1 //===- ARMInstrNEON.td - NEON support for ARM -----------------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the ARM NEON instruction set.
12 //===----------------------------------------------------------------------===//
15 //===----------------------------------------------------------------------===//
16 // NEON-specific Operands.
17 //===----------------------------------------------------------------------===//
18 def nModImm : Operand<i32> {
19 let PrintMethod = "printNEONModImmOperand";
22 def nImmSplatI8AsmOperand : AsmOperandClass { let Name = "NEONi8splat"; }
23 def nImmSplatI8 : Operand<i32> {
24 let PrintMethod = "printNEONModImmOperand";
25 let ParserMatchClass = nImmSplatI8AsmOperand;
27 def nImmSplatI16AsmOperand : AsmOperandClass { let Name = "NEONi16splat"; }
28 def nImmSplatI16 : Operand<i32> {
29 let PrintMethod = "printNEONModImmOperand";
30 let ParserMatchClass = nImmSplatI16AsmOperand;
32 def nImmSplatI32AsmOperand : AsmOperandClass { let Name = "NEONi32splat"; }
33 def nImmSplatI32 : Operand<i32> {
34 let PrintMethod = "printNEONModImmOperand";
35 let ParserMatchClass = nImmSplatI32AsmOperand;
37 def nImmVMOVI32AsmOperand : AsmOperandClass { let Name = "NEONi32vmov"; }
38 def nImmVMOVI32 : Operand<i32> {
39 let PrintMethod = "printNEONModImmOperand";
40 let ParserMatchClass = nImmVMOVI32AsmOperand;
42 def nImmSplatI64AsmOperand : AsmOperandClass { let Name = "NEONi64splat"; }
43 def nImmSplatI64 : Operand<i32> {
44 let PrintMethod = "printNEONModImmOperand";
45 let ParserMatchClass = nImmSplatI64AsmOperand;
48 def VectorIndex8Operand : AsmOperandClass { let Name = "VectorIndex8"; }
49 def VectorIndex16Operand : AsmOperandClass { let Name = "VectorIndex16"; }
50 def VectorIndex32Operand : AsmOperandClass { let Name = "VectorIndex32"; }
51 def VectorIndex8 : Operand<i32>, ImmLeaf<i32, [{
52 return ((uint64_t)Imm) < 8;
54 let ParserMatchClass = VectorIndex8Operand;
55 let PrintMethod = "printVectorIndex";
56 let MIOperandInfo = (ops i32imm);
58 def VectorIndex16 : Operand<i32>, ImmLeaf<i32, [{
59 return ((uint64_t)Imm) < 4;
61 let ParserMatchClass = VectorIndex16Operand;
62 let PrintMethod = "printVectorIndex";
63 let MIOperandInfo = (ops i32imm);
65 def VectorIndex32 : Operand<i32>, ImmLeaf<i32, [{
66 return ((uint64_t)Imm) < 2;
68 let ParserMatchClass = VectorIndex32Operand;
69 let PrintMethod = "printVectorIndex";
70 let MIOperandInfo = (ops i32imm);
73 def VecListOneDAsmOperand : AsmOperandClass {
74 let Name = "VecListOneD";
75 let ParserMethod = "parseVectorList";
77 def VecListOneD : RegisterOperand<DPR, "printVectorListOne"> {
78 let ParserMatchClass = VecListOneDAsmOperand;
80 // Register list of two sequential D registers.
81 def VecListTwoDAsmOperand : AsmOperandClass {
82 let Name = "VecListTwoD";
83 let ParserMethod = "parseVectorList";
85 def VecListTwoD : RegisterOperand<DPR, "printVectorListTwo"> {
86 let ParserMatchClass = VecListTwoDAsmOperand;
88 // Register list of three sequential D registers.
89 def VecListThreeDAsmOperand : AsmOperandClass {
90 let Name = "VecListThreeD";
91 let ParserMethod = "parseVectorList";
93 def VecListThreeD : RegisterOperand<DPR, "printVectorListThree"> {
94 let ParserMatchClass = VecListThreeDAsmOperand;
96 // Register list of four sequential D registers.
97 def VecListFourDAsmOperand : AsmOperandClass {
98 let Name = "VecListFourD";
99 let ParserMethod = "parseVectorList";
101 def VecListFourD : RegisterOperand<DPR, "printVectorListFour"> {
102 let ParserMatchClass = VecListFourDAsmOperand;
104 // Register list of two D registers spaced by 2 (two sequential Q registers).
105 def VecListTwoQAsmOperand : AsmOperandClass {
106 let Name = "VecListTwoQ";
107 let ParserMethod = "parseVectorList";
109 def VecListTwoQ : RegisterOperand<DPR, "printVectorListTwo"> {
110 let ParserMatchClass = VecListTwoQAsmOperand;
113 //===----------------------------------------------------------------------===//
114 // NEON-specific DAG Nodes.
115 //===----------------------------------------------------------------------===//
117 def SDTARMVCMP : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<1, 2>]>;
118 def SDTARMVCMPZ : SDTypeProfile<1, 1, []>;
120 def NEONvceq : SDNode<"ARMISD::VCEQ", SDTARMVCMP>;
121 def NEONvceqz : SDNode<"ARMISD::VCEQZ", SDTARMVCMPZ>;
122 def NEONvcge : SDNode<"ARMISD::VCGE", SDTARMVCMP>;
123 def NEONvcgez : SDNode<"ARMISD::VCGEZ", SDTARMVCMPZ>;
124 def NEONvclez : SDNode<"ARMISD::VCLEZ", SDTARMVCMPZ>;
125 def NEONvcgeu : SDNode<"ARMISD::VCGEU", SDTARMVCMP>;
126 def NEONvcgt : SDNode<"ARMISD::VCGT", SDTARMVCMP>;
127 def NEONvcgtz : SDNode<"ARMISD::VCGTZ", SDTARMVCMPZ>;
128 def NEONvcltz : SDNode<"ARMISD::VCLTZ", SDTARMVCMPZ>;
129 def NEONvcgtu : SDNode<"ARMISD::VCGTU", SDTARMVCMP>;
130 def NEONvtst : SDNode<"ARMISD::VTST", SDTARMVCMP>;
132 // Types for vector shift by immediates. The "SHX" version is for long and
133 // narrow operations where the source and destination vectors have different
134 // types. The "SHINS" version is for shift and insert operations.
135 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
137 def SDTARMVSHX : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
139 def SDTARMVSHINS : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
140 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
142 def NEONvshl : SDNode<"ARMISD::VSHL", SDTARMVSH>;
143 def NEONvshrs : SDNode<"ARMISD::VSHRs", SDTARMVSH>;
144 def NEONvshru : SDNode<"ARMISD::VSHRu", SDTARMVSH>;
145 def NEONvshlls : SDNode<"ARMISD::VSHLLs", SDTARMVSHX>;
146 def NEONvshllu : SDNode<"ARMISD::VSHLLu", SDTARMVSHX>;
147 def NEONvshlli : SDNode<"ARMISD::VSHLLi", SDTARMVSHX>;
148 def NEONvshrn : SDNode<"ARMISD::VSHRN", SDTARMVSHX>;
150 def NEONvrshrs : SDNode<"ARMISD::VRSHRs", SDTARMVSH>;
151 def NEONvrshru : SDNode<"ARMISD::VRSHRu", SDTARMVSH>;
152 def NEONvrshrn : SDNode<"ARMISD::VRSHRN", SDTARMVSHX>;
154 def NEONvqshls : SDNode<"ARMISD::VQSHLs", SDTARMVSH>;
155 def NEONvqshlu : SDNode<"ARMISD::VQSHLu", SDTARMVSH>;
156 def NEONvqshlsu : SDNode<"ARMISD::VQSHLsu", SDTARMVSH>;
157 def NEONvqshrns : SDNode<"ARMISD::VQSHRNs", SDTARMVSHX>;
158 def NEONvqshrnu : SDNode<"ARMISD::VQSHRNu", SDTARMVSHX>;
159 def NEONvqshrnsu : SDNode<"ARMISD::VQSHRNsu", SDTARMVSHX>;
161 def NEONvqrshrns : SDNode<"ARMISD::VQRSHRNs", SDTARMVSHX>;
162 def NEONvqrshrnu : SDNode<"ARMISD::VQRSHRNu", SDTARMVSHX>;
163 def NEONvqrshrnsu : SDNode<"ARMISD::VQRSHRNsu", SDTARMVSHX>;
165 def NEONvsli : SDNode<"ARMISD::VSLI", SDTARMVSHINS>;
166 def NEONvsri : SDNode<"ARMISD::VSRI", SDTARMVSHINS>;
168 def SDTARMVGETLN : SDTypeProfile<1, 2, [SDTCisVT<0, i32>, SDTCisInt<1>,
170 def NEONvgetlaneu : SDNode<"ARMISD::VGETLANEu", SDTARMVGETLN>;
171 def NEONvgetlanes : SDNode<"ARMISD::VGETLANEs", SDTARMVGETLN>;
173 def SDTARMVMOVIMM : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisVT<1, i32>]>;
174 def NEONvmovImm : SDNode<"ARMISD::VMOVIMM", SDTARMVMOVIMM>;
175 def NEONvmvnImm : SDNode<"ARMISD::VMVNIMM", SDTARMVMOVIMM>;
177 def SDTARMVORRIMM : SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
179 def NEONvorrImm : SDNode<"ARMISD::VORRIMM", SDTARMVORRIMM>;
180 def NEONvbicImm : SDNode<"ARMISD::VBICIMM", SDTARMVORRIMM>;
182 def NEONvbsl : SDNode<"ARMISD::VBSL",
183 SDTypeProfile<1, 3, [SDTCisVec<0>,
186 SDTCisSameAs<0, 3>]>>;
188 def NEONvdup : SDNode<"ARMISD::VDUP", SDTypeProfile<1, 1, [SDTCisVec<0>]>>;
190 // VDUPLANE can produce a quad-register result from a double-register source,
191 // so the result is not constrained to match the source.
192 def NEONvduplane : SDNode<"ARMISD::VDUPLANE",
193 SDTypeProfile<1, 2, [SDTCisVec<0>, SDTCisVec<1>,
196 def SDTARMVEXT : SDTypeProfile<1, 3, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
197 SDTCisSameAs<0, 2>, SDTCisVT<3, i32>]>;
198 def NEONvext : SDNode<"ARMISD::VEXT", SDTARMVEXT>;
200 def SDTARMVSHUF : SDTypeProfile<1, 1, [SDTCisVec<0>, SDTCisSameAs<0, 1>]>;
201 def NEONvrev64 : SDNode<"ARMISD::VREV64", SDTARMVSHUF>;
202 def NEONvrev32 : SDNode<"ARMISD::VREV32", SDTARMVSHUF>;
203 def NEONvrev16 : SDNode<"ARMISD::VREV16", SDTARMVSHUF>;
205 def SDTARMVSHUF2 : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
207 SDTCisSameAs<0, 3>]>;
208 def NEONzip : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
209 def NEONuzp : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
210 def NEONtrn : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
212 def SDTARMVMULL : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
213 SDTCisSameAs<1, 2>]>;
214 def NEONvmulls : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
215 def NEONvmullu : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
217 def SDTARMFMAX : SDTypeProfile<1, 2, [SDTCisVT<0, f32>, SDTCisSameAs<0, 1>,
218 SDTCisSameAs<0, 2>]>;
219 def NEONfmax : SDNode<"ARMISD::FMAX", SDTARMFMAX>;
220 def NEONfmin : SDNode<"ARMISD::FMIN", SDTARMFMAX>;
222 def NEONimmAllZerosV: PatLeaf<(NEONvmovImm (i32 timm)), [{
223 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
224 unsigned EltBits = 0;
225 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
226 return (EltBits == 32 && EltVal == 0);
229 def NEONimmAllOnesV: PatLeaf<(NEONvmovImm (i32 timm)), [{
230 ConstantSDNode *ConstVal = cast<ConstantSDNode>(N->getOperand(0));
231 unsigned EltBits = 0;
232 uint64_t EltVal = ARM_AM::decodeNEONModImm(ConstVal->getZExtValue(), EltBits);
233 return (EltBits == 8 && EltVal == 0xff);
236 //===----------------------------------------------------------------------===//
237 // NEON load / store instructions
238 //===----------------------------------------------------------------------===//
240 // Use VLDM to load a Q register as a D register pair.
241 // This is a pseudo instruction that is expanded to VLDMD after reg alloc.
243 : PseudoVFPLdStM<(outs QPR:$dst), (ins GPR:$Rn),
245 [(set QPR:$dst, (v2f64 (load GPR:$Rn)))]>;
247 // Use VSTM to store a Q register as a D register pair.
248 // This is a pseudo instruction that is expanded to VSTMD after reg alloc.
250 : PseudoVFPLdStM<(outs), (ins QPR:$src, GPR:$Rn),
252 [(store (v2f64 QPR:$src), GPR:$Rn)]>;
254 // Classes for VLD* pseudo-instructions with multi-register operands.
255 // These are expanded to real instructions after register allocation.
256 class VLDQPseudo<InstrItinClass itin>
257 : PseudoNLdSt<(outs QPR:$dst), (ins addrmode6:$addr), itin, "">;
258 class VLDQWBPseudo<InstrItinClass itin>
259 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
260 (ins addrmode6:$addr, am6offset:$offset), itin,
262 class VLDQWBfixedPseudo<InstrItinClass itin>
263 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
264 (ins addrmode6:$addr), itin,
266 class VLDQWBregisterPseudo<InstrItinClass itin>
267 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
268 (ins addrmode6:$addr, rGPR:$offset), itin,
270 class VLDQQPseudo<InstrItinClass itin>
271 : PseudoNLdSt<(outs QQPR:$dst), (ins addrmode6:$addr), itin, "">;
272 class VLDQQWBPseudo<InstrItinClass itin>
273 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
274 (ins addrmode6:$addr, am6offset:$offset), itin,
276 class VLDQQQQPseudo<InstrItinClass itin>
277 : PseudoNLdSt<(outs QQQQPR:$dst), (ins addrmode6:$addr, QQQQPR:$src),itin,
279 class VLDQQQQWBPseudo<InstrItinClass itin>
280 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
281 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
282 "$addr.addr = $wb, $src = $dst">;
284 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
286 // VLD1 : Vector Load (multiple single elements)
287 class VLD1D<bits<4> op7_4, string Dt>
288 : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd),
289 (ins addrmode6:$Rn), IIC_VLD1,
290 "vld1", Dt, "$Vd, $Rn", "", []> {
293 let DecoderMethod = "DecodeVLDInstruction";
295 class VLD1Q<bits<4> op7_4, string Dt>
296 : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd),
297 (ins addrmode6:$Rn), IIC_VLD1x2,
298 "vld1", Dt, "$Vd, $Rn", "", []> {
300 let Inst{5-4} = Rn{5-4};
301 let DecoderMethod = "DecodeVLDInstruction";
304 def VLD1d8 : VLD1D<{0,0,0,?}, "8">;
305 def VLD1d16 : VLD1D<{0,1,0,?}, "16">;
306 def VLD1d32 : VLD1D<{1,0,0,?}, "32">;
307 def VLD1d64 : VLD1D<{1,1,0,?}, "64">;
309 def VLD1q8 : VLD1Q<{0,0,?,?}, "8">;
310 def VLD1q16 : VLD1Q<{0,1,?,?}, "16">;
311 def VLD1q32 : VLD1Q<{1,0,?,?}, "32">;
312 def VLD1q64 : VLD1Q<{1,1,?,?}, "64">;
314 def VLD1q8Pseudo : VLDQPseudo<IIC_VLD1x2>;
315 def VLD1q16Pseudo : VLDQPseudo<IIC_VLD1x2>;
316 def VLD1q32Pseudo : VLDQPseudo<IIC_VLD1x2>;
317 def VLD1q64Pseudo : VLDQPseudo<IIC_VLD1x2>;
319 // ...with address register writeback:
320 multiclass VLD1DWB<bits<4> op7_4, string Dt> {
321 def _fixed : NLdSt<0,0b10, 0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
322 (ins addrmode6:$Rn), IIC_VLD1u,
323 "vld1", Dt, "$Vd, $Rn!",
324 "$Rn.addr = $wb", []> {
325 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
327 let DecoderMethod = "DecodeVLDInstruction";
328 let AsmMatchConverter = "cvtVLDwbFixed";
330 def _register : NLdSt<0,0b10,0b0111,op7_4, (outs VecListOneD:$Vd, GPR:$wb),
331 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1u,
332 "vld1", Dt, "$Vd, $Rn, $Rm",
333 "$Rn.addr = $wb", []> {
335 let DecoderMethod = "DecodeVLDInstruction";
336 let AsmMatchConverter = "cvtVLDwbRegister";
339 multiclass VLD1QWB<bits<4> op7_4, string Dt> {
340 def _fixed : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
341 (ins addrmode6:$Rn), IIC_VLD1x2u,
342 "vld1", Dt, "$Vd, $Rn!",
343 "$Rn.addr = $wb", []> {
344 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
345 let Inst{5-4} = Rn{5-4};
346 let DecoderMethod = "DecodeVLDInstruction";
347 let AsmMatchConverter = "cvtVLDwbFixed";
349 def _register : NLdSt<0,0b10,0b1010,op7_4, (outs VecListTwoD:$Vd, GPR:$wb),
350 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
351 "vld1", Dt, "$Vd, $Rn, $Rm",
352 "$Rn.addr = $wb", []> {
353 let Inst{5-4} = Rn{5-4};
354 let DecoderMethod = "DecodeVLDInstruction";
355 let AsmMatchConverter = "cvtVLDwbRegister";
359 defm VLD1d8wb : VLD1DWB<{0,0,0,?}, "8">;
360 defm VLD1d16wb : VLD1DWB<{0,1,0,?}, "16">;
361 defm VLD1d32wb : VLD1DWB<{1,0,0,?}, "32">;
362 defm VLD1d64wb : VLD1DWB<{1,1,0,?}, "64">;
363 defm VLD1q8wb : VLD1QWB<{0,0,?,?}, "8">;
364 defm VLD1q16wb : VLD1QWB<{0,1,?,?}, "16">;
365 defm VLD1q32wb : VLD1QWB<{1,0,?,?}, "32">;
366 defm VLD1q64wb : VLD1QWB<{1,1,?,?}, "64">;
368 def VLD1q8PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
369 def VLD1q16PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
370 def VLD1q32PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
371 def VLD1q64PseudoWB_fixed : VLDQWBfixedPseudo<IIC_VLD1x2u>;
372 def VLD1q8PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
373 def VLD1q16PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
374 def VLD1q32PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
375 def VLD1q64PseudoWB_register : VLDQWBregisterPseudo<IIC_VLD1x2u>;
377 // ...with 3 registers
378 class VLD1D3<bits<4> op7_4, string Dt>
379 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd),
380 (ins addrmode6:$Rn), IIC_VLD1x3, "vld1", Dt,
381 "$Vd, $Rn", "", []> {
384 let DecoderMethod = "DecodeVLDInstruction";
386 class VLD1D3WB<bits<4> op7_4, string Dt>
387 : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
388 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x3u, "vld1", Dt,
389 "$Vd, $Rn$Rm", "$Rn.addr = $wb", []> {
391 let DecoderMethod = "DecodeVLDInstruction";
394 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
395 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
396 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
397 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
399 def VLD1d8T_UPD : VLD1D3WB<{0,0,0,?}, "8">;
400 def VLD1d16T_UPD : VLD1D3WB<{0,1,0,?}, "16">;
401 def VLD1d32T_UPD : VLD1D3WB<{1,0,0,?}, "32">;
402 def VLD1d64T_UPD : VLD1D3WB<{1,1,0,?}, "64">;
404 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
405 def VLD1d64TPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x3u>;
407 // ...with 4 registers
408 class VLD1D4<bits<4> op7_4, string Dt>
409 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
410 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
411 "$Vd, $Rn", "", []> {
413 let Inst{5-4} = Rn{5-4};
414 let DecoderMethod = "DecodeVLDInstruction";
416 class VLD1D4WB<bits<4> op7_4, string Dt>
417 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
418 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD1x4u, "vld1", Dt,
419 "$Vd, $Rn$Rm", "$Rn.addr = $wb",
421 let Inst{5-4} = Rn{5-4};
422 let DecoderMethod = "DecodeVLDInstruction";
425 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
426 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
427 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
428 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
430 def VLD1d8Q_UPD : VLD1D4WB<{0,0,?,?}, "8">;
431 def VLD1d16Q_UPD : VLD1D4WB<{0,1,?,?}, "16">;
432 def VLD1d32Q_UPD : VLD1D4WB<{1,0,?,?}, "32">;
433 def VLD1d64Q_UPD : VLD1D4WB<{1,1,?,?}, "64">;
435 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
436 def VLD1d64QPseudo_UPD : VLDQQWBPseudo<IIC_VLD1x4u>;
438 // VLD2 : Vector Load (multiple 2-element structures)
439 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
440 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
441 (ins addrmode6:$Rn), IIC_VLD2,
442 "vld2", Dt, "$Vd, $Rn", "", []> {
444 let Inst{5-4} = Rn{5-4};
445 let DecoderMethod = "DecodeVLDInstruction";
447 class VLD2Q<bits<4> op7_4, string Dt, RegisterOperand VdTy>
448 : NLdSt<0, 0b10, 0b0011, op7_4,
450 (ins addrmode6:$Rn), IIC_VLD2x2,
451 "vld2", Dt, "$Vd, $Rn", "", []> {
453 let Inst{5-4} = Rn{5-4};
454 let DecoderMethod = "DecodeVLDInstruction";
457 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
458 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
459 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
461 def VLD2q8 : VLD2Q<{0,0,?,?}, "8", VecListFourD>;
462 def VLD2q16 : VLD2Q<{0,1,?,?}, "16", VecListFourD>;
463 def VLD2q32 : VLD2Q<{1,0,?,?}, "32", VecListFourD>;
465 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
466 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
467 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
469 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
470 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
471 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
473 // ...with address register writeback:
474 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
475 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
476 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
477 "vld2", Dt, "$Vd, $Rn$Rm",
478 "$Rn.addr = $wb", []> {
479 let Inst{5-4} = Rn{5-4};
480 let DecoderMethod = "DecodeVLDInstruction";
482 class VLD2QWB<bits<4> op7_4, string Dt, RegisterOperand VdTy>
483 : NLdSt<0, 0b10, 0b0011, op7_4,
484 (outs VdTy:$Vd, GPR:$wb),
485 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
486 "vld2", Dt, "$Vd, $Rn$Rm",
487 "$Rn.addr = $wb", []> {
488 let Inst{5-4} = Rn{5-4};
489 let DecoderMethod = "DecodeVLDInstruction";
492 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
493 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
494 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
496 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8", VecListFourD>;
497 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16", VecListFourD>;
498 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32", VecListFourD>;
500 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
501 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
502 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
504 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
505 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
506 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
508 // ...with double-spaced registers
509 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
510 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
511 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
512 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
513 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
514 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
516 // VLD3 : Vector Load (multiple 3-element structures)
517 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
518 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
519 (ins addrmode6:$Rn), IIC_VLD3,
520 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
523 let DecoderMethod = "DecodeVLDInstruction";
526 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
527 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
528 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
530 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
531 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
532 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
534 // ...with address register writeback:
535 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
536 : NLdSt<0, 0b10, op11_8, op7_4,
537 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
538 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
539 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
540 "$Rn.addr = $wb", []> {
542 let DecoderMethod = "DecodeVLDInstruction";
545 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
546 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
547 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
549 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
550 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
551 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
553 // ...with double-spaced registers:
554 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
555 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
556 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
557 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
558 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
559 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
561 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
562 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
563 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
565 // ...alternate versions to be allocated odd register numbers:
566 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
567 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
568 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
570 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
571 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
572 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
574 // VLD4 : Vector Load (multiple 4-element structures)
575 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
576 : NLdSt<0, 0b10, op11_8, op7_4,
577 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
578 (ins addrmode6:$Rn), IIC_VLD4,
579 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
581 let Inst{5-4} = Rn{5-4};
582 let DecoderMethod = "DecodeVLDInstruction";
585 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
586 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
587 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
589 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
590 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
591 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
593 // ...with address register writeback:
594 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
595 : NLdSt<0, 0b10, op11_8, op7_4,
596 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
597 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
598 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
599 "$Rn.addr = $wb", []> {
600 let Inst{5-4} = Rn{5-4};
601 let DecoderMethod = "DecodeVLDInstruction";
604 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
605 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
606 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
608 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
609 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
610 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
612 // ...with double-spaced registers:
613 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
614 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
615 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
616 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
617 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
618 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
620 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
621 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
622 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
624 // ...alternate versions to be allocated odd register numbers:
625 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
626 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
627 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
629 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
630 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
631 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
633 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
635 // Classes for VLD*LN pseudo-instructions with multi-register operands.
636 // These are expanded to real instructions after register allocation.
637 class VLDQLNPseudo<InstrItinClass itin>
638 : PseudoNLdSt<(outs QPR:$dst),
639 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
640 itin, "$src = $dst">;
641 class VLDQLNWBPseudo<InstrItinClass itin>
642 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
643 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
644 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
645 class VLDQQLNPseudo<InstrItinClass itin>
646 : PseudoNLdSt<(outs QQPR:$dst),
647 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
648 itin, "$src = $dst">;
649 class VLDQQLNWBPseudo<InstrItinClass itin>
650 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
651 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
652 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
653 class VLDQQQQLNPseudo<InstrItinClass itin>
654 : PseudoNLdSt<(outs QQQQPR:$dst),
655 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
656 itin, "$src = $dst">;
657 class VLDQQQQLNWBPseudo<InstrItinClass itin>
658 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
659 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
660 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
662 // VLD1LN : Vector Load (single element to one lane)
663 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
665 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
666 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
667 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
669 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
670 (i32 (LoadOp addrmode6:$Rn)),
673 let DecoderMethod = "DecodeVLD1LN";
675 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
677 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
678 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
679 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
681 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
682 (i32 (LoadOp addrmode6oneL32:$Rn)),
685 let DecoderMethod = "DecodeVLD1LN";
687 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
688 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
689 (i32 (LoadOp addrmode6:$addr)),
693 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
694 let Inst{7-5} = lane{2-0};
696 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
697 let Inst{7-6} = lane{1-0};
700 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
701 let Inst{7} = lane{0};
706 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
707 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
708 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
710 def : Pat<(vector_insert (v2f32 DPR:$src),
711 (f32 (load addrmode6:$addr)), imm:$lane),
712 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
713 def : Pat<(vector_insert (v4f32 QPR:$src),
714 (f32 (load addrmode6:$addr)), imm:$lane),
715 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
717 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
719 // ...with address register writeback:
720 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
721 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
722 (ins addrmode6:$Rn, am6offset:$Rm,
723 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
724 "\\{$Vd[$lane]\\}, $Rn$Rm",
725 "$src = $Vd, $Rn.addr = $wb", []> {
726 let DecoderMethod = "DecodeVLD1LN";
729 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
730 let Inst{7-5} = lane{2-0};
732 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
733 let Inst{7-6} = lane{1-0};
736 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
737 let Inst{7} = lane{0};
742 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
743 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
744 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
746 // VLD2LN : Vector Load (single 2-element structure to one lane)
747 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
748 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
749 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
750 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
751 "$src1 = $Vd, $src2 = $dst2", []> {
754 let DecoderMethod = "DecodeVLD2LN";
757 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
758 let Inst{7-5} = lane{2-0};
760 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
761 let Inst{7-6} = lane{1-0};
763 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
764 let Inst{7} = lane{0};
767 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
768 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
769 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
771 // ...with double-spaced registers:
772 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
773 let Inst{7-6} = lane{1-0};
775 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
776 let Inst{7} = lane{0};
779 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
780 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
782 // ...with address register writeback:
783 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
784 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
785 (ins addrmode6:$Rn, am6offset:$Rm,
786 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
787 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
788 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
790 let DecoderMethod = "DecodeVLD2LN";
793 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
794 let Inst{7-5} = lane{2-0};
796 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
797 let Inst{7-6} = lane{1-0};
799 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
800 let Inst{7} = lane{0};
803 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
804 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
805 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
807 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
808 let Inst{7-6} = lane{1-0};
810 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
811 let Inst{7} = lane{0};
814 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
815 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
817 // VLD3LN : Vector Load (single 3-element structure to one lane)
818 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
819 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
820 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
821 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
822 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
823 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
825 let DecoderMethod = "DecodeVLD3LN";
828 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
829 let Inst{7-5} = lane{2-0};
831 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
832 let Inst{7-6} = lane{1-0};
834 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
835 let Inst{7} = lane{0};
838 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
839 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
840 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
842 // ...with double-spaced registers:
843 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
844 let Inst{7-6} = lane{1-0};
846 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
847 let Inst{7} = lane{0};
850 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
851 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
853 // ...with address register writeback:
854 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
855 : NLdStLn<1, 0b10, op11_8, op7_4,
856 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
857 (ins addrmode6:$Rn, am6offset:$Rm,
858 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
859 IIC_VLD3lnu, "vld3", Dt,
860 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
861 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
863 let DecoderMethod = "DecodeVLD3LN";
866 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
867 let Inst{7-5} = lane{2-0};
869 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
870 let Inst{7-6} = lane{1-0};
872 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
873 let Inst{7} = lane{0};
876 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
877 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
878 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
880 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
881 let Inst{7-6} = lane{1-0};
883 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
884 let Inst{7} = lane{0};
887 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
888 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
890 // VLD4LN : Vector Load (single 4-element structure to one lane)
891 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
892 : NLdStLn<1, 0b10, op11_8, op7_4,
893 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
894 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
895 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
896 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
897 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
900 let DecoderMethod = "DecodeVLD4LN";
903 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
904 let Inst{7-5} = lane{2-0};
906 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
907 let Inst{7-6} = lane{1-0};
909 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
910 let Inst{7} = lane{0};
914 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
915 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
916 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
918 // ...with double-spaced registers:
919 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
920 let Inst{7-6} = lane{1-0};
922 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
923 let Inst{7} = lane{0};
927 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
928 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
930 // ...with address register writeback:
931 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
932 : NLdStLn<1, 0b10, op11_8, op7_4,
933 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
934 (ins addrmode6:$Rn, am6offset:$Rm,
935 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
936 IIC_VLD4lnu, "vld4", Dt,
937 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
938 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
941 let DecoderMethod = "DecodeVLD4LN" ;
944 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
945 let Inst{7-5} = lane{2-0};
947 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
948 let Inst{7-6} = lane{1-0};
950 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
951 let Inst{7} = lane{0};
955 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
956 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
957 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
959 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
960 let Inst{7-6} = lane{1-0};
962 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
963 let Inst{7} = lane{0};
967 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
968 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
970 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
972 // VLD1DUP : Vector Load (single element to all lanes)
973 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
974 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
975 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
976 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
979 let DecoderMethod = "DecodeVLD1DupInstruction";
981 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
982 let Pattern = [(set QPR:$dst,
983 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
986 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
987 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
988 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
990 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
991 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
992 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
994 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
995 (VLD1DUPd32 addrmode6:$addr)>;
996 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
997 (VLD1DUPq32Pseudo addrmode6:$addr)>;
999 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1001 class VLD1QDUP<bits<4> op7_4, string Dt>
1002 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
1003 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1004 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1006 let Inst{4} = Rn{4};
1007 let DecoderMethod = "DecodeVLD1DupInstruction";
1010 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1011 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1012 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1014 // ...with address register writeback:
1015 class VLD1DUPWB<bits<4> op7_4, string Dt>
1016 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
1017 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1018 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1019 let Inst{4} = Rn{4};
1020 let DecoderMethod = "DecodeVLD1DupInstruction";
1022 class VLD1QDUPWB<bits<4> op7_4, string Dt>
1023 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1024 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1025 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1026 let Inst{4} = Rn{4};
1027 let DecoderMethod = "DecodeVLD1DupInstruction";
1030 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
1031 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
1032 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
1034 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
1035 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
1036 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
1038 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1039 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1040 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1042 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1043 class VLD2DUP<bits<4> op7_4, string Dt>
1044 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1045 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1046 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1048 let Inst{4} = Rn{4};
1049 let DecoderMethod = "DecodeVLD2DupInstruction";
1052 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1053 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1054 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1056 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1057 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1058 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1060 // ...with double-spaced registers (not used for codegen):
1061 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1062 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1063 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1065 // ...with address register writeback:
1066 class VLD2DUPWB<bits<4> op7_4, string Dt>
1067 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1068 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1069 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1070 let Inst{4} = Rn{4};
1071 let DecoderMethod = "DecodeVLD2DupInstruction";
1074 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1075 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1076 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1078 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1079 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1080 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1082 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1083 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1084 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1086 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1087 class VLD3DUP<bits<4> op7_4, string Dt>
1088 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1089 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1090 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1093 let DecoderMethod = "DecodeVLD3DupInstruction";
1096 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1097 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1098 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1100 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1101 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1102 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1104 // ...with double-spaced registers (not used for codegen):
1105 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1106 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1107 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1109 // ...with address register writeback:
1110 class VLD3DUPWB<bits<4> op7_4, string Dt>
1111 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1112 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1113 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1114 "$Rn.addr = $wb", []> {
1116 let DecoderMethod = "DecodeVLD3DupInstruction";
1119 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1120 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1121 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1123 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1124 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1125 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1127 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1128 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1129 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1131 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1132 class VLD4DUP<bits<4> op7_4, string Dt>
1133 : NLdSt<1, 0b10, 0b1111, op7_4,
1134 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1135 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1136 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1138 let Inst{4} = Rn{4};
1139 let DecoderMethod = "DecodeVLD4DupInstruction";
1142 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1143 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1144 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1146 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1147 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1148 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1150 // ...with double-spaced registers (not used for codegen):
1151 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1152 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1153 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1155 // ...with address register writeback:
1156 class VLD4DUPWB<bits<4> op7_4, string Dt>
1157 : NLdSt<1, 0b10, 0b1111, op7_4,
1158 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1159 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1160 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1161 "$Rn.addr = $wb", []> {
1162 let Inst{4} = Rn{4};
1163 let DecoderMethod = "DecodeVLD4DupInstruction";
1166 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1167 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1168 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1170 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1171 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1172 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1174 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1175 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1176 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1178 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1180 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1182 // Classes for VST* pseudo-instructions with multi-register operands.
1183 // These are expanded to real instructions after register allocation.
1184 class VSTQPseudo<InstrItinClass itin>
1185 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1186 class VSTQWBPseudo<InstrItinClass itin>
1187 : PseudoNLdSt<(outs GPR:$wb),
1188 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1189 "$addr.addr = $wb">;
1190 class VSTQQPseudo<InstrItinClass itin>
1191 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1192 class VSTQQWBPseudo<InstrItinClass itin>
1193 : PseudoNLdSt<(outs GPR:$wb),
1194 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1195 "$addr.addr = $wb">;
1196 class VSTQQQQPseudo<InstrItinClass itin>
1197 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1198 class VSTQQQQWBPseudo<InstrItinClass itin>
1199 : PseudoNLdSt<(outs GPR:$wb),
1200 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1201 "$addr.addr = $wb">;
1203 // VST1 : Vector Store (multiple single elements)
1204 class VST1D<bits<4> op7_4, string Dt>
1205 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1206 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1208 let Inst{4} = Rn{4};
1209 let DecoderMethod = "DecodeVSTInstruction";
1211 class VST1Q<bits<4> op7_4, string Dt>
1212 : NLdSt<0,0b00,0b1010,op7_4, (outs),
1213 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
1214 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1216 let Inst{5-4} = Rn{5-4};
1217 let DecoderMethod = "DecodeVSTInstruction";
1220 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1221 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1222 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1223 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1225 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1226 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1227 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1228 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1230 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1231 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1232 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1233 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1235 // ...with address register writeback:
1236 class VST1DWB<bits<4> op7_4, string Dt>
1237 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
1238 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1239 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1240 let Inst{4} = Rn{4};
1241 let DecoderMethod = "DecodeVSTInstruction";
1243 class VST1QWB<bits<4> op7_4, string Dt>
1244 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1245 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1246 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1247 "$Rn.addr = $wb", []> {
1248 let Inst{5-4} = Rn{5-4};
1249 let DecoderMethod = "DecodeVSTInstruction";
1252 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1253 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1254 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1255 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1257 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1258 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1259 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1260 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1262 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1263 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1264 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1265 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1267 // ...with 3 registers
1268 class VST1D3<bits<4> op7_4, string Dt>
1269 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1270 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1271 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1273 let Inst{4} = Rn{4};
1274 let DecoderMethod = "DecodeVSTInstruction";
1276 class VST1D3WB<bits<4> op7_4, string Dt>
1277 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1278 (ins addrmode6:$Rn, am6offset:$Rm,
1279 DPR:$Vd, DPR:$src2, DPR:$src3),
1280 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1281 "$Rn.addr = $wb", []> {
1282 let Inst{4} = Rn{4};
1283 let DecoderMethod = "DecodeVSTInstruction";
1286 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1287 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1288 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1289 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1291 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1292 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1293 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1294 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1296 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1297 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1299 // ...with 4 registers
1300 class VST1D4<bits<4> op7_4, string Dt>
1301 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1302 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1303 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1306 let Inst{5-4} = Rn{5-4};
1307 let DecoderMethod = "DecodeVSTInstruction";
1309 class VST1D4WB<bits<4> op7_4, string Dt>
1310 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1311 (ins addrmode6:$Rn, am6offset:$Rm,
1312 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1313 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1314 "$Rn.addr = $wb", []> {
1315 let Inst{5-4} = Rn{5-4};
1316 let DecoderMethod = "DecodeVSTInstruction";
1319 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1320 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1321 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1322 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1324 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1325 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1326 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1327 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1329 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1330 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1332 // VST2 : Vector Store (multiple 2-element structures)
1333 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1334 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1335 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1336 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1338 let Inst{5-4} = Rn{5-4};
1339 let DecoderMethod = "DecodeVSTInstruction";
1341 class VST2Q<bits<4> op7_4, string Dt>
1342 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1343 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1344 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1347 let Inst{5-4} = Rn{5-4};
1348 let DecoderMethod = "DecodeVSTInstruction";
1351 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1352 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1353 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1355 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1356 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1357 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1359 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1360 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1361 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1363 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1364 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1365 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1367 // ...with address register writeback:
1368 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1369 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1370 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1371 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1372 "$Rn.addr = $wb", []> {
1373 let Inst{5-4} = Rn{5-4};
1374 let DecoderMethod = "DecodeVSTInstruction";
1376 class VST2QWB<bits<4> op7_4, string Dt>
1377 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1378 (ins addrmode6:$Rn, am6offset:$Rm,
1379 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1380 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1381 "$Rn.addr = $wb", []> {
1382 let Inst{5-4} = Rn{5-4};
1383 let DecoderMethod = "DecodeVSTInstruction";
1386 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1387 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1388 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1390 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1391 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1392 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1394 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1395 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1396 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1398 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1399 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1400 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1402 // ...with double-spaced registers
1403 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1404 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1405 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1406 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1407 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1408 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1410 // VST3 : Vector Store (multiple 3-element structures)
1411 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1412 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1413 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1414 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1416 let Inst{4} = Rn{4};
1417 let DecoderMethod = "DecodeVSTInstruction";
1420 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1421 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1422 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1424 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1425 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1426 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1428 // ...with address register writeback:
1429 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1430 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1431 (ins addrmode6:$Rn, am6offset:$Rm,
1432 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1433 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1434 "$Rn.addr = $wb", []> {
1435 let Inst{4} = Rn{4};
1436 let DecoderMethod = "DecodeVSTInstruction";
1439 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1440 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1441 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1443 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1444 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1445 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1447 // ...with double-spaced registers:
1448 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1449 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1450 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1451 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1452 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1453 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1455 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1456 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1457 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1459 // ...alternate versions to be allocated odd register numbers:
1460 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1461 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1462 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1464 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1465 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1466 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1468 // VST4 : Vector Store (multiple 4-element structures)
1469 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1470 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1471 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1472 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1475 let Inst{5-4} = Rn{5-4};
1476 let DecoderMethod = "DecodeVSTInstruction";
1479 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1480 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1481 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1483 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1484 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1485 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1487 // ...with address register writeback:
1488 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1489 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1490 (ins addrmode6:$Rn, am6offset:$Rm,
1491 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1492 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1493 "$Rn.addr = $wb", []> {
1494 let Inst{5-4} = Rn{5-4};
1495 let DecoderMethod = "DecodeVSTInstruction";
1498 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1499 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1500 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1502 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1503 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1504 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1506 // ...with double-spaced registers:
1507 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1508 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1509 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1510 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1511 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1512 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1514 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1515 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1516 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1518 // ...alternate versions to be allocated odd register numbers:
1519 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1520 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1521 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1523 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1524 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1525 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1527 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1529 // Classes for VST*LN pseudo-instructions with multi-register operands.
1530 // These are expanded to real instructions after register allocation.
1531 class VSTQLNPseudo<InstrItinClass itin>
1532 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1534 class VSTQLNWBPseudo<InstrItinClass itin>
1535 : PseudoNLdSt<(outs GPR:$wb),
1536 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1537 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1538 class VSTQQLNPseudo<InstrItinClass itin>
1539 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1541 class VSTQQLNWBPseudo<InstrItinClass itin>
1542 : PseudoNLdSt<(outs GPR:$wb),
1543 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1544 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1545 class VSTQQQQLNPseudo<InstrItinClass itin>
1546 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1548 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1549 : PseudoNLdSt<(outs GPR:$wb),
1550 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1551 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1553 // VST1LN : Vector Store (single element from one lane)
1554 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1555 PatFrag StoreOp, SDNode ExtractOp>
1556 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1557 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1558 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1559 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1561 let DecoderMethod = "DecodeVST1LN";
1563 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1564 PatFrag StoreOp, SDNode ExtractOp>
1565 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1566 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1567 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1568 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1570 let DecoderMethod = "DecodeVST1LN";
1572 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1573 : VSTQLNPseudo<IIC_VST1ln> {
1574 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1578 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1580 let Inst{7-5} = lane{2-0};
1582 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1584 let Inst{7-6} = lane{1-0};
1585 let Inst{4} = Rn{5};
1588 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1589 let Inst{7} = lane{0};
1590 let Inst{5-4} = Rn{5-4};
1593 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1594 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1595 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1597 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1598 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1599 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1600 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1602 // ...with address register writeback:
1603 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1604 PatFrag StoreOp, SDNode ExtractOp>
1605 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1606 (ins addrmode6:$Rn, am6offset:$Rm,
1607 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1608 "\\{$Vd[$lane]\\}, $Rn$Rm",
1610 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1611 addrmode6:$Rn, am6offset:$Rm))]> {
1612 let DecoderMethod = "DecodeVST1LN";
1614 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1615 : VSTQLNWBPseudo<IIC_VST1lnu> {
1616 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1617 addrmode6:$addr, am6offset:$offset))];
1620 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1622 let Inst{7-5} = lane{2-0};
1624 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1626 let Inst{7-6} = lane{1-0};
1627 let Inst{4} = Rn{5};
1629 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1631 let Inst{7} = lane{0};
1632 let Inst{5-4} = Rn{5-4};
1635 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1636 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1637 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1639 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1641 // VST2LN : Vector Store (single 2-element structure from one lane)
1642 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1643 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1644 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1645 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1648 let Inst{4} = Rn{4};
1649 let DecoderMethod = "DecodeVST2LN";
1652 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1653 let Inst{7-5} = lane{2-0};
1655 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1656 let Inst{7-6} = lane{1-0};
1658 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1659 let Inst{7} = lane{0};
1662 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1663 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1664 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1666 // ...with double-spaced registers:
1667 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1668 let Inst{7-6} = lane{1-0};
1669 let Inst{4} = Rn{4};
1671 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1672 let Inst{7} = lane{0};
1673 let Inst{4} = Rn{4};
1676 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1677 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1679 // ...with address register writeback:
1680 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1681 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1682 (ins addrmode6:$addr, am6offset:$offset,
1683 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1684 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1685 "$addr.addr = $wb", []> {
1686 let Inst{4} = Rn{4};
1687 let DecoderMethod = "DecodeVST2LN";
1690 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1691 let Inst{7-5} = lane{2-0};
1693 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1694 let Inst{7-6} = lane{1-0};
1696 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1697 let Inst{7} = lane{0};
1700 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1701 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1702 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1704 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1705 let Inst{7-6} = lane{1-0};
1707 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1708 let Inst{7} = lane{0};
1711 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1712 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1714 // VST3LN : Vector Store (single 3-element structure from one lane)
1715 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1716 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1717 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1718 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1719 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1721 let DecoderMethod = "DecodeVST3LN";
1724 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1725 let Inst{7-5} = lane{2-0};
1727 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1728 let Inst{7-6} = lane{1-0};
1730 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1731 let Inst{7} = lane{0};
1734 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1735 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1736 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1738 // ...with double-spaced registers:
1739 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1740 let Inst{7-6} = lane{1-0};
1742 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1743 let Inst{7} = lane{0};
1746 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1747 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1749 // ...with address register writeback:
1750 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1751 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1752 (ins addrmode6:$Rn, am6offset:$Rm,
1753 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1754 IIC_VST3lnu, "vst3", Dt,
1755 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1756 "$Rn.addr = $wb", []> {
1757 let DecoderMethod = "DecodeVST3LN";
1760 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1761 let Inst{7-5} = lane{2-0};
1763 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1764 let Inst{7-6} = lane{1-0};
1766 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1767 let Inst{7} = lane{0};
1770 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1771 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1772 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1774 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1775 let Inst{7-6} = lane{1-0};
1777 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1778 let Inst{7} = lane{0};
1781 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1782 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1784 // VST4LN : Vector Store (single 4-element structure from one lane)
1785 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1786 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1787 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1788 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1789 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1792 let Inst{4} = Rn{4};
1793 let DecoderMethod = "DecodeVST4LN";
1796 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1797 let Inst{7-5} = lane{2-0};
1799 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1800 let Inst{7-6} = lane{1-0};
1802 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1803 let Inst{7} = lane{0};
1804 let Inst{5} = Rn{5};
1807 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1808 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1809 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1811 // ...with double-spaced registers:
1812 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1813 let Inst{7-6} = lane{1-0};
1815 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1816 let Inst{7} = lane{0};
1817 let Inst{5} = Rn{5};
1820 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1821 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1823 // ...with address register writeback:
1824 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1825 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1826 (ins addrmode6:$Rn, am6offset:$Rm,
1827 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1828 IIC_VST4lnu, "vst4", Dt,
1829 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1830 "$Rn.addr = $wb", []> {
1831 let Inst{4} = Rn{4};
1832 let DecoderMethod = "DecodeVST4LN";
1835 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1836 let Inst{7-5} = lane{2-0};
1838 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1839 let Inst{7-6} = lane{1-0};
1841 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1842 let Inst{7} = lane{0};
1843 let Inst{5} = Rn{5};
1846 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1847 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1848 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1850 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1851 let Inst{7-6} = lane{1-0};
1853 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1854 let Inst{7} = lane{0};
1855 let Inst{5} = Rn{5};
1858 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1859 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1861 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1864 //===----------------------------------------------------------------------===//
1865 // NEON pattern fragments
1866 //===----------------------------------------------------------------------===//
1868 // Extract D sub-registers of Q registers.
1869 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1870 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1871 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1873 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1874 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1875 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1877 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1878 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1879 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1881 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1882 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1883 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1886 // Extract S sub-registers of Q/D registers.
1887 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1888 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1889 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1892 // Translate lane numbers from Q registers to D subregs.
1893 def SubReg_i8_lane : SDNodeXForm<imm, [{
1894 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1896 def SubReg_i16_lane : SDNodeXForm<imm, [{
1897 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1899 def SubReg_i32_lane : SDNodeXForm<imm, [{
1900 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1903 //===----------------------------------------------------------------------===//
1904 // Instruction Classes
1905 //===----------------------------------------------------------------------===//
1907 // Basic 2-register operations: double- and quad-register.
1908 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1909 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1910 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1911 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1912 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1913 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1914 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1915 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1916 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1917 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1918 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1919 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1921 // Basic 2-register intrinsics, both double- and quad-register.
1922 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1923 bits<2> op17_16, bits<5> op11_7, bit op4,
1924 InstrItinClass itin, string OpcodeStr, string Dt,
1925 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1926 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1927 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1928 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1929 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1930 bits<2> op17_16, bits<5> op11_7, bit op4,
1931 InstrItinClass itin, string OpcodeStr, string Dt,
1932 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1933 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1934 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1935 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1937 // Narrow 2-register operations.
1938 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1939 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1940 InstrItinClass itin, string OpcodeStr, string Dt,
1941 ValueType TyD, ValueType TyQ, SDNode OpNode>
1942 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1943 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1944 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1946 // Narrow 2-register intrinsics.
1947 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1948 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1949 InstrItinClass itin, string OpcodeStr, string Dt,
1950 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1951 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1952 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1953 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1955 // Long 2-register operations (currently only used for VMOVL).
1956 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1957 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1958 InstrItinClass itin, string OpcodeStr, string Dt,
1959 ValueType TyQ, ValueType TyD, SDNode OpNode>
1960 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1961 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1962 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1964 // Long 2-register intrinsics.
1965 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1966 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1967 InstrItinClass itin, string OpcodeStr, string Dt,
1968 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1969 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1970 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1971 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
1973 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1974 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1975 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1976 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1977 OpcodeStr, Dt, "$Vd, $Vm",
1978 "$src1 = $Vd, $src2 = $Vm", []>;
1979 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
1980 InstrItinClass itin, string OpcodeStr, string Dt>
1981 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
1982 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
1983 "$src1 = $Vd, $src2 = $Vm", []>;
1985 // Basic 3-register operations: double- and quad-register.
1986 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1987 InstrItinClass itin, string OpcodeStr, string Dt,
1988 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
1989 : N3V<op24, op23, op21_20, op11_8, 0, op4,
1990 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
1991 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
1992 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
1993 let isCommutable = Commutable;
1995 // Same as N3VD but no data type.
1996 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
1997 InstrItinClass itin, string OpcodeStr,
1998 ValueType ResTy, ValueType OpTy,
1999 SDNode OpNode, bit Commutable>
2000 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2001 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2002 OpcodeStr, "$Vd, $Vn, $Vm", "",
2003 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2004 let isCommutable = Commutable;
2007 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2008 InstrItinClass itin, string OpcodeStr, string Dt,
2009 ValueType Ty, SDNode ShOp>
2010 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2011 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2012 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2014 (Ty (ShOp (Ty DPR:$Vn),
2015 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2016 let isCommutable = 0;
2018 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2019 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2020 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2021 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2022 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2024 (Ty (ShOp (Ty DPR:$Vn),
2025 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2026 let isCommutable = 0;
2029 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2030 InstrItinClass itin, string OpcodeStr, string Dt,
2031 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2032 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2033 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2034 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2035 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2036 let isCommutable = Commutable;
2038 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2039 InstrItinClass itin, string OpcodeStr,
2040 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2041 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2042 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2043 OpcodeStr, "$Vd, $Vn, $Vm", "",
2044 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2045 let isCommutable = Commutable;
2047 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2048 InstrItinClass itin, string OpcodeStr, string Dt,
2049 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2050 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2051 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2052 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2053 [(set (ResTy QPR:$Vd),
2054 (ResTy (ShOp (ResTy QPR:$Vn),
2055 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2057 let isCommutable = 0;
2059 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2060 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2061 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2062 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2063 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2064 [(set (ResTy QPR:$Vd),
2065 (ResTy (ShOp (ResTy QPR:$Vn),
2066 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2068 let isCommutable = 0;
2071 // Basic 3-register intrinsics, both double- and quad-register.
2072 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2073 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2074 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2075 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2076 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2077 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2078 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2079 let isCommutable = Commutable;
2081 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2082 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2083 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2084 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2085 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2087 (Ty (IntOp (Ty DPR:$Vn),
2088 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2090 let isCommutable = 0;
2092 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2093 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2094 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2095 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2096 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2098 (Ty (IntOp (Ty DPR:$Vn),
2099 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2100 let isCommutable = 0;
2102 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2103 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2104 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2105 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2106 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2107 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2108 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2109 let isCommutable = 0;
2112 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2113 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2114 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2115 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2116 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2117 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2118 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2119 let isCommutable = Commutable;
2121 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2122 string OpcodeStr, string Dt,
2123 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2124 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2125 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2126 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2127 [(set (ResTy QPR:$Vd),
2128 (ResTy (IntOp (ResTy QPR:$Vn),
2129 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2131 let isCommutable = 0;
2133 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2134 string OpcodeStr, string Dt,
2135 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2136 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2137 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2138 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2139 [(set (ResTy QPR:$Vd),
2140 (ResTy (IntOp (ResTy QPR:$Vn),
2141 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2143 let isCommutable = 0;
2145 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2146 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2147 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2148 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2149 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2150 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2151 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2152 let isCommutable = 0;
2155 // Multiply-Add/Sub operations: double- and quad-register.
2156 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2157 InstrItinClass itin, string OpcodeStr, string Dt,
2158 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2159 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2160 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2161 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2162 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2163 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2165 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2166 string OpcodeStr, string Dt,
2167 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2168 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2170 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2172 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2174 (Ty (ShOp (Ty DPR:$src1),
2176 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2178 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2179 string OpcodeStr, string Dt,
2180 ValueType Ty, SDNode MulOp, SDNode ShOp>
2181 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2183 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2185 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2187 (Ty (ShOp (Ty DPR:$src1),
2189 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2192 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2193 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2194 SDPatternOperator MulOp, SDPatternOperator OpNode>
2195 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2196 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2197 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2198 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2199 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2200 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2201 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2202 SDPatternOperator MulOp, SDPatternOperator ShOp>
2203 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2205 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2207 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2208 [(set (ResTy QPR:$Vd),
2209 (ResTy (ShOp (ResTy QPR:$src1),
2210 (ResTy (MulOp QPR:$Vn,
2211 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2213 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2214 string OpcodeStr, string Dt,
2215 ValueType ResTy, ValueType OpTy,
2216 SDNode MulOp, SDNode ShOp>
2217 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2219 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2221 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2222 [(set (ResTy QPR:$Vd),
2223 (ResTy (ShOp (ResTy QPR:$src1),
2224 (ResTy (MulOp QPR:$Vn,
2225 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2228 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2229 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2230 InstrItinClass itin, string OpcodeStr, string Dt,
2231 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2232 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2233 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2234 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2235 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2236 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2237 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2238 InstrItinClass itin, string OpcodeStr, string Dt,
2239 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2240 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2241 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2242 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2243 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2244 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2246 // Neon 3-argument intrinsics, both double- and quad-register.
2247 // The destination register is also used as the first source operand register.
2248 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2249 InstrItinClass itin, string OpcodeStr, string Dt,
2250 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2251 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2252 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2253 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2254 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2255 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2256 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2257 InstrItinClass itin, string OpcodeStr, string Dt,
2258 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2259 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2260 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2261 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2262 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2263 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2265 // Long Multiply-Add/Sub operations.
2266 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2267 InstrItinClass itin, string OpcodeStr, string Dt,
2268 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2269 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2270 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2271 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2272 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2273 (TyQ (MulOp (TyD DPR:$Vn),
2274 (TyD DPR:$Vm)))))]>;
2275 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2276 InstrItinClass itin, string OpcodeStr, string Dt,
2277 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2278 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2279 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2281 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2283 (OpNode (TyQ QPR:$src1),
2284 (TyQ (MulOp (TyD DPR:$Vn),
2285 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2287 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2288 InstrItinClass itin, string OpcodeStr, string Dt,
2289 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2290 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2291 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2293 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2295 (OpNode (TyQ QPR:$src1),
2296 (TyQ (MulOp (TyD DPR:$Vn),
2297 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2300 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2301 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2302 InstrItinClass itin, string OpcodeStr, string Dt,
2303 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2305 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2306 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2307 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2308 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2309 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2310 (TyD DPR:$Vm)))))))]>;
2312 // Neon Long 3-argument intrinsic. The destination register is
2313 // a quad-register and is also used as the first source operand register.
2314 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2315 InstrItinClass itin, string OpcodeStr, string Dt,
2316 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2317 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2318 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2319 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2321 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2322 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2323 string OpcodeStr, string Dt,
2324 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2325 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2327 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2329 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2330 [(set (ResTy QPR:$Vd),
2331 (ResTy (IntOp (ResTy QPR:$src1),
2333 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2335 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2336 InstrItinClass itin, string OpcodeStr, string Dt,
2337 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2338 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2340 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2342 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2343 [(set (ResTy QPR:$Vd),
2344 (ResTy (IntOp (ResTy QPR:$src1),
2346 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2349 // Narrowing 3-register intrinsics.
2350 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2351 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2352 Intrinsic IntOp, bit Commutable>
2353 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2354 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2355 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2356 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2357 let isCommutable = Commutable;
2360 // Long 3-register operations.
2361 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2362 InstrItinClass itin, string OpcodeStr, string Dt,
2363 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2364 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2365 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2366 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2367 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2368 let isCommutable = Commutable;
2370 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2371 InstrItinClass itin, string OpcodeStr, string Dt,
2372 ValueType TyQ, ValueType TyD, SDNode OpNode>
2373 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2374 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2375 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2377 (TyQ (OpNode (TyD DPR:$Vn),
2378 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2379 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2380 InstrItinClass itin, string OpcodeStr, string Dt,
2381 ValueType TyQ, ValueType TyD, SDNode OpNode>
2382 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2383 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2384 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2386 (TyQ (OpNode (TyD DPR:$Vn),
2387 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2389 // Long 3-register operations with explicitly extended operands.
2390 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2391 InstrItinClass itin, string OpcodeStr, string Dt,
2392 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2394 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2395 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2396 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2397 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2398 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2399 let isCommutable = Commutable;
2402 // Long 3-register intrinsics with explicit extend (VABDL).
2403 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2404 InstrItinClass itin, string OpcodeStr, string Dt,
2405 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2407 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2408 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2409 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2410 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2411 (TyD DPR:$Vm))))))]> {
2412 let isCommutable = Commutable;
2415 // Long 3-register intrinsics.
2416 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2417 InstrItinClass itin, string OpcodeStr, string Dt,
2418 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2419 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2420 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2421 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2422 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2423 let isCommutable = Commutable;
2425 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2426 string OpcodeStr, string Dt,
2427 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2428 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2429 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2430 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2431 [(set (ResTy QPR:$Vd),
2432 (ResTy (IntOp (OpTy DPR:$Vn),
2433 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2435 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2436 InstrItinClass itin, string OpcodeStr, string Dt,
2437 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2438 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2439 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2440 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2441 [(set (ResTy QPR:$Vd),
2442 (ResTy (IntOp (OpTy DPR:$Vn),
2443 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2446 // Wide 3-register operations.
2447 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2448 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2449 SDNode OpNode, SDNode ExtOp, bit Commutable>
2450 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2451 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2452 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2453 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2454 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2455 let isCommutable = Commutable;
2458 // Pairwise long 2-register intrinsics, both double- and quad-register.
2459 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2460 bits<2> op17_16, bits<5> op11_7, bit op4,
2461 string OpcodeStr, string Dt,
2462 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2463 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2464 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2465 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2466 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2467 bits<2> op17_16, bits<5> op11_7, bit op4,
2468 string OpcodeStr, string Dt,
2469 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2470 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2471 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2472 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2474 // Pairwise long 2-register accumulate intrinsics,
2475 // both double- and quad-register.
2476 // The destination register is also used as the first source operand register.
2477 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2478 bits<2> op17_16, bits<5> op11_7, bit op4,
2479 string OpcodeStr, string Dt,
2480 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2481 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2482 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2483 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2484 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2485 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2486 bits<2> op17_16, bits<5> op11_7, bit op4,
2487 string OpcodeStr, string Dt,
2488 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2489 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2490 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2491 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2492 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2494 // Shift by immediate,
2495 // both double- and quad-register.
2496 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2497 Format f, InstrItinClass itin, Operand ImmTy,
2498 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2499 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2500 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2501 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2502 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2503 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2504 Format f, InstrItinClass itin, Operand ImmTy,
2505 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2506 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2507 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2508 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2509 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2511 // Long shift by immediate.
2512 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2513 string OpcodeStr, string Dt,
2514 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2515 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2516 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2517 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2518 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2519 (i32 imm:$SIMM))))]>;
2521 // Narrow shift by immediate.
2522 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2523 InstrItinClass itin, string OpcodeStr, string Dt,
2524 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2525 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2526 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2527 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2528 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2529 (i32 imm:$SIMM))))]>;
2531 // Shift right by immediate and accumulate,
2532 // both double- and quad-register.
2533 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2534 Operand ImmTy, string OpcodeStr, string Dt,
2535 ValueType Ty, SDNode ShOp>
2536 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2537 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2538 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2539 [(set DPR:$Vd, (Ty (add DPR:$src1,
2540 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2541 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2542 Operand ImmTy, string OpcodeStr, string Dt,
2543 ValueType Ty, SDNode ShOp>
2544 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2545 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2546 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2547 [(set QPR:$Vd, (Ty (add QPR:$src1,
2548 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2550 // Shift by immediate and insert,
2551 // both double- and quad-register.
2552 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2553 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2554 ValueType Ty,SDNode ShOp>
2555 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2556 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2557 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2558 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2559 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2560 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2561 ValueType Ty,SDNode ShOp>
2562 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2563 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2564 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2565 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2567 // Convert, with fractional bits immediate,
2568 // both double- and quad-register.
2569 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2570 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2572 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2573 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2574 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2575 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2576 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2577 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2579 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2580 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2581 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2582 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2584 //===----------------------------------------------------------------------===//
2586 //===----------------------------------------------------------------------===//
2588 // Abbreviations used in multiclass suffixes:
2589 // Q = quarter int (8 bit) elements
2590 // H = half int (16 bit) elements
2591 // S = single int (32 bit) elements
2592 // D = double int (64 bit) elements
2594 // Neon 2-register vector operations and intrinsics.
2596 // Neon 2-register comparisons.
2597 // source operand element sizes of 8, 16 and 32 bits:
2598 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2599 bits<5> op11_7, bit op4, string opc, string Dt,
2600 string asm, SDNode OpNode> {
2601 // 64-bit vector types.
2602 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2603 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2604 opc, !strconcat(Dt, "8"), asm, "",
2605 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2606 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2607 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2608 opc, !strconcat(Dt, "16"), asm, "",
2609 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2610 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2611 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2612 opc, !strconcat(Dt, "32"), asm, "",
2613 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2614 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2615 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2616 opc, "f32", asm, "",
2617 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2618 let Inst{10} = 1; // overwrite F = 1
2621 // 128-bit vector types.
2622 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2623 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2624 opc, !strconcat(Dt, "8"), asm, "",
2625 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2626 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2627 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2628 opc, !strconcat(Dt, "16"), asm, "",
2629 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2630 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2631 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2632 opc, !strconcat(Dt, "32"), asm, "",
2633 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2634 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2635 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2636 opc, "f32", asm, "",
2637 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2638 let Inst{10} = 1; // overwrite F = 1
2643 // Neon 2-register vector intrinsics,
2644 // element sizes of 8, 16 and 32 bits:
2645 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2646 bits<5> op11_7, bit op4,
2647 InstrItinClass itinD, InstrItinClass itinQ,
2648 string OpcodeStr, string Dt, Intrinsic IntOp> {
2649 // 64-bit vector types.
2650 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2651 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2652 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2653 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2654 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2655 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2657 // 128-bit vector types.
2658 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2659 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2660 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2661 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2662 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2663 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2667 // Neon Narrowing 2-register vector operations,
2668 // source operand element sizes of 16, 32 and 64 bits:
2669 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2670 bits<5> op11_7, bit op6, bit op4,
2671 InstrItinClass itin, string OpcodeStr, string Dt,
2673 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2674 itin, OpcodeStr, !strconcat(Dt, "16"),
2675 v8i8, v8i16, OpNode>;
2676 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2677 itin, OpcodeStr, !strconcat(Dt, "32"),
2678 v4i16, v4i32, OpNode>;
2679 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2680 itin, OpcodeStr, !strconcat(Dt, "64"),
2681 v2i32, v2i64, OpNode>;
2684 // Neon Narrowing 2-register vector intrinsics,
2685 // source operand element sizes of 16, 32 and 64 bits:
2686 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2687 bits<5> op11_7, bit op6, bit op4,
2688 InstrItinClass itin, string OpcodeStr, string Dt,
2690 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2691 itin, OpcodeStr, !strconcat(Dt, "16"),
2692 v8i8, v8i16, IntOp>;
2693 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2694 itin, OpcodeStr, !strconcat(Dt, "32"),
2695 v4i16, v4i32, IntOp>;
2696 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2697 itin, OpcodeStr, !strconcat(Dt, "64"),
2698 v2i32, v2i64, IntOp>;
2702 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2703 // source operand element sizes of 16, 32 and 64 bits:
2704 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2705 string OpcodeStr, string Dt, SDNode OpNode> {
2706 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2707 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2708 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2709 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2710 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2711 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2715 // Neon 3-register vector operations.
2717 // First with only element sizes of 8, 16 and 32 bits:
2718 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2719 InstrItinClass itinD16, InstrItinClass itinD32,
2720 InstrItinClass itinQ16, InstrItinClass itinQ32,
2721 string OpcodeStr, string Dt,
2722 SDNode OpNode, bit Commutable = 0> {
2723 // 64-bit vector types.
2724 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2725 OpcodeStr, !strconcat(Dt, "8"),
2726 v8i8, v8i8, OpNode, Commutable>;
2727 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2728 OpcodeStr, !strconcat(Dt, "16"),
2729 v4i16, v4i16, OpNode, Commutable>;
2730 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2731 OpcodeStr, !strconcat(Dt, "32"),
2732 v2i32, v2i32, OpNode, Commutable>;
2734 // 128-bit vector types.
2735 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2736 OpcodeStr, !strconcat(Dt, "8"),
2737 v16i8, v16i8, OpNode, Commutable>;
2738 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2739 OpcodeStr, !strconcat(Dt, "16"),
2740 v8i16, v8i16, OpNode, Commutable>;
2741 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2742 OpcodeStr, !strconcat(Dt, "32"),
2743 v4i32, v4i32, OpNode, Commutable>;
2746 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2747 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2749 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2751 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2752 v8i16, v4i16, ShOp>;
2753 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2754 v4i32, v2i32, ShOp>;
2757 // ....then also with element size 64 bits:
2758 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2759 InstrItinClass itinD, InstrItinClass itinQ,
2760 string OpcodeStr, string Dt,
2761 SDNode OpNode, bit Commutable = 0>
2762 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2763 OpcodeStr, Dt, OpNode, Commutable> {
2764 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2765 OpcodeStr, !strconcat(Dt, "64"),
2766 v1i64, v1i64, OpNode, Commutable>;
2767 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2768 OpcodeStr, !strconcat(Dt, "64"),
2769 v2i64, v2i64, OpNode, Commutable>;
2773 // Neon 3-register vector intrinsics.
2775 // First with only element sizes of 16 and 32 bits:
2776 multiclass N3VInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2777 InstrItinClass itinD16, InstrItinClass itinD32,
2778 InstrItinClass itinQ16, InstrItinClass itinQ32,
2779 string OpcodeStr, string Dt,
2780 Intrinsic IntOp, bit Commutable = 0> {
2781 // 64-bit vector types.
2782 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2783 OpcodeStr, !strconcat(Dt, "16"),
2784 v4i16, v4i16, IntOp, Commutable>;
2785 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2786 OpcodeStr, !strconcat(Dt, "32"),
2787 v2i32, v2i32, IntOp, Commutable>;
2789 // 128-bit vector types.
2790 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2791 OpcodeStr, !strconcat(Dt, "16"),
2792 v8i16, v8i16, IntOp, Commutable>;
2793 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2794 OpcodeStr, !strconcat(Dt, "32"),
2795 v4i32, v4i32, IntOp, Commutable>;
2797 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2798 InstrItinClass itinD16, InstrItinClass itinD32,
2799 InstrItinClass itinQ16, InstrItinClass itinQ32,
2800 string OpcodeStr, string Dt,
2802 // 64-bit vector types.
2803 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2804 OpcodeStr, !strconcat(Dt, "16"),
2805 v4i16, v4i16, IntOp>;
2806 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2807 OpcodeStr, !strconcat(Dt, "32"),
2808 v2i32, v2i32, IntOp>;
2810 // 128-bit vector types.
2811 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2812 OpcodeStr, !strconcat(Dt, "16"),
2813 v8i16, v8i16, IntOp>;
2814 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2815 OpcodeStr, !strconcat(Dt, "32"),
2816 v4i32, v4i32, IntOp>;
2819 multiclass N3VIntSL_HS<bits<4> op11_8,
2820 InstrItinClass itinD16, InstrItinClass itinD32,
2821 InstrItinClass itinQ16, InstrItinClass itinQ32,
2822 string OpcodeStr, string Dt, Intrinsic IntOp> {
2823 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2824 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2825 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2826 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2827 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2828 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2829 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2830 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2833 // ....then also with element size of 8 bits:
2834 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2835 InstrItinClass itinD16, InstrItinClass itinD32,
2836 InstrItinClass itinQ16, InstrItinClass itinQ32,
2837 string OpcodeStr, string Dt,
2838 Intrinsic IntOp, bit Commutable = 0>
2839 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2840 OpcodeStr, Dt, IntOp, Commutable> {
2841 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2842 OpcodeStr, !strconcat(Dt, "8"),
2843 v8i8, v8i8, IntOp, Commutable>;
2844 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2845 OpcodeStr, !strconcat(Dt, "8"),
2846 v16i8, v16i8, IntOp, Commutable>;
2848 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2849 InstrItinClass itinD16, InstrItinClass itinD32,
2850 InstrItinClass itinQ16, InstrItinClass itinQ32,
2851 string OpcodeStr, string Dt,
2853 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2854 OpcodeStr, Dt, IntOp> {
2855 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2856 OpcodeStr, !strconcat(Dt, "8"),
2858 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2859 OpcodeStr, !strconcat(Dt, "8"),
2860 v16i8, v16i8, IntOp>;
2864 // ....then also with element size of 64 bits:
2865 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2866 InstrItinClass itinD16, InstrItinClass itinD32,
2867 InstrItinClass itinQ16, InstrItinClass itinQ32,
2868 string OpcodeStr, string Dt,
2869 Intrinsic IntOp, bit Commutable = 0>
2870 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2871 OpcodeStr, Dt, IntOp, Commutable> {
2872 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2873 OpcodeStr, !strconcat(Dt, "64"),
2874 v1i64, v1i64, IntOp, Commutable>;
2875 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2876 OpcodeStr, !strconcat(Dt, "64"),
2877 v2i64, v2i64, IntOp, Commutable>;
2879 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2880 InstrItinClass itinD16, InstrItinClass itinD32,
2881 InstrItinClass itinQ16, InstrItinClass itinQ32,
2882 string OpcodeStr, string Dt,
2884 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2885 OpcodeStr, Dt, IntOp> {
2886 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2887 OpcodeStr, !strconcat(Dt, "64"),
2888 v1i64, v1i64, IntOp>;
2889 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2890 OpcodeStr, !strconcat(Dt, "64"),
2891 v2i64, v2i64, IntOp>;
2894 // Neon Narrowing 3-register vector intrinsics,
2895 // source operand element sizes of 16, 32 and 64 bits:
2896 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2897 string OpcodeStr, string Dt,
2898 Intrinsic IntOp, bit Commutable = 0> {
2899 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2900 OpcodeStr, !strconcat(Dt, "16"),
2901 v8i8, v8i16, IntOp, Commutable>;
2902 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2903 OpcodeStr, !strconcat(Dt, "32"),
2904 v4i16, v4i32, IntOp, Commutable>;
2905 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2906 OpcodeStr, !strconcat(Dt, "64"),
2907 v2i32, v2i64, IntOp, Commutable>;
2911 // Neon Long 3-register vector operations.
2913 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2914 InstrItinClass itin16, InstrItinClass itin32,
2915 string OpcodeStr, string Dt,
2916 SDNode OpNode, bit Commutable = 0> {
2917 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2918 OpcodeStr, !strconcat(Dt, "8"),
2919 v8i16, v8i8, OpNode, Commutable>;
2920 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2921 OpcodeStr, !strconcat(Dt, "16"),
2922 v4i32, v4i16, OpNode, Commutable>;
2923 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2924 OpcodeStr, !strconcat(Dt, "32"),
2925 v2i64, v2i32, OpNode, Commutable>;
2928 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2929 InstrItinClass itin, string OpcodeStr, string Dt,
2931 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2932 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2933 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2934 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2937 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2938 InstrItinClass itin16, InstrItinClass itin32,
2939 string OpcodeStr, string Dt,
2940 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2941 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2942 OpcodeStr, !strconcat(Dt, "8"),
2943 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2944 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2945 OpcodeStr, !strconcat(Dt, "16"),
2946 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2947 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2948 OpcodeStr, !strconcat(Dt, "32"),
2949 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2952 // Neon Long 3-register vector intrinsics.
2954 // First with only element sizes of 16 and 32 bits:
2955 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2956 InstrItinClass itin16, InstrItinClass itin32,
2957 string OpcodeStr, string Dt,
2958 Intrinsic IntOp, bit Commutable = 0> {
2959 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2960 OpcodeStr, !strconcat(Dt, "16"),
2961 v4i32, v4i16, IntOp, Commutable>;
2962 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2963 OpcodeStr, !strconcat(Dt, "32"),
2964 v2i64, v2i32, IntOp, Commutable>;
2967 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2968 InstrItinClass itin, string OpcodeStr, string Dt,
2970 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2971 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2972 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2973 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2976 // ....then also with element size of 8 bits:
2977 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2978 InstrItinClass itin16, InstrItinClass itin32,
2979 string OpcodeStr, string Dt,
2980 Intrinsic IntOp, bit Commutable = 0>
2981 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
2982 IntOp, Commutable> {
2983 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
2984 OpcodeStr, !strconcat(Dt, "8"),
2985 v8i16, v8i8, IntOp, Commutable>;
2988 // ....with explicit extend (VABDL).
2989 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2990 InstrItinClass itin, string OpcodeStr, string Dt,
2991 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
2992 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
2993 OpcodeStr, !strconcat(Dt, "8"),
2994 v8i16, v8i8, IntOp, ExtOp, Commutable>;
2995 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
2996 OpcodeStr, !strconcat(Dt, "16"),
2997 v4i32, v4i16, IntOp, ExtOp, Commutable>;
2998 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
2999 OpcodeStr, !strconcat(Dt, "32"),
3000 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3004 // Neon Wide 3-register vector intrinsics,
3005 // source operand element sizes of 8, 16 and 32 bits:
3006 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3007 string OpcodeStr, string Dt,
3008 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3009 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3010 OpcodeStr, !strconcat(Dt, "8"),
3011 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3012 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3013 OpcodeStr, !strconcat(Dt, "16"),
3014 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3015 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3016 OpcodeStr, !strconcat(Dt, "32"),
3017 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3021 // Neon Multiply-Op vector operations,
3022 // element sizes of 8, 16 and 32 bits:
3023 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3024 InstrItinClass itinD16, InstrItinClass itinD32,
3025 InstrItinClass itinQ16, InstrItinClass itinQ32,
3026 string OpcodeStr, string Dt, SDNode OpNode> {
3027 // 64-bit vector types.
3028 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3029 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3030 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3031 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3032 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3033 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3035 // 128-bit vector types.
3036 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3037 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3038 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3039 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3040 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3041 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3044 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3045 InstrItinClass itinD16, InstrItinClass itinD32,
3046 InstrItinClass itinQ16, InstrItinClass itinQ32,
3047 string OpcodeStr, string Dt, SDNode ShOp> {
3048 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3049 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3050 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3051 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3052 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3053 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3055 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3056 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3060 // Neon Intrinsic-Op vector operations,
3061 // element sizes of 8, 16 and 32 bits:
3062 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3063 InstrItinClass itinD, InstrItinClass itinQ,
3064 string OpcodeStr, string Dt, Intrinsic IntOp,
3066 // 64-bit vector types.
3067 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3068 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3069 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3070 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3071 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3072 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3074 // 128-bit vector types.
3075 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3076 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3077 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3078 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3079 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3080 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3083 // Neon 3-argument intrinsics,
3084 // element sizes of 8, 16 and 32 bits:
3085 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3086 InstrItinClass itinD, InstrItinClass itinQ,
3087 string OpcodeStr, string Dt, Intrinsic IntOp> {
3088 // 64-bit vector types.
3089 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3090 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3091 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3092 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3093 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3094 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3096 // 128-bit vector types.
3097 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3098 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3099 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3100 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3101 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3102 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3106 // Neon Long Multiply-Op vector operations,
3107 // element sizes of 8, 16 and 32 bits:
3108 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3109 InstrItinClass itin16, InstrItinClass itin32,
3110 string OpcodeStr, string Dt, SDNode MulOp,
3112 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3113 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3114 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3115 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3116 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3117 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3120 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3121 string Dt, SDNode MulOp, SDNode OpNode> {
3122 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3123 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3124 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3125 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3129 // Neon Long 3-argument intrinsics.
3131 // First with only element sizes of 16 and 32 bits:
3132 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3133 InstrItinClass itin16, InstrItinClass itin32,
3134 string OpcodeStr, string Dt, Intrinsic IntOp> {
3135 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3136 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3137 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3138 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3141 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3142 string OpcodeStr, string Dt, Intrinsic IntOp> {
3143 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3144 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3145 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3146 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3149 // ....then also with element size of 8 bits:
3150 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3151 InstrItinClass itin16, InstrItinClass itin32,
3152 string OpcodeStr, string Dt, Intrinsic IntOp>
3153 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3154 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3155 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3158 // ....with explicit extend (VABAL).
3159 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3160 InstrItinClass itin, string OpcodeStr, string Dt,
3161 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3162 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3163 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3164 IntOp, ExtOp, OpNode>;
3165 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3166 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3167 IntOp, ExtOp, OpNode>;
3168 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3169 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3170 IntOp, ExtOp, OpNode>;
3174 // Neon Pairwise long 2-register intrinsics,
3175 // element sizes of 8, 16 and 32 bits:
3176 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3177 bits<5> op11_7, bit op4,
3178 string OpcodeStr, string Dt, Intrinsic IntOp> {
3179 // 64-bit vector types.
3180 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3181 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3182 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3183 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3184 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3185 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3187 // 128-bit vector types.
3188 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3189 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3190 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3191 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3192 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3193 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3197 // Neon Pairwise long 2-register accumulate intrinsics,
3198 // element sizes of 8, 16 and 32 bits:
3199 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3200 bits<5> op11_7, bit op4,
3201 string OpcodeStr, string Dt, Intrinsic IntOp> {
3202 // 64-bit vector types.
3203 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3204 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3205 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3206 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3207 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3208 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3210 // 128-bit vector types.
3211 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3212 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3213 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3214 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3215 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3216 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3220 // Neon 2-register vector shift by immediate,
3221 // with f of either N2RegVShLFrm or N2RegVShRFrm
3222 // element sizes of 8, 16, 32 and 64 bits:
3223 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3224 InstrItinClass itin, string OpcodeStr, string Dt,
3226 // 64-bit vector types.
3227 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3228 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3229 let Inst{21-19} = 0b001; // imm6 = 001xxx
3231 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3232 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3233 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3235 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3236 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3237 let Inst{21} = 0b1; // imm6 = 1xxxxx
3239 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3240 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3243 // 128-bit vector types.
3244 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3245 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3246 let Inst{21-19} = 0b001; // imm6 = 001xxx
3248 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3249 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3250 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3252 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3253 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3254 let Inst{21} = 0b1; // imm6 = 1xxxxx
3256 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3257 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3260 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3261 InstrItinClass itin, string OpcodeStr, string Dt,
3263 // 64-bit vector types.
3264 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3265 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3266 let Inst{21-19} = 0b001; // imm6 = 001xxx
3268 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3269 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3270 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3272 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3273 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3274 let Inst{21} = 0b1; // imm6 = 1xxxxx
3276 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3277 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3280 // 128-bit vector types.
3281 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3282 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3283 let Inst{21-19} = 0b001; // imm6 = 001xxx
3285 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3286 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3287 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3289 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3290 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3291 let Inst{21} = 0b1; // imm6 = 1xxxxx
3293 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3294 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3298 // Neon Shift-Accumulate vector operations,
3299 // element sizes of 8, 16, 32 and 64 bits:
3300 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3301 string OpcodeStr, string Dt, SDNode ShOp> {
3302 // 64-bit vector types.
3303 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3304 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3305 let Inst{21-19} = 0b001; // imm6 = 001xxx
3307 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3308 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3309 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3311 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3312 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3313 let Inst{21} = 0b1; // imm6 = 1xxxxx
3315 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3316 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3319 // 128-bit vector types.
3320 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3321 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3322 let Inst{21-19} = 0b001; // imm6 = 001xxx
3324 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3325 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3326 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3328 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3329 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3330 let Inst{21} = 0b1; // imm6 = 1xxxxx
3332 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3333 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3337 // Neon Shift-Insert vector operations,
3338 // with f of either N2RegVShLFrm or N2RegVShRFrm
3339 // element sizes of 8, 16, 32 and 64 bits:
3340 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3342 // 64-bit vector types.
3343 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3344 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3345 let Inst{21-19} = 0b001; // imm6 = 001xxx
3347 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3348 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3349 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3351 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3352 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3353 let Inst{21} = 0b1; // imm6 = 1xxxxx
3355 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3356 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3359 // 128-bit vector types.
3360 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3361 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3362 let Inst{21-19} = 0b001; // imm6 = 001xxx
3364 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3365 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3366 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3368 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3369 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3370 let Inst{21} = 0b1; // imm6 = 1xxxxx
3372 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3373 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3376 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3378 // 64-bit vector types.
3379 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3380 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3381 let Inst{21-19} = 0b001; // imm6 = 001xxx
3383 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3384 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3385 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3387 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3388 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3389 let Inst{21} = 0b1; // imm6 = 1xxxxx
3391 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3392 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3395 // 128-bit vector types.
3396 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3397 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3398 let Inst{21-19} = 0b001; // imm6 = 001xxx
3400 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3401 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3402 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3404 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3405 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3406 let Inst{21} = 0b1; // imm6 = 1xxxxx
3408 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3409 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3413 // Neon Shift Long operations,
3414 // element sizes of 8, 16, 32 bits:
3415 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3416 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3417 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3418 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3419 let Inst{21-19} = 0b001; // imm6 = 001xxx
3421 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3422 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3423 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3425 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3426 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3427 let Inst{21} = 0b1; // imm6 = 1xxxxx
3431 // Neon Shift Narrow operations,
3432 // element sizes of 16, 32, 64 bits:
3433 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3434 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3436 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3437 OpcodeStr, !strconcat(Dt, "16"),
3438 v8i8, v8i16, shr_imm8, OpNode> {
3439 let Inst{21-19} = 0b001; // imm6 = 001xxx
3441 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3442 OpcodeStr, !strconcat(Dt, "32"),
3443 v4i16, v4i32, shr_imm16, OpNode> {
3444 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3446 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3447 OpcodeStr, !strconcat(Dt, "64"),
3448 v2i32, v2i64, shr_imm32, OpNode> {
3449 let Inst{21} = 0b1; // imm6 = 1xxxxx
3453 //===----------------------------------------------------------------------===//
3454 // Instruction Definitions.
3455 //===----------------------------------------------------------------------===//
3457 // Vector Add Operations.
3459 // VADD : Vector Add (integer and floating-point)
3460 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3462 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3463 v2f32, v2f32, fadd, 1>;
3464 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3465 v4f32, v4f32, fadd, 1>;
3466 // VADDL : Vector Add Long (Q = D + D)
3467 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3468 "vaddl", "s", add, sext, 1>;
3469 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3470 "vaddl", "u", add, zext, 1>;
3471 // VADDW : Vector Add Wide (Q = Q + D)
3472 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3473 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3474 // VHADD : Vector Halving Add
3475 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3476 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3477 "vhadd", "s", int_arm_neon_vhadds, 1>;
3478 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3479 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3480 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3481 // VRHADD : Vector Rounding Halving Add
3482 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3483 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3484 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3485 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3486 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3487 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3488 // VQADD : Vector Saturating Add
3489 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3490 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3491 "vqadd", "s", int_arm_neon_vqadds, 1>;
3492 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3493 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3494 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3495 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3496 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3497 int_arm_neon_vaddhn, 1>;
3498 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3499 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3500 int_arm_neon_vraddhn, 1>;
3502 // Vector Multiply Operations.
3504 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3505 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3506 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3507 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3508 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3509 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3510 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3511 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3512 v2f32, v2f32, fmul, 1>;
3513 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3514 v4f32, v4f32, fmul, 1>;
3515 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3516 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3517 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3520 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3521 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3522 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3523 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3524 (DSubReg_i16_reg imm:$lane))),
3525 (SubReg_i16_lane imm:$lane)))>;
3526 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3527 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3528 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3529 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3530 (DSubReg_i32_reg imm:$lane))),
3531 (SubReg_i32_lane imm:$lane)))>;
3532 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3533 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3534 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3535 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3536 (DSubReg_i32_reg imm:$lane))),
3537 (SubReg_i32_lane imm:$lane)))>;
3539 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3540 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3541 IIC_VMULi16Q, IIC_VMULi32Q,
3542 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3543 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3544 IIC_VMULi16Q, IIC_VMULi32Q,
3545 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3546 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3547 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3549 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3550 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3551 (DSubReg_i16_reg imm:$lane))),
3552 (SubReg_i16_lane imm:$lane)))>;
3553 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3554 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3556 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3557 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3558 (DSubReg_i32_reg imm:$lane))),
3559 (SubReg_i32_lane imm:$lane)))>;
3561 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3562 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3563 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3564 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3565 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3566 IIC_VMULi16Q, IIC_VMULi32Q,
3567 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3568 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3569 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3571 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3572 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3573 (DSubReg_i16_reg imm:$lane))),
3574 (SubReg_i16_lane imm:$lane)))>;
3575 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3576 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3578 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3579 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3580 (DSubReg_i32_reg imm:$lane))),
3581 (SubReg_i32_lane imm:$lane)))>;
3583 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3584 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3585 "vmull", "s", NEONvmulls, 1>;
3586 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3587 "vmull", "u", NEONvmullu, 1>;
3588 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3589 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3590 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3591 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3593 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3594 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3595 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3596 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3597 "vqdmull", "s", int_arm_neon_vqdmull>;
3599 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3601 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3602 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3603 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3604 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3605 v2f32, fmul_su, fadd_mlx>,
3606 Requires<[HasNEON, UseFPVMLx]>;
3607 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3608 v4f32, fmul_su, fadd_mlx>,
3609 Requires<[HasNEON, UseFPVMLx]>;
3610 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3611 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3612 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3613 v2f32, fmul_su, fadd_mlx>,
3614 Requires<[HasNEON, UseFPVMLx]>;
3615 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3616 v4f32, v2f32, fmul_su, fadd_mlx>,
3617 Requires<[HasNEON, UseFPVMLx]>;
3619 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3620 (mul (v8i16 QPR:$src2),
3621 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3622 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3623 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3624 (DSubReg_i16_reg imm:$lane))),
3625 (SubReg_i16_lane imm:$lane)))>;
3627 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3628 (mul (v4i32 QPR:$src2),
3629 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3630 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3631 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3632 (DSubReg_i32_reg imm:$lane))),
3633 (SubReg_i32_lane imm:$lane)))>;
3635 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3636 (fmul_su (v4f32 QPR:$src2),
3637 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3638 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3640 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3641 (DSubReg_i32_reg imm:$lane))),
3642 (SubReg_i32_lane imm:$lane)))>,
3643 Requires<[HasNEON, UseFPVMLx]>;
3645 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3646 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3647 "vmlal", "s", NEONvmulls, add>;
3648 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3649 "vmlal", "u", NEONvmullu, add>;
3651 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3652 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3654 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3655 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3656 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3657 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3659 // VMLS : Vector Multiply Subtract (integer and floating-point)
3660 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3661 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3662 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3663 v2f32, fmul_su, fsub_mlx>,
3664 Requires<[HasNEON, UseFPVMLx]>;
3665 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3666 v4f32, fmul_su, fsub_mlx>,
3667 Requires<[HasNEON, UseFPVMLx]>;
3668 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3669 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3670 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3671 v2f32, fmul_su, fsub_mlx>,
3672 Requires<[HasNEON, UseFPVMLx]>;
3673 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3674 v4f32, v2f32, fmul_su, fsub_mlx>,
3675 Requires<[HasNEON, UseFPVMLx]>;
3677 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3678 (mul (v8i16 QPR:$src2),
3679 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3680 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3681 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3682 (DSubReg_i16_reg imm:$lane))),
3683 (SubReg_i16_lane imm:$lane)))>;
3685 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3686 (mul (v4i32 QPR:$src2),
3687 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3688 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3689 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3690 (DSubReg_i32_reg imm:$lane))),
3691 (SubReg_i32_lane imm:$lane)))>;
3693 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3694 (fmul_su (v4f32 QPR:$src2),
3695 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3696 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3697 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3698 (DSubReg_i32_reg imm:$lane))),
3699 (SubReg_i32_lane imm:$lane)))>,
3700 Requires<[HasNEON, UseFPVMLx]>;
3702 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3703 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3704 "vmlsl", "s", NEONvmulls, sub>;
3705 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3706 "vmlsl", "u", NEONvmullu, sub>;
3708 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3709 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3711 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3712 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3713 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3714 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3716 // Vector Subtract Operations.
3718 // VSUB : Vector Subtract (integer and floating-point)
3719 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3720 "vsub", "i", sub, 0>;
3721 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3722 v2f32, v2f32, fsub, 0>;
3723 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3724 v4f32, v4f32, fsub, 0>;
3725 // VSUBL : Vector Subtract Long (Q = D - D)
3726 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3727 "vsubl", "s", sub, sext, 0>;
3728 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3729 "vsubl", "u", sub, zext, 0>;
3730 // VSUBW : Vector Subtract Wide (Q = Q - D)
3731 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3732 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3733 // VHSUB : Vector Halving Subtract
3734 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3735 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3736 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3737 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3738 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3739 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3740 // VQSUB : Vector Saturing Subtract
3741 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3742 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3743 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3744 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3745 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3746 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3747 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3748 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3749 int_arm_neon_vsubhn, 0>;
3750 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3751 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3752 int_arm_neon_vrsubhn, 0>;
3754 // Vector Comparisons.
3756 // VCEQ : Vector Compare Equal
3757 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3758 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3759 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3761 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3764 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3765 "$Vd, $Vm, #0", NEONvceqz>;
3767 // VCGE : Vector Compare Greater Than or Equal
3768 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3769 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3770 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3771 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3772 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3774 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3777 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3778 "$Vd, $Vm, #0", NEONvcgez>;
3779 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3780 "$Vd, $Vm, #0", NEONvclez>;
3782 // VCGT : Vector Compare Greater Than
3783 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3784 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3785 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3786 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3787 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3789 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3792 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3793 "$Vd, $Vm, #0", NEONvcgtz>;
3794 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3795 "$Vd, $Vm, #0", NEONvcltz>;
3797 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3798 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3799 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3800 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3801 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3802 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3803 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3804 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3805 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3806 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3807 // VTST : Vector Test Bits
3808 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3809 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3811 // Vector Bitwise Operations.
3813 def vnotd : PatFrag<(ops node:$in),
3814 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3815 def vnotq : PatFrag<(ops node:$in),
3816 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3819 // VAND : Vector Bitwise AND
3820 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3821 v2i32, v2i32, and, 1>;
3822 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3823 v4i32, v4i32, and, 1>;
3825 // VEOR : Vector Bitwise Exclusive OR
3826 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3827 v2i32, v2i32, xor, 1>;
3828 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3829 v4i32, v4i32, xor, 1>;
3831 // VORR : Vector Bitwise OR
3832 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3833 v2i32, v2i32, or, 1>;
3834 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3835 v4i32, v4i32, or, 1>;
3837 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3838 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3840 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3842 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3843 let Inst{9} = SIMM{9};
3846 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3847 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3849 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3851 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3852 let Inst{10-9} = SIMM{10-9};
3855 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3856 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3858 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3860 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3861 let Inst{9} = SIMM{9};
3864 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3865 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3867 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3869 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3870 let Inst{10-9} = SIMM{10-9};
3874 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3875 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3876 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3877 "vbic", "$Vd, $Vn, $Vm", "",
3878 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3879 (vnotd DPR:$Vm))))]>;
3880 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3881 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3882 "vbic", "$Vd, $Vn, $Vm", "",
3883 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3884 (vnotq QPR:$Vm))))]>;
3886 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3887 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3889 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3891 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3892 let Inst{9} = SIMM{9};
3895 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3896 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3898 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3900 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3901 let Inst{10-9} = SIMM{10-9};
3904 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3905 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3907 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3909 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3910 let Inst{9} = SIMM{9};
3913 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3914 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3916 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3918 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3919 let Inst{10-9} = SIMM{10-9};
3922 // VORN : Vector Bitwise OR NOT
3923 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3924 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3925 "vorn", "$Vd, $Vn, $Vm", "",
3926 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3927 (vnotd DPR:$Vm))))]>;
3928 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3929 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3930 "vorn", "$Vd, $Vn, $Vm", "",
3931 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3932 (vnotq QPR:$Vm))))]>;
3934 // VMVN : Vector Bitwise NOT (Immediate)
3936 let isReMaterializable = 1 in {
3938 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3939 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3940 "vmvn", "i16", "$Vd, $SIMM", "",
3941 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3942 let Inst{9} = SIMM{9};
3945 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3946 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3947 "vmvn", "i16", "$Vd, $SIMM", "",
3948 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3949 let Inst{9} = SIMM{9};
3952 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3953 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3954 "vmvn", "i32", "$Vd, $SIMM", "",
3955 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3956 let Inst{11-8} = SIMM{11-8};
3959 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3960 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3961 "vmvn", "i32", "$Vd, $SIMM", "",
3962 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3963 let Inst{11-8} = SIMM{11-8};
3967 // VMVN : Vector Bitwise NOT
3968 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3969 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3970 "vmvn", "$Vd, $Vm", "",
3971 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3972 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3973 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3974 "vmvn", "$Vd, $Vm", "",
3975 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3976 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3977 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
3979 // VBSL : Vector Bitwise Select
3980 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3981 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
3982 N3RegFrm, IIC_VCNTiD,
3983 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3985 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
3987 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
3988 (and DPR:$Vm, (vnotd DPR:$Vd)))),
3989 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
3991 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3992 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
3993 N3RegFrm, IIC_VCNTiQ,
3994 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
3996 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
3998 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
3999 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4000 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4002 // VBIF : Vector Bitwise Insert if False
4003 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4004 // FIXME: This instruction's encoding MAY NOT BE correct.
4005 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4006 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4007 N3RegFrm, IIC_VBINiD,
4008 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4010 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4011 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4012 N3RegFrm, IIC_VBINiQ,
4013 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4016 // VBIT : Vector Bitwise Insert if True
4017 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4018 // FIXME: This instruction's encoding MAY NOT BE correct.
4019 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4020 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4021 N3RegFrm, IIC_VBINiD,
4022 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4024 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4025 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4026 N3RegFrm, IIC_VBINiQ,
4027 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4030 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4031 // for equivalent operations with different register constraints; it just
4034 // Vector Absolute Differences.
4036 // VABD : Vector Absolute Difference
4037 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4038 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4039 "vabd", "s", int_arm_neon_vabds, 1>;
4040 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4041 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4042 "vabd", "u", int_arm_neon_vabdu, 1>;
4043 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4044 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4045 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4046 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4048 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4049 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4050 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4051 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4052 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4054 // VABA : Vector Absolute Difference and Accumulate
4055 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4056 "vaba", "s", int_arm_neon_vabds, add>;
4057 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4058 "vaba", "u", int_arm_neon_vabdu, add>;
4060 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4061 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4062 "vabal", "s", int_arm_neon_vabds, zext, add>;
4063 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4064 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4066 // Vector Maximum and Minimum.
4068 // VMAX : Vector Maximum
4069 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4070 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4071 "vmax", "s", int_arm_neon_vmaxs, 1>;
4072 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4073 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4074 "vmax", "u", int_arm_neon_vmaxu, 1>;
4075 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4077 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4078 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4080 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4082 // VMIN : Vector Minimum
4083 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4084 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4085 "vmin", "s", int_arm_neon_vmins, 1>;
4086 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4087 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4088 "vmin", "u", int_arm_neon_vminu, 1>;
4089 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4091 v2f32, v2f32, int_arm_neon_vmins, 1>;
4092 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4094 v4f32, v4f32, int_arm_neon_vmins, 1>;
4096 // Vector Pairwise Operations.
4098 // VPADD : Vector Pairwise Add
4099 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4101 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4102 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4104 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4105 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4107 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4108 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4109 IIC_VPBIND, "vpadd", "f32",
4110 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4112 // VPADDL : Vector Pairwise Add Long
4113 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4114 int_arm_neon_vpaddls>;
4115 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4116 int_arm_neon_vpaddlu>;
4118 // VPADAL : Vector Pairwise Add and Accumulate Long
4119 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4120 int_arm_neon_vpadals>;
4121 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4122 int_arm_neon_vpadalu>;
4124 // VPMAX : Vector Pairwise Maximum
4125 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4126 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4127 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4128 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4129 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4130 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4131 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4132 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4133 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4134 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4135 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4136 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4137 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4138 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4140 // VPMIN : Vector Pairwise Minimum
4141 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4142 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4143 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4144 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4145 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4146 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4147 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4148 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4149 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4150 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4151 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4152 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4153 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4154 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4156 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4158 // VRECPE : Vector Reciprocal Estimate
4159 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4160 IIC_VUNAD, "vrecpe", "u32",
4161 v2i32, v2i32, int_arm_neon_vrecpe>;
4162 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4163 IIC_VUNAQ, "vrecpe", "u32",
4164 v4i32, v4i32, int_arm_neon_vrecpe>;
4165 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4166 IIC_VUNAD, "vrecpe", "f32",
4167 v2f32, v2f32, int_arm_neon_vrecpe>;
4168 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4169 IIC_VUNAQ, "vrecpe", "f32",
4170 v4f32, v4f32, int_arm_neon_vrecpe>;
4172 // VRECPS : Vector Reciprocal Step
4173 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4174 IIC_VRECSD, "vrecps", "f32",
4175 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4176 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4177 IIC_VRECSQ, "vrecps", "f32",
4178 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4180 // VRSQRTE : Vector Reciprocal Square Root Estimate
4181 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4182 IIC_VUNAD, "vrsqrte", "u32",
4183 v2i32, v2i32, int_arm_neon_vrsqrte>;
4184 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4185 IIC_VUNAQ, "vrsqrte", "u32",
4186 v4i32, v4i32, int_arm_neon_vrsqrte>;
4187 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4188 IIC_VUNAD, "vrsqrte", "f32",
4189 v2f32, v2f32, int_arm_neon_vrsqrte>;
4190 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4191 IIC_VUNAQ, "vrsqrte", "f32",
4192 v4f32, v4f32, int_arm_neon_vrsqrte>;
4194 // VRSQRTS : Vector Reciprocal Square Root Step
4195 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4196 IIC_VRECSD, "vrsqrts", "f32",
4197 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4198 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4199 IIC_VRECSQ, "vrsqrts", "f32",
4200 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4204 // VSHL : Vector Shift
4205 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4206 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4207 "vshl", "s", int_arm_neon_vshifts>;
4208 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4209 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4210 "vshl", "u", int_arm_neon_vshiftu>;
4212 // VSHL : Vector Shift Left (Immediate)
4213 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4215 // VSHR : Vector Shift Right (Immediate)
4216 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4217 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4219 // VSHLL : Vector Shift Left Long
4220 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4221 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4223 // VSHLL : Vector Shift Left Long (with maximum shift count)
4224 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4225 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4226 ValueType OpTy, SDNode OpNode>
4227 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4228 ResTy, OpTy, OpNode> {
4229 let Inst{21-16} = op21_16;
4230 let DecoderMethod = "DecodeVSHLMaxInstruction";
4232 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4233 v8i16, v8i8, NEONvshlli>;
4234 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4235 v4i32, v4i16, NEONvshlli>;
4236 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4237 v2i64, v2i32, NEONvshlli>;
4239 // VSHRN : Vector Shift Right and Narrow
4240 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4243 // VRSHL : Vector Rounding Shift
4244 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4245 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4246 "vrshl", "s", int_arm_neon_vrshifts>;
4247 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4248 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4249 "vrshl", "u", int_arm_neon_vrshiftu>;
4250 // VRSHR : Vector Rounding Shift Right
4251 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4252 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4254 // VRSHRN : Vector Rounding Shift Right and Narrow
4255 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4258 // VQSHL : Vector Saturating Shift
4259 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4260 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4261 "vqshl", "s", int_arm_neon_vqshifts>;
4262 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4263 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4264 "vqshl", "u", int_arm_neon_vqshiftu>;
4265 // VQSHL : Vector Saturating Shift Left (Immediate)
4266 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4267 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4269 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4270 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4272 // VQSHRN : Vector Saturating Shift Right and Narrow
4273 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4275 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4278 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4279 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4282 // VQRSHL : Vector Saturating Rounding Shift
4283 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4284 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4285 "vqrshl", "s", int_arm_neon_vqrshifts>;
4286 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4287 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4288 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4290 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4291 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4293 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4296 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4297 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4300 // VSRA : Vector Shift Right and Accumulate
4301 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4302 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4303 // VRSRA : Vector Rounding Shift Right and Accumulate
4304 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4305 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4307 // VSLI : Vector Shift Left and Insert
4308 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4310 // VSRI : Vector Shift Right and Insert
4311 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4313 // Vector Absolute and Saturating Absolute.
4315 // VABS : Vector Absolute Value
4316 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4317 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4319 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4320 IIC_VUNAD, "vabs", "f32",
4321 v2f32, v2f32, int_arm_neon_vabs>;
4322 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4323 IIC_VUNAQ, "vabs", "f32",
4324 v4f32, v4f32, int_arm_neon_vabs>;
4326 // VQABS : Vector Saturating Absolute Value
4327 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4328 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4329 int_arm_neon_vqabs>;
4333 def vnegd : PatFrag<(ops node:$in),
4334 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4335 def vnegq : PatFrag<(ops node:$in),
4336 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4338 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4339 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4340 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4341 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4342 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4343 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4344 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4345 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4347 // VNEG : Vector Negate (integer)
4348 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4349 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4350 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4351 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4352 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4353 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4355 // VNEG : Vector Negate (floating-point)
4356 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4357 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4358 "vneg", "f32", "$Vd, $Vm", "",
4359 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4360 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4361 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4362 "vneg", "f32", "$Vd, $Vm", "",
4363 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4365 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4366 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4367 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4368 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4369 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4370 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4372 // VQNEG : Vector Saturating Negate
4373 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4374 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4375 int_arm_neon_vqneg>;
4377 // Vector Bit Counting Operations.
4379 // VCLS : Vector Count Leading Sign Bits
4380 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4381 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4383 // VCLZ : Vector Count Leading Zeros
4384 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4385 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4387 // VCNT : Vector Count One Bits
4388 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4389 IIC_VCNTiD, "vcnt", "8",
4390 v8i8, v8i8, int_arm_neon_vcnt>;
4391 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4392 IIC_VCNTiQ, "vcnt", "8",
4393 v16i8, v16i8, int_arm_neon_vcnt>;
4396 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4397 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4398 "vswp", "$Vd, $Vm", "", []>;
4399 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4400 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4401 "vswp", "$Vd, $Vm", "", []>;
4403 // Vector Move Operations.
4405 // VMOV : Vector Move (Register)
4406 def : InstAlias<"vmov${p} $Vd, $Vm",
4407 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4408 def : InstAlias<"vmov${p} $Vd, $Vm",
4409 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4411 // VMOV : Vector Move (Immediate)
4413 let isReMaterializable = 1 in {
4414 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4415 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4416 "vmov", "i8", "$Vd, $SIMM", "",
4417 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4418 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4419 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4420 "vmov", "i8", "$Vd, $SIMM", "",
4421 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4423 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4424 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4425 "vmov", "i16", "$Vd, $SIMM", "",
4426 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4427 let Inst{9} = SIMM{9};
4430 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4431 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4432 "vmov", "i16", "$Vd, $SIMM", "",
4433 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4434 let Inst{9} = SIMM{9};
4437 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4438 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4439 "vmov", "i32", "$Vd, $SIMM", "",
4440 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4441 let Inst{11-8} = SIMM{11-8};
4444 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4445 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4446 "vmov", "i32", "$Vd, $SIMM", "",
4447 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4448 let Inst{11-8} = SIMM{11-8};
4451 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4452 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4453 "vmov", "i64", "$Vd, $SIMM", "",
4454 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4455 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4456 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4457 "vmov", "i64", "$Vd, $SIMM", "",
4458 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4459 } // isReMaterializable
4461 // VMOV : Vector Get Lane (move scalar to ARM core register)
4463 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4464 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4465 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4466 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4468 let Inst{21} = lane{2};
4469 let Inst{6-5} = lane{1-0};
4471 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4472 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4473 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4474 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4476 let Inst{21} = lane{1};
4477 let Inst{6} = lane{0};
4479 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4480 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4481 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4482 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4484 let Inst{21} = lane{2};
4485 let Inst{6-5} = lane{1-0};
4487 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4488 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4489 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4490 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4492 let Inst{21} = lane{1};
4493 let Inst{6} = lane{0};
4495 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4496 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4497 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4498 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4500 let Inst{21} = lane{0};
4502 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4503 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4504 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4505 (DSubReg_i8_reg imm:$lane))),
4506 (SubReg_i8_lane imm:$lane))>;
4507 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4508 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4509 (DSubReg_i16_reg imm:$lane))),
4510 (SubReg_i16_lane imm:$lane))>;
4511 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4512 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4513 (DSubReg_i8_reg imm:$lane))),
4514 (SubReg_i8_lane imm:$lane))>;
4515 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4516 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4517 (DSubReg_i16_reg imm:$lane))),
4518 (SubReg_i16_lane imm:$lane))>;
4519 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4520 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4521 (DSubReg_i32_reg imm:$lane))),
4522 (SubReg_i32_lane imm:$lane))>;
4523 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4524 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4525 (SSubReg_f32_reg imm:$src2))>;
4526 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4527 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4528 (SSubReg_f32_reg imm:$src2))>;
4529 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4530 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4531 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4532 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4535 // VMOV : Vector Set Lane (move ARM core register to scalar)
4537 let Constraints = "$src1 = $V" in {
4538 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4539 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4540 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4541 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4542 GPR:$R, imm:$lane))]> {
4543 let Inst{21} = lane{2};
4544 let Inst{6-5} = lane{1-0};
4546 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4547 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4548 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4549 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4550 GPR:$R, imm:$lane))]> {
4551 let Inst{21} = lane{1};
4552 let Inst{6} = lane{0};
4554 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4555 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4556 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4557 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4558 GPR:$R, imm:$lane))]> {
4559 let Inst{21} = lane{0};
4562 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4563 (v16i8 (INSERT_SUBREG QPR:$src1,
4564 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4565 (DSubReg_i8_reg imm:$lane))),
4566 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4567 (DSubReg_i8_reg imm:$lane)))>;
4568 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4569 (v8i16 (INSERT_SUBREG QPR:$src1,
4570 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4571 (DSubReg_i16_reg imm:$lane))),
4572 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4573 (DSubReg_i16_reg imm:$lane)))>;
4574 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4575 (v4i32 (INSERT_SUBREG QPR:$src1,
4576 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4577 (DSubReg_i32_reg imm:$lane))),
4578 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4579 (DSubReg_i32_reg imm:$lane)))>;
4581 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4582 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4583 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4584 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4585 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4586 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4588 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4589 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4590 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4591 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4593 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4594 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4595 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4596 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4597 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4598 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4600 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4601 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4602 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4603 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4604 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4605 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4607 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4608 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4609 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4611 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4612 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4613 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4615 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4616 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4617 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4620 // VDUP : Vector Duplicate (from ARM core register to all elements)
4622 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4623 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4624 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4625 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4626 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4627 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4628 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4629 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4631 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4632 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4633 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4634 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4635 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4636 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4638 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4639 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4641 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4643 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4644 ValueType Ty, Operand IdxTy>
4645 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4646 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4647 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4649 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4650 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4651 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4652 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4653 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4654 VectorIndex32:$lane)))]>;
4656 // Inst{19-16} is partially specified depending on the element size.
4658 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4660 let Inst{19-17} = lane{2-0};
4662 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4664 let Inst{19-18} = lane{1-0};
4666 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4668 let Inst{19} = lane{0};
4670 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4672 let Inst{19-17} = lane{2-0};
4674 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4676 let Inst{19-18} = lane{1-0};
4678 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4680 let Inst{19} = lane{0};
4683 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4684 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4686 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4687 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4689 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4690 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4691 (DSubReg_i8_reg imm:$lane))),
4692 (SubReg_i8_lane imm:$lane)))>;
4693 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4694 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4695 (DSubReg_i16_reg imm:$lane))),
4696 (SubReg_i16_lane imm:$lane)))>;
4697 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4698 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4699 (DSubReg_i32_reg imm:$lane))),
4700 (SubReg_i32_lane imm:$lane)))>;
4701 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4702 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4703 (DSubReg_i32_reg imm:$lane))),
4704 (SubReg_i32_lane imm:$lane)))>;
4706 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4707 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4708 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4709 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4711 // VMOVN : Vector Narrowing Move
4712 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4713 "vmovn", "i", trunc>;
4714 // VQMOVN : Vector Saturating Narrowing Move
4715 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4716 "vqmovn", "s", int_arm_neon_vqmovns>;
4717 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4718 "vqmovn", "u", int_arm_neon_vqmovnu>;
4719 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4720 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4721 // VMOVL : Vector Lengthening Move
4722 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4723 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4725 // Vector Conversions.
4727 // VCVT : Vector Convert Between Floating-Point and Integers
4728 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4729 v2i32, v2f32, fp_to_sint>;
4730 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4731 v2i32, v2f32, fp_to_uint>;
4732 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4733 v2f32, v2i32, sint_to_fp>;
4734 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4735 v2f32, v2i32, uint_to_fp>;
4737 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4738 v4i32, v4f32, fp_to_sint>;
4739 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4740 v4i32, v4f32, fp_to_uint>;
4741 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4742 v4f32, v4i32, sint_to_fp>;
4743 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4744 v4f32, v4i32, uint_to_fp>;
4746 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4747 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4748 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4749 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4750 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4751 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4752 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4753 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4754 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4756 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4757 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4758 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4759 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4760 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4761 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4762 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4763 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4765 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4766 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4767 IIC_VUNAQ, "vcvt", "f16.f32",
4768 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4769 Requires<[HasNEON, HasFP16]>;
4770 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4771 IIC_VUNAQ, "vcvt", "f32.f16",
4772 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4773 Requires<[HasNEON, HasFP16]>;
4777 // VREV64 : Vector Reverse elements within 64-bit doublewords
4779 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4780 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4781 (ins DPR:$Vm), IIC_VMOVD,
4782 OpcodeStr, Dt, "$Vd, $Vm", "",
4783 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4784 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4785 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4786 (ins QPR:$Vm), IIC_VMOVQ,
4787 OpcodeStr, Dt, "$Vd, $Vm", "",
4788 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4790 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4791 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4792 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4793 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4795 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4796 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4797 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4798 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4800 // VREV32 : Vector Reverse elements within 32-bit words
4802 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4803 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4804 (ins DPR:$Vm), IIC_VMOVD,
4805 OpcodeStr, Dt, "$Vd, $Vm", "",
4806 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4807 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4808 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4809 (ins QPR:$Vm), IIC_VMOVQ,
4810 OpcodeStr, Dt, "$Vd, $Vm", "",
4811 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4813 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4814 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4816 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4817 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4819 // VREV16 : Vector Reverse elements within 16-bit halfwords
4821 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4822 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4823 (ins DPR:$Vm), IIC_VMOVD,
4824 OpcodeStr, Dt, "$Vd, $Vm", "",
4825 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4826 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4827 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4828 (ins QPR:$Vm), IIC_VMOVQ,
4829 OpcodeStr, Dt, "$Vd, $Vm", "",
4830 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4832 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4833 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4835 // Other Vector Shuffles.
4837 // Aligned extractions: really just dropping registers
4839 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4840 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4841 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4843 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4845 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4847 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4849 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4851 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4854 // VEXT : Vector Extract
4856 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4857 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4858 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4859 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4860 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4861 (Ty DPR:$Vm), imm:$index)))]> {
4863 let Inst{11-8} = index{3-0};
4866 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4867 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4868 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4869 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4870 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4871 (Ty QPR:$Vm), imm:$index)))]> {
4873 let Inst{11-8} = index{3-0};
4876 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4877 let Inst{11-8} = index{3-0};
4879 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4880 let Inst{11-9} = index{2-0};
4883 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4884 let Inst{11-10} = index{1-0};
4885 let Inst{9-8} = 0b00;
4887 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4890 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
4892 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4893 let Inst{11-8} = index{3-0};
4895 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4896 let Inst{11-9} = index{2-0};
4899 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4900 let Inst{11-10} = index{1-0};
4901 let Inst{9-8} = 0b00;
4903 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
4906 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
4908 // VTRN : Vector Transpose
4910 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4911 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4912 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4914 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4915 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4916 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4918 // VUZP : Vector Unzip (Deinterleave)
4920 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4921 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4922 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4924 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4925 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4926 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4928 // VZIP : Vector Zip (Interleave)
4930 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4931 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4932 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4934 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4935 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4936 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4938 // Vector Table Lookup and Table Extension.
4940 // VTBL : Vector Table Lookup
4941 let DecoderMethod = "DecodeTBLInstruction" in {
4943 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4944 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4945 "vtbl", "8", "$Vd, $Vn, $Vm", "",
4946 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
4947 let hasExtraSrcRegAllocReq = 1 in {
4949 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4950 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4951 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4953 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4954 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4955 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4957 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4958 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4960 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4961 } // hasExtraSrcRegAllocReq = 1
4964 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4966 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4968 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4970 // VTBX : Vector Table Extension
4972 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4973 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4974 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
4975 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4976 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
4977 let hasExtraSrcRegAllocReq = 1 in {
4979 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
4980 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
4981 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
4983 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
4984 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
4985 NVTBLFrm, IIC_VTBX3,
4986 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
4989 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
4990 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
4991 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
4993 } // hasExtraSrcRegAllocReq = 1
4996 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
4997 IIC_VTBX2, "$orig = $dst", []>;
4999 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5000 IIC_VTBX3, "$orig = $dst", []>;
5002 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5003 IIC_VTBX4, "$orig = $dst", []>;
5004 } // DecoderMethod = "DecodeTBLInstruction"
5006 //===----------------------------------------------------------------------===//
5007 // NEON instructions for single-precision FP math
5008 //===----------------------------------------------------------------------===//
5010 class N2VSPat<SDNode OpNode, NeonI Inst>
5011 : NEONFPPat<(f32 (OpNode SPR:$a)),
5013 (v2f32 (COPY_TO_REGCLASS (Inst
5015 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5016 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5018 class N3VSPat<SDNode OpNode, NeonI Inst>
5019 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5021 (v2f32 (COPY_TO_REGCLASS (Inst
5023 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5026 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5027 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5029 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5030 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5032 (v2f32 (COPY_TO_REGCLASS (Inst
5034 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5037 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5040 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5041 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5043 def : N3VSPat<fadd, VADDfd>;
5044 def : N3VSPat<fsub, VSUBfd>;
5045 def : N3VSPat<fmul, VMULfd>;
5046 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5047 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5048 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5049 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5050 def : N2VSPat<fabs, VABSfd>;
5051 def : N2VSPat<fneg, VNEGfd>;
5052 def : N3VSPat<NEONfmax, VMAXfd>;
5053 def : N3VSPat<NEONfmin, VMINfd>;
5054 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5055 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5056 def : N2VSPat<arm_sitof, VCVTs2fd>;
5057 def : N2VSPat<arm_uitof, VCVTu2fd>;
5059 //===----------------------------------------------------------------------===//
5060 // Non-Instruction Patterns
5061 //===----------------------------------------------------------------------===//
5064 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5065 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5066 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5067 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5068 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5069 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5070 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5071 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5072 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5073 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5074 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5075 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5076 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5077 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5078 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5079 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5080 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5081 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5082 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5083 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5084 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5085 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5086 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5087 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5088 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5089 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5090 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5091 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5092 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5093 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5095 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5096 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5097 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5098 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5099 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5100 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5101 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5102 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5103 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5104 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5105 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5106 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5107 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5108 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5109 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5110 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5111 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5112 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5113 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5114 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5115 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5116 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5117 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5118 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5119 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5120 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5121 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5122 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5123 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5124 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;