte>;
4286 // VRSQRTS : Vector Reciprocal Square Root Step
4287 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4288 IIC_VRECSD, "vrsqrts", "f32",
4289 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4290 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4291 IIC_VRECSQ, "vrsqrts", "f32",
4292 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4296 // VSHL : Vector Shift
4297 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4298 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4299 "vshl", "s", int_arm_neon_vshifts>;
4300 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4301 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4302 "vshl", "u", int_arm_neon_vshiftu>;
4304 // VSHL : Vector Shift Left (Immediate)
4305 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4307 // VSHR : Vector Shift Right (Immediate)
4308 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4309 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4311 // VSHLL : Vector Shift Left Long
4312 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4313 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4315 // VSHLL : Vector Shift Left Long (with maximum shift count)
4316 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4317 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4318 ValueType OpTy, SDNode OpNode>
4319 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4320 ResTy, OpTy, OpNode> {
4321 let Inst{21-16} = op21_16;
4322 let DecoderMethod = "DecodeVSHLMaxInstruction";
4324 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4325 v8i16, v8i8, NEONvshlli>;
4326 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4327 v4i32, v4i16, NEONvshlli>;
4328 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4329 v2i64, v2i32, NEONvshlli>;
4331 // VSHRN : Vector Shift Right and Narrow
4332 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4335 // VRSHL : Vector Rounding Shift
4336 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4337 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4338 "vrshl", "s", int_arm_neon_vrshifts>;
4339 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4340 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4341 "vrshl", "u", int_arm_neon_vrshiftu>;
4342 // VRSHR : Vector Rounding Shift Right
4343 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4344 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4346 // VRSHRN : Vector Rounding Shift Right and Narrow
4347 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4350 // VQSHL : Vector Saturating Shift
4351 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4352 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4353 "vqshl", "s", int_arm_neon_vqshifts>;
4354 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4355 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4356 "vqshl", "u", int_arm_neon_vqshiftu>;
4357 // VQSHL : Vector Saturating Shift Left (Immediate)
4358 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4359 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4361 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4362 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4364 // VQSHRN : Vector Saturating Shift Right and Narrow
4365 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4367 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4370 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4371 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4374 // VQRSHL : Vector Saturating Rounding Shift
4375 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4376 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4377 "vqrshl", "s", int_arm_neon_vqrshifts>;
4378 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4379 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4380 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4382 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4383 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4385 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4388 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4389 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4392 // VSRA : Vector Shift Right and Accumulate
4393 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4394 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4395 // VRSRA : Vector Rounding Shift Right and Accumulate
4396 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4397 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4399 // VSLI : Vector Shift Left and Insert
4400 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4402 // VSRI : Vector Shift Right and Insert
4403 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4405 // Vector Absolute and Saturating Absolute.
4407 // VABS : Vector Absolute Value
4408 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4409 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4411 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4412 IIC_VUNAD, "vabs", "f32",
4413 v2f32, v2f32, int_arm_neon_vabs>;
4414 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4415 IIC_VUNAQ, "vabs", "f32",
4416 v4f32, v4f32, int_arm_neon_vabs>;
4418 // VQABS : Vector Saturating Absolute Value
4419 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4420 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4421 int_arm_neon_vqabs>;
4425 def vnegd : PatFrag<(ops node:$in),
4426 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4427 def vnegq : PatFrag<(ops node:$in),
4428 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4430 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4431 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4432 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4433 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4434 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4435 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4436 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4437 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4439 // VNEG : Vector Negate (integer)
4440 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4441 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4442 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4443 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4444 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4445 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4447 // VNEG : Vector Negate (floating-point)
4448 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4449 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4450 "vneg", "f32", "$Vd, $Vm", "",
4451 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4452 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4453 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4454 "vneg", "f32", "$Vd, $Vm", "",
4455 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4457 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4458 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4459 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4460 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4461 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4462 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4464 // VQNEG : Vector Saturating Negate
4465 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4466 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4467 int_arm_neon_vqneg>;
4469 // Vector Bit Counting Operations.
4471 // VCLS : Vector Count Leading Sign Bits
4472 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4473 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4475 // VCLZ : Vector Count Leading Zeros
4476 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4477 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4479 // VCNT : Vector Count One Bits
4480 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4481 IIC_VCNTiD, "vcnt", "8",
4482 v8i8, v8i8, int_arm_neon_vcnt>;
4483 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4484 IIC_VCNTiQ, "vcnt", "8",
4485 v16i8, v16i8, int_arm_neon_vcnt>;
4488 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4489 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4490 "vswp", "$Vd, $Vm", "", []>;
4491 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4492 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4493 "vswp", "$Vd, $Vm", "", []>;
4495 // Vector Move Operations.
4497 // VMOV : Vector Move (Register)
4498 def : InstAlias<"vmov${p} $Vd, $Vm",
4499 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4500 def : InstAlias<"vmov${p} $Vd, $Vm",
4501 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4502 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4503 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4504 defm : VFPDTAnyNoF64InstAlias<"vmov${p}", "$Vd, $Vm",
4505 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4507 // VMOV : Vector Move (Immediate)
4509 let isReMaterializable = 1 in {
4510 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4511 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4512 "vmov", "i8", "$Vd, $SIMM", "",
4513 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4514 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4515 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4516 "vmov", "i8", "$Vd, $SIMM", "",
4517 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4519 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4520 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4521 "vmov", "i16", "$Vd, $SIMM", "",
4522 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4523 let Inst{9} = SIMM{9};
4526 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4527 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4528 "vmov", "i16", "$Vd, $SIMM", "",
4529 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4530 let Inst{9} = SIMM{9};
4533 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4534 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4535 "vmov", "i32", "$Vd, $SIMM", "",
4536 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4537 let Inst{11-8} = SIMM{11-8};
4540 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4541 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4542 "vmov", "i32", "$Vd, $SIMM", "",
4543 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4544 let Inst{11-8} = SIMM{11-8};
4547 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4548 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4549 "vmov", "i64", "$Vd, $SIMM", "",
4550 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4551 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4552 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4553 "vmov", "i64", "$Vd, $SIMM", "",
4554 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4556 def VMOVv2f32 : N1ModImm<1, 0b000, 0b1111, 0, 0, 0, 1, (outs DPR:$Vd),
4557 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4558 "vmov", "f32", "$Vd, $SIMM", "",
4559 [(set DPR:$Vd, (v2f32 (NEONvmovFPImm timm:$SIMM)))]>;
4560 def VMOVv4f32 : N1ModImm<1, 0b000, 0b1111, 0, 1, 0, 1, (outs QPR:$Vd),
4561 (ins nImmVMOVF32:$SIMM), IIC_VMOVImm,
4562 "vmov", "f32", "$Vd, $SIMM", "",
4563 [(set QPR:$Vd, (v4f32 (NEONvmovFPImm timm:$SIMM)))]>;
4564 } // isReMaterializable
4566 // VMOV : Vector Get Lane (move scalar to ARM core register)
4568 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4569 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4570 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4571 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4573 let Inst{21} = lane{2};
4574 let Inst{6-5} = lane{1-0};
4576 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4577 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4578 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4579 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4581 let Inst{21} = lane{1};
4582 let Inst{6} = lane{0};
4584 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4585 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4586 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4587 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4589 let Inst{21} = lane{2};
4590 let Inst{6-5} = lane{1-0};
4592 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4593 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4594 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4595 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4597 let Inst{21} = lane{1};
4598 let Inst{6} = lane{0};
4600 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4601 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4602 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4603 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4605 let Inst{21} = lane{0};
4607 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4608 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4609 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4610 (DSubReg_i8_reg imm:$lane))),
4611 (SubReg_i8_lane imm:$lane))>;
4612 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4613 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4614 (DSubReg_i16_reg imm:$lane))),
4615 (SubReg_i16_lane imm:$lane))>;
4616 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4617 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4618 (DSubReg_i8_reg imm:$lane))),
4619 (SubReg_i8_lane imm:$lane))>;
4620 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4621 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4622 (DSubReg_i16_reg imm:$lane))),
4623 (SubReg_i16_lane imm:$lane))>;
4624 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4625 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4626 (DSubReg_i32_reg imm:$lane))),
4627 (SubReg_i32_lane imm:$lane))>;
4628 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4629 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4630 (SSubReg_f32_reg imm:$src2))>;
4631 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4632 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4633 (SSubReg_f32_reg imm:$src2))>;
4634 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4635 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4636 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4637 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4640 // VMOV : Vector Set Lane (move ARM core register to scalar)
4642 let Constraints = "$src1 = $V" in {
4643 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4644 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4645 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4646 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4647 GPR:$R, imm:$lane))]> {
4648 let Inst{21} = lane{2};
4649 let Inst{6-5} = lane{1-0};
4651 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4652 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4653 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4654 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4655 GPR:$R, imm:$lane))]> {
4656 let Inst{21} = lane{1};
4657 let Inst{6} = lane{0};
4659 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4660 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4661 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4662 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4663 GPR:$R, imm:$lane))]> {
4664 let Inst{21} = lane{0};
4667 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4668 (v16i8 (INSERT_SUBREG QPR:$src1,
4669 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4670 (DSubReg_i8_reg imm:$lane))),
4671 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4672 (DSubReg_i8_reg imm:$lane)))>;
4673 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4674 (v8i16 (INSERT_SUBREG QPR:$src1,
4675 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4676 (DSubReg_i16_reg imm:$lane))),
4677 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4678 (DSubReg_i16_reg imm:$lane)))>;
4679 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4680 (v4i32 (INSERT_SUBREG QPR:$src1,
4681 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4682 (DSubReg_i32_reg imm:$lane))),
4683 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4684 (DSubReg_i32_reg imm:$lane)))>;
4686 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4687 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4688 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4689 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4690 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4691 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4693 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4694 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4695 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4696 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4698 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4699 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4700 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4701 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4702 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4703 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4705 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4706 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4707 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4708 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4709 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4710 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4712 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4713 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4714 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4716 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4717 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4718 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4720 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4721 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4722 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4725 // VDUP : Vector Duplicate (from ARM core register to all elements)
4727 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4728 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4729 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4730 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4731 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4732 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4733 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4734 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4736 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4737 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4738 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4739 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4740 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4741 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4743 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4744 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4746 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4748 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4749 ValueType Ty, Operand IdxTy>
4750 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4751 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4752 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4754 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4755 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4756 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4757 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4758 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4759 VectorIndex32:$lane)))]>;
4761 // Inst{19-16} is partially specified depending on the element size.
4763 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4765 let Inst{19-17} = lane{2-0};
4767 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4769 let Inst{19-18} = lane{1-0};
4771 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4773 let Inst{19} = lane{0};
4775 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4777 let Inst{19-17} = lane{2-0};
4779 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4781 let Inst{19-18} = lane{1-0};
4783 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4785 let Inst{19} = lane{0};
4788 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4789 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4791 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4792 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4794 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4795 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4796 (DSubReg_i8_reg imm:$lane))),
4797 (SubReg_i8_lane imm:$lane)))>;
4798 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4799 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4800 (DSubReg_i16_reg imm:$lane))),
4801 (SubReg_i16_lane imm:$lane)))>;
4802 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4803 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4804 (DSubReg_i32_reg imm:$lane))),
4805 (SubReg_i32_lane imm:$lane)))>;
4806 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4807 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4808 (DSubReg_i32_reg imm:$lane))),
4809 (SubReg_i32_lane imm:$lane)))>;
4811 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4812 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4813 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4814 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4816 // VMOVN : Vector Narrowing Move
4817 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4818 "vmovn", "i", trunc>;
4819 // VQMOVN : Vector Saturating Narrowing Move
4820 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4821 "vqmovn", "s", int_arm_neon_vqmovns>;
4822 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4823 "vqmovn", "u", int_arm_neon_vqmovnu>;
4824 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4825 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4826 // VMOVL : Vector Lengthening Move
4827 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4828 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4830 // Vector Conversions.
4832 // VCVT : Vector Convert Between Floating-Point and Integers
4833 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4834 v2i32, v2f32, fp_to_sint>;
4835 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4836 v2i32, v2f32, fp_to_uint>;
4837 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4838 v2f32, v2i32, sint_to_fp>;
4839 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4840 v2f32, v2i32, uint_to_fp>;
4842 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4843 v4i32, v4f32, fp_to_sint>;
4844 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4845 v4i32, v4f32, fp_to_uint>;
4846 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4847 v4f32, v4i32, sint_to_fp>;
4848 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4849 v4f32, v4i32, uint_to_fp>;
4851 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4852 let DecoderMethod = "DecodeVCVTD" in {
4853 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4854 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4855 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4856 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4857 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4858 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4859 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4860 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4863 let DecoderMethod = "DecodeVCVTQ" in {
4864 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4865 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4866 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4867 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4868 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4869 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4870 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4871 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4874 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4875 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4876 IIC_VUNAQ, "vcvt", "f16.f32",
4877 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4878 Requires<[HasNEON, HasFP16]>;
4879 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4880 IIC_VUNAQ, "vcvt", "f32.f16",
4881 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4882 Requires<[HasNEON, HasFP16]>;
4886 // VREV64 : Vector Reverse elements within 64-bit doublewords
4888 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4889 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4890 (ins DPR:$Vm), IIC_VMOVD,
4891 OpcodeStr, Dt, "$Vd, $Vm", "",
4892 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4893 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4894 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4895 (ins QPR:$Vm), IIC_VMOVQ,
4896 OpcodeStr, Dt, "$Vd, $Vm", "",
4897 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4899 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4900 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4901 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4902 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4904 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4905 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4906 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4907 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4909 // VREV32 : Vector Reverse elements within 32-bit words
4911 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4912 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4913 (ins DPR:$Vm), IIC_VMOVD,
4914 OpcodeStr, Dt, "$Vd, $Vm", "",
4915 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4916 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4917 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4918 (ins QPR:$Vm), IIC_VMOVQ,
4919 OpcodeStr, Dt, "$Vd, $Vm", "",
4920 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4922 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4923 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4925 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4926 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4928 // VREV16 : Vector Reverse elements within 16-bit halfwords
4930 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4931 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4932 (ins DPR:$Vm), IIC_VMOVD,
4933 OpcodeStr, Dt, "$Vd, $Vm", "",
4934 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4935 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4936 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4937 (ins QPR:$Vm), IIC_VMOVQ,
4938 OpcodeStr, Dt, "$Vd, $Vm", "",
4939 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4941 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4942 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4944 // Other Vector Shuffles.
4946 // Aligned extractions: really just dropping registers
4948 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4949 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4950 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4952 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4954 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4956 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4958 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4960 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4963 // VEXT : Vector Extract
4965 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4966 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4967 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4968 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4969 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4970 (Ty DPR:$Vm), imm:$index)))]> {
4972 let Inst{11-8} = index{3-0};
4975 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4976 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4977 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4978 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4979 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4980 (Ty QPR:$Vm), imm:$index)))]> {
4982 let Inst{11-8} = index{3-0};
4985 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4986 let Inst{11-8} = index{3-0};
4988 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4989 let Inst{11-9} = index{2-0};
4992 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4993 let Inst{11-10} = index{1-0};
4994 let Inst{9-8} = 0b00;
4996 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4999 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
5001 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
5002 let Inst{11-8} = index{3-0};
5004 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
5005 let Inst{11-9} = index{2-0};
5008 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
5009 let Inst{11-10} = index{1-0};
5010 let Inst{9-8} = 0b00;
5012 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
5015 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
5017 // VTRN : Vector Transpose
5019 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
5020 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
5021 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
5023 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
5024 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
5025 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
5027 // VUZP : Vector Unzip (Deinterleave)
5029 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
5030 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
5031 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
5033 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
5034 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
5035 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
5037 // VZIP : Vector Zip (Interleave)
5039 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
5040 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
5041 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
5043 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
5044 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
5045 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
5047 // Vector Table Lookup and Table Extension.
5049 // VTBL : Vector Table Lookup
5050 let DecoderMethod = "DecodeTBLInstruction" in {
5052 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
5053 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
5054 "vtbl", "8", "$Vd, $Vn, $Vm", "",
5055 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
5056 let hasExtraSrcRegAllocReq = 1 in {
5058 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
5059 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
5060 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
5062 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
5063 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
5064 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
5066 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
5067 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
5069 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
5070 } // hasExtraSrcRegAllocReq = 1
5073 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
5075 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
5077 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
5079 // VTBX : Vector Table Extension
5081 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
5082 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
5083 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
5084 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
5085 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
5086 let hasExtraSrcRegAllocReq = 1 in {
5088 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5089 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5090 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5092 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5093 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5094 NVTBLFrm, IIC_VTBX3,
5095 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5098 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5099 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5100 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5102 } // hasExtraSrcRegAllocReq = 1
5105 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5106 IIC_VTBX2, "$orig = $dst", []>;
5108 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5109 IIC_VTBX3, "$orig = $dst", []>;
5111 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5112 IIC_VTBX4, "$orig = $dst", []>;
5113 } // DecoderMethod = "DecodeTBLInstruction"
5115 //===----------------------------------------------------------------------===//
5116 // NEON instructions for single-precision FP math
5117 //===----------------------------------------------------------------------===//
5119 class N2VSPat<SDNode OpNode, NeonI Inst>
5120 : NEONFPPat<(f32 (OpNode SPR:$a)),
5122 (v2f32 (COPY_TO_REGCLASS (Inst
5124 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5125 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5127 class N3VSPat<SDNode OpNode, NeonI Inst>
5128 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5130 (v2f32 (COPY_TO_REGCLASS (Inst
5132 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5135 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5136 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5138 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5139 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5141 (v2f32 (COPY_TO_REGCLASS (Inst
5143 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5146 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5149 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5150 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5152 def : N3VSPat<fadd, VADDfd>;
5153 def : N3VSPat<fsub, VSUBfd>;
5154 def : N3VSPat<fmul, VMULfd>;
5155 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5156 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5157 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5158 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5159 def : N2VSPat<fabs, VABSfd>;
5160 def : N2VSPat<fneg, VNEGfd>;
5161 def : N3VSPat<NEONfmax, VMAXfd>;
5162 def : N3VSPat<NEONfmin, VMINfd>;
5163 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5164 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5165 def : N2VSPat<arm_sitof, VCVTs2fd>;
5166 def : N2VSPat<arm_uitof, VCVTu2fd>;
5168 //===----------------------------------------------------------------------===//
5169 // Non-Instruction Patterns
5170 //===----------------------------------------------------------------------===//
5173 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5174 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5175 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5176 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5177 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5178 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5179 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5180 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5181 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5182 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5183 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5184 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5185 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5186 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5187 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5188 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5189 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5190 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5191 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5192 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5193 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5194 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5195 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5196 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5197 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5198 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5199 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5200 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5201 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5202 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5204 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5205 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5206 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5207 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5208 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5209 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5210 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5211 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5212 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5213 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5214 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5215 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5216 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5217 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5218 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5219 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5220 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5221 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5222 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5223 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5224 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5225 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5226 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5227 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5228 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5229 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5230 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5231 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5232 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5233 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;
5236 //===----------------------------------------------------------------------===//
5237 // Assembler aliases
5240 // VAND/VEOR/VORR accept but do not require a type suffix.
5241 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5242 (VANDd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5243 defm : VFPDTAnyInstAlias<"vand${p}", "$Vd, $Vn, $Vm",
5244 (VANDq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5245 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5246 (VEORd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5247 defm : VFPDTAnyInstAlias<"veor${p}", "$Vd, $Vn, $Vm",
5248 (VEORq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5249 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5250 (VORRd DPR:$Vd, DPR:$Vn, DPR:$Vm, pred:$p)>;
5251 defm : VFPDTAnyInstAlias<"vorr${p}", "$Vd, $Vn, $Vm",
5252 (VORRq QPR:$Vd, QPR:$Vn, QPR:$Vm, pred:$p)>;
5254 // VLD1 requires a size suffix, but also accepts type specific variants.
5255 // Load one D register.
5256 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5257 (VLD1d8 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5258 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5259 (VLD1d16 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5260 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5261 (VLD1d32 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5262 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5263 (VLD1d64 VecListOneD:$Vd, addrmode6:$Rn, pred:$p)>;
5264 // with writeback, fixed stride
5265 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5266 (VLD1d8wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5267 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5268 (VLD1d16wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5269 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5270 (VLD1d32wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5271 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5272 (VLD1d64wb_fixed VecListOneD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5273 // with writeback, register stride
5274 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5275 (VLD1d8wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5276 rGPR:$Rm, pred:$p)>;
5277 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5278 (VLD1d16wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5279 rGPR:$Rm, pred:$p)>;
5280 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5281 (VLD1d32wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5282 rGPR:$Rm, pred:$p)>;
5283 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5284 (VLD1d64wb_register VecListOneD:$Vd, zero_reg, addrmode6:$Rn,
5285 rGPR:$Rm, pred:$p)>;
5287 // Load two D registers.
5288 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5289 (VLD1q8 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5290 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5291 (VLD1q16 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5292 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5293 (VLD1q32 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5294 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5295 (VLD1q64 VecListTwoD:$Vd, addrmode6:$Rn, pred:$p)>;
5296 // with writeback, fixed stride
5297 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5298 (VLD1q8wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5299 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5300 (VLD1q16wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5301 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5302 (VLD1q32wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5303 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5304 (VLD1q64wb_fixed VecListTwoD:$Vd, zero_reg, addrmode6:$Rn, pred:$p)>;
5305 // with writeback, register stride
5306 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5307 (VLD1q8wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5308 rGPR:$Rm, pred:$p)>;
5309 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5310 (VLD1q16wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5311 rGPR:$Rm, pred:$p)>;
5312 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5313 (VLD1q32wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5314 rGPR:$Rm, pred:$p)>;
5315 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5316 (VLD1q64wb_register VecListTwoD:$Vd, zero_reg, addrmode6:$Rn,
5317 rGPR:$Rm, pred:$p)>;
5319 // Load three D registers.
5320 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5321 (VLD1d8T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5322 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5323 (VLD1d16T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5324 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5325 (VLD1d32T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5326 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5327 (VLD1d64T VecListThreeD:$Vd, addrmode6:$Rn, pred:$p)>;
5328 // with writeback, fixed stride
5329 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5330 (VLD1d8Twb_fixed VecListThreeD:$Vd, zero_reg,
5331 addrmode6:$Rn, pred:$p)>;
5332 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5333 (VLD1d16Twb_fixed VecListThreeD:$Vd, zero_reg,
5334 addrmode6:$Rn, pred:$p)>;
5335 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5336 (VLD1d32Twb_fixed VecListThreeD:$Vd, zero_reg,
5337 addrmode6:$Rn, pred:$p)>;
5338 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5339 (VLD1d64Twb_fixed VecListThreeD:$Vd, zero_reg,
5340 addrmode6:$Rn, pred:$p)>;
5341 // with writeback, register stride
5342 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5343 (VLD1d8Twb_register VecListThreeD:$Vd, zero_reg,
5344 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5345 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5346 (VLD1d16Twb_register VecListThreeD:$Vd, zero_reg,
5347 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5348 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5349 (VLD1d32Twb_register VecListThreeD:$Vd, zero_reg,
5350 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5351 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5352 (VLD1d64Twb_register VecListThreeD:$Vd, zero_reg,
5353 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5356 // Load four D registers.
5357 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5358 (VLD1d8Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5359 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5360 (VLD1d16Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5361 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5362 (VLD1d32Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5363 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn",
5364 (VLD1d64Q VecListFourD:$Vd, addrmode6:$Rn, pred:$p)>;
5365 // with writeback, fixed stride
5366 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5367 (VLD1d8Qwb_fixed VecListFourD:$Vd, zero_reg,
5368 addrmode6:$Rn, pred:$p)>;
5369 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5370 (VLD1d16Qwb_fixed VecListFourD:$Vd, zero_reg,
5371 addrmode6:$Rn, pred:$p)>;
5372 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5373 (VLD1d32Qwb_fixed VecListFourD:$Vd, zero_reg,
5374 addrmode6:$Rn, pred:$p)>;
5375 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn!",
5376 (VLD1d64Qwb_fixed VecListFourD:$Vd, zero_reg,
5377 addrmode6:$Rn, pred:$p)>;
5378 // with writeback, register stride
5379 defm : VFPDT8ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5380 (VLD1d8Qwb_register VecListFourD:$Vd, zero_reg,
5381 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5382 defm : VFPDT16ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5383 (VLD1d16Qwb_register VecListFourD:$Vd, zero_reg,
5384 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5385 defm : VFPDT32ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5386 (VLD1d32Qwb_register VecListFourD:$Vd, zero_reg,
5387 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5388 defm : VFPDT64ReqInstAlias<"vld1${p}", "$Vd, $Rn, $Rm",
5389 (VLD1d64Qwb_register VecListFourD:$Vd, zero_reg,
5390 addrmode6:$Rn, rGPR:$Rm, pred:$p)>;
5392 // VST1 requires a size suffix, but also accepts type specific variants.
5393 // Store one D register.
5394 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5395 (VST1d8 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5396 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5397 (VST1d16 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5398 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5399 (VST1d32 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5400 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5401 (VST1d64 addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5402 // with writeback, fixed stride
5403 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5404 (VST1d8wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5405 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5406 (VST1d16wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5407 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5408 (VST1d32wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5409 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5410 (VST1d64wb_fixed zero_reg, addrmode6:$Rn, VecListOneD:$Vd, pred:$p)>;
5411 // with writeback, register stride
5412 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5413 (VST1d8wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5414 VecListOneD:$Vd, pred:$p)>;
5415 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5416 (VST1d16wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5417 VecListOneD:$Vd, pred:$p)>;
5418 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5419 (VST1d32wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5420 VecListOneD:$Vd, pred:$p)>;
5421 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5422 (VST1d64wb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5423 VecListOneD:$Vd, pred:$p)>;
5425 // Store two D registers.
5426 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5427 (VST1q8 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5428 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5429 (VST1q16 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5430 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5431 (VST1q32 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5432 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5433 (VST1q64 addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5434 // with writeback, fixed stride
5435 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5436 (VST1q8wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5437 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5438 (VST1q16wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5439 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5440 (VST1q32wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5441 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5442 (VST1q64wb_fixed zero_reg, addrmode6:$Rn, VecListTwoD:$Vd, pred:$p)>;
5443 // with writeback, register stride
5444 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5445 (VST1q8wb_register zero_reg, addrmode6:$Rn,
5446 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5447 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5448 (VST1q16wb_register zero_reg, addrmode6:$Rn,
5449 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5450 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5451 (VST1q32wb_register zero_reg, addrmode6:$Rn,
5452 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5453 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5454 (VST1q64wb_register zero_reg, addrmode6:$Rn,
5455 rGPR:$Rm, VecListTwoD:$Vd, pred:$p)>;
5457 // Load three D registers.
5458 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5459 (VST1d8T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5460 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5461 (VST1d16T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5462 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5463 (VST1d32T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5464 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5465 (VST1d64T addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5466 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5467 (VST1d8Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5468 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5469 (VST1d16Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5470 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5471 (VST1d32Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5472 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5473 (VST1d64Twb_fixed zero_reg, addrmode6:$Rn, VecListThreeD:$Vd, pred:$p)>;
5474 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5475 (VST1d8Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5476 VecListThreeD:$Vd, pred:$p)>;
5477 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5478 (VST1d16Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5479 VecListThreeD:$Vd, pred:$p)>;
5480 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5481 (VST1d32Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5482 VecListThreeD:$Vd, pred:$p)>;
5483 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5484 (VST1d64Twb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5485 VecListThreeD:$Vd, pred:$p)>;
5487 // Load four D registers.
5488 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5489 (VST1d8Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5490 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5491 (VST1d16Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5492 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5493 (VST1d32Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5494 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn",
5495 (VST1d64Q addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5496 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5497 (VST1d8Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5498 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5499 (VST1d16Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5500 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5501 (VST1d32Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5502 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn!",
5503 (VST1d64Qwb_fixed zero_reg, addrmode6:$Rn, VecListFourD:$Vd, pred:$p)>;
5504 defm : VFPDT8ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5505 (VST1d8Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5506 VecListFourD:$Vd, pred:$p)>;
5507 defm : VFPDT16ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5508 (VST1d16Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5509 VecListFourD:$Vd, pred:$p)>;
5510 defm : VFPDT32ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5511 (VST1d32Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5512 VecListFourD:$Vd, pred:$p)>;
5513 defm : VFPDT64ReqInstAlias<"vst1${p}", "$Vd, $Rn, $Rm",
5514 (VST1d64Qwb_register zero_reg, addrmode6:$Rn, rGPR:$Rm,
5515 VecListFourD:$Vd, pred:$p)>;
5518 // VTRN instructions data type suffix aliases for more-specific types.
5519 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Dd, $Dm",
5520 (VTRNd8 DPR:$Dd, DPR:$Dm, pred:$p)>;
5521 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5522 (VTRNd16 DPR:$Dd, DPR:$Dm, pred:$p)>;
5523 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Dd, $Dm",
5524 (VTRNd32 DPR:$Dd, DPR:$Dm, pred:$p)>;
5526 defm : VFPDT8ReqInstAlias <"vtrn${p}", "$Qd, $Qm",
5527 (VTRNq8 QPR:$Qd, QPR:$Qm, pred:$p)>;
5528 defm : VFPDT16ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5529 (VTRNq16 QPR:$Qd, QPR:$Qm, pred:$p)>;
5530 defm : VFPDT32ReqInstAlias<"vtrn${p}", "$Qd, $Qm",
5531 (VTRNq32 QPR:$Qd, QPR:$Qm, pred:$p)>;