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, $Rn", "", []> {
384 let DecoderMethod = "DecodeVLDInstruction";
386 multiclass VLD1D3WB<bits<4> op7_4, string Dt> {
387 def _fixed : NLdSt<0,0b10,0b0110, op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
388 (ins addrmode6:$Rn), IIC_VLD1x2u,
389 "vld1", Dt, "$Vd, $Rn!",
390 "$Rn.addr = $wb", []> {
391 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
392 let Inst{5-4} = Rn{5-4};
393 let DecoderMethod = "DecodeVLDInstruction";
394 let AsmMatchConverter = "cvtVLDwbFixed";
396 def _register : NLdSt<0,0b10,0b0110,op7_4, (outs VecListThreeD:$Vd, GPR:$wb),
397 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
398 "vld1", Dt, "$Vd, $Rn, $Rm",
399 "$Rn.addr = $wb", []> {
400 let Inst{5-4} = Rn{5-4};
401 let DecoderMethod = "DecodeVLDInstruction";
402 let AsmMatchConverter = "cvtVLDwbRegister";
406 def VLD1d8T : VLD1D3<{0,0,0,?}, "8">;
407 def VLD1d16T : VLD1D3<{0,1,0,?}, "16">;
408 def VLD1d32T : VLD1D3<{1,0,0,?}, "32">;
409 def VLD1d64T : VLD1D3<{1,1,0,?}, "64">;
411 defm VLD1d8Twb : VLD1D3WB<{0,0,0,?}, "8">;
412 defm VLD1d16Twb : VLD1D3WB<{0,1,0,?}, "16">;
413 defm VLD1d32Twb : VLD1D3WB<{1,0,0,?}, "32">;
414 defm VLD1d64Twb : VLD1D3WB<{1,1,0,?}, "64">;
416 def VLD1d64TPseudo : VLDQQPseudo<IIC_VLD1x3>;
418 // ...with 4 registers
419 class VLD1D4<bits<4> op7_4, string Dt>
420 : NLdSt<0, 0b10, 0b0010, op7_4, (outs VecListFourD:$Vd),
421 (ins addrmode6:$Rn), IIC_VLD1x4, "vld1", Dt,
422 "$Vd, $Rn", "", []> {
424 let Inst{5-4} = Rn{5-4};
425 let DecoderMethod = "DecodeVLDInstruction";
427 multiclass VLD1D4WB<bits<4> op7_4, string Dt> {
428 def _fixed : NLdSt<0,0b10,0b0010, op7_4, (outs VecListFourD:$Vd, GPR:$wb),
429 (ins addrmode6:$Rn), IIC_VLD1x2u,
430 "vld1", Dt, "$Vd, $Rn!",
431 "$Rn.addr = $wb", []> {
432 let Rm = 0b1101; // NLdSt will assign to the right encoding bits.
433 let Inst{5-4} = Rn{5-4};
434 let DecoderMethod = "DecodeVLDInstruction";
435 let AsmMatchConverter = "cvtVLDwbFixed";
437 def _register : NLdSt<0,0b10,0b0010,op7_4, (outs VecListFourD:$Vd, GPR:$wb),
438 (ins addrmode6:$Rn, rGPR:$Rm), IIC_VLD1x2u,
439 "vld1", Dt, "$Vd, $Rn, $Rm",
440 "$Rn.addr = $wb", []> {
441 let Inst{5-4} = Rn{5-4};
442 let DecoderMethod = "DecodeVLDInstruction";
443 let AsmMatchConverter = "cvtVLDwbRegister";
447 def VLD1d8Q : VLD1D4<{0,0,?,?}, "8">;
448 def VLD1d16Q : VLD1D4<{0,1,?,?}, "16">;
449 def VLD1d32Q : VLD1D4<{1,0,?,?}, "32">;
450 def VLD1d64Q : VLD1D4<{1,1,?,?}, "64">;
452 defm VLD1d8Qwb : VLD1D4WB<{0,0,?,?}, "8">;
453 defm VLD1d16Qwb : VLD1D4WB<{0,1,?,?}, "16">;
454 defm VLD1d32Qwb : VLD1D4WB<{1,0,?,?}, "32">;
455 defm VLD1d64Qwb : VLD1D4WB<{1,1,?,?}, "64">;
457 def VLD1d64QPseudo : VLDQQPseudo<IIC_VLD1x4>;
459 // VLD2 : Vector Load (multiple 2-element structures)
460 class VLD2D<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
461 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd),
462 (ins addrmode6:$Rn), IIC_VLD2,
463 "vld2", Dt, "$Vd, $Rn", "", []> {
465 let Inst{5-4} = Rn{5-4};
466 let DecoderMethod = "DecodeVLDInstruction";
468 class VLD2Q<bits<4> op7_4, string Dt, RegisterOperand VdTy>
469 : NLdSt<0, 0b10, 0b0011, op7_4,
471 (ins addrmode6:$Rn), IIC_VLD2x2,
472 "vld2", Dt, "$Vd, $Rn", "", []> {
474 let Inst{5-4} = Rn{5-4};
475 let DecoderMethod = "DecodeVLDInstruction";
478 def VLD2d8 : VLD2D<0b1000, {0,0,?,?}, "8", VecListTwoD>;
479 def VLD2d16 : VLD2D<0b1000, {0,1,?,?}, "16", VecListTwoD>;
480 def VLD2d32 : VLD2D<0b1000, {1,0,?,?}, "32", VecListTwoD>;
482 def VLD2q8 : VLD2Q<{0,0,?,?}, "8", VecListFourD>;
483 def VLD2q16 : VLD2Q<{0,1,?,?}, "16", VecListFourD>;
484 def VLD2q32 : VLD2Q<{1,0,?,?}, "32", VecListFourD>;
486 def VLD2d8Pseudo : VLDQPseudo<IIC_VLD2>;
487 def VLD2d16Pseudo : VLDQPseudo<IIC_VLD2>;
488 def VLD2d32Pseudo : VLDQPseudo<IIC_VLD2>;
490 def VLD2q8Pseudo : VLDQQPseudo<IIC_VLD2x2>;
491 def VLD2q16Pseudo : VLDQQPseudo<IIC_VLD2x2>;
492 def VLD2q32Pseudo : VLDQQPseudo<IIC_VLD2x2>;
494 // ...with address register writeback:
495 class VLD2DWB<bits<4> op11_8, bits<4> op7_4, string Dt, RegisterOperand VdTy>
496 : NLdSt<0, 0b10, op11_8, op7_4, (outs VdTy:$Vd, GPR:$wb),
497 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2u,
498 "vld2", Dt, "$Vd, $Rn$Rm",
499 "$Rn.addr = $wb", []> {
500 let Inst{5-4} = Rn{5-4};
501 let DecoderMethod = "DecodeVLDInstruction";
503 class VLD2QWB<bits<4> op7_4, string Dt, RegisterOperand VdTy>
504 : NLdSt<0, 0b10, 0b0011, op7_4,
505 (outs VdTy:$Vd, GPR:$wb),
506 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD2x2u,
507 "vld2", Dt, "$Vd, $Rn$Rm",
508 "$Rn.addr = $wb", []> {
509 let Inst{5-4} = Rn{5-4};
510 let DecoderMethod = "DecodeVLDInstruction";
513 def VLD2d8_UPD : VLD2DWB<0b1000, {0,0,?,?}, "8", VecListTwoD>;
514 def VLD2d16_UPD : VLD2DWB<0b1000, {0,1,?,?}, "16", VecListTwoD>;
515 def VLD2d32_UPD : VLD2DWB<0b1000, {1,0,?,?}, "32", VecListTwoD>;
517 def VLD2q8_UPD : VLD2QWB<{0,0,?,?}, "8", VecListFourD>;
518 def VLD2q16_UPD : VLD2QWB<{0,1,?,?}, "16", VecListFourD>;
519 def VLD2q32_UPD : VLD2QWB<{1,0,?,?}, "32", VecListFourD>;
521 def VLD2d8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
522 def VLD2d16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
523 def VLD2d32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2u>;
525 def VLD2q8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
526 def VLD2q16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
527 def VLD2q32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD2x2u>;
529 // ...with double-spaced registers
530 def VLD2b8 : VLD2D<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
531 def VLD2b16 : VLD2D<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
532 def VLD2b32 : VLD2D<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
533 def VLD2b8_UPD : VLD2DWB<0b1001, {0,0,?,?}, "8", VecListTwoQ>;
534 def VLD2b16_UPD : VLD2DWB<0b1001, {0,1,?,?}, "16", VecListTwoQ>;
535 def VLD2b32_UPD : VLD2DWB<0b1001, {1,0,?,?}, "32", VecListTwoQ>;
537 // VLD3 : Vector Load (multiple 3-element structures)
538 class VLD3D<bits<4> op11_8, bits<4> op7_4, string Dt>
539 : NLdSt<0, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
540 (ins addrmode6:$Rn), IIC_VLD3,
541 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn", "", []> {
544 let DecoderMethod = "DecodeVLDInstruction";
547 def VLD3d8 : VLD3D<0b0100, {0,0,0,?}, "8">;
548 def VLD3d16 : VLD3D<0b0100, {0,1,0,?}, "16">;
549 def VLD3d32 : VLD3D<0b0100, {1,0,0,?}, "32">;
551 def VLD3d8Pseudo : VLDQQPseudo<IIC_VLD3>;
552 def VLD3d16Pseudo : VLDQQPseudo<IIC_VLD3>;
553 def VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
555 // ...with address register writeback:
556 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
557 : NLdSt<0, 0b10, op11_8, op7_4,
558 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
559 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD3u,
560 "vld3", Dt, "\\{$Vd, $dst2, $dst3\\}, $Rn$Rm",
561 "$Rn.addr = $wb", []> {
563 let DecoderMethod = "DecodeVLDInstruction";
566 def VLD3d8_UPD : VLD3DWB<0b0100, {0,0,0,?}, "8">;
567 def VLD3d16_UPD : VLD3DWB<0b0100, {0,1,0,?}, "16">;
568 def VLD3d32_UPD : VLD3DWB<0b0100, {1,0,0,?}, "32">;
570 def VLD3d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
571 def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
572 def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
574 // ...with double-spaced registers:
575 def VLD3q8 : VLD3D<0b0101, {0,0,0,?}, "8">;
576 def VLD3q16 : VLD3D<0b0101, {0,1,0,?}, "16">;
577 def VLD3q32 : VLD3D<0b0101, {1,0,0,?}, "32">;
578 def VLD3q8_UPD : VLD3DWB<0b0101, {0,0,0,?}, "8">;
579 def VLD3q16_UPD : VLD3DWB<0b0101, {0,1,0,?}, "16">;
580 def VLD3q32_UPD : VLD3DWB<0b0101, {1,0,0,?}, "32">;
582 def VLD3q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
583 def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
584 def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
586 // ...alternate versions to be allocated odd register numbers:
587 def VLD3q8oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
588 def VLD3q16oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
589 def VLD3q32oddPseudo : VLDQQQQPseudo<IIC_VLD3>;
591 def VLD3q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
592 def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
593 def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
595 // VLD4 : Vector Load (multiple 4-element structures)
596 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
597 : NLdSt<0, 0b10, op11_8, op7_4,
598 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
599 (ins addrmode6:$Rn), IIC_VLD4,
600 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn", "", []> {
602 let Inst{5-4} = Rn{5-4};
603 let DecoderMethod = "DecodeVLDInstruction";
606 def VLD4d8 : VLD4D<0b0000, {0,0,?,?}, "8">;
607 def VLD4d16 : VLD4D<0b0000, {0,1,?,?}, "16">;
608 def VLD4d32 : VLD4D<0b0000, {1,0,?,?}, "32">;
610 def VLD4d8Pseudo : VLDQQPseudo<IIC_VLD4>;
611 def VLD4d16Pseudo : VLDQQPseudo<IIC_VLD4>;
612 def VLD4d32Pseudo : VLDQQPseudo<IIC_VLD4>;
614 // ...with address register writeback:
615 class VLD4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
616 : NLdSt<0, 0b10, op11_8, op7_4,
617 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
618 (ins addrmode6:$Rn, am6offset:$Rm), IIC_VLD4u,
619 "vld4", Dt, "\\{$Vd, $dst2, $dst3, $dst4\\}, $Rn$Rm",
620 "$Rn.addr = $wb", []> {
621 let Inst{5-4} = Rn{5-4};
622 let DecoderMethod = "DecodeVLDInstruction";
625 def VLD4d8_UPD : VLD4DWB<0b0000, {0,0,?,?}, "8">;
626 def VLD4d16_UPD : VLD4DWB<0b0000, {0,1,?,?}, "16">;
627 def VLD4d32_UPD : VLD4DWB<0b0000, {1,0,?,?}, "32">;
629 def VLD4d8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
630 def VLD4d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
631 def VLD4d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4u>;
633 // ...with double-spaced registers:
634 def VLD4q8 : VLD4D<0b0001, {0,0,?,?}, "8">;
635 def VLD4q16 : VLD4D<0b0001, {0,1,?,?}, "16">;
636 def VLD4q32 : VLD4D<0b0001, {1,0,?,?}, "32">;
637 def VLD4q8_UPD : VLD4DWB<0b0001, {0,0,?,?}, "8">;
638 def VLD4q16_UPD : VLD4DWB<0b0001, {0,1,?,?}, "16">;
639 def VLD4q32_UPD : VLD4DWB<0b0001, {1,0,?,?}, "32">;
641 def VLD4q8Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
642 def VLD4q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
643 def VLD4q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
645 // ...alternate versions to be allocated odd register numbers:
646 def VLD4q8oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
647 def VLD4q16oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
648 def VLD4q32oddPseudo : VLDQQQQPseudo<IIC_VLD4>;
650 def VLD4q8oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
651 def VLD4q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
652 def VLD4q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD4u>;
654 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
656 // Classes for VLD*LN pseudo-instructions with multi-register operands.
657 // These are expanded to real instructions after register allocation.
658 class VLDQLNPseudo<InstrItinClass itin>
659 : PseudoNLdSt<(outs QPR:$dst),
660 (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
661 itin, "$src = $dst">;
662 class VLDQLNWBPseudo<InstrItinClass itin>
663 : PseudoNLdSt<(outs QPR:$dst, GPR:$wb),
664 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
665 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
666 class VLDQQLNPseudo<InstrItinClass itin>
667 : PseudoNLdSt<(outs QQPR:$dst),
668 (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
669 itin, "$src = $dst">;
670 class VLDQQLNWBPseudo<InstrItinClass itin>
671 : PseudoNLdSt<(outs QQPR:$dst, GPR:$wb),
672 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
673 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
674 class VLDQQQQLNPseudo<InstrItinClass itin>
675 : PseudoNLdSt<(outs QQQQPR:$dst),
676 (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
677 itin, "$src = $dst">;
678 class VLDQQQQLNWBPseudo<InstrItinClass itin>
679 : PseudoNLdSt<(outs QQQQPR:$dst, GPR:$wb),
680 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
681 nohash_imm:$lane), itin, "$addr.addr = $wb, $src = $dst">;
683 // VLD1LN : Vector Load (single element to one lane)
684 class VLD1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
686 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
687 (ins addrmode6:$Rn, DPR:$src, nohash_imm:$lane),
688 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
690 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
691 (i32 (LoadOp addrmode6:$Rn)),
694 let DecoderMethod = "DecodeVLD1LN";
696 class VLD1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
698 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd),
699 (ins addrmode6oneL32:$Rn, DPR:$src, nohash_imm:$lane),
700 IIC_VLD1ln, "vld1", Dt, "\\{$Vd[$lane]\\}, $Rn",
702 [(set DPR:$Vd, (vector_insert (Ty DPR:$src),
703 (i32 (LoadOp addrmode6oneL32:$Rn)),
706 let DecoderMethod = "DecodeVLD1LN";
708 class VLD1QLNPseudo<ValueType Ty, PatFrag LoadOp> : VLDQLNPseudo<IIC_VLD1ln> {
709 let Pattern = [(set QPR:$dst, (vector_insert (Ty QPR:$src),
710 (i32 (LoadOp addrmode6:$addr)),
714 def VLD1LNd8 : VLD1LN<0b0000, {?,?,?,0}, "8", v8i8, extloadi8> {
715 let Inst{7-5} = lane{2-0};
717 def VLD1LNd16 : VLD1LN<0b0100, {?,?,0,?}, "16", v4i16, extloadi16> {
718 let Inst{7-6} = lane{1-0};
721 def VLD1LNd32 : VLD1LN32<0b1000, {?,0,?,?}, "32", v2i32, load> {
722 let Inst{7} = lane{0};
727 def VLD1LNq8Pseudo : VLD1QLNPseudo<v16i8, extloadi8>;
728 def VLD1LNq16Pseudo : VLD1QLNPseudo<v8i16, extloadi16>;
729 def VLD1LNq32Pseudo : VLD1QLNPseudo<v4i32, load>;
731 def : Pat<(vector_insert (v2f32 DPR:$src),
732 (f32 (load addrmode6:$addr)), imm:$lane),
733 (VLD1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
734 def : Pat<(vector_insert (v4f32 QPR:$src),
735 (f32 (load addrmode6:$addr)), imm:$lane),
736 (VLD1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
738 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
740 // ...with address register writeback:
741 class VLD1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
742 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, GPR:$wb),
743 (ins addrmode6:$Rn, am6offset:$Rm,
744 DPR:$src, nohash_imm:$lane), IIC_VLD1lnu, "vld1", Dt,
745 "\\{$Vd[$lane]\\}, $Rn$Rm",
746 "$src = $Vd, $Rn.addr = $wb", []> {
747 let DecoderMethod = "DecodeVLD1LN";
750 def VLD1LNd8_UPD : VLD1LNWB<0b0000, {?,?,?,0}, "8"> {
751 let Inst{7-5} = lane{2-0};
753 def VLD1LNd16_UPD : VLD1LNWB<0b0100, {?,?,0,?}, "16"> {
754 let Inst{7-6} = lane{1-0};
757 def VLD1LNd32_UPD : VLD1LNWB<0b1000, {?,0,?,?}, "32"> {
758 let Inst{7} = lane{0};
763 def VLD1LNq8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
764 def VLD1LNq16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
765 def VLD1LNq32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD1lnu>;
767 // VLD2LN : Vector Load (single 2-element structure to one lane)
768 class VLD2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
769 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2),
770 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, nohash_imm:$lane),
771 IIC_VLD2ln, "vld2", Dt, "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn",
772 "$src1 = $Vd, $src2 = $dst2", []> {
775 let DecoderMethod = "DecodeVLD2LN";
778 def VLD2LNd8 : VLD2LN<0b0001, {?,?,?,?}, "8"> {
779 let Inst{7-5} = lane{2-0};
781 def VLD2LNd16 : VLD2LN<0b0101, {?,?,0,?}, "16"> {
782 let Inst{7-6} = lane{1-0};
784 def VLD2LNd32 : VLD2LN<0b1001, {?,0,0,?}, "32"> {
785 let Inst{7} = lane{0};
788 def VLD2LNd8Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
789 def VLD2LNd16Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
790 def VLD2LNd32Pseudo : VLDQLNPseudo<IIC_VLD2ln>;
792 // ...with double-spaced registers:
793 def VLD2LNq16 : VLD2LN<0b0101, {?,?,1,?}, "16"> {
794 let Inst{7-6} = lane{1-0};
796 def VLD2LNq32 : VLD2LN<0b1001, {?,1,0,?}, "32"> {
797 let Inst{7} = lane{0};
800 def VLD2LNq16Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
801 def VLD2LNq32Pseudo : VLDQQLNPseudo<IIC_VLD2ln>;
803 // ...with address register writeback:
804 class VLD2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
805 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
806 (ins addrmode6:$Rn, am6offset:$Rm,
807 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VLD2lnu, "vld2", Dt,
808 "\\{$Vd[$lane], $dst2[$lane]\\}, $Rn$Rm",
809 "$src1 = $Vd, $src2 = $dst2, $Rn.addr = $wb", []> {
811 let DecoderMethod = "DecodeVLD2LN";
814 def VLD2LNd8_UPD : VLD2LNWB<0b0001, {?,?,?,?}, "8"> {
815 let Inst{7-5} = lane{2-0};
817 def VLD2LNd16_UPD : VLD2LNWB<0b0101, {?,?,0,?}, "16"> {
818 let Inst{7-6} = lane{1-0};
820 def VLD2LNd32_UPD : VLD2LNWB<0b1001, {?,0,0,?}, "32"> {
821 let Inst{7} = lane{0};
824 def VLD2LNd8Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
825 def VLD2LNd16Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
826 def VLD2LNd32Pseudo_UPD : VLDQLNWBPseudo<IIC_VLD2lnu>;
828 def VLD2LNq16_UPD : VLD2LNWB<0b0101, {?,?,1,?}, "16"> {
829 let Inst{7-6} = lane{1-0};
831 def VLD2LNq32_UPD : VLD2LNWB<0b1001, {?,1,0,?}, "32"> {
832 let Inst{7} = lane{0};
835 def VLD2LNq16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
836 def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
838 // VLD3LN : Vector Load (single 3-element structure to one lane)
839 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
840 : NLdStLn<1, 0b10, op11_8, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
841 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3,
842 nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
843 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn",
844 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3", []> {
846 let DecoderMethod = "DecodeVLD3LN";
849 def VLD3LNd8 : VLD3LN<0b0010, {?,?,?,0}, "8"> {
850 let Inst{7-5} = lane{2-0};
852 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16"> {
853 let Inst{7-6} = lane{1-0};
855 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32"> {
856 let Inst{7} = lane{0};
859 def VLD3LNd8Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
860 def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
861 def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
863 // ...with double-spaced registers:
864 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16"> {
865 let Inst{7-6} = lane{1-0};
867 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32"> {
868 let Inst{7} = lane{0};
871 def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
872 def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
874 // ...with address register writeback:
875 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
876 : NLdStLn<1, 0b10, op11_8, op7_4,
877 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
878 (ins addrmode6:$Rn, am6offset:$Rm,
879 DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
880 IIC_VLD3lnu, "vld3", Dt,
881 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane]\\}, $Rn$Rm",
882 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $Rn.addr = $wb",
884 let DecoderMethod = "DecodeVLD3LN";
887 def VLD3LNd8_UPD : VLD3LNWB<0b0010, {?,?,?,0}, "8"> {
888 let Inst{7-5} = lane{2-0};
890 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16"> {
891 let Inst{7-6} = lane{1-0};
893 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32"> {
894 let Inst{7} = lane{0};
897 def VLD3LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
898 def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
899 def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
901 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16"> {
902 let Inst{7-6} = lane{1-0};
904 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32"> {
905 let Inst{7} = lane{0};
908 def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
909 def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
911 // VLD4LN : Vector Load (single 4-element structure to one lane)
912 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
913 : NLdStLn<1, 0b10, op11_8, op7_4,
914 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
915 (ins addrmode6:$Rn, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4,
916 nohash_imm:$lane), IIC_VLD4ln, "vld4", Dt,
917 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn",
918 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4", []> {
921 let DecoderMethod = "DecodeVLD4LN";
924 def VLD4LNd8 : VLD4LN<0b0011, {?,?,?,?}, "8"> {
925 let Inst{7-5} = lane{2-0};
927 def VLD4LNd16 : VLD4LN<0b0111, {?,?,0,?}, "16"> {
928 let Inst{7-6} = lane{1-0};
930 def VLD4LNd32 : VLD4LN<0b1011, {?,0,?,?}, "32"> {
931 let Inst{7} = lane{0};
935 def VLD4LNd8Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
936 def VLD4LNd16Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
937 def VLD4LNd32Pseudo : VLDQQLNPseudo<IIC_VLD4ln>;
939 // ...with double-spaced registers:
940 def VLD4LNq16 : VLD4LN<0b0111, {?,?,1,?}, "16"> {
941 let Inst{7-6} = lane{1-0};
943 def VLD4LNq32 : VLD4LN<0b1011, {?,1,?,?}, "32"> {
944 let Inst{7} = lane{0};
948 def VLD4LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
949 def VLD4LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD4ln>;
951 // ...with address register writeback:
952 class VLD4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
953 : NLdStLn<1, 0b10, op11_8, op7_4,
954 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
955 (ins addrmode6:$Rn, am6offset:$Rm,
956 DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
957 IIC_VLD4lnu, "vld4", Dt,
958 "\\{$Vd[$lane], $dst2[$lane], $dst3[$lane], $dst4[$lane]\\}, $Rn$Rm",
959 "$src1 = $Vd, $src2 = $dst2, $src3 = $dst3, $src4 = $dst4, $Rn.addr = $wb",
962 let DecoderMethod = "DecodeVLD4LN" ;
965 def VLD4LNd8_UPD : VLD4LNWB<0b0011, {?,?,?,?}, "8"> {
966 let Inst{7-5} = lane{2-0};
968 def VLD4LNd16_UPD : VLD4LNWB<0b0111, {?,?,0,?}, "16"> {
969 let Inst{7-6} = lane{1-0};
971 def VLD4LNd32_UPD : VLD4LNWB<0b1011, {?,0,?,?}, "32"> {
972 let Inst{7} = lane{0};
976 def VLD4LNd8Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
977 def VLD4LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
978 def VLD4LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD4lnu>;
980 def VLD4LNq16_UPD : VLD4LNWB<0b0111, {?,?,1,?}, "16"> {
981 let Inst{7-6} = lane{1-0};
983 def VLD4LNq32_UPD : VLD4LNWB<0b1011, {?,1,?,?}, "32"> {
984 let Inst{7} = lane{0};
988 def VLD4LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
989 def VLD4LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD4lnu>;
991 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
993 // VLD1DUP : Vector Load (single element to all lanes)
994 class VLD1DUP<bits<4> op7_4, string Dt, ValueType Ty, PatFrag LoadOp>
995 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd), (ins addrmode6dup:$Rn),
996 IIC_VLD1dup, "vld1", Dt, "\\{$Vd[]\\}, $Rn", "",
997 [(set DPR:$Vd, (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$Rn)))))]> {
1000 let DecoderMethod = "DecodeVLD1DupInstruction";
1002 class VLD1QDUPPseudo<ValueType Ty, PatFrag LoadOp> : VLDQPseudo<IIC_VLD1dup> {
1003 let Pattern = [(set QPR:$dst,
1004 (Ty (NEONvdup (i32 (LoadOp addrmode6dup:$addr)))))];
1007 def VLD1DUPd8 : VLD1DUP<{0,0,0,?}, "8", v8i8, extloadi8>;
1008 def VLD1DUPd16 : VLD1DUP<{0,1,0,?}, "16", v4i16, extloadi16>;
1009 def VLD1DUPd32 : VLD1DUP<{1,0,0,?}, "32", v2i32, load>;
1011 def VLD1DUPq8Pseudo : VLD1QDUPPseudo<v16i8, extloadi8>;
1012 def VLD1DUPq16Pseudo : VLD1QDUPPseudo<v8i16, extloadi16>;
1013 def VLD1DUPq32Pseudo : VLD1QDUPPseudo<v4i32, load>;
1015 def : Pat<(v2f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1016 (VLD1DUPd32 addrmode6:$addr)>;
1017 def : Pat<(v4f32 (NEONvdup (f32 (load addrmode6dup:$addr)))),
1018 (VLD1DUPq32Pseudo addrmode6:$addr)>;
1020 let mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1 in {
1022 class VLD1QDUP<bits<4> op7_4, string Dt>
1023 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2),
1024 (ins addrmode6dup:$Rn), IIC_VLD1dup,
1025 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1027 let Inst{4} = Rn{4};
1028 let DecoderMethod = "DecodeVLD1DupInstruction";
1031 def VLD1DUPq8 : VLD1QDUP<{0,0,1,0}, "8">;
1032 def VLD1DUPq16 : VLD1QDUP<{0,1,1,?}, "16">;
1033 def VLD1DUPq32 : VLD1QDUP<{1,0,1,?}, "32">;
1035 // ...with address register writeback:
1036 class VLD1DUPWB<bits<4> op7_4, string Dt>
1037 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, GPR:$wb),
1038 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1039 "vld1", Dt, "\\{$Vd[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1040 let Inst{4} = Rn{4};
1041 let DecoderMethod = "DecodeVLD1DupInstruction";
1043 class VLD1QDUPWB<bits<4> op7_4, string Dt>
1044 : NLdSt<1, 0b10, 0b1100, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1045 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD1dupu,
1046 "vld1", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1047 let Inst{4} = Rn{4};
1048 let DecoderMethod = "DecodeVLD1DupInstruction";
1051 def VLD1DUPd8_UPD : VLD1DUPWB<{0,0,0,0}, "8">;
1052 def VLD1DUPd16_UPD : VLD1DUPWB<{0,1,0,?}, "16">;
1053 def VLD1DUPd32_UPD : VLD1DUPWB<{1,0,0,?}, "32">;
1055 def VLD1DUPq8_UPD : VLD1QDUPWB<{0,0,1,0}, "8">;
1056 def VLD1DUPq16_UPD : VLD1QDUPWB<{0,1,1,?}, "16">;
1057 def VLD1DUPq32_UPD : VLD1QDUPWB<{1,0,1,?}, "32">;
1059 def VLD1DUPq8Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1060 def VLD1DUPq16Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1061 def VLD1DUPq32Pseudo_UPD : VLDQWBPseudo<IIC_VLD1dupu>;
1063 // VLD2DUP : Vector Load (single 2-element structure to all lanes)
1064 class VLD2DUP<bits<4> op7_4, string Dt>
1065 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2),
1066 (ins addrmode6dup:$Rn), IIC_VLD2dup,
1067 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn", "", []> {
1069 let Inst{4} = Rn{4};
1070 let DecoderMethod = "DecodeVLD2DupInstruction";
1073 def VLD2DUPd8 : VLD2DUP<{0,0,0,?}, "8">;
1074 def VLD2DUPd16 : VLD2DUP<{0,1,0,?}, "16">;
1075 def VLD2DUPd32 : VLD2DUP<{1,0,0,?}, "32">;
1077 def VLD2DUPd8Pseudo : VLDQPseudo<IIC_VLD2dup>;
1078 def VLD2DUPd16Pseudo : VLDQPseudo<IIC_VLD2dup>;
1079 def VLD2DUPd32Pseudo : VLDQPseudo<IIC_VLD2dup>;
1081 // ...with double-spaced registers (not used for codegen):
1082 def VLD2DUPd8x2 : VLD2DUP<{0,0,1,?}, "8">;
1083 def VLD2DUPd16x2 : VLD2DUP<{0,1,1,?}, "16">;
1084 def VLD2DUPd32x2 : VLD2DUP<{1,0,1,?}, "32">;
1086 // ...with address register writeback:
1087 class VLD2DUPWB<bits<4> op7_4, string Dt>
1088 : NLdSt<1, 0b10, 0b1101, op7_4, (outs DPR:$Vd, DPR:$dst2, GPR:$wb),
1089 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD2dupu,
1090 "vld2", Dt, "\\{$Vd[], $dst2[]\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1091 let Inst{4} = Rn{4};
1092 let DecoderMethod = "DecodeVLD2DupInstruction";
1095 def VLD2DUPd8_UPD : VLD2DUPWB<{0,0,0,0}, "8">;
1096 def VLD2DUPd16_UPD : VLD2DUPWB<{0,1,0,?}, "16">;
1097 def VLD2DUPd32_UPD : VLD2DUPWB<{1,0,0,?}, "32">;
1099 def VLD2DUPd8x2_UPD : VLD2DUPWB<{0,0,1,0}, "8">;
1100 def VLD2DUPd16x2_UPD : VLD2DUPWB<{0,1,1,?}, "16">;
1101 def VLD2DUPd32x2_UPD : VLD2DUPWB<{1,0,1,?}, "32">;
1103 def VLD2DUPd8Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1104 def VLD2DUPd16Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1105 def VLD2DUPd32Pseudo_UPD : VLDQWBPseudo<IIC_VLD2dupu>;
1107 // VLD3DUP : Vector Load (single 3-element structure to all lanes)
1108 class VLD3DUP<bits<4> op7_4, string Dt>
1109 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3),
1110 (ins addrmode6dup:$Rn), IIC_VLD3dup,
1111 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn", "", []> {
1114 let DecoderMethod = "DecodeVLD3DupInstruction";
1117 def VLD3DUPd8 : VLD3DUP<{0,0,0,?}, "8">;
1118 def VLD3DUPd16 : VLD3DUP<{0,1,0,?}, "16">;
1119 def VLD3DUPd32 : VLD3DUP<{1,0,0,?}, "32">;
1121 def VLD3DUPd8Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1122 def VLD3DUPd16Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1123 def VLD3DUPd32Pseudo : VLDQQPseudo<IIC_VLD3dup>;
1125 // ...with double-spaced registers (not used for codegen):
1126 def VLD3DUPd8x2 : VLD3DUP<{0,0,1,?}, "8">;
1127 def VLD3DUPd16x2 : VLD3DUP<{0,1,1,?}, "16">;
1128 def VLD3DUPd32x2 : VLD3DUP<{1,0,1,?}, "32">;
1130 // ...with address register writeback:
1131 class VLD3DUPWB<bits<4> op7_4, string Dt>
1132 : NLdSt<1, 0b10, 0b1110, op7_4, (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, GPR:$wb),
1133 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD3dupu,
1134 "vld3", Dt, "\\{$Vd[], $dst2[], $dst3[]\\}, $Rn$Rm",
1135 "$Rn.addr = $wb", []> {
1137 let DecoderMethod = "DecodeVLD3DupInstruction";
1140 def VLD3DUPd8_UPD : VLD3DUPWB<{0,0,0,0}, "8">;
1141 def VLD3DUPd16_UPD : VLD3DUPWB<{0,1,0,?}, "16">;
1142 def VLD3DUPd32_UPD : VLD3DUPWB<{1,0,0,?}, "32">;
1144 def VLD3DUPd8x2_UPD : VLD3DUPWB<{0,0,1,0}, "8">;
1145 def VLD3DUPd16x2_UPD : VLD3DUPWB<{0,1,1,?}, "16">;
1146 def VLD3DUPd32x2_UPD : VLD3DUPWB<{1,0,1,?}, "32">;
1148 def VLD3DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1149 def VLD3DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1150 def VLD3DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3dupu>;
1152 // VLD4DUP : Vector Load (single 4-element structure to all lanes)
1153 class VLD4DUP<bits<4> op7_4, string Dt>
1154 : NLdSt<1, 0b10, 0b1111, op7_4,
1155 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4),
1156 (ins addrmode6dup:$Rn), IIC_VLD4dup,
1157 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn", "", []> {
1159 let Inst{4} = Rn{4};
1160 let DecoderMethod = "DecodeVLD4DupInstruction";
1163 def VLD4DUPd8 : VLD4DUP<{0,0,0,?}, "8">;
1164 def VLD4DUPd16 : VLD4DUP<{0,1,0,?}, "16">;
1165 def VLD4DUPd32 : VLD4DUP<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1167 def VLD4DUPd8Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1168 def VLD4DUPd16Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1169 def VLD4DUPd32Pseudo : VLDQQPseudo<IIC_VLD4dup>;
1171 // ...with double-spaced registers (not used for codegen):
1172 def VLD4DUPd8x2 : VLD4DUP<{0,0,1,?}, "8">;
1173 def VLD4DUPd16x2 : VLD4DUP<{0,1,1,?}, "16">;
1174 def VLD4DUPd32x2 : VLD4DUP<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1176 // ...with address register writeback:
1177 class VLD4DUPWB<bits<4> op7_4, string Dt>
1178 : NLdSt<1, 0b10, 0b1111, op7_4,
1179 (outs DPR:$Vd, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
1180 (ins addrmode6dup:$Rn, am6offset:$Rm), IIC_VLD4dupu,
1181 "vld4", Dt, "\\{$Vd[], $dst2[], $dst3[], $dst4[]\\}, $Rn$Rm",
1182 "$Rn.addr = $wb", []> {
1183 let Inst{4} = Rn{4};
1184 let DecoderMethod = "DecodeVLD4DupInstruction";
1187 def VLD4DUPd8_UPD : VLD4DUPWB<{0,0,0,0}, "8">;
1188 def VLD4DUPd16_UPD : VLD4DUPWB<{0,1,0,?}, "16">;
1189 def VLD4DUPd32_UPD : VLD4DUPWB<{1,?,0,?}, "32"> { let Inst{6} = Rn{5}; }
1191 def VLD4DUPd8x2_UPD : VLD4DUPWB<{0,0,1,0}, "8">;
1192 def VLD4DUPd16x2_UPD : VLD4DUPWB<{0,1,1,?}, "16">;
1193 def VLD4DUPd32x2_UPD : VLD4DUPWB<{1,?,1,?}, "32"> { let Inst{6} = Rn{5}; }
1195 def VLD4DUPd8Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1196 def VLD4DUPd16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1197 def VLD4DUPd32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD4dupu>;
1199 } // mayLoad = 1, neverHasSideEffects = 1, hasExtraDefRegAllocReq = 1
1201 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1203 // Classes for VST* pseudo-instructions with multi-register operands.
1204 // These are expanded to real instructions after register allocation.
1205 class VSTQPseudo<InstrItinClass itin>
1206 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src), itin, "">;
1207 class VSTQWBPseudo<InstrItinClass itin>
1208 : PseudoNLdSt<(outs GPR:$wb),
1209 (ins addrmode6:$addr, am6offset:$offset, QPR:$src), itin,
1210 "$addr.addr = $wb">;
1211 class VSTQQPseudo<InstrItinClass itin>
1212 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src), itin, "">;
1213 class VSTQQWBPseudo<InstrItinClass itin>
1214 : PseudoNLdSt<(outs GPR:$wb),
1215 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src), itin,
1216 "$addr.addr = $wb">;
1217 class VSTQQQQPseudo<InstrItinClass itin>
1218 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src), itin, "">;
1219 class VSTQQQQWBPseudo<InstrItinClass itin>
1220 : PseudoNLdSt<(outs GPR:$wb),
1221 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src), itin,
1222 "$addr.addr = $wb">;
1224 // VST1 : Vector Store (multiple single elements)
1225 class VST1D<bits<4> op7_4, string Dt>
1226 : NLdSt<0,0b00,0b0111,op7_4, (outs), (ins addrmode6:$Rn, VecListOneD:$Vd),
1227 IIC_VST1, "vst1", Dt, "$Vd, $Rn", "", []> {
1229 let Inst{4} = Rn{4};
1230 let DecoderMethod = "DecodeVSTInstruction";
1232 class VST1Q<bits<4> op7_4, string Dt>
1233 : NLdSt<0,0b00,0b1010,op7_4, (outs),
1234 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2), IIC_VST1x2,
1235 "vst1", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1237 let Inst{5-4} = Rn{5-4};
1238 let DecoderMethod = "DecodeVSTInstruction";
1241 def VST1d8 : VST1D<{0,0,0,?}, "8">;
1242 def VST1d16 : VST1D<{0,1,0,?}, "16">;
1243 def VST1d32 : VST1D<{1,0,0,?}, "32">;
1244 def VST1d64 : VST1D<{1,1,0,?}, "64">;
1246 def VST1q8 : VST1Q<{0,0,?,?}, "8">;
1247 def VST1q16 : VST1Q<{0,1,?,?}, "16">;
1248 def VST1q32 : VST1Q<{1,0,?,?}, "32">;
1249 def VST1q64 : VST1Q<{1,1,?,?}, "64">;
1251 def VST1q8Pseudo : VSTQPseudo<IIC_VST1x2>;
1252 def VST1q16Pseudo : VSTQPseudo<IIC_VST1x2>;
1253 def VST1q32Pseudo : VSTQPseudo<IIC_VST1x2>;
1254 def VST1q64Pseudo : VSTQPseudo<IIC_VST1x2>;
1256 // ...with address register writeback:
1257 class VST1DWB<bits<4> op7_4, string Dt>
1258 : NLdSt<0, 0b00, 0b0111, op7_4, (outs GPR:$wb),
1259 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd), IIC_VST1u,
1260 "vst1", Dt, "\\{$Vd\\}, $Rn$Rm", "$Rn.addr = $wb", []> {
1261 let Inst{4} = Rn{4};
1262 let DecoderMethod = "DecodeVSTInstruction";
1264 class VST1QWB<bits<4> op7_4, string Dt>
1265 : NLdSt<0, 0b00, 0b1010, op7_4, (outs GPR:$wb),
1266 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1267 IIC_VST1x2u, "vst1", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1268 "$Rn.addr = $wb", []> {
1269 let Inst{5-4} = Rn{5-4};
1270 let DecoderMethod = "DecodeVSTInstruction";
1273 def VST1d8_UPD : VST1DWB<{0,0,0,?}, "8">;
1274 def VST1d16_UPD : VST1DWB<{0,1,0,?}, "16">;
1275 def VST1d32_UPD : VST1DWB<{1,0,0,?}, "32">;
1276 def VST1d64_UPD : VST1DWB<{1,1,0,?}, "64">;
1278 def VST1q8_UPD : VST1QWB<{0,0,?,?}, "8">;
1279 def VST1q16_UPD : VST1QWB<{0,1,?,?}, "16">;
1280 def VST1q32_UPD : VST1QWB<{1,0,?,?}, "32">;
1281 def VST1q64_UPD : VST1QWB<{1,1,?,?}, "64">;
1283 def VST1q8Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1284 def VST1q16Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1285 def VST1q32Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1286 def VST1q64Pseudo_UPD : VSTQWBPseudo<IIC_VST1x2u>;
1288 // ...with 3 registers
1289 class VST1D3<bits<4> op7_4, string Dt>
1290 : NLdSt<0, 0b00, 0b0110, op7_4, (outs),
1291 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3),
1292 IIC_VST1x3, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1294 let Inst{4} = Rn{4};
1295 let DecoderMethod = "DecodeVSTInstruction";
1297 class VST1D3WB<bits<4> op7_4, string Dt>
1298 : NLdSt<0, 0b00, 0b0110, op7_4, (outs GPR:$wb),
1299 (ins addrmode6:$Rn, am6offset:$Rm,
1300 DPR:$Vd, DPR:$src2, DPR:$src3),
1301 IIC_VST1x3u, "vst1", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1302 "$Rn.addr = $wb", []> {
1303 let Inst{4} = Rn{4};
1304 let DecoderMethod = "DecodeVSTInstruction";
1307 def VST1d8T : VST1D3<{0,0,0,?}, "8">;
1308 def VST1d16T : VST1D3<{0,1,0,?}, "16">;
1309 def VST1d32T : VST1D3<{1,0,0,?}, "32">;
1310 def VST1d64T : VST1D3<{1,1,0,?}, "64">;
1312 def VST1d8T_UPD : VST1D3WB<{0,0,0,?}, "8">;
1313 def VST1d16T_UPD : VST1D3WB<{0,1,0,?}, "16">;
1314 def VST1d32T_UPD : VST1D3WB<{1,0,0,?}, "32">;
1315 def VST1d64T_UPD : VST1D3WB<{1,1,0,?}, "64">;
1317 def VST1d64TPseudo : VSTQQPseudo<IIC_VST1x3>;
1318 def VST1d64TPseudo_UPD : VSTQQWBPseudo<IIC_VST1x3u>;
1320 // ...with 4 registers
1321 class VST1D4<bits<4> op7_4, string Dt>
1322 : NLdSt<0, 0b00, 0b0010, op7_4, (outs),
1323 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1324 IIC_VST1x4, "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn", "",
1327 let Inst{5-4} = Rn{5-4};
1328 let DecoderMethod = "DecodeVSTInstruction";
1330 class VST1D4WB<bits<4> op7_4, string Dt>
1331 : NLdSt<0, 0b00, 0b0010, op7_4, (outs GPR:$wb),
1332 (ins addrmode6:$Rn, am6offset:$Rm,
1333 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST1x4u,
1334 "vst1", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1335 "$Rn.addr = $wb", []> {
1336 let Inst{5-4} = Rn{5-4};
1337 let DecoderMethod = "DecodeVSTInstruction";
1340 def VST1d8Q : VST1D4<{0,0,?,?}, "8">;
1341 def VST1d16Q : VST1D4<{0,1,?,?}, "16">;
1342 def VST1d32Q : VST1D4<{1,0,?,?}, "32">;
1343 def VST1d64Q : VST1D4<{1,1,?,?}, "64">;
1345 def VST1d8Q_UPD : VST1D4WB<{0,0,?,?}, "8">;
1346 def VST1d16Q_UPD : VST1D4WB<{0,1,?,?}, "16">;
1347 def VST1d32Q_UPD : VST1D4WB<{1,0,?,?}, "32">;
1348 def VST1d64Q_UPD : VST1D4WB<{1,1,?,?}, "64">;
1350 def VST1d64QPseudo : VSTQQPseudo<IIC_VST1x4>;
1351 def VST1d64QPseudo_UPD : VSTQQWBPseudo<IIC_VST1x4u>;
1353 // VST2 : Vector Store (multiple 2-element structures)
1354 class VST2D<bits<4> op11_8, bits<4> op7_4, string Dt>
1355 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1356 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2),
1357 IIC_VST2, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn", "", []> {
1359 let Inst{5-4} = Rn{5-4};
1360 let DecoderMethod = "DecodeVSTInstruction";
1362 class VST2Q<bits<4> op7_4, string Dt>
1363 : NLdSt<0, 0b00, 0b0011, op7_4, (outs),
1364 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1365 IIC_VST2x2, "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1368 let Inst{5-4} = Rn{5-4};
1369 let DecoderMethod = "DecodeVSTInstruction";
1372 def VST2d8 : VST2D<0b1000, {0,0,?,?}, "8">;
1373 def VST2d16 : VST2D<0b1000, {0,1,?,?}, "16">;
1374 def VST2d32 : VST2D<0b1000, {1,0,?,?}, "32">;
1376 def VST2q8 : VST2Q<{0,0,?,?}, "8">;
1377 def VST2q16 : VST2Q<{0,1,?,?}, "16">;
1378 def VST2q32 : VST2Q<{1,0,?,?}, "32">;
1380 def VST2d8Pseudo : VSTQPseudo<IIC_VST2>;
1381 def VST2d16Pseudo : VSTQPseudo<IIC_VST2>;
1382 def VST2d32Pseudo : VSTQPseudo<IIC_VST2>;
1384 def VST2q8Pseudo : VSTQQPseudo<IIC_VST2x2>;
1385 def VST2q16Pseudo : VSTQQPseudo<IIC_VST2x2>;
1386 def VST2q32Pseudo : VSTQQPseudo<IIC_VST2x2>;
1388 // ...with address register writeback:
1389 class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1390 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1391 (ins addrmode6:$Rn, am6offset:$Rm, DPR:$Vd, DPR:$src2),
1392 IIC_VST2u, "vst2", Dt, "\\{$Vd, $src2\\}, $Rn$Rm",
1393 "$Rn.addr = $wb", []> {
1394 let Inst{5-4} = Rn{5-4};
1395 let DecoderMethod = "DecodeVSTInstruction";
1397 class VST2QWB<bits<4> op7_4, string Dt>
1398 : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
1399 (ins addrmode6:$Rn, am6offset:$Rm,
1400 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST2x2u,
1401 "vst2", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1402 "$Rn.addr = $wb", []> {
1403 let Inst{5-4} = Rn{5-4};
1404 let DecoderMethod = "DecodeVSTInstruction";
1407 def VST2d8_UPD : VST2DWB<0b1000, {0,0,?,?}, "8">;
1408 def VST2d16_UPD : VST2DWB<0b1000, {0,1,?,?}, "16">;
1409 def VST2d32_UPD : VST2DWB<0b1000, {1,0,?,?}, "32">;
1411 def VST2q8_UPD : VST2QWB<{0,0,?,?}, "8">;
1412 def VST2q16_UPD : VST2QWB<{0,1,?,?}, "16">;
1413 def VST2q32_UPD : VST2QWB<{1,0,?,?}, "32">;
1415 def VST2d8Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1416 def VST2d16Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1417 def VST2d32Pseudo_UPD : VSTQWBPseudo<IIC_VST2u>;
1419 def VST2q8Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1420 def VST2q16Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1421 def VST2q32Pseudo_UPD : VSTQQWBPseudo<IIC_VST2x2u>;
1423 // ...with double-spaced registers
1424 def VST2b8 : VST2D<0b1001, {0,0,?,?}, "8">;
1425 def VST2b16 : VST2D<0b1001, {0,1,?,?}, "16">;
1426 def VST2b32 : VST2D<0b1001, {1,0,?,?}, "32">;
1427 def VST2b8_UPD : VST2DWB<0b1001, {0,0,?,?}, "8">;
1428 def VST2b16_UPD : VST2DWB<0b1001, {0,1,?,?}, "16">;
1429 def VST2b32_UPD : VST2DWB<0b1001, {1,0,?,?}, "32">;
1431 // VST3 : Vector Store (multiple 3-element structures)
1432 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
1433 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1434 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3,
1435 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn", "", []> {
1437 let Inst{4} = Rn{4};
1438 let DecoderMethod = "DecodeVSTInstruction";
1441 def VST3d8 : VST3D<0b0100, {0,0,0,?}, "8">;
1442 def VST3d16 : VST3D<0b0100, {0,1,0,?}, "16">;
1443 def VST3d32 : VST3D<0b0100, {1,0,0,?}, "32">;
1445 def VST3d8Pseudo : VSTQQPseudo<IIC_VST3>;
1446 def VST3d16Pseudo : VSTQQPseudo<IIC_VST3>;
1447 def VST3d32Pseudo : VSTQQPseudo<IIC_VST3>;
1449 // ...with address register writeback:
1450 class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1451 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1452 (ins addrmode6:$Rn, am6offset:$Rm,
1453 DPR:$Vd, DPR:$src2, DPR:$src3), IIC_VST3u,
1454 "vst3", Dt, "\\{$Vd, $src2, $src3\\}, $Rn$Rm",
1455 "$Rn.addr = $wb", []> {
1456 let Inst{4} = Rn{4};
1457 let DecoderMethod = "DecodeVSTInstruction";
1460 def VST3d8_UPD : VST3DWB<0b0100, {0,0,0,?}, "8">;
1461 def VST3d16_UPD : VST3DWB<0b0100, {0,1,0,?}, "16">;
1462 def VST3d32_UPD : VST3DWB<0b0100, {1,0,0,?}, "32">;
1464 def VST3d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1465 def VST3d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1466 def VST3d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST3u>;
1468 // ...with double-spaced registers:
1469 def VST3q8 : VST3D<0b0101, {0,0,0,?}, "8">;
1470 def VST3q16 : VST3D<0b0101, {0,1,0,?}, "16">;
1471 def VST3q32 : VST3D<0b0101, {1,0,0,?}, "32">;
1472 def VST3q8_UPD : VST3DWB<0b0101, {0,0,0,?}, "8">;
1473 def VST3q16_UPD : VST3DWB<0b0101, {0,1,0,?}, "16">;
1474 def VST3q32_UPD : VST3DWB<0b0101, {1,0,0,?}, "32">;
1476 def VST3q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1477 def VST3q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1478 def VST3q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1480 // ...alternate versions to be allocated odd register numbers:
1481 def VST3q8oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1482 def VST3q16oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1483 def VST3q32oddPseudo : VSTQQQQPseudo<IIC_VST3>;
1485 def VST3q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1486 def VST3q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1487 def VST3q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST3u>;
1489 // VST4 : Vector Store (multiple 4-element structures)
1490 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
1491 : NLdSt<0, 0b00, op11_8, op7_4, (outs),
1492 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4),
1493 IIC_VST4, "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn",
1496 let Inst{5-4} = Rn{5-4};
1497 let DecoderMethod = "DecodeVSTInstruction";
1500 def VST4d8 : VST4D<0b0000, {0,0,?,?}, "8">;
1501 def VST4d16 : VST4D<0b0000, {0,1,?,?}, "16">;
1502 def VST4d32 : VST4D<0b0000, {1,0,?,?}, "32">;
1504 def VST4d8Pseudo : VSTQQPseudo<IIC_VST4>;
1505 def VST4d16Pseudo : VSTQQPseudo<IIC_VST4>;
1506 def VST4d32Pseudo : VSTQQPseudo<IIC_VST4>;
1508 // ...with address register writeback:
1509 class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1510 : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
1511 (ins addrmode6:$Rn, am6offset:$Rm,
1512 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST4u,
1513 "vst4", Dt, "\\{$Vd, $src2, $src3, $src4\\}, $Rn$Rm",
1514 "$Rn.addr = $wb", []> {
1515 let Inst{5-4} = Rn{5-4};
1516 let DecoderMethod = "DecodeVSTInstruction";
1519 def VST4d8_UPD : VST4DWB<0b0000, {0,0,?,?}, "8">;
1520 def VST4d16_UPD : VST4DWB<0b0000, {0,1,?,?}, "16">;
1521 def VST4d32_UPD : VST4DWB<0b0000, {1,0,?,?}, "32">;
1523 def VST4d8Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1524 def VST4d16Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1525 def VST4d32Pseudo_UPD : VSTQQWBPseudo<IIC_VST4u>;
1527 // ...with double-spaced registers:
1528 def VST4q8 : VST4D<0b0001, {0,0,?,?}, "8">;
1529 def VST4q16 : VST4D<0b0001, {0,1,?,?}, "16">;
1530 def VST4q32 : VST4D<0b0001, {1,0,?,?}, "32">;
1531 def VST4q8_UPD : VST4DWB<0b0001, {0,0,?,?}, "8">;
1532 def VST4q16_UPD : VST4DWB<0b0001, {0,1,?,?}, "16">;
1533 def VST4q32_UPD : VST4DWB<0b0001, {1,0,?,?}, "32">;
1535 def VST4q8Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1536 def VST4q16Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1537 def VST4q32Pseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1539 // ...alternate versions to be allocated odd register numbers:
1540 def VST4q8oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1541 def VST4q16oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1542 def VST4q32oddPseudo : VSTQQQQPseudo<IIC_VST4>;
1544 def VST4q8oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1545 def VST4q16oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1546 def VST4q32oddPseudo_UPD : VSTQQQQWBPseudo<IIC_VST4u>;
1548 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1550 // Classes for VST*LN pseudo-instructions with multi-register operands.
1551 // These are expanded to real instructions after register allocation.
1552 class VSTQLNPseudo<InstrItinClass itin>
1553 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QPR:$src, nohash_imm:$lane),
1555 class VSTQLNWBPseudo<InstrItinClass itin>
1556 : PseudoNLdSt<(outs GPR:$wb),
1557 (ins addrmode6:$addr, am6offset:$offset, QPR:$src,
1558 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1559 class VSTQQLNPseudo<InstrItinClass itin>
1560 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQPR:$src, nohash_imm:$lane),
1562 class VSTQQLNWBPseudo<InstrItinClass itin>
1563 : PseudoNLdSt<(outs GPR:$wb),
1564 (ins addrmode6:$addr, am6offset:$offset, QQPR:$src,
1565 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1566 class VSTQQQQLNPseudo<InstrItinClass itin>
1567 : PseudoNLdSt<(outs), (ins addrmode6:$addr, QQQQPR:$src, nohash_imm:$lane),
1569 class VSTQQQQLNWBPseudo<InstrItinClass itin>
1570 : PseudoNLdSt<(outs GPR:$wb),
1571 (ins addrmode6:$addr, am6offset:$offset, QQQQPR:$src,
1572 nohash_imm:$lane), itin, "$addr.addr = $wb">;
1574 // VST1LN : Vector Store (single element from one lane)
1575 class VST1LN<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1576 PatFrag StoreOp, SDNode ExtractOp>
1577 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1578 (ins addrmode6:$Rn, DPR:$Vd, nohash_imm:$lane),
1579 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1580 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6:$Rn)]> {
1582 let DecoderMethod = "DecodeVST1LN";
1584 class VST1LN32<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1585 PatFrag StoreOp, SDNode ExtractOp>
1586 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1587 (ins addrmode6oneL32:$Rn, DPR:$Vd, nohash_imm:$lane),
1588 IIC_VST1ln, "vst1", Dt, "\\{$Vd[$lane]\\}, $Rn", "",
1589 [(StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane), addrmode6oneL32:$Rn)]>{
1591 let DecoderMethod = "DecodeVST1LN";
1593 class VST1QLNPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1594 : VSTQLNPseudo<IIC_VST1ln> {
1595 let Pattern = [(StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1599 def VST1LNd8 : VST1LN<0b0000, {?,?,?,0}, "8", v8i8, truncstorei8,
1601 let Inst{7-5} = lane{2-0};
1603 def VST1LNd16 : VST1LN<0b0100, {?,?,0,?}, "16", v4i16, truncstorei16,
1605 let Inst{7-6} = lane{1-0};
1606 let Inst{4} = Rn{5};
1609 def VST1LNd32 : VST1LN32<0b1000, {?,0,?,?}, "32", v2i32, store, extractelt> {
1610 let Inst{7} = lane{0};
1611 let Inst{5-4} = Rn{5-4};
1614 def VST1LNq8Pseudo : VST1QLNPseudo<v16i8, truncstorei8, NEONvgetlaneu>;
1615 def VST1LNq16Pseudo : VST1QLNPseudo<v8i16, truncstorei16, NEONvgetlaneu>;
1616 def VST1LNq32Pseudo : VST1QLNPseudo<v4i32, store, extractelt>;
1618 def : Pat<(store (extractelt (v2f32 DPR:$src), imm:$lane), addrmode6:$addr),
1619 (VST1LNd32 addrmode6:$addr, DPR:$src, imm:$lane)>;
1620 def : Pat<(store (extractelt (v4f32 QPR:$src), imm:$lane), addrmode6:$addr),
1621 (VST1LNq32Pseudo addrmode6:$addr, QPR:$src, imm:$lane)>;
1623 // ...with address register writeback:
1624 class VST1LNWB<bits<4> op11_8, bits<4> op7_4, string Dt, ValueType Ty,
1625 PatFrag StoreOp, SDNode ExtractOp>
1626 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1627 (ins addrmode6:$Rn, am6offset:$Rm,
1628 DPR:$Vd, nohash_imm:$lane), IIC_VST1lnu, "vst1", Dt,
1629 "\\{$Vd[$lane]\\}, $Rn$Rm",
1631 [(set GPR:$wb, (StoreOp (ExtractOp (Ty DPR:$Vd), imm:$lane),
1632 addrmode6:$Rn, am6offset:$Rm))]> {
1633 let DecoderMethod = "DecodeVST1LN";
1635 class VST1QLNWBPseudo<ValueType Ty, PatFrag StoreOp, SDNode ExtractOp>
1636 : VSTQLNWBPseudo<IIC_VST1lnu> {
1637 let Pattern = [(set GPR:$wb, (StoreOp (ExtractOp (Ty QPR:$src), imm:$lane),
1638 addrmode6:$addr, am6offset:$offset))];
1641 def VST1LNd8_UPD : VST1LNWB<0b0000, {?,?,?,0}, "8", v8i8, post_truncsti8,
1643 let Inst{7-5} = lane{2-0};
1645 def VST1LNd16_UPD : VST1LNWB<0b0100, {?,?,0,?}, "16", v4i16, post_truncsti16,
1647 let Inst{7-6} = lane{1-0};
1648 let Inst{4} = Rn{5};
1650 def VST1LNd32_UPD : VST1LNWB<0b1000, {?,0,?,?}, "32", v2i32, post_store,
1652 let Inst{7} = lane{0};
1653 let Inst{5-4} = Rn{5-4};
1656 def VST1LNq8Pseudo_UPD : VST1QLNWBPseudo<v16i8, post_truncsti8, NEONvgetlaneu>;
1657 def VST1LNq16Pseudo_UPD : VST1QLNWBPseudo<v8i16, post_truncsti16,NEONvgetlaneu>;
1658 def VST1LNq32Pseudo_UPD : VST1QLNWBPseudo<v4i32, post_store, extractelt>;
1660 let mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1 in {
1662 // VST2LN : Vector Store (single 2-element structure from one lane)
1663 class VST2LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1664 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1665 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, nohash_imm:$lane),
1666 IIC_VST2ln, "vst2", Dt, "\\{$Vd[$lane], $src2[$lane]\\}, $Rn",
1669 let Inst{4} = Rn{4};
1670 let DecoderMethod = "DecodeVST2LN";
1673 def VST2LNd8 : VST2LN<0b0001, {?,?,?,?}, "8"> {
1674 let Inst{7-5} = lane{2-0};
1676 def VST2LNd16 : VST2LN<0b0101, {?,?,0,?}, "16"> {
1677 let Inst{7-6} = lane{1-0};
1679 def VST2LNd32 : VST2LN<0b1001, {?,0,0,?}, "32"> {
1680 let Inst{7} = lane{0};
1683 def VST2LNd8Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1684 def VST2LNd16Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1685 def VST2LNd32Pseudo : VSTQLNPseudo<IIC_VST2ln>;
1687 // ...with double-spaced registers:
1688 def VST2LNq16 : VST2LN<0b0101, {?,?,1,?}, "16"> {
1689 let Inst{7-6} = lane{1-0};
1690 let Inst{4} = Rn{4};
1692 def VST2LNq32 : VST2LN<0b1001, {?,1,0,?}, "32"> {
1693 let Inst{7} = lane{0};
1694 let Inst{4} = Rn{4};
1697 def VST2LNq16Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1698 def VST2LNq32Pseudo : VSTQQLNPseudo<IIC_VST2ln>;
1700 // ...with address register writeback:
1701 class VST2LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1702 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1703 (ins addrmode6:$addr, am6offset:$offset,
1704 DPR:$src1, DPR:$src2, nohash_imm:$lane), IIC_VST2lnu, "vst2", Dt,
1705 "\\{$src1[$lane], $src2[$lane]\\}, $addr$offset",
1706 "$addr.addr = $wb", []> {
1707 let Inst{4} = Rn{4};
1708 let DecoderMethod = "DecodeVST2LN";
1711 def VST2LNd8_UPD : VST2LNWB<0b0001, {?,?,?,?}, "8"> {
1712 let Inst{7-5} = lane{2-0};
1714 def VST2LNd16_UPD : VST2LNWB<0b0101, {?,?,0,?}, "16"> {
1715 let Inst{7-6} = lane{1-0};
1717 def VST2LNd32_UPD : VST2LNWB<0b1001, {?,0,0,?}, "32"> {
1718 let Inst{7} = lane{0};
1721 def VST2LNd8Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1722 def VST2LNd16Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1723 def VST2LNd32Pseudo_UPD : VSTQLNWBPseudo<IIC_VST2lnu>;
1725 def VST2LNq16_UPD : VST2LNWB<0b0101, {?,?,1,?}, "16"> {
1726 let Inst{7-6} = lane{1-0};
1728 def VST2LNq32_UPD : VST2LNWB<0b1001, {?,1,0,?}, "32"> {
1729 let Inst{7} = lane{0};
1732 def VST2LNq16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1733 def VST2LNq32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST2lnu>;
1735 // VST3LN : Vector Store (single 3-element structure from one lane)
1736 class VST3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1737 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1738 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3,
1739 nohash_imm:$lane), IIC_VST3ln, "vst3", Dt,
1740 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn", "", []> {
1742 let DecoderMethod = "DecodeVST3LN";
1745 def VST3LNd8 : VST3LN<0b0010, {?,?,?,0}, "8"> {
1746 let Inst{7-5} = lane{2-0};
1748 def VST3LNd16 : VST3LN<0b0110, {?,?,0,0}, "16"> {
1749 let Inst{7-6} = lane{1-0};
1751 def VST3LNd32 : VST3LN<0b1010, {?,0,0,0}, "32"> {
1752 let Inst{7} = lane{0};
1755 def VST3LNd8Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1756 def VST3LNd16Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1757 def VST3LNd32Pseudo : VSTQQLNPseudo<IIC_VST3ln>;
1759 // ...with double-spaced registers:
1760 def VST3LNq16 : VST3LN<0b0110, {?,?,1,0}, "16"> {
1761 let Inst{7-6} = lane{1-0};
1763 def VST3LNq32 : VST3LN<0b1010, {?,1,0,0}, "32"> {
1764 let Inst{7} = lane{0};
1767 def VST3LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1768 def VST3LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST3ln>;
1770 // ...with address register writeback:
1771 class VST3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1772 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1773 (ins addrmode6:$Rn, am6offset:$Rm,
1774 DPR:$Vd, DPR:$src2, DPR:$src3, nohash_imm:$lane),
1775 IIC_VST3lnu, "vst3", Dt,
1776 "\\{$Vd[$lane], $src2[$lane], $src3[$lane]\\}, $Rn$Rm",
1777 "$Rn.addr = $wb", []> {
1778 let DecoderMethod = "DecodeVST3LN";
1781 def VST3LNd8_UPD : VST3LNWB<0b0010, {?,?,?,0}, "8"> {
1782 let Inst{7-5} = lane{2-0};
1784 def VST3LNd16_UPD : VST3LNWB<0b0110, {?,?,0,0}, "16"> {
1785 let Inst{7-6} = lane{1-0};
1787 def VST3LNd32_UPD : VST3LNWB<0b1010, {?,0,0,0}, "32"> {
1788 let Inst{7} = lane{0};
1791 def VST3LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1792 def VST3LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1793 def VST3LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST3lnu>;
1795 def VST3LNq16_UPD : VST3LNWB<0b0110, {?,?,1,0}, "16"> {
1796 let Inst{7-6} = lane{1-0};
1798 def VST3LNq32_UPD : VST3LNWB<0b1010, {?,1,0,0}, "32"> {
1799 let Inst{7} = lane{0};
1802 def VST3LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1803 def VST3LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST3lnu>;
1805 // VST4LN : Vector Store (single 4-element structure from one lane)
1806 class VST4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
1807 : NLdStLn<1, 0b00, op11_8, op7_4, (outs),
1808 (ins addrmode6:$Rn, DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4,
1809 nohash_imm:$lane), IIC_VST4ln, "vst4", Dt,
1810 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn",
1813 let Inst{4} = Rn{4};
1814 let DecoderMethod = "DecodeVST4LN";
1817 def VST4LNd8 : VST4LN<0b0011, {?,?,?,?}, "8"> {
1818 let Inst{7-5} = lane{2-0};
1820 def VST4LNd16 : VST4LN<0b0111, {?,?,0,?}, "16"> {
1821 let Inst{7-6} = lane{1-0};
1823 def VST4LNd32 : VST4LN<0b1011, {?,0,?,?}, "32"> {
1824 let Inst{7} = lane{0};
1825 let Inst{5} = Rn{5};
1828 def VST4LNd8Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1829 def VST4LNd16Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1830 def VST4LNd32Pseudo : VSTQQLNPseudo<IIC_VST4ln>;
1832 // ...with double-spaced registers:
1833 def VST4LNq16 : VST4LN<0b0111, {?,?,1,?}, "16"> {
1834 let Inst{7-6} = lane{1-0};
1836 def VST4LNq32 : VST4LN<0b1011, {?,1,?,?}, "32"> {
1837 let Inst{7} = lane{0};
1838 let Inst{5} = Rn{5};
1841 def VST4LNq16Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1842 def VST4LNq32Pseudo : VSTQQQQLNPseudo<IIC_VST4ln>;
1844 // ...with address register writeback:
1845 class VST4LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
1846 : NLdStLn<1, 0b00, op11_8, op7_4, (outs GPR:$wb),
1847 (ins addrmode6:$Rn, am6offset:$Rm,
1848 DPR:$Vd, DPR:$src2, DPR:$src3, DPR:$src4, nohash_imm:$lane),
1849 IIC_VST4lnu, "vst4", Dt,
1850 "\\{$Vd[$lane], $src2[$lane], $src3[$lane], $src4[$lane]\\}, $Rn$Rm",
1851 "$Rn.addr = $wb", []> {
1852 let Inst{4} = Rn{4};
1853 let DecoderMethod = "DecodeVST4LN";
1856 def VST4LNd8_UPD : VST4LNWB<0b0011, {?,?,?,?}, "8"> {
1857 let Inst{7-5} = lane{2-0};
1859 def VST4LNd16_UPD : VST4LNWB<0b0111, {?,?,0,?}, "16"> {
1860 let Inst{7-6} = lane{1-0};
1862 def VST4LNd32_UPD : VST4LNWB<0b1011, {?,0,?,?}, "32"> {
1863 let Inst{7} = lane{0};
1864 let Inst{5} = Rn{5};
1867 def VST4LNd8Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1868 def VST4LNd16Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1869 def VST4LNd32Pseudo_UPD : VSTQQLNWBPseudo<IIC_VST4lnu>;
1871 def VST4LNq16_UPD : VST4LNWB<0b0111, {?,?,1,?}, "16"> {
1872 let Inst{7-6} = lane{1-0};
1874 def VST4LNq32_UPD : VST4LNWB<0b1011, {?,1,?,?}, "32"> {
1875 let Inst{7} = lane{0};
1876 let Inst{5} = Rn{5};
1879 def VST4LNq16Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1880 def VST4LNq32Pseudo_UPD : VSTQQQQLNWBPseudo<IIC_VST4lnu>;
1882 } // mayStore = 1, neverHasSideEffects = 1, hasExtraSrcRegAllocReq = 1
1885 //===----------------------------------------------------------------------===//
1886 // NEON pattern fragments
1887 //===----------------------------------------------------------------------===//
1889 // Extract D sub-registers of Q registers.
1890 def DSubReg_i8_reg : SDNodeXForm<imm, [{
1891 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1892 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/8, MVT::i32);
1894 def DSubReg_i16_reg : SDNodeXForm<imm, [{
1895 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1896 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/4, MVT::i32);
1898 def DSubReg_i32_reg : SDNodeXForm<imm, [{
1899 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1900 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue()/2, MVT::i32);
1902 def DSubReg_f64_reg : SDNodeXForm<imm, [{
1903 assert(ARM::dsub_7 == ARM::dsub_0+7 && "Unexpected subreg numbering");
1904 return CurDAG->getTargetConstant(ARM::dsub_0 + N->getZExtValue(), MVT::i32);
1907 // Extract S sub-registers of Q/D registers.
1908 def SSubReg_f32_reg : SDNodeXForm<imm, [{
1909 assert(ARM::ssub_3 == ARM::ssub_0+3 && "Unexpected subreg numbering");
1910 return CurDAG->getTargetConstant(ARM::ssub_0 + N->getZExtValue(), MVT::i32);
1913 // Translate lane numbers from Q registers to D subregs.
1914 def SubReg_i8_lane : SDNodeXForm<imm, [{
1915 return CurDAG->getTargetConstant(N->getZExtValue() & 7, MVT::i32);
1917 def SubReg_i16_lane : SDNodeXForm<imm, [{
1918 return CurDAG->getTargetConstant(N->getZExtValue() & 3, MVT::i32);
1920 def SubReg_i32_lane : SDNodeXForm<imm, [{
1921 return CurDAG->getTargetConstant(N->getZExtValue() & 1, MVT::i32);
1924 //===----------------------------------------------------------------------===//
1925 // Instruction Classes
1926 //===----------------------------------------------------------------------===//
1928 // Basic 2-register operations: double- and quad-register.
1929 class N2VD<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1930 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1931 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1932 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1933 (ins DPR:$Vm), IIC_VUNAD, OpcodeStr, Dt,"$Vd, $Vm", "",
1934 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm))))]>;
1935 class N2VQ<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1936 bits<2> op17_16, bits<5> op11_7, bit op4, string OpcodeStr,
1937 string Dt, ValueType ResTy, ValueType OpTy, SDNode OpNode>
1938 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1939 (ins QPR:$Vm), IIC_VUNAQ, OpcodeStr, Dt,"$Vd, $Vm", "",
1940 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm))))]>;
1942 // Basic 2-register intrinsics, both double- and quad-register.
1943 class N2VDInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1944 bits<2> op17_16, bits<5> op11_7, bit op4,
1945 InstrItinClass itin, string OpcodeStr, string Dt,
1946 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1947 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
1948 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1949 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
1950 class N2VQInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1951 bits<2> op17_16, bits<5> op11_7, bit op4,
1952 InstrItinClass itin, string OpcodeStr, string Dt,
1953 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
1954 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
1955 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1956 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
1958 // Narrow 2-register operations.
1959 class N2VN<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1960 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1961 InstrItinClass itin, string OpcodeStr, string Dt,
1962 ValueType TyD, ValueType TyQ, SDNode OpNode>
1963 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1964 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1965 [(set DPR:$Vd, (TyD (OpNode (TyQ QPR:$Vm))))]>;
1967 // Narrow 2-register intrinsics.
1968 class N2VNInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1969 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1970 InstrItinClass itin, string OpcodeStr, string Dt,
1971 ValueType TyD, ValueType TyQ, Intrinsic IntOp>
1972 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs DPR:$Vd),
1973 (ins QPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1974 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vm))))]>;
1976 // Long 2-register operations (currently only used for VMOVL).
1977 class N2VL<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1978 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1979 InstrItinClass itin, string OpcodeStr, string Dt,
1980 ValueType TyQ, ValueType TyD, SDNode OpNode>
1981 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1982 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1983 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vm))))]>;
1985 // Long 2-register intrinsics.
1986 class N2VLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
1987 bits<2> op17_16, bits<5> op11_7, bit op6, bit op4,
1988 InstrItinClass itin, string OpcodeStr, string Dt,
1989 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
1990 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, op6, op4, (outs QPR:$Vd),
1991 (ins DPR:$Vm), itin, OpcodeStr, Dt, "$Vd, $Vm", "",
1992 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vm))))]>;
1994 // 2-register shuffles (VTRN/VZIP/VUZP), both double- and quad-register.
1995 class N2VDShuffle<bits<2> op19_18, bits<5> op11_7, string OpcodeStr, string Dt>
1996 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 0, 0, (outs DPR:$Vd, DPR:$Vm),
1997 (ins DPR:$src1, DPR:$src2), IIC_VPERMD,
1998 OpcodeStr, Dt, "$Vd, $Vm",
1999 "$src1 = $Vd, $src2 = $Vm", []>;
2000 class N2VQShuffle<bits<2> op19_18, bits<5> op11_7,
2001 InstrItinClass itin, string OpcodeStr, string Dt>
2002 : N2V<0b11, 0b11, op19_18, 0b10, op11_7, 1, 0, (outs QPR:$Vd, QPR:$Vm),
2003 (ins QPR:$src1, QPR:$src2), itin, OpcodeStr, Dt, "$Vd, $Vm",
2004 "$src1 = $Vd, $src2 = $Vm", []>;
2006 // Basic 3-register operations: double- and quad-register.
2007 class N3VD<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2008 InstrItinClass itin, string OpcodeStr, string Dt,
2009 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2010 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2011 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2012 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2013 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2014 let isCommutable = Commutable;
2016 // Same as N3VD but no data type.
2017 class N3VDX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2018 InstrItinClass itin, string OpcodeStr,
2019 ValueType ResTy, ValueType OpTy,
2020 SDNode OpNode, bit Commutable>
2021 : N3VX<op24, op23, op21_20, op11_8, 0, op4,
2022 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2023 OpcodeStr, "$Vd, $Vn, $Vm", "",
2024 [(set DPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>{
2025 let isCommutable = Commutable;
2028 class N3VDSL<bits<2> op21_20, bits<4> op11_8,
2029 InstrItinClass itin, string OpcodeStr, string Dt,
2030 ValueType Ty, SDNode ShOp>
2031 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2032 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2033 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2035 (Ty (ShOp (Ty DPR:$Vn),
2036 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),imm:$lane)))))]> {
2037 let isCommutable = 0;
2039 class N3VDSL16<bits<2> op21_20, bits<4> op11_8,
2040 string OpcodeStr, string Dt, ValueType Ty, SDNode ShOp>
2041 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2042 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2043 NVMulSLFrm, IIC_VMULi16D, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane","",
2045 (Ty (ShOp (Ty DPR:$Vn),
2046 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2047 let isCommutable = 0;
2050 class N3VQ<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2051 InstrItinClass itin, string OpcodeStr, string Dt,
2052 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2053 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2054 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2055 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2056 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2057 let isCommutable = Commutable;
2059 class N3VQX<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2060 InstrItinClass itin, string OpcodeStr,
2061 ValueType ResTy, ValueType OpTy, SDNode OpNode, bit Commutable>
2062 : N3VX<op24, op23, op21_20, op11_8, 1, op4,
2063 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2064 OpcodeStr, "$Vd, $Vn, $Vm", "",
2065 [(set QPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>{
2066 let isCommutable = Commutable;
2068 class N3VQSL<bits<2> op21_20, bits<4> op11_8,
2069 InstrItinClass itin, string OpcodeStr, string Dt,
2070 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2071 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2072 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2073 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2074 [(set (ResTy QPR:$Vd),
2075 (ResTy (ShOp (ResTy QPR:$Vn),
2076 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2078 let isCommutable = 0;
2080 class N3VQSL16<bits<2> op21_20, bits<4> op11_8, string OpcodeStr, string Dt,
2081 ValueType ResTy, ValueType OpTy, SDNode ShOp>
2082 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2083 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2084 NVMulSLFrm, IIC_VMULi16Q, OpcodeStr, Dt,"$Vd, $Vn, $Vm$lane", "",
2085 [(set (ResTy QPR:$Vd),
2086 (ResTy (ShOp (ResTy QPR:$Vn),
2087 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2089 let isCommutable = 0;
2092 // Basic 3-register intrinsics, both double- and quad-register.
2093 class N3VDInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2094 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2095 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2096 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2097 (outs DPR:$Vd), (ins DPR:$Vn, DPR:$Vm), f, itin,
2098 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2099 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]> {
2100 let isCommutable = Commutable;
2102 class N3VDIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2103 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2104 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2105 (outs DPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2106 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2108 (Ty (IntOp (Ty DPR:$Vn),
2109 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2111 let isCommutable = 0;
2113 class N3VDIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2114 string OpcodeStr, string Dt, ValueType Ty, Intrinsic IntOp>
2115 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2116 (outs DPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2117 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2119 (Ty (IntOp (Ty DPR:$Vn),
2120 (Ty (NEONvduplane (Ty DPR_8:$Vm), imm:$lane)))))]> {
2121 let isCommutable = 0;
2123 class N3VDIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2124 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2125 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2126 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2127 (outs DPR:$Vd), (ins DPR:$Vm, DPR:$Vn), f, itin,
2128 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2129 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (OpTy DPR:$Vn))))]> {
2130 let isCommutable = 0;
2133 class N3VQInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2134 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2135 ValueType ResTy, ValueType OpTy, Intrinsic IntOp, bit Commutable>
2136 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2137 (outs QPR:$Vd), (ins QPR:$Vn, QPR:$Vm), f, itin,
2138 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2139 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]> {
2140 let isCommutable = Commutable;
2142 class N3VQIntSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2143 string OpcodeStr, string Dt,
2144 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2145 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2146 (outs QPR:$Vd), (ins QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2147 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2148 [(set (ResTy QPR:$Vd),
2149 (ResTy (IntOp (ResTy QPR:$Vn),
2150 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2152 let isCommutable = 0;
2154 class N3VQIntSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2155 string OpcodeStr, string Dt,
2156 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2157 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2158 (outs QPR:$Vd), (ins QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2159 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2160 [(set (ResTy QPR:$Vd),
2161 (ResTy (IntOp (ResTy QPR:$Vn),
2162 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2164 let isCommutable = 0;
2166 class N3VQIntSh<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2167 Format f, InstrItinClass itin, string OpcodeStr, string Dt,
2168 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2169 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2170 (outs QPR:$Vd), (ins QPR:$Vm, QPR:$Vn), f, itin,
2171 OpcodeStr, Dt, "$Vd, $Vm, $Vn", "",
2172 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (OpTy QPR:$Vn))))]> {
2173 let isCommutable = 0;
2176 // Multiply-Add/Sub operations: double- and quad-register.
2177 class N3VDMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2178 InstrItinClass itin, string OpcodeStr, string Dt,
2179 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator OpNode>
2180 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2181 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2182 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2183 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2184 (Ty (MulOp DPR:$Vn, DPR:$Vm)))))]>;
2186 class N3VDMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2187 string OpcodeStr, string Dt,
2188 ValueType Ty, SDPatternOperator MulOp, SDPatternOperator ShOp>
2189 : N3VLane32<0, 1, op21_20, op11_8, 1, 0,
2191 (ins DPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2193 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2195 (Ty (ShOp (Ty DPR:$src1),
2197 (Ty (NEONvduplane (Ty DPR_VFP2:$Vm),
2199 class N3VDMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2200 string OpcodeStr, string Dt,
2201 ValueType Ty, SDNode MulOp, SDNode ShOp>
2202 : N3VLane16<0, 1, op21_20, op11_8, 1, 0,
2204 (ins DPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2206 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2208 (Ty (ShOp (Ty DPR:$src1),
2210 (Ty (NEONvduplane (Ty DPR_8:$Vm),
2213 class N3VQMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2214 InstrItinClass itin, string OpcodeStr, string Dt, ValueType Ty,
2215 SDPatternOperator MulOp, SDPatternOperator OpNode>
2216 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2217 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2218 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2219 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2220 (Ty (MulOp QPR:$Vn, QPR:$Vm)))))]>;
2221 class N3VQMulOpSL<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2222 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2223 SDPatternOperator MulOp, SDPatternOperator ShOp>
2224 : N3VLane32<1, 1, op21_20, op11_8, 1, 0,
2226 (ins QPR:$src1, QPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2228 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2229 [(set (ResTy QPR:$Vd),
2230 (ResTy (ShOp (ResTy QPR:$src1),
2231 (ResTy (MulOp QPR:$Vn,
2232 (ResTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2234 class N3VQMulOpSL16<bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2235 string OpcodeStr, string Dt,
2236 ValueType ResTy, ValueType OpTy,
2237 SDNode MulOp, SDNode ShOp>
2238 : N3VLane16<1, 1, op21_20, op11_8, 1, 0,
2240 (ins QPR:$src1, QPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2242 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2243 [(set (ResTy QPR:$Vd),
2244 (ResTy (ShOp (ResTy QPR:$src1),
2245 (ResTy (MulOp QPR:$Vn,
2246 (ResTy (NEONvduplane (OpTy DPR_8:$Vm),
2249 // Neon Intrinsic-Op instructions (VABA): double- and quad-register.
2250 class N3VDIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2251 InstrItinClass itin, string OpcodeStr, string Dt,
2252 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2253 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2254 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2255 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2256 [(set DPR:$Vd, (Ty (OpNode DPR:$src1,
2257 (Ty (IntOp (Ty DPR:$Vn), (Ty DPR:$Vm))))))]>;
2258 class N3VQIntOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2259 InstrItinClass itin, string OpcodeStr, string Dt,
2260 ValueType Ty, Intrinsic IntOp, SDNode OpNode>
2261 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2262 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2263 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2264 [(set QPR:$Vd, (Ty (OpNode QPR:$src1,
2265 (Ty (IntOp (Ty QPR:$Vn), (Ty QPR:$Vm))))))]>;
2267 // Neon 3-argument intrinsics, both double- and quad-register.
2268 // The destination register is also used as the first source operand register.
2269 class N3VDInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2270 InstrItinClass itin, string OpcodeStr, string Dt,
2271 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2272 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2273 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2274 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2275 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$src1),
2276 (OpTy DPR:$Vn), (OpTy DPR:$Vm))))]>;
2277 class N3VQInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2278 InstrItinClass itin, string OpcodeStr, string Dt,
2279 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2280 : N3V<op24, op23, op21_20, op11_8, 1, op4,
2281 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm), N3RegFrm, itin,
2282 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2283 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$src1),
2284 (OpTy QPR:$Vn), (OpTy QPR:$Vm))))]>;
2286 // Long Multiply-Add/Sub operations.
2287 class N3VLMulOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2288 InstrItinClass itin, string OpcodeStr, string Dt,
2289 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2290 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2291 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2292 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2293 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2294 (TyQ (MulOp (TyD DPR:$Vn),
2295 (TyD DPR:$Vm)))))]>;
2296 class N3VLMulOpSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2297 InstrItinClass itin, string OpcodeStr, string Dt,
2298 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2299 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2300 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2302 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2304 (OpNode (TyQ QPR:$src1),
2305 (TyQ (MulOp (TyD DPR:$Vn),
2306 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),
2308 class N3VLMulOpSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2309 InstrItinClass itin, string OpcodeStr, string Dt,
2310 ValueType TyQ, ValueType TyD, SDNode MulOp, SDNode OpNode>
2311 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0, (outs QPR:$Vd),
2312 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2314 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2316 (OpNode (TyQ QPR:$src1),
2317 (TyQ (MulOp (TyD DPR:$Vn),
2318 (TyD (NEONvduplane (TyD DPR_8:$Vm),
2321 // Long Intrinsic-Op vector operations with explicit extend (VABAL).
2322 class N3VLIntExtOp<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2323 InstrItinClass itin, string OpcodeStr, string Dt,
2324 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2326 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2327 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2328 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2329 [(set QPR:$Vd, (OpNode (TyQ QPR:$src1),
2330 (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2331 (TyD DPR:$Vm)))))))]>;
2333 // Neon Long 3-argument intrinsic. The destination register is
2334 // a quad-register and is also used as the first source operand register.
2335 class N3VLInt3<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2336 InstrItinClass itin, string OpcodeStr, string Dt,
2337 ValueType TyQ, ValueType TyD, Intrinsic IntOp>
2338 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2339 (outs QPR:$Vd), (ins QPR:$src1, DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2340 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "$src1 = $Vd",
2342 (TyQ (IntOp (TyQ QPR:$src1), (TyD DPR:$Vn), (TyD DPR:$Vm))))]>;
2343 class N3VLInt3SL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2344 string OpcodeStr, string Dt,
2345 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2346 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2348 (ins QPR:$src1, DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2350 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2351 [(set (ResTy QPR:$Vd),
2352 (ResTy (IntOp (ResTy QPR:$src1),
2354 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2356 class N3VLInt3SL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2357 InstrItinClass itin, string OpcodeStr, string Dt,
2358 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2359 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2361 (ins QPR:$src1, DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2363 OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "$src1 = $Vd",
2364 [(set (ResTy QPR:$Vd),
2365 (ResTy (IntOp (ResTy QPR:$src1),
2367 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2370 // Narrowing 3-register intrinsics.
2371 class N3VNInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2372 string OpcodeStr, string Dt, ValueType TyD, ValueType TyQ,
2373 Intrinsic IntOp, bit Commutable>
2374 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2375 (outs DPR:$Vd), (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINi4D,
2376 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2377 [(set DPR:$Vd, (TyD (IntOp (TyQ QPR:$Vn), (TyQ QPR:$Vm))))]> {
2378 let isCommutable = Commutable;
2381 // Long 3-register operations.
2382 class N3VL<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2383 InstrItinClass itin, string OpcodeStr, string Dt,
2384 ValueType TyQ, ValueType TyD, SDNode OpNode, bit Commutable>
2385 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2386 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2387 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2388 [(set QPR:$Vd, (TyQ (OpNode (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2389 let isCommutable = Commutable;
2391 class N3VLSL<bit op24, bits<2> op21_20, bits<4> op11_8,
2392 InstrItinClass itin, string OpcodeStr, string Dt,
2393 ValueType TyQ, ValueType TyD, SDNode OpNode>
2394 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2395 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2396 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2398 (TyQ (OpNode (TyD DPR:$Vn),
2399 (TyD (NEONvduplane (TyD DPR_VFP2:$Vm),imm:$lane)))))]>;
2400 class N3VLSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2401 InstrItinClass itin, string OpcodeStr, string Dt,
2402 ValueType TyQ, ValueType TyD, SDNode OpNode>
2403 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2404 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2405 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2407 (TyQ (OpNode (TyD DPR:$Vn),
2408 (TyD (NEONvduplane (TyD DPR_8:$Vm), imm:$lane)))))]>;
2410 // Long 3-register operations with explicitly extended operands.
2411 class N3VLExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2412 InstrItinClass itin, string OpcodeStr, string Dt,
2413 ValueType TyQ, ValueType TyD, SDNode OpNode, SDNode ExtOp,
2415 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2416 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2417 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2418 [(set QPR:$Vd, (OpNode (TyQ (ExtOp (TyD DPR:$Vn))),
2419 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2420 let isCommutable = Commutable;
2423 // Long 3-register intrinsics with explicit extend (VABDL).
2424 class N3VLIntExt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2425 InstrItinClass itin, string OpcodeStr, string Dt,
2426 ValueType TyQ, ValueType TyD, Intrinsic IntOp, SDNode ExtOp,
2428 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2429 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2430 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2431 [(set QPR:$Vd, (TyQ (ExtOp (TyD (IntOp (TyD DPR:$Vn),
2432 (TyD DPR:$Vm))))))]> {
2433 let isCommutable = Commutable;
2436 // Long 3-register intrinsics.
2437 class N3VLInt<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2438 InstrItinClass itin, string OpcodeStr, string Dt,
2439 ValueType TyQ, ValueType TyD, Intrinsic IntOp, bit Commutable>
2440 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2441 (outs QPR:$Vd), (ins DPR:$Vn, DPR:$Vm), N3RegFrm, itin,
2442 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2443 [(set QPR:$Vd, (TyQ (IntOp (TyD DPR:$Vn), (TyD DPR:$Vm))))]> {
2444 let isCommutable = Commutable;
2446 class N3VLIntSL<bit op24, bits<2> op21_20, bits<4> op11_8, InstrItinClass itin,
2447 string OpcodeStr, string Dt,
2448 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2449 : N3VLane32<op24, 1, op21_20, op11_8, 1, 0,
2450 (outs QPR:$Vd), (ins DPR:$Vn, DPR_VFP2:$Vm, VectorIndex32:$lane),
2451 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2452 [(set (ResTy QPR:$Vd),
2453 (ResTy (IntOp (OpTy DPR:$Vn),
2454 (OpTy (NEONvduplane (OpTy DPR_VFP2:$Vm),
2456 class N3VLIntSL16<bit op24, bits<2> op21_20, bits<4> op11_8,
2457 InstrItinClass itin, string OpcodeStr, string Dt,
2458 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2459 : N3VLane16<op24, 1, op21_20, op11_8, 1, 0,
2460 (outs QPR:$Vd), (ins DPR:$Vn, DPR_8:$Vm, VectorIndex16:$lane),
2461 NVMulSLFrm, itin, OpcodeStr, Dt, "$Vd, $Vn, $Vm$lane", "",
2462 [(set (ResTy QPR:$Vd),
2463 (ResTy (IntOp (OpTy DPR:$Vn),
2464 (OpTy (NEONvduplane (OpTy DPR_8:$Vm),
2467 // Wide 3-register operations.
2468 class N3VW<bit op24, bit op23, bits<2> op21_20, bits<4> op11_8, bit op4,
2469 string OpcodeStr, string Dt, ValueType TyQ, ValueType TyD,
2470 SDNode OpNode, SDNode ExtOp, bit Commutable>
2471 : N3V<op24, op23, op21_20, op11_8, 0, op4,
2472 (outs QPR:$Vd), (ins QPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VSUBiD,
2473 OpcodeStr, Dt, "$Vd, $Vn, $Vm", "",
2474 [(set QPR:$Vd, (OpNode (TyQ QPR:$Vn),
2475 (TyQ (ExtOp (TyD DPR:$Vm)))))]> {
2476 let isCommutable = Commutable;
2479 // Pairwise long 2-register intrinsics, both double- and quad-register.
2480 class N2VDPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2481 bits<2> op17_16, bits<5> op11_7, bit op4,
2482 string OpcodeStr, string Dt,
2483 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2484 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4, (outs DPR:$Vd),
2485 (ins DPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2486 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm))))]>;
2487 class N2VQPLInt<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2488 bits<2> op17_16, bits<5> op11_7, bit op4,
2489 string OpcodeStr, string Dt,
2490 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2491 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4, (outs QPR:$Vd),
2492 (ins QPR:$Vm), IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
2493 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm))))]>;
2495 // Pairwise long 2-register accumulate intrinsics,
2496 // both double- and quad-register.
2497 // The destination register is also used as the first source operand register.
2498 class N2VDPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2499 bits<2> op17_16, bits<5> op11_7, bit op4,
2500 string OpcodeStr, string Dt,
2501 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2502 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 0, op4,
2503 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vm), IIC_VPALiD,
2504 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2505 [(set DPR:$Vd, (ResTy (IntOp (ResTy DPR:$src1), (OpTy DPR:$Vm))))]>;
2506 class N2VQPLInt2<bits<2> op24_23, bits<2> op21_20, bits<2> op19_18,
2507 bits<2> op17_16, bits<5> op11_7, bit op4,
2508 string OpcodeStr, string Dt,
2509 ValueType ResTy, ValueType OpTy, Intrinsic IntOp>
2510 : N2V<op24_23, op21_20, op19_18, op17_16, op11_7, 1, op4,
2511 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vm), IIC_VPALiQ,
2512 OpcodeStr, Dt, "$Vd, $Vm", "$src1 = $Vd",
2513 [(set QPR:$Vd, (ResTy (IntOp (ResTy QPR:$src1), (OpTy QPR:$Vm))))]>;
2515 // Shift by immediate,
2516 // both double- and quad-register.
2517 class N2VDSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2518 Format f, InstrItinClass itin, Operand ImmTy,
2519 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2520 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2521 (outs DPR:$Vd), (ins DPR:$Vm, ImmTy:$SIMM), f, itin,
2522 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2523 [(set DPR:$Vd, (Ty (OpNode (Ty DPR:$Vm), (i32 imm:$SIMM))))]>;
2524 class N2VQSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2525 Format f, InstrItinClass itin, Operand ImmTy,
2526 string OpcodeStr, string Dt, ValueType Ty, SDNode OpNode>
2527 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2528 (outs QPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), f, itin,
2529 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2530 [(set QPR:$Vd, (Ty (OpNode (Ty QPR:$Vm), (i32 imm:$SIMM))))]>;
2532 // Long shift by immediate.
2533 class N2VLSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2534 string OpcodeStr, string Dt,
2535 ValueType ResTy, ValueType OpTy, SDNode OpNode>
2536 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2537 (outs QPR:$Vd), (ins DPR:$Vm, i32imm:$SIMM), N2RegVShLFrm,
2538 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2539 [(set QPR:$Vd, (ResTy (OpNode (OpTy DPR:$Vm),
2540 (i32 imm:$SIMM))))]>;
2542 // Narrow shift by immediate.
2543 class N2VNSh<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6, bit op4,
2544 InstrItinClass itin, string OpcodeStr, string Dt,
2545 ValueType ResTy, ValueType OpTy, Operand ImmTy, SDNode OpNode>
2546 : N2VImm<op24, op23, op11_8, op7, op6, op4,
2547 (outs DPR:$Vd), (ins QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, itin,
2548 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2549 [(set DPR:$Vd, (ResTy (OpNode (OpTy QPR:$Vm),
2550 (i32 imm:$SIMM))))]>;
2552 // Shift right by immediate and accumulate,
2553 // both double- and quad-register.
2554 class N2VDShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2555 Operand ImmTy, string OpcodeStr, string Dt,
2556 ValueType Ty, SDNode ShOp>
2557 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2558 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2559 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2560 [(set DPR:$Vd, (Ty (add DPR:$src1,
2561 (Ty (ShOp DPR:$Vm, (i32 imm:$SIMM))))))]>;
2562 class N2VQShAdd<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2563 Operand ImmTy, string OpcodeStr, string Dt,
2564 ValueType Ty, SDNode ShOp>
2565 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2566 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), N2RegVShRFrm, IIC_VPALiD,
2567 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2568 [(set QPR:$Vd, (Ty (add QPR:$src1,
2569 (Ty (ShOp QPR:$Vm, (i32 imm:$SIMM))))))]>;
2571 // Shift by immediate and insert,
2572 // both double- and quad-register.
2573 class N2VDShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2574 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2575 ValueType Ty,SDNode ShOp>
2576 : N2VImm<op24, op23, op11_8, op7, 0, op4, (outs DPR:$Vd),
2577 (ins DPR:$src1, DPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiD,
2578 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2579 [(set DPR:$Vd, (Ty (ShOp DPR:$src1, DPR:$Vm, (i32 imm:$SIMM))))]>;
2580 class N2VQShIns<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2581 Operand ImmTy, Format f, string OpcodeStr, string Dt,
2582 ValueType Ty,SDNode ShOp>
2583 : N2VImm<op24, op23, op11_8, op7, 1, op4, (outs QPR:$Vd),
2584 (ins QPR:$src1, QPR:$Vm, ImmTy:$SIMM), f, IIC_VSHLiQ,
2585 OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "$src1 = $Vd",
2586 [(set QPR:$Vd, (Ty (ShOp QPR:$src1, QPR:$Vm, (i32 imm:$SIMM))))]>;
2588 // Convert, with fractional bits immediate,
2589 // both double- and quad-register.
2590 class N2VCvtD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2591 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2593 : N2VImm<op24, op23, op11_8, op7, 0, op4,
2594 (outs DPR:$Vd), (ins DPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2595 IIC_VUNAD, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2596 [(set DPR:$Vd, (ResTy (IntOp (OpTy DPR:$Vm), (i32 imm:$SIMM))))]>;
2597 class N2VCvtQ<bit op24, bit op23, bits<4> op11_8, bit op7, bit op4,
2598 string OpcodeStr, string Dt, ValueType ResTy, ValueType OpTy,
2600 : N2VImm<op24, op23, op11_8, op7, 1, op4,
2601 (outs QPR:$Vd), (ins QPR:$Vm, neon_vcvt_imm32:$SIMM), NVCVTFrm,
2602 IIC_VUNAQ, OpcodeStr, Dt, "$Vd, $Vm, $SIMM", "",
2603 [(set QPR:$Vd, (ResTy (IntOp (OpTy QPR:$Vm), (i32 imm:$SIMM))))]>;
2605 //===----------------------------------------------------------------------===//
2607 //===----------------------------------------------------------------------===//
2609 // Abbreviations used in multiclass suffixes:
2610 // Q = quarter int (8 bit) elements
2611 // H = half int (16 bit) elements
2612 // S = single int (32 bit) elements
2613 // D = double int (64 bit) elements
2615 // Neon 2-register vector operations and intrinsics.
2617 // Neon 2-register comparisons.
2618 // source operand element sizes of 8, 16 and 32 bits:
2619 multiclass N2V_QHS_cmp<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2620 bits<5> op11_7, bit op4, string opc, string Dt,
2621 string asm, SDNode OpNode> {
2622 // 64-bit vector types.
2623 def v8i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 0, op4,
2624 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2625 opc, !strconcat(Dt, "8"), asm, "",
2626 [(set DPR:$Vd, (v8i8 (OpNode (v8i8 DPR:$Vm))))]>;
2627 def v4i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 0, op4,
2628 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2629 opc, !strconcat(Dt, "16"), asm, "",
2630 [(set DPR:$Vd, (v4i16 (OpNode (v4i16 DPR:$Vm))))]>;
2631 def v2i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2632 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2633 opc, !strconcat(Dt, "32"), asm, "",
2634 [(set DPR:$Vd, (v2i32 (OpNode (v2i32 DPR:$Vm))))]>;
2635 def v2f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 0, op4,
2636 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
2637 opc, "f32", asm, "",
2638 [(set DPR:$Vd, (v2i32 (OpNode (v2f32 DPR:$Vm))))]> {
2639 let Inst{10} = 1; // overwrite F = 1
2642 // 128-bit vector types.
2643 def v16i8 : N2V<op24_23, op21_20, 0b00, op17_16, op11_7, 1, op4,
2644 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2645 opc, !strconcat(Dt, "8"), asm, "",
2646 [(set QPR:$Vd, (v16i8 (OpNode (v16i8 QPR:$Vm))))]>;
2647 def v8i16 : N2V<op24_23, op21_20, 0b01, op17_16, op11_7, 1, op4,
2648 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2649 opc, !strconcat(Dt, "16"), asm, "",
2650 [(set QPR:$Vd, (v8i16 (OpNode (v8i16 QPR:$Vm))))]>;
2651 def v4i32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2652 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2653 opc, !strconcat(Dt, "32"), asm, "",
2654 [(set QPR:$Vd, (v4i32 (OpNode (v4i32 QPR:$Vm))))]>;
2655 def v4f32 : N2V<op24_23, op21_20, 0b10, op17_16, op11_7, 1, op4,
2656 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
2657 opc, "f32", asm, "",
2658 [(set QPR:$Vd, (v4i32 (OpNode (v4f32 QPR:$Vm))))]> {
2659 let Inst{10} = 1; // overwrite F = 1
2664 // Neon 2-register vector intrinsics,
2665 // element sizes of 8, 16 and 32 bits:
2666 multiclass N2VInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2667 bits<5> op11_7, bit op4,
2668 InstrItinClass itinD, InstrItinClass itinQ,
2669 string OpcodeStr, string Dt, Intrinsic IntOp> {
2670 // 64-bit vector types.
2671 def v8i8 : N2VDInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2672 itinD, OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
2673 def v4i16 : N2VDInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2674 itinD, OpcodeStr, !strconcat(Dt, "16"),v4i16,v4i16,IntOp>;
2675 def v2i32 : N2VDInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2676 itinD, OpcodeStr, !strconcat(Dt, "32"),v2i32,v2i32,IntOp>;
2678 // 128-bit vector types.
2679 def v16i8 : N2VQInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
2680 itinQ, OpcodeStr, !strconcat(Dt, "8"), v16i8,v16i8,IntOp>;
2681 def v8i16 : N2VQInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
2682 itinQ, OpcodeStr, !strconcat(Dt, "16"),v8i16,v8i16,IntOp>;
2683 def v4i32 : N2VQInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
2684 itinQ, OpcodeStr, !strconcat(Dt, "32"),v4i32,v4i32,IntOp>;
2688 // Neon Narrowing 2-register vector operations,
2689 // source operand element sizes of 16, 32 and 64 bits:
2690 multiclass N2VN_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2691 bits<5> op11_7, bit op6, bit op4,
2692 InstrItinClass itin, string OpcodeStr, string Dt,
2694 def v8i8 : N2VN<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2695 itin, OpcodeStr, !strconcat(Dt, "16"),
2696 v8i8, v8i16, OpNode>;
2697 def v4i16 : N2VN<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2698 itin, OpcodeStr, !strconcat(Dt, "32"),
2699 v4i16, v4i32, OpNode>;
2700 def v2i32 : N2VN<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2701 itin, OpcodeStr, !strconcat(Dt, "64"),
2702 v2i32, v2i64, OpNode>;
2705 // Neon Narrowing 2-register vector intrinsics,
2706 // source operand element sizes of 16, 32 and 64 bits:
2707 multiclass N2VNInt_HSD<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
2708 bits<5> op11_7, bit op6, bit op4,
2709 InstrItinClass itin, string OpcodeStr, string Dt,
2711 def v8i8 : N2VNInt<op24_23, op21_20, 0b00, op17_16, op11_7, op6, op4,
2712 itin, OpcodeStr, !strconcat(Dt, "16"),
2713 v8i8, v8i16, IntOp>;
2714 def v4i16 : N2VNInt<op24_23, op21_20, 0b01, op17_16, op11_7, op6, op4,
2715 itin, OpcodeStr, !strconcat(Dt, "32"),
2716 v4i16, v4i32, IntOp>;
2717 def v2i32 : N2VNInt<op24_23, op21_20, 0b10, op17_16, op11_7, op6, op4,
2718 itin, OpcodeStr, !strconcat(Dt, "64"),
2719 v2i32, v2i64, IntOp>;
2723 // Neon Lengthening 2-register vector intrinsic (currently specific to VMOVL).
2724 // source operand element sizes of 16, 32 and 64 bits:
2725 multiclass N2VL_QHS<bits<2> op24_23, bits<5> op11_7, bit op6, bit op4,
2726 string OpcodeStr, string Dt, SDNode OpNode> {
2727 def v8i16 : N2VL<op24_23, 0b00, 0b10, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2728 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode>;
2729 def v4i32 : N2VL<op24_23, 0b01, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2730 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2731 def v2i64 : N2VL<op24_23, 0b10, 0b00, 0b00, op11_7, op6, op4, IIC_VQUNAiD,
2732 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2736 // Neon 3-register vector operations.
2738 // First with only element sizes of 8, 16 and 32 bits:
2739 multiclass N3V_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2740 InstrItinClass itinD16, InstrItinClass itinD32,
2741 InstrItinClass itinQ16, InstrItinClass itinQ32,
2742 string OpcodeStr, string Dt,
2743 SDNode OpNode, bit Commutable = 0> {
2744 // 64-bit vector types.
2745 def v8i8 : N3VD<op24, op23, 0b00, op11_8, op4, itinD16,
2746 OpcodeStr, !strconcat(Dt, "8"),
2747 v8i8, v8i8, OpNode, Commutable>;
2748 def v4i16 : N3VD<op24, op23, 0b01, op11_8, op4, itinD16,
2749 OpcodeStr, !strconcat(Dt, "16"),
2750 v4i16, v4i16, OpNode, Commutable>;
2751 def v2i32 : N3VD<op24, op23, 0b10, op11_8, op4, itinD32,
2752 OpcodeStr, !strconcat(Dt, "32"),
2753 v2i32, v2i32, OpNode, Commutable>;
2755 // 128-bit vector types.
2756 def v16i8 : N3VQ<op24, op23, 0b00, op11_8, op4, itinQ16,
2757 OpcodeStr, !strconcat(Dt, "8"),
2758 v16i8, v16i8, OpNode, Commutable>;
2759 def v8i16 : N3VQ<op24, op23, 0b01, op11_8, op4, itinQ16,
2760 OpcodeStr, !strconcat(Dt, "16"),
2761 v8i16, v8i16, OpNode, Commutable>;
2762 def v4i32 : N3VQ<op24, op23, 0b10, op11_8, op4, itinQ32,
2763 OpcodeStr, !strconcat(Dt, "32"),
2764 v4i32, v4i32, OpNode, Commutable>;
2767 multiclass N3VSL_HS<bits<4> op11_8, string OpcodeStr, string Dt, SDNode ShOp> {
2768 def v4i16 : N3VDSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2770 def v2i32 : N3VDSL<0b10, op11_8, IIC_VMULi32D, OpcodeStr, !strconcat(Dt,"32"),
2772 def v8i16 : N3VQSL16<0b01, op11_8, OpcodeStr, !strconcat(Dt, "16"),
2773 v8i16, v4i16, ShOp>;
2774 def v4i32 : N3VQSL<0b10, op11_8, IIC_VMULi32Q, OpcodeStr, !strconcat(Dt,"32"),
2775 v4i32, v2i32, ShOp>;
2778 // ....then also with element size 64 bits:
2779 multiclass N3V_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2780 InstrItinClass itinD, InstrItinClass itinQ,
2781 string OpcodeStr, string Dt,
2782 SDNode OpNode, bit Commutable = 0>
2783 : N3V_QHS<op24, op23, op11_8, op4, itinD, itinD, itinQ, itinQ,
2784 OpcodeStr, Dt, OpNode, Commutable> {
2785 def v1i64 : N3VD<op24, op23, 0b11, op11_8, op4, itinD,
2786 OpcodeStr, !strconcat(Dt, "64"),
2787 v1i64, v1i64, OpNode, Commutable>;
2788 def v2i64 : N3VQ<op24, op23, 0b11, op11_8, op4, itinQ,
2789 OpcodeStr, !strconcat(Dt, "64"),
2790 v2i64, v2i64, OpNode, Commutable>;
2794 // Neon 3-register vector intrinsics.
2796 // First with only element sizes of 16 and 32 bits:
2797 multiclass N3VInt_HS<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,
2801 Intrinsic IntOp, bit Commutable = 0> {
2802 // 64-bit vector types.
2803 def v4i16 : N3VDInt<op24, op23, 0b01, op11_8, op4, f, itinD16,
2804 OpcodeStr, !strconcat(Dt, "16"),
2805 v4i16, v4i16, IntOp, Commutable>;
2806 def v2i32 : N3VDInt<op24, op23, 0b10, op11_8, op4, f, itinD32,
2807 OpcodeStr, !strconcat(Dt, "32"),
2808 v2i32, v2i32, IntOp, Commutable>;
2810 // 128-bit vector types.
2811 def v8i16 : N3VQInt<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2812 OpcodeStr, !strconcat(Dt, "16"),
2813 v8i16, v8i16, IntOp, Commutable>;
2814 def v4i32 : N3VQInt<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2815 OpcodeStr, !strconcat(Dt, "32"),
2816 v4i32, v4i32, IntOp, Commutable>;
2818 multiclass N3VInt_HSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2819 InstrItinClass itinD16, InstrItinClass itinD32,
2820 InstrItinClass itinQ16, InstrItinClass itinQ32,
2821 string OpcodeStr, string Dt,
2823 // 64-bit vector types.
2824 def v4i16 : N3VDIntSh<op24, op23, 0b01, op11_8, op4, f, itinD16,
2825 OpcodeStr, !strconcat(Dt, "16"),
2826 v4i16, v4i16, IntOp>;
2827 def v2i32 : N3VDIntSh<op24, op23, 0b10, op11_8, op4, f, itinD32,
2828 OpcodeStr, !strconcat(Dt, "32"),
2829 v2i32, v2i32, IntOp>;
2831 // 128-bit vector types.
2832 def v8i16 : N3VQIntSh<op24, op23, 0b01, op11_8, op4, f, itinQ16,
2833 OpcodeStr, !strconcat(Dt, "16"),
2834 v8i16, v8i16, IntOp>;
2835 def v4i32 : N3VQIntSh<op24, op23, 0b10, op11_8, op4, f, itinQ32,
2836 OpcodeStr, !strconcat(Dt, "32"),
2837 v4i32, v4i32, IntOp>;
2840 multiclass N3VIntSL_HS<bits<4> op11_8,
2841 InstrItinClass itinD16, InstrItinClass itinD32,
2842 InstrItinClass itinQ16, InstrItinClass itinQ32,
2843 string OpcodeStr, string Dt, Intrinsic IntOp> {
2844 def v4i16 : N3VDIntSL16<0b01, op11_8, itinD16,
2845 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp>;
2846 def v2i32 : N3VDIntSL<0b10, op11_8, itinD32,
2847 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp>;
2848 def v8i16 : N3VQIntSL16<0b01, op11_8, itinQ16,
2849 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16, IntOp>;
2850 def v4i32 : N3VQIntSL<0b10, op11_8, itinQ32,
2851 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32, IntOp>;
2854 // ....then also with element size of 8 bits:
2855 multiclass N3VInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2856 InstrItinClass itinD16, InstrItinClass itinD32,
2857 InstrItinClass itinQ16, InstrItinClass itinQ32,
2858 string OpcodeStr, string Dt,
2859 Intrinsic IntOp, bit Commutable = 0>
2860 : N3VInt_HS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2861 OpcodeStr, Dt, IntOp, Commutable> {
2862 def v8i8 : N3VDInt<op24, op23, 0b00, op11_8, op4, f, itinD16,
2863 OpcodeStr, !strconcat(Dt, "8"),
2864 v8i8, v8i8, IntOp, Commutable>;
2865 def v16i8 : N3VQInt<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2866 OpcodeStr, !strconcat(Dt, "8"),
2867 v16i8, v16i8, IntOp, Commutable>;
2869 multiclass N3VInt_QHSSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2870 InstrItinClass itinD16, InstrItinClass itinD32,
2871 InstrItinClass itinQ16, InstrItinClass itinQ32,
2872 string OpcodeStr, string Dt,
2874 : N3VInt_HSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2875 OpcodeStr, Dt, IntOp> {
2876 def v8i8 : N3VDIntSh<op24, op23, 0b00, op11_8, op4, f, itinD16,
2877 OpcodeStr, !strconcat(Dt, "8"),
2879 def v16i8 : N3VQIntSh<op24, op23, 0b00, op11_8, op4, f, itinQ16,
2880 OpcodeStr, !strconcat(Dt, "8"),
2881 v16i8, v16i8, IntOp>;
2885 // ....then also with element size of 64 bits:
2886 multiclass N3VInt_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2887 InstrItinClass itinD16, InstrItinClass itinD32,
2888 InstrItinClass itinQ16, InstrItinClass itinQ32,
2889 string OpcodeStr, string Dt,
2890 Intrinsic IntOp, bit Commutable = 0>
2891 : N3VInt_QHS<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2892 OpcodeStr, Dt, IntOp, Commutable> {
2893 def v1i64 : N3VDInt<op24, op23, 0b11, op11_8, op4, f, itinD32,
2894 OpcodeStr, !strconcat(Dt, "64"),
2895 v1i64, v1i64, IntOp, Commutable>;
2896 def v2i64 : N3VQInt<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2897 OpcodeStr, !strconcat(Dt, "64"),
2898 v2i64, v2i64, IntOp, Commutable>;
2900 multiclass N3VInt_QHSDSh<bit op24, bit op23, bits<4> op11_8, bit op4, Format f,
2901 InstrItinClass itinD16, InstrItinClass itinD32,
2902 InstrItinClass itinQ16, InstrItinClass itinQ32,
2903 string OpcodeStr, string Dt,
2905 : N3VInt_QHSSh<op24, op23, op11_8, op4, f, itinD16, itinD32, itinQ16, itinQ32,
2906 OpcodeStr, Dt, IntOp> {
2907 def v1i64 : N3VDIntSh<op24, op23, 0b11, op11_8, op4, f, itinD32,
2908 OpcodeStr, !strconcat(Dt, "64"),
2909 v1i64, v1i64, IntOp>;
2910 def v2i64 : N3VQIntSh<op24, op23, 0b11, op11_8, op4, f, itinQ32,
2911 OpcodeStr, !strconcat(Dt, "64"),
2912 v2i64, v2i64, IntOp>;
2915 // Neon Narrowing 3-register vector intrinsics,
2916 // source operand element sizes of 16, 32 and 64 bits:
2917 multiclass N3VNInt_HSD<bit op24, bit op23, bits<4> op11_8, bit op4,
2918 string OpcodeStr, string Dt,
2919 Intrinsic IntOp, bit Commutable = 0> {
2920 def v8i8 : N3VNInt<op24, op23, 0b00, op11_8, op4,
2921 OpcodeStr, !strconcat(Dt, "16"),
2922 v8i8, v8i16, IntOp, Commutable>;
2923 def v4i16 : N3VNInt<op24, op23, 0b01, op11_8, op4,
2924 OpcodeStr, !strconcat(Dt, "32"),
2925 v4i16, v4i32, IntOp, Commutable>;
2926 def v2i32 : N3VNInt<op24, op23, 0b10, op11_8, op4,
2927 OpcodeStr, !strconcat(Dt, "64"),
2928 v2i32, v2i64, IntOp, Commutable>;
2932 // Neon Long 3-register vector operations.
2934 multiclass N3VL_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2935 InstrItinClass itin16, InstrItinClass itin32,
2936 string OpcodeStr, string Dt,
2937 SDNode OpNode, bit Commutable = 0> {
2938 def v8i16 : N3VL<op24, op23, 0b00, op11_8, op4, itin16,
2939 OpcodeStr, !strconcat(Dt, "8"),
2940 v8i16, v8i8, OpNode, Commutable>;
2941 def v4i32 : N3VL<op24, op23, 0b01, op11_8, op4, itin16,
2942 OpcodeStr, !strconcat(Dt, "16"),
2943 v4i32, v4i16, OpNode, Commutable>;
2944 def v2i64 : N3VL<op24, op23, 0b10, op11_8, op4, itin32,
2945 OpcodeStr, !strconcat(Dt, "32"),
2946 v2i64, v2i32, OpNode, Commutable>;
2949 multiclass N3VLSL_HS<bit op24, bits<4> op11_8,
2950 InstrItinClass itin, string OpcodeStr, string Dt,
2952 def v4i16 : N3VLSL16<op24, 0b01, op11_8, itin, OpcodeStr,
2953 !strconcat(Dt, "16"), v4i32, v4i16, OpNode>;
2954 def v2i32 : N3VLSL<op24, 0b10, op11_8, itin, OpcodeStr,
2955 !strconcat(Dt, "32"), v2i64, v2i32, OpNode>;
2958 multiclass N3VLExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2959 InstrItinClass itin16, InstrItinClass itin32,
2960 string OpcodeStr, string Dt,
2961 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
2962 def v8i16 : N3VLExt<op24, op23, 0b00, op11_8, op4, itin16,
2963 OpcodeStr, !strconcat(Dt, "8"),
2964 v8i16, v8i8, OpNode, ExtOp, Commutable>;
2965 def v4i32 : N3VLExt<op24, op23, 0b01, op11_8, op4, itin16,
2966 OpcodeStr, !strconcat(Dt, "16"),
2967 v4i32, v4i16, OpNode, ExtOp, Commutable>;
2968 def v2i64 : N3VLExt<op24, op23, 0b10, op11_8, op4, itin32,
2969 OpcodeStr, !strconcat(Dt, "32"),
2970 v2i64, v2i32, OpNode, ExtOp, Commutable>;
2973 // Neon Long 3-register vector intrinsics.
2975 // First with only element sizes of 16 and 32 bits:
2976 multiclass N3VLInt_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
2977 InstrItinClass itin16, InstrItinClass itin32,
2978 string OpcodeStr, string Dt,
2979 Intrinsic IntOp, bit Commutable = 0> {
2980 def v4i32 : N3VLInt<op24, op23, 0b01, op11_8, op4, itin16,
2981 OpcodeStr, !strconcat(Dt, "16"),
2982 v4i32, v4i16, IntOp, Commutable>;
2983 def v2i64 : N3VLInt<op24, op23, 0b10, op11_8, op4, itin32,
2984 OpcodeStr, !strconcat(Dt, "32"),
2985 v2i64, v2i32, IntOp, Commutable>;
2988 multiclass N3VLIntSL_HS<bit op24, bits<4> op11_8,
2989 InstrItinClass itin, string OpcodeStr, string Dt,
2991 def v4i16 : N3VLIntSL16<op24, 0b01, op11_8, itin,
2992 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
2993 def v2i32 : N3VLIntSL<op24, 0b10, op11_8, itin,
2994 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
2997 // ....then also with element size of 8 bits:
2998 multiclass N3VLInt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
2999 InstrItinClass itin16, InstrItinClass itin32,
3000 string OpcodeStr, string Dt,
3001 Intrinsic IntOp, bit Commutable = 0>
3002 : N3VLInt_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt,
3003 IntOp, Commutable> {
3004 def v8i16 : N3VLInt<op24, op23, 0b00, op11_8, op4, itin16,
3005 OpcodeStr, !strconcat(Dt, "8"),
3006 v8i16, v8i8, IntOp, Commutable>;
3009 // ....with explicit extend (VABDL).
3010 multiclass N3VLIntExt_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3011 InstrItinClass itin, string OpcodeStr, string Dt,
3012 Intrinsic IntOp, SDNode ExtOp, bit Commutable = 0> {
3013 def v8i16 : N3VLIntExt<op24, op23, 0b00, op11_8, op4, itin,
3014 OpcodeStr, !strconcat(Dt, "8"),
3015 v8i16, v8i8, IntOp, ExtOp, Commutable>;
3016 def v4i32 : N3VLIntExt<op24, op23, 0b01, op11_8, op4, itin,
3017 OpcodeStr, !strconcat(Dt, "16"),
3018 v4i32, v4i16, IntOp, ExtOp, Commutable>;
3019 def v2i64 : N3VLIntExt<op24, op23, 0b10, op11_8, op4, itin,
3020 OpcodeStr, !strconcat(Dt, "32"),
3021 v2i64, v2i32, IntOp, ExtOp, Commutable>;
3025 // Neon Wide 3-register vector intrinsics,
3026 // source operand element sizes of 8, 16 and 32 bits:
3027 multiclass N3VW_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3028 string OpcodeStr, string Dt,
3029 SDNode OpNode, SDNode ExtOp, bit Commutable = 0> {
3030 def v8i16 : N3VW<op24, op23, 0b00, op11_8, op4,
3031 OpcodeStr, !strconcat(Dt, "8"),
3032 v8i16, v8i8, OpNode, ExtOp, Commutable>;
3033 def v4i32 : N3VW<op24, op23, 0b01, op11_8, op4,
3034 OpcodeStr, !strconcat(Dt, "16"),
3035 v4i32, v4i16, OpNode, ExtOp, Commutable>;
3036 def v2i64 : N3VW<op24, op23, 0b10, op11_8, op4,
3037 OpcodeStr, !strconcat(Dt, "32"),
3038 v2i64, v2i32, OpNode, ExtOp, Commutable>;
3042 // Neon Multiply-Op vector operations,
3043 // element sizes of 8, 16 and 32 bits:
3044 multiclass N3VMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3045 InstrItinClass itinD16, InstrItinClass itinD32,
3046 InstrItinClass itinQ16, InstrItinClass itinQ32,
3047 string OpcodeStr, string Dt, SDNode OpNode> {
3048 // 64-bit vector types.
3049 def v8i8 : N3VDMulOp<op24, op23, 0b00, op11_8, op4, itinD16,
3050 OpcodeStr, !strconcat(Dt, "8"), v8i8, mul, OpNode>;
3051 def v4i16 : N3VDMulOp<op24, op23, 0b01, op11_8, op4, itinD16,
3052 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, OpNode>;
3053 def v2i32 : N3VDMulOp<op24, op23, 0b10, op11_8, op4, itinD32,
3054 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, OpNode>;
3056 // 128-bit vector types.
3057 def v16i8 : N3VQMulOp<op24, op23, 0b00, op11_8, op4, itinQ16,
3058 OpcodeStr, !strconcat(Dt, "8"), v16i8, mul, OpNode>;
3059 def v8i16 : N3VQMulOp<op24, op23, 0b01, op11_8, op4, itinQ16,
3060 OpcodeStr, !strconcat(Dt, "16"), v8i16, mul, OpNode>;
3061 def v4i32 : N3VQMulOp<op24, op23, 0b10, op11_8, op4, itinQ32,
3062 OpcodeStr, !strconcat(Dt, "32"), v4i32, mul, OpNode>;
3065 multiclass N3VMulOpSL_HS<bits<4> op11_8,
3066 InstrItinClass itinD16, InstrItinClass itinD32,
3067 InstrItinClass itinQ16, InstrItinClass itinQ32,
3068 string OpcodeStr, string Dt, SDNode ShOp> {
3069 def v4i16 : N3VDMulOpSL16<0b01, op11_8, itinD16,
3070 OpcodeStr, !strconcat(Dt, "16"), v4i16, mul, ShOp>;
3071 def v2i32 : N3VDMulOpSL<0b10, op11_8, itinD32,
3072 OpcodeStr, !strconcat(Dt, "32"), v2i32, mul, ShOp>;
3073 def v8i16 : N3VQMulOpSL16<0b01, op11_8, itinQ16,
3074 OpcodeStr, !strconcat(Dt, "16"), v8i16, v4i16,
3076 def v4i32 : N3VQMulOpSL<0b10, op11_8, itinQ32,
3077 OpcodeStr, !strconcat(Dt, "32"), v4i32, v2i32,
3081 // Neon Intrinsic-Op vector operations,
3082 // element sizes of 8, 16 and 32 bits:
3083 multiclass N3VIntOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3084 InstrItinClass itinD, InstrItinClass itinQ,
3085 string OpcodeStr, string Dt, Intrinsic IntOp,
3087 // 64-bit vector types.
3088 def v8i8 : N3VDIntOp<op24, op23, 0b00, op11_8, op4, itinD,
3089 OpcodeStr, !strconcat(Dt, "8"), v8i8, IntOp, OpNode>;
3090 def v4i16 : N3VDIntOp<op24, op23, 0b01, op11_8, op4, itinD,
3091 OpcodeStr, !strconcat(Dt, "16"), v4i16, IntOp, OpNode>;
3092 def v2i32 : N3VDIntOp<op24, op23, 0b10, op11_8, op4, itinD,
3093 OpcodeStr, !strconcat(Dt, "32"), v2i32, IntOp, OpNode>;
3095 // 128-bit vector types.
3096 def v16i8 : N3VQIntOp<op24, op23, 0b00, op11_8, op4, itinQ,
3097 OpcodeStr, !strconcat(Dt, "8"), v16i8, IntOp, OpNode>;
3098 def v8i16 : N3VQIntOp<op24, op23, 0b01, op11_8, op4, itinQ,
3099 OpcodeStr, !strconcat(Dt, "16"), v8i16, IntOp, OpNode>;
3100 def v4i32 : N3VQIntOp<op24, op23, 0b10, op11_8, op4, itinQ,
3101 OpcodeStr, !strconcat(Dt, "32"), v4i32, IntOp, OpNode>;
3104 // Neon 3-argument intrinsics,
3105 // element sizes of 8, 16 and 32 bits:
3106 multiclass N3VInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3107 InstrItinClass itinD, InstrItinClass itinQ,
3108 string OpcodeStr, string Dt, Intrinsic IntOp> {
3109 // 64-bit vector types.
3110 def v8i8 : N3VDInt3<op24, op23, 0b00, op11_8, op4, itinD,
3111 OpcodeStr, !strconcat(Dt, "8"), v8i8, v8i8, IntOp>;
3112 def v4i16 : N3VDInt3<op24, op23, 0b01, op11_8, op4, itinD,
3113 OpcodeStr, !strconcat(Dt, "16"), v4i16, v4i16, IntOp>;
3114 def v2i32 : N3VDInt3<op24, op23, 0b10, op11_8, op4, itinD,
3115 OpcodeStr, !strconcat(Dt, "32"), v2i32, v2i32, IntOp>;
3117 // 128-bit vector types.
3118 def v16i8 : N3VQInt3<op24, op23, 0b00, op11_8, op4, itinQ,
3119 OpcodeStr, !strconcat(Dt, "8"), v16i8, v16i8, IntOp>;
3120 def v8i16 : N3VQInt3<op24, op23, 0b01, op11_8, op4, itinQ,
3121 OpcodeStr, !strconcat(Dt, "16"), v8i16, v8i16, IntOp>;
3122 def v4i32 : N3VQInt3<op24, op23, 0b10, op11_8, op4, itinQ,
3123 OpcodeStr, !strconcat(Dt, "32"), v4i32, v4i32, IntOp>;
3127 // Neon Long Multiply-Op vector operations,
3128 // element sizes of 8, 16 and 32 bits:
3129 multiclass N3VLMulOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3130 InstrItinClass itin16, InstrItinClass itin32,
3131 string OpcodeStr, string Dt, SDNode MulOp,
3133 def v8i16 : N3VLMulOp<op24, op23, 0b00, op11_8, op4, itin16, OpcodeStr,
3134 !strconcat(Dt, "8"), v8i16, v8i8, MulOp, OpNode>;
3135 def v4i32 : N3VLMulOp<op24, op23, 0b01, op11_8, op4, itin16, OpcodeStr,
3136 !strconcat(Dt, "16"), v4i32, v4i16, MulOp, OpNode>;
3137 def v2i64 : N3VLMulOp<op24, op23, 0b10, op11_8, op4, itin32, OpcodeStr,
3138 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3141 multiclass N3VLMulOpSL_HS<bit op24, bits<4> op11_8, string OpcodeStr,
3142 string Dt, SDNode MulOp, SDNode OpNode> {
3143 def v4i16 : N3VLMulOpSL16<op24, 0b01, op11_8, IIC_VMACi16D, OpcodeStr,
3144 !strconcat(Dt,"16"), v4i32, v4i16, MulOp, OpNode>;
3145 def v2i32 : N3VLMulOpSL<op24, 0b10, op11_8, IIC_VMACi32D, OpcodeStr,
3146 !strconcat(Dt, "32"), v2i64, v2i32, MulOp, OpNode>;
3150 // Neon Long 3-argument intrinsics.
3152 // First with only element sizes of 16 and 32 bits:
3153 multiclass N3VLInt3_HS<bit op24, bit op23, bits<4> op11_8, bit op4,
3154 InstrItinClass itin16, InstrItinClass itin32,
3155 string OpcodeStr, string Dt, Intrinsic IntOp> {
3156 def v4i32 : N3VLInt3<op24, op23, 0b01, op11_8, op4, itin16,
3157 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, IntOp>;
3158 def v2i64 : N3VLInt3<op24, op23, 0b10, op11_8, op4, itin32,
3159 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3162 multiclass N3VLInt3SL_HS<bit op24, bits<4> op11_8,
3163 string OpcodeStr, string Dt, Intrinsic IntOp> {
3164 def v4i16 : N3VLInt3SL16<op24, 0b01, op11_8, IIC_VMACi16D,
3165 OpcodeStr, !strconcat(Dt,"16"), v4i32, v4i16, IntOp>;
3166 def v2i32 : N3VLInt3SL<op24, 0b10, op11_8, IIC_VMACi32D,
3167 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, IntOp>;
3170 // ....then also with element size of 8 bits:
3171 multiclass N3VLInt3_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3172 InstrItinClass itin16, InstrItinClass itin32,
3173 string OpcodeStr, string Dt, Intrinsic IntOp>
3174 : N3VLInt3_HS<op24, op23, op11_8, op4, itin16, itin32, OpcodeStr, Dt, IntOp> {
3175 def v8i16 : N3VLInt3<op24, op23, 0b00, op11_8, op4, itin16,
3176 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, IntOp>;
3179 // ....with explicit extend (VABAL).
3180 multiclass N3VLIntExtOp_QHS<bit op24, bit op23, bits<4> op11_8, bit op4,
3181 InstrItinClass itin, string OpcodeStr, string Dt,
3182 Intrinsic IntOp, SDNode ExtOp, SDNode OpNode> {
3183 def v8i16 : N3VLIntExtOp<op24, op23, 0b00, op11_8, op4, itin,
3184 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8,
3185 IntOp, ExtOp, OpNode>;
3186 def v4i32 : N3VLIntExtOp<op24, op23, 0b01, op11_8, op4, itin,
3187 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16,
3188 IntOp, ExtOp, OpNode>;
3189 def v2i64 : N3VLIntExtOp<op24, op23, 0b10, op11_8, op4, itin,
3190 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32,
3191 IntOp, ExtOp, OpNode>;
3195 // Neon Pairwise long 2-register intrinsics,
3196 // element sizes of 8, 16 and 32 bits:
3197 multiclass N2VPLInt_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3198 bits<5> op11_7, bit op4,
3199 string OpcodeStr, string Dt, Intrinsic IntOp> {
3200 // 64-bit vector types.
3201 def v8i8 : N2VDPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3202 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3203 def v4i16 : N2VDPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3204 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3205 def v2i32 : N2VDPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3206 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3208 // 128-bit vector types.
3209 def v16i8 : N2VQPLInt<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3210 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3211 def v8i16 : N2VQPLInt<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3212 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3213 def v4i32 : N2VQPLInt<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3214 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3218 // Neon Pairwise long 2-register accumulate intrinsics,
3219 // element sizes of 8, 16 and 32 bits:
3220 multiclass N2VPLInt2_QHS<bits<2> op24_23, bits<2> op21_20, bits<2> op17_16,
3221 bits<5> op11_7, bit op4,
3222 string OpcodeStr, string Dt, Intrinsic IntOp> {
3223 // 64-bit vector types.
3224 def v8i8 : N2VDPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3225 OpcodeStr, !strconcat(Dt, "8"), v4i16, v8i8, IntOp>;
3226 def v4i16 : N2VDPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3227 OpcodeStr, !strconcat(Dt, "16"), v2i32, v4i16, IntOp>;
3228 def v2i32 : N2VDPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3229 OpcodeStr, !strconcat(Dt, "32"), v1i64, v2i32, IntOp>;
3231 // 128-bit vector types.
3232 def v16i8 : N2VQPLInt2<op24_23, op21_20, 0b00, op17_16, op11_7, op4,
3233 OpcodeStr, !strconcat(Dt, "8"), v8i16, v16i8, IntOp>;
3234 def v8i16 : N2VQPLInt2<op24_23, op21_20, 0b01, op17_16, op11_7, op4,
3235 OpcodeStr, !strconcat(Dt, "16"), v4i32, v8i16, IntOp>;
3236 def v4i32 : N2VQPLInt2<op24_23, op21_20, 0b10, op17_16, op11_7, op4,
3237 OpcodeStr, !strconcat(Dt, "32"), v2i64, v4i32, IntOp>;
3241 // Neon 2-register vector shift by immediate,
3242 // with f of either N2RegVShLFrm or N2RegVShRFrm
3243 // element sizes of 8, 16, 32 and 64 bits:
3244 multiclass N2VShL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3245 InstrItinClass itin, string OpcodeStr, string Dt,
3247 // 64-bit vector types.
3248 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3249 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3250 let Inst{21-19} = 0b001; // imm6 = 001xxx
3252 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3253 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3254 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3256 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3257 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3258 let Inst{21} = 0b1; // imm6 = 1xxxxx
3260 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3261 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3264 // 128-bit vector types.
3265 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3266 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3267 let Inst{21-19} = 0b001; // imm6 = 001xxx
3269 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3270 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3271 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3273 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShLFrm, itin, i32imm,
3274 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3275 let Inst{21} = 0b1; // imm6 = 1xxxxx
3277 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShLFrm, itin, i32imm,
3278 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3281 multiclass N2VShR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3282 InstrItinClass itin, string OpcodeStr, string Dt,
3284 // 64-bit vector types.
3285 def v8i8 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3286 OpcodeStr, !strconcat(Dt, "8"), v8i8, OpNode> {
3287 let Inst{21-19} = 0b001; // imm6 = 001xxx
3289 def v4i16 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3290 OpcodeStr, !strconcat(Dt, "16"), v4i16, OpNode> {
3291 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3293 def v2i32 : N2VDSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3294 OpcodeStr, !strconcat(Dt, "32"), v2i32, OpNode> {
3295 let Inst{21} = 0b1; // imm6 = 1xxxxx
3297 def v1i64 : N2VDSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3298 OpcodeStr, !strconcat(Dt, "64"), v1i64, OpNode>;
3301 // 128-bit vector types.
3302 def v16i8 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm8,
3303 OpcodeStr, !strconcat(Dt, "8"), v16i8, OpNode> {
3304 let Inst{21-19} = 0b001; // imm6 = 001xxx
3306 def v8i16 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm16,
3307 OpcodeStr, !strconcat(Dt, "16"), v8i16, OpNode> {
3308 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3310 def v4i32 : N2VQSh<op24, op23, op11_8, 0, op4, N2RegVShRFrm, itin, shr_imm32,
3311 OpcodeStr, !strconcat(Dt, "32"), v4i32, OpNode> {
3312 let Inst{21} = 0b1; // imm6 = 1xxxxx
3314 def v2i64 : N2VQSh<op24, op23, op11_8, 1, op4, N2RegVShRFrm, itin, shr_imm64,
3315 OpcodeStr, !strconcat(Dt, "64"), v2i64, OpNode>;
3319 // Neon Shift-Accumulate vector operations,
3320 // element sizes of 8, 16, 32 and 64 bits:
3321 multiclass N2VShAdd_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3322 string OpcodeStr, string Dt, SDNode ShOp> {
3323 // 64-bit vector types.
3324 def v8i8 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3325 OpcodeStr, !strconcat(Dt, "8"), v8i8, ShOp> {
3326 let Inst{21-19} = 0b001; // imm6 = 001xxx
3328 def v4i16 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3329 OpcodeStr, !strconcat(Dt, "16"), v4i16, ShOp> {
3330 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3332 def v2i32 : N2VDShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3333 OpcodeStr, !strconcat(Dt, "32"), v2i32, ShOp> {
3334 let Inst{21} = 0b1; // imm6 = 1xxxxx
3336 def v1i64 : N2VDShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3337 OpcodeStr, !strconcat(Dt, "64"), v1i64, ShOp>;
3340 // 128-bit vector types.
3341 def v16i8 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm8,
3342 OpcodeStr, !strconcat(Dt, "8"), v16i8, ShOp> {
3343 let Inst{21-19} = 0b001; // imm6 = 001xxx
3345 def v8i16 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm16,
3346 OpcodeStr, !strconcat(Dt, "16"), v8i16, ShOp> {
3347 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3349 def v4i32 : N2VQShAdd<op24, op23, op11_8, 0, op4, shr_imm32,
3350 OpcodeStr, !strconcat(Dt, "32"), v4i32, ShOp> {
3351 let Inst{21} = 0b1; // imm6 = 1xxxxx
3353 def v2i64 : N2VQShAdd<op24, op23, op11_8, 1, op4, shr_imm64,
3354 OpcodeStr, !strconcat(Dt, "64"), v2i64, ShOp>;
3358 // Neon Shift-Insert vector operations,
3359 // with f of either N2RegVShLFrm or N2RegVShRFrm
3360 // element sizes of 8, 16, 32 and 64 bits:
3361 multiclass N2VShInsL_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3363 // 64-bit vector types.
3364 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3365 N2RegVShLFrm, OpcodeStr, "8", v8i8, NEONvsli> {
3366 let Inst{21-19} = 0b001; // imm6 = 001xxx
3368 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3369 N2RegVShLFrm, OpcodeStr, "16", v4i16, NEONvsli> {
3370 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3372 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, i32imm,
3373 N2RegVShLFrm, OpcodeStr, "32", v2i32, NEONvsli> {
3374 let Inst{21} = 0b1; // imm6 = 1xxxxx
3376 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, i32imm,
3377 N2RegVShLFrm, OpcodeStr, "64", v1i64, NEONvsli>;
3380 // 128-bit vector types.
3381 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3382 N2RegVShLFrm, OpcodeStr, "8", v16i8, NEONvsli> {
3383 let Inst{21-19} = 0b001; // imm6 = 001xxx
3385 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3386 N2RegVShLFrm, OpcodeStr, "16", v8i16, NEONvsli> {
3387 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3389 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, i32imm,
3390 N2RegVShLFrm, OpcodeStr, "32", v4i32, NEONvsli> {
3391 let Inst{21} = 0b1; // imm6 = 1xxxxx
3393 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, i32imm,
3394 N2RegVShLFrm, OpcodeStr, "64", v2i64, NEONvsli>;
3397 multiclass N2VShInsR_QHSD<bit op24, bit op23, bits<4> op11_8, bit op4,
3399 // 64-bit vector types.
3400 def v8i8 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3401 N2RegVShRFrm, OpcodeStr, "8", v8i8, NEONvsri> {
3402 let Inst{21-19} = 0b001; // imm6 = 001xxx
3404 def v4i16 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3405 N2RegVShRFrm, OpcodeStr, "16", v4i16, NEONvsri> {
3406 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3408 def v2i32 : N2VDShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3409 N2RegVShRFrm, OpcodeStr, "32", v2i32, NEONvsri> {
3410 let Inst{21} = 0b1; // imm6 = 1xxxxx
3412 def v1i64 : N2VDShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3413 N2RegVShRFrm, OpcodeStr, "64", v1i64, NEONvsri>;
3416 // 128-bit vector types.
3417 def v16i8 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm8,
3418 N2RegVShRFrm, OpcodeStr, "8", v16i8, NEONvsri> {
3419 let Inst{21-19} = 0b001; // imm6 = 001xxx
3421 def v8i16 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm16,
3422 N2RegVShRFrm, OpcodeStr, "16", v8i16, NEONvsri> {
3423 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3425 def v4i32 : N2VQShIns<op24, op23, op11_8, 0, op4, shr_imm32,
3426 N2RegVShRFrm, OpcodeStr, "32", v4i32, NEONvsri> {
3427 let Inst{21} = 0b1; // imm6 = 1xxxxx
3429 def v2i64 : N2VQShIns<op24, op23, op11_8, 1, op4, shr_imm64,
3430 N2RegVShRFrm, OpcodeStr, "64", v2i64, NEONvsri>;
3434 // Neon Shift Long operations,
3435 // element sizes of 8, 16, 32 bits:
3436 multiclass N2VLSh_QHS<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3437 bit op4, string OpcodeStr, string Dt, SDNode OpNode> {
3438 def v8i16 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3439 OpcodeStr, !strconcat(Dt, "8"), v8i16, v8i8, OpNode> {
3440 let Inst{21-19} = 0b001; // imm6 = 001xxx
3442 def v4i32 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3443 OpcodeStr, !strconcat(Dt, "16"), v4i32, v4i16, OpNode> {
3444 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3446 def v2i64 : N2VLSh<op24, op23, op11_8, op7, op6, op4,
3447 OpcodeStr, !strconcat(Dt, "32"), v2i64, v2i32, OpNode> {
3448 let Inst{21} = 0b1; // imm6 = 1xxxxx
3452 // Neon Shift Narrow operations,
3453 // element sizes of 16, 32, 64 bits:
3454 multiclass N2VNSh_HSD<bit op24, bit op23, bits<4> op11_8, bit op7, bit op6,
3455 bit op4, InstrItinClass itin, string OpcodeStr, string Dt,
3457 def v8i8 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3458 OpcodeStr, !strconcat(Dt, "16"),
3459 v8i8, v8i16, shr_imm8, OpNode> {
3460 let Inst{21-19} = 0b001; // imm6 = 001xxx
3462 def v4i16 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3463 OpcodeStr, !strconcat(Dt, "32"),
3464 v4i16, v4i32, shr_imm16, OpNode> {
3465 let Inst{21-20} = 0b01; // imm6 = 01xxxx
3467 def v2i32 : N2VNSh<op24, op23, op11_8, op7, op6, op4, itin,
3468 OpcodeStr, !strconcat(Dt, "64"),
3469 v2i32, v2i64, shr_imm32, OpNode> {
3470 let Inst{21} = 0b1; // imm6 = 1xxxxx
3474 //===----------------------------------------------------------------------===//
3475 // Instruction Definitions.
3476 //===----------------------------------------------------------------------===//
3478 // Vector Add Operations.
3480 // VADD : Vector Add (integer and floating-point)
3481 defm VADD : N3V_QHSD<0, 0, 0b1000, 0, IIC_VBINiD, IIC_VBINiQ, "vadd", "i",
3483 def VADDfd : N3VD<0, 0, 0b00, 0b1101, 0, IIC_VBIND, "vadd", "f32",
3484 v2f32, v2f32, fadd, 1>;
3485 def VADDfq : N3VQ<0, 0, 0b00, 0b1101, 0, IIC_VBINQ, "vadd", "f32",
3486 v4f32, v4f32, fadd, 1>;
3487 // VADDL : Vector Add Long (Q = D + D)
3488 defm VADDLs : N3VLExt_QHS<0,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3489 "vaddl", "s", add, sext, 1>;
3490 defm VADDLu : N3VLExt_QHS<1,1,0b0000,0, IIC_VSHLiD, IIC_VSHLiD,
3491 "vaddl", "u", add, zext, 1>;
3492 // VADDW : Vector Add Wide (Q = Q + D)
3493 defm VADDWs : N3VW_QHS<0,1,0b0001,0, "vaddw", "s", add, sext, 0>;
3494 defm VADDWu : N3VW_QHS<1,1,0b0001,0, "vaddw", "u", add, zext, 0>;
3495 // VHADD : Vector Halving Add
3496 defm VHADDs : N3VInt_QHS<0, 0, 0b0000, 0, N3RegFrm,
3497 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3498 "vhadd", "s", int_arm_neon_vhadds, 1>;
3499 defm VHADDu : N3VInt_QHS<1, 0, 0b0000, 0, N3RegFrm,
3500 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3501 "vhadd", "u", int_arm_neon_vhaddu, 1>;
3502 // VRHADD : Vector Rounding Halving Add
3503 defm VRHADDs : N3VInt_QHS<0, 0, 0b0001, 0, N3RegFrm,
3504 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3505 "vrhadd", "s", int_arm_neon_vrhadds, 1>;
3506 defm VRHADDu : N3VInt_QHS<1, 0, 0b0001, 0, N3RegFrm,
3507 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3508 "vrhadd", "u", int_arm_neon_vrhaddu, 1>;
3509 // VQADD : Vector Saturating Add
3510 defm VQADDs : N3VInt_QHSD<0, 0, 0b0000, 1, N3RegFrm,
3511 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3512 "vqadd", "s", int_arm_neon_vqadds, 1>;
3513 defm VQADDu : N3VInt_QHSD<1, 0, 0b0000, 1, N3RegFrm,
3514 IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q, IIC_VBINi4Q,
3515 "vqadd", "u", int_arm_neon_vqaddu, 1>;
3516 // VADDHN : Vector Add and Narrow Returning High Half (D = Q + Q)
3517 defm VADDHN : N3VNInt_HSD<0,1,0b0100,0, "vaddhn", "i",
3518 int_arm_neon_vaddhn, 1>;
3519 // VRADDHN : Vector Rounding Add and Narrow Returning High Half (D = Q + Q)
3520 defm VRADDHN : N3VNInt_HSD<1,1,0b0100,0, "vraddhn", "i",
3521 int_arm_neon_vraddhn, 1>;
3523 // Vector Multiply Operations.
3525 // VMUL : Vector Multiply (integer, polynomial and floating-point)
3526 defm VMUL : N3V_QHS<0, 0, 0b1001, 1, IIC_VMULi16D, IIC_VMULi32D,
3527 IIC_VMULi16Q, IIC_VMULi32Q, "vmul", "i", mul, 1>;
3528 def VMULpd : N3VDInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16D, "vmul",
3529 "p8", v8i8, v8i8, int_arm_neon_vmulp, 1>;
3530 def VMULpq : N3VQInt<1, 0, 0b00, 0b1001, 1, N3RegFrm, IIC_VMULi16Q, "vmul",
3531 "p8", v16i8, v16i8, int_arm_neon_vmulp, 1>;
3532 def VMULfd : N3VD<1, 0, 0b00, 0b1101, 1, IIC_VFMULD, "vmul", "f32",
3533 v2f32, v2f32, fmul, 1>;
3534 def VMULfq : N3VQ<1, 0, 0b00, 0b1101, 1, IIC_VFMULQ, "vmul", "f32",
3535 v4f32, v4f32, fmul, 1>;
3536 defm VMULsl : N3VSL_HS<0b1000, "vmul", "i", mul>;
3537 def VMULslfd : N3VDSL<0b10, 0b1001, IIC_VBIND, "vmul", "f32", v2f32, fmul>;
3538 def VMULslfq : N3VQSL<0b10, 0b1001, IIC_VBINQ, "vmul", "f32", v4f32,
3541 def : Pat<(v8i16 (mul (v8i16 QPR:$src1),
3542 (v8i16 (NEONvduplane (v8i16 QPR:$src2), imm:$lane)))),
3543 (v8i16 (VMULslv8i16 (v8i16 QPR:$src1),
3544 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3545 (DSubReg_i16_reg imm:$lane))),
3546 (SubReg_i16_lane imm:$lane)))>;
3547 def : Pat<(v4i32 (mul (v4i32 QPR:$src1),
3548 (v4i32 (NEONvduplane (v4i32 QPR:$src2), imm:$lane)))),
3549 (v4i32 (VMULslv4i32 (v4i32 QPR:$src1),
3550 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3551 (DSubReg_i32_reg imm:$lane))),
3552 (SubReg_i32_lane imm:$lane)))>;
3553 def : Pat<(v4f32 (fmul (v4f32 QPR:$src1),
3554 (v4f32 (NEONvduplane (v4f32 QPR:$src2), imm:$lane)))),
3555 (v4f32 (VMULslfq (v4f32 QPR:$src1),
3556 (v2f32 (EXTRACT_SUBREG QPR:$src2,
3557 (DSubReg_i32_reg imm:$lane))),
3558 (SubReg_i32_lane imm:$lane)))>;
3560 // VQDMULH : Vector Saturating Doubling Multiply Returning High Half
3561 defm VQDMULH : N3VInt_HS<0, 0, 0b1011, 0, N3RegFrm, IIC_VMULi16D, IIC_VMULi32D,
3562 IIC_VMULi16Q, IIC_VMULi32Q,
3563 "vqdmulh", "s", int_arm_neon_vqdmulh, 1>;
3564 defm VQDMULHsl: N3VIntSL_HS<0b1100, IIC_VMULi16D, IIC_VMULi32D,
3565 IIC_VMULi16Q, IIC_VMULi32Q,
3566 "vqdmulh", "s", int_arm_neon_vqdmulh>;
3567 def : Pat<(v8i16 (int_arm_neon_vqdmulh (v8i16 QPR:$src1),
3568 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3570 (v8i16 (VQDMULHslv8i16 (v8i16 QPR:$src1),
3571 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3572 (DSubReg_i16_reg imm:$lane))),
3573 (SubReg_i16_lane imm:$lane)))>;
3574 def : Pat<(v4i32 (int_arm_neon_vqdmulh (v4i32 QPR:$src1),
3575 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3577 (v4i32 (VQDMULHslv4i32 (v4i32 QPR:$src1),
3578 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3579 (DSubReg_i32_reg imm:$lane))),
3580 (SubReg_i32_lane imm:$lane)))>;
3582 // VQRDMULH : Vector Rounding Saturating Doubling Multiply Returning High Half
3583 defm VQRDMULH : N3VInt_HS<1, 0, 0b1011, 0, N3RegFrm,
3584 IIC_VMULi16D,IIC_VMULi32D,IIC_VMULi16Q,IIC_VMULi32Q,
3585 "vqrdmulh", "s", int_arm_neon_vqrdmulh, 1>;
3586 defm VQRDMULHsl : N3VIntSL_HS<0b1101, IIC_VMULi16D, IIC_VMULi32D,
3587 IIC_VMULi16Q, IIC_VMULi32Q,
3588 "vqrdmulh", "s", int_arm_neon_vqrdmulh>;
3589 def : Pat<(v8i16 (int_arm_neon_vqrdmulh (v8i16 QPR:$src1),
3590 (v8i16 (NEONvduplane (v8i16 QPR:$src2),
3592 (v8i16 (VQRDMULHslv8i16 (v8i16 QPR:$src1),
3593 (v4i16 (EXTRACT_SUBREG QPR:$src2,
3594 (DSubReg_i16_reg imm:$lane))),
3595 (SubReg_i16_lane imm:$lane)))>;
3596 def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
3597 (v4i32 (NEONvduplane (v4i32 QPR:$src2),
3599 (v4i32 (VQRDMULHslv4i32 (v4i32 QPR:$src1),
3600 (v2i32 (EXTRACT_SUBREG QPR:$src2,
3601 (DSubReg_i32_reg imm:$lane))),
3602 (SubReg_i32_lane imm:$lane)))>;
3604 // VMULL : Vector Multiply Long (integer and polynomial) (Q = D * D)
3605 defm VMULLs : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3606 "vmull", "s", NEONvmulls, 1>;
3607 defm VMULLu : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
3608 "vmull", "u", NEONvmullu, 1>;
3609 def VMULLp : N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
3610 v8i16, v8i8, int_arm_neon_vmullp, 1>;
3611 defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
3612 defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
3614 // VQDMULL : Vector Saturating Doubling Multiply Long (Q = D * D)
3615 defm VQDMULL : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
3616 "vqdmull", "s", int_arm_neon_vqdmull, 1>;
3617 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
3618 "vqdmull", "s", int_arm_neon_vqdmull>;
3620 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
3622 // VMLA : Vector Multiply Accumulate (integer and floating-point)
3623 defm VMLA : N3VMulOp_QHS<0, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3624 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3625 def VMLAfd : N3VDMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACD, "vmla", "f32",
3626 v2f32, fmul_su, fadd_mlx>,
3627 Requires<[HasNEON, UseFPVMLx]>;
3628 def VMLAfq : N3VQMulOp<0, 0, 0b00, 0b1101, 1, IIC_VMACQ, "vmla", "f32",
3629 v4f32, fmul_su, fadd_mlx>,
3630 Requires<[HasNEON, UseFPVMLx]>;
3631 defm VMLAsl : N3VMulOpSL_HS<0b0000, IIC_VMACi16D, IIC_VMACi32D,
3632 IIC_VMACi16Q, IIC_VMACi32Q, "vmla", "i", add>;
3633 def VMLAslfd : N3VDMulOpSL<0b10, 0b0001, IIC_VMACD, "vmla", "f32",
3634 v2f32, fmul_su, fadd_mlx>,
3635 Requires<[HasNEON, UseFPVMLx]>;
3636 def VMLAslfq : N3VQMulOpSL<0b10, 0b0001, IIC_VMACQ, "vmla", "f32",
3637 v4f32, v2f32, fmul_su, fadd_mlx>,
3638 Requires<[HasNEON, UseFPVMLx]>;
3640 def : Pat<(v8i16 (add (v8i16 QPR:$src1),
3641 (mul (v8i16 QPR:$src2),
3642 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3643 (v8i16 (VMLAslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3644 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3645 (DSubReg_i16_reg imm:$lane))),
3646 (SubReg_i16_lane imm:$lane)))>;
3648 def : Pat<(v4i32 (add (v4i32 QPR:$src1),
3649 (mul (v4i32 QPR:$src2),
3650 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3651 (v4i32 (VMLAslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3652 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3653 (DSubReg_i32_reg imm:$lane))),
3654 (SubReg_i32_lane imm:$lane)))>;
3656 def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
3657 (fmul_su (v4f32 QPR:$src2),
3658 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3659 (v4f32 (VMLAslfq (v4f32 QPR:$src1),
3661 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3662 (DSubReg_i32_reg imm:$lane))),
3663 (SubReg_i32_lane imm:$lane)))>,
3664 Requires<[HasNEON, UseFPVMLx]>;
3666 // VMLAL : Vector Multiply Accumulate Long (Q += D * D)
3667 defm VMLALs : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3668 "vmlal", "s", NEONvmulls, add>;
3669 defm VMLALu : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
3670 "vmlal", "u", NEONvmullu, add>;
3672 defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
3673 defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
3675 // VQDMLAL : Vector Saturating Doubling Multiply Accumulate Long (Q += D * D)
3676 defm VQDMLAL : N3VLInt3_HS<0, 1, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3677 "vqdmlal", "s", int_arm_neon_vqdmlal>;
3678 defm VQDMLALsl: N3VLInt3SL_HS<0, 0b0011, "vqdmlal", "s", int_arm_neon_vqdmlal>;
3680 // VMLS : Vector Multiply Subtract (integer and floating-point)
3681 defm VMLS : N3VMulOp_QHS<1, 0, 0b1001, 0, IIC_VMACi16D, IIC_VMACi32D,
3682 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3683 def VMLSfd : N3VDMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACD, "vmls", "f32",
3684 v2f32, fmul_su, fsub_mlx>,
3685 Requires<[HasNEON, UseFPVMLx]>;
3686 def VMLSfq : N3VQMulOp<0, 0, 0b10, 0b1101, 1, IIC_VMACQ, "vmls", "f32",
3687 v4f32, fmul_su, fsub_mlx>,
3688 Requires<[HasNEON, UseFPVMLx]>;
3689 defm VMLSsl : N3VMulOpSL_HS<0b0100, IIC_VMACi16D, IIC_VMACi32D,
3690 IIC_VMACi16Q, IIC_VMACi32Q, "vmls", "i", sub>;
3691 def VMLSslfd : N3VDMulOpSL<0b10, 0b0101, IIC_VMACD, "vmls", "f32",
3692 v2f32, fmul_su, fsub_mlx>,
3693 Requires<[HasNEON, UseFPVMLx]>;
3694 def VMLSslfq : N3VQMulOpSL<0b10, 0b0101, IIC_VMACQ, "vmls", "f32",
3695 v4f32, v2f32, fmul_su, fsub_mlx>,
3696 Requires<[HasNEON, UseFPVMLx]>;
3698 def : Pat<(v8i16 (sub (v8i16 QPR:$src1),
3699 (mul (v8i16 QPR:$src2),
3700 (v8i16 (NEONvduplane (v8i16 QPR:$src3), imm:$lane))))),
3701 (v8i16 (VMLSslv8i16 (v8i16 QPR:$src1), (v8i16 QPR:$src2),
3702 (v4i16 (EXTRACT_SUBREG QPR:$src3,
3703 (DSubReg_i16_reg imm:$lane))),
3704 (SubReg_i16_lane imm:$lane)))>;
3706 def : Pat<(v4i32 (sub (v4i32 QPR:$src1),
3707 (mul (v4i32 QPR:$src2),
3708 (v4i32 (NEONvduplane (v4i32 QPR:$src3), imm:$lane))))),
3709 (v4i32 (VMLSslv4i32 (v4i32 QPR:$src1), (v4i32 QPR:$src2),
3710 (v2i32 (EXTRACT_SUBREG QPR:$src3,
3711 (DSubReg_i32_reg imm:$lane))),
3712 (SubReg_i32_lane imm:$lane)))>;
3714 def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
3715 (fmul_su (v4f32 QPR:$src2),
3716 (v4f32 (NEONvduplane (v4f32 QPR:$src3), imm:$lane))))),
3717 (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
3718 (v2f32 (EXTRACT_SUBREG QPR:$src3,
3719 (DSubReg_i32_reg imm:$lane))),
3720 (SubReg_i32_lane imm:$lane)))>,
3721 Requires<[HasNEON, UseFPVMLx]>;
3723 // VMLSL : Vector Multiply Subtract Long (Q -= D * D)
3724 defm VMLSLs : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3725 "vmlsl", "s", NEONvmulls, sub>;
3726 defm VMLSLu : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
3727 "vmlsl", "u", NEONvmullu, sub>;
3729 defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
3730 defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
3732 // VQDMLSL : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
3733 defm VQDMLSL : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
3734 "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3735 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b111, "vqdmlsl", "s", int_arm_neon_vqdmlsl>;
3737 // Vector Subtract Operations.
3739 // VSUB : Vector Subtract (integer and floating-point)
3740 defm VSUB : N3V_QHSD<1, 0, 0b1000, 0, IIC_VSUBiD, IIC_VSUBiQ,
3741 "vsub", "i", sub, 0>;
3742 def VSUBfd : N3VD<0, 0, 0b10, 0b1101, 0, IIC_VBIND, "vsub", "f32",
3743 v2f32, v2f32, fsub, 0>;
3744 def VSUBfq : N3VQ<0, 0, 0b10, 0b1101, 0, IIC_VBINQ, "vsub", "f32",
3745 v4f32, v4f32, fsub, 0>;
3746 // VSUBL : Vector Subtract Long (Q = D - D)
3747 defm VSUBLs : N3VLExt_QHS<0,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3748 "vsubl", "s", sub, sext, 0>;
3749 defm VSUBLu : N3VLExt_QHS<1,1,0b0010,0, IIC_VSHLiD, IIC_VSHLiD,
3750 "vsubl", "u", sub, zext, 0>;
3751 // VSUBW : Vector Subtract Wide (Q = Q - D)
3752 defm VSUBWs : N3VW_QHS<0,1,0b0011,0, "vsubw", "s", sub, sext, 0>;
3753 defm VSUBWu : N3VW_QHS<1,1,0b0011,0, "vsubw", "u", sub, zext, 0>;
3754 // VHSUB : Vector Halving Subtract
3755 defm VHSUBs : N3VInt_QHS<0, 0, 0b0010, 0, N3RegFrm,
3756 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3757 "vhsub", "s", int_arm_neon_vhsubs, 0>;
3758 defm VHSUBu : N3VInt_QHS<1, 0, 0b0010, 0, N3RegFrm,
3759 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3760 "vhsub", "u", int_arm_neon_vhsubu, 0>;
3761 // VQSUB : Vector Saturing Subtract
3762 defm VQSUBs : N3VInt_QHSD<0, 0, 0b0010, 1, N3RegFrm,
3763 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3764 "vqsub", "s", int_arm_neon_vqsubs, 0>;
3765 defm VQSUBu : N3VInt_QHSD<1, 0, 0b0010, 1, N3RegFrm,
3766 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
3767 "vqsub", "u", int_arm_neon_vqsubu, 0>;
3768 // VSUBHN : Vector Subtract and Narrow Returning High Half (D = Q - Q)
3769 defm VSUBHN : N3VNInt_HSD<0,1,0b0110,0, "vsubhn", "i",
3770 int_arm_neon_vsubhn, 0>;
3771 // VRSUBHN : Vector Rounding Subtract and Narrow Returning High Half (D=Q-Q)
3772 defm VRSUBHN : N3VNInt_HSD<1,1,0b0110,0, "vrsubhn", "i",
3773 int_arm_neon_vrsubhn, 0>;
3775 // Vector Comparisons.
3777 // VCEQ : Vector Compare Equal
3778 defm VCEQ : N3V_QHS<1, 0, 0b1000, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3779 IIC_VSUBi4Q, "vceq", "i", NEONvceq, 1>;
3780 def VCEQfd : N3VD<0,0,0b00,0b1110,0, IIC_VBIND, "vceq", "f32", v2i32, v2f32,
3782 def VCEQfq : N3VQ<0,0,0b00,0b1110,0, IIC_VBINQ, "vceq", "f32", v4i32, v4f32,
3785 defm VCEQz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00010, 0, "vceq", "i",
3786 "$Vd, $Vm, #0", NEONvceqz>;
3788 // VCGE : Vector Compare Greater Than or Equal
3789 defm VCGEs : N3V_QHS<0, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3790 IIC_VSUBi4Q, "vcge", "s", NEONvcge, 0>;
3791 defm VCGEu : N3V_QHS<1, 0, 0b0011, 1, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3792 IIC_VSUBi4Q, "vcge", "u", NEONvcgeu, 0>;
3793 def VCGEfd : N3VD<1,0,0b00,0b1110,0, IIC_VBIND, "vcge", "f32", v2i32, v2f32,
3795 def VCGEfq : N3VQ<1,0,0b00,0b1110,0, IIC_VBINQ, "vcge", "f32", v4i32, v4f32,
3798 defm VCGEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00001, 0, "vcge", "s",
3799 "$Vd, $Vm, #0", NEONvcgez>;
3800 defm VCLEz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00011, 0, "vcle", "s",
3801 "$Vd, $Vm, #0", NEONvclez>;
3803 // VCGT : Vector Compare Greater Than
3804 defm VCGTs : N3V_QHS<0, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3805 IIC_VSUBi4Q, "vcgt", "s", NEONvcgt, 0>;
3806 defm VCGTu : N3V_QHS<1, 0, 0b0011, 0, IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q,
3807 IIC_VSUBi4Q, "vcgt", "u", NEONvcgtu, 0>;
3808 def VCGTfd : N3VD<1,0,0b10,0b1110,0, IIC_VBIND, "vcgt", "f32", v2i32, v2f32,
3810 def VCGTfq : N3VQ<1,0,0b10,0b1110,0, IIC_VBINQ, "vcgt", "f32", v4i32, v4f32,
3813 defm VCGTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00000, 0, "vcgt", "s",
3814 "$Vd, $Vm, #0", NEONvcgtz>;
3815 defm VCLTz : N2V_QHS_cmp<0b11, 0b11, 0b01, 0b00100, 0, "vclt", "s",
3816 "$Vd, $Vm, #0", NEONvcltz>;
3818 // VACGE : Vector Absolute Compare Greater Than or Equal (aka VCAGE)
3819 def VACGEd : N3VDInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacge",
3820 "f32", v2i32, v2f32, int_arm_neon_vacged, 0>;
3821 def VACGEq : N3VQInt<1, 0, 0b00, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacge",
3822 "f32", v4i32, v4f32, int_arm_neon_vacgeq, 0>;
3823 // VACGT : Vector Absolute Compare Greater Than (aka VCAGT)
3824 def VACGTd : N3VDInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBIND, "vacgt",
3825 "f32", v2i32, v2f32, int_arm_neon_vacgtd, 0>;
3826 def VACGTq : N3VQInt<1, 0, 0b10, 0b1110, 1, N3RegFrm, IIC_VBINQ, "vacgt",
3827 "f32", v4i32, v4f32, int_arm_neon_vacgtq, 0>;
3828 // VTST : Vector Test Bits
3829 defm VTST : N3V_QHS<0, 0, 0b1000, 1, IIC_VBINi4D, IIC_VBINi4D, IIC_VBINi4Q,
3830 IIC_VBINi4Q, "vtst", "", NEONvtst, 1>;
3832 // Vector Bitwise Operations.
3834 def vnotd : PatFrag<(ops node:$in),
3835 (xor node:$in, (bitconvert (v8i8 NEONimmAllOnesV)))>;
3836 def vnotq : PatFrag<(ops node:$in),
3837 (xor node:$in, (bitconvert (v16i8 NEONimmAllOnesV)))>;
3840 // VAND : Vector Bitwise AND
3841 def VANDd : N3VDX<0, 0, 0b00, 0b0001, 1, IIC_VBINiD, "vand",
3842 v2i32, v2i32, and, 1>;
3843 def VANDq : N3VQX<0, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "vand",
3844 v4i32, v4i32, and, 1>;
3846 // VEOR : Vector Bitwise Exclusive OR
3847 def VEORd : N3VDX<1, 0, 0b00, 0b0001, 1, IIC_VBINiD, "veor",
3848 v2i32, v2i32, xor, 1>;
3849 def VEORq : N3VQX<1, 0, 0b00, 0b0001, 1, IIC_VBINiQ, "veor",
3850 v4i32, v4i32, xor, 1>;
3852 // VORR : Vector Bitwise OR
3853 def VORRd : N3VDX<0, 0, 0b10, 0b0001, 1, IIC_VBINiD, "vorr",
3854 v2i32, v2i32, or, 1>;
3855 def VORRq : N3VQX<0, 0, 0b10, 0b0001, 1, IIC_VBINiQ, "vorr",
3856 v4i32, v4i32, or, 1>;
3858 def VORRiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 0, 1,
3859 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3861 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3863 (v4i16 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3864 let Inst{9} = SIMM{9};
3867 def VORRiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 0, 1,
3868 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3870 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3872 (v2i32 (NEONvorrImm DPR:$src, timm:$SIMM)))]> {
3873 let Inst{10-9} = SIMM{10-9};
3876 def VORRiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 0, 1,
3877 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3879 "vorr", "i16", "$Vd, $SIMM", "$src = $Vd",
3881 (v8i16 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3882 let Inst{9} = SIMM{9};
3885 def VORRiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 0, 1,
3886 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3888 "vorr", "i32", "$Vd, $SIMM", "$src = $Vd",
3890 (v4i32 (NEONvorrImm QPR:$src, timm:$SIMM)))]> {
3891 let Inst{10-9} = SIMM{10-9};
3895 // VBIC : Vector Bitwise Bit Clear (AND NOT)
3896 def VBICd : N3VX<0, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
3897 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3898 "vbic", "$Vd, $Vn, $Vm", "",
3899 [(set DPR:$Vd, (v2i32 (and DPR:$Vn,
3900 (vnotd DPR:$Vm))))]>;
3901 def VBICq : N3VX<0, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
3902 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3903 "vbic", "$Vd, $Vn, $Vm", "",
3904 [(set QPR:$Vd, (v4i32 (and QPR:$Vn,
3905 (vnotq QPR:$Vm))))]>;
3907 def VBICiv4i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 0, 1, 1,
3908 (outs DPR:$Vd), (ins nImmSplatI16:$SIMM, DPR:$src),
3910 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3912 (v4i16 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3913 let Inst{9} = SIMM{9};
3916 def VBICiv2i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 0, 1, 1,
3917 (outs DPR:$Vd), (ins nImmSplatI32:$SIMM, DPR:$src),
3919 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3921 (v2i32 (NEONvbicImm DPR:$src, timm:$SIMM)))]> {
3922 let Inst{10-9} = SIMM{10-9};
3925 def VBICiv8i16 : N1ModImm<1, 0b000, {1,0,?,1}, 0, 1, 1, 1,
3926 (outs QPR:$Vd), (ins nImmSplatI16:$SIMM, QPR:$src),
3928 "vbic", "i16", "$Vd, $SIMM", "$src = $Vd",
3930 (v8i16 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3931 let Inst{9} = SIMM{9};
3934 def VBICiv4i32 : N1ModImm<1, 0b000, {0,?,?,1}, 0, 1, 1, 1,
3935 (outs QPR:$Vd), (ins nImmSplatI32:$SIMM, QPR:$src),
3937 "vbic", "i32", "$Vd, $SIMM", "$src = $Vd",
3939 (v4i32 (NEONvbicImm QPR:$src, timm:$SIMM)))]> {
3940 let Inst{10-9} = SIMM{10-9};
3943 // VORN : Vector Bitwise OR NOT
3944 def VORNd : N3VX<0, 0, 0b11, 0b0001, 0, 1, (outs DPR:$Vd),
3945 (ins DPR:$Vn, DPR:$Vm), N3RegFrm, IIC_VBINiD,
3946 "vorn", "$Vd, $Vn, $Vm", "",
3947 [(set DPR:$Vd, (v2i32 (or DPR:$Vn,
3948 (vnotd DPR:$Vm))))]>;
3949 def VORNq : N3VX<0, 0, 0b11, 0b0001, 1, 1, (outs QPR:$Vd),
3950 (ins QPR:$Vn, QPR:$Vm), N3RegFrm, IIC_VBINiQ,
3951 "vorn", "$Vd, $Vn, $Vm", "",
3952 [(set QPR:$Vd, (v4i32 (or QPR:$Vn,
3953 (vnotq QPR:$Vm))))]>;
3955 // VMVN : Vector Bitwise NOT (Immediate)
3957 let isReMaterializable = 1 in {
3959 def VMVNv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 1, 1, (outs DPR:$Vd),
3960 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3961 "vmvn", "i16", "$Vd, $SIMM", "",
3962 [(set DPR:$Vd, (v4i16 (NEONvmvnImm timm:$SIMM)))]> {
3963 let Inst{9} = SIMM{9};
3966 def VMVNv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 1, 1, (outs QPR:$Vd),
3967 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
3968 "vmvn", "i16", "$Vd, $SIMM", "",
3969 [(set QPR:$Vd, (v8i16 (NEONvmvnImm timm:$SIMM)))]> {
3970 let Inst{9} = SIMM{9};
3973 def VMVNv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 1, 1, (outs DPR:$Vd),
3974 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3975 "vmvn", "i32", "$Vd, $SIMM", "",
3976 [(set DPR:$Vd, (v2i32 (NEONvmvnImm timm:$SIMM)))]> {
3977 let Inst{11-8} = SIMM{11-8};
3980 def VMVNv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 1, 1, (outs QPR:$Vd),
3981 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
3982 "vmvn", "i32", "$Vd, $SIMM", "",
3983 [(set QPR:$Vd, (v4i32 (NEONvmvnImm timm:$SIMM)))]> {
3984 let Inst{11-8} = SIMM{11-8};
3988 // VMVN : Vector Bitwise NOT
3989 def VMVNd : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 0, 0,
3990 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VSUBiD,
3991 "vmvn", "$Vd, $Vm", "",
3992 [(set DPR:$Vd, (v2i32 (vnotd DPR:$Vm)))]>;
3993 def VMVNq : N2VX<0b11, 0b11, 0b00, 0b00, 0b01011, 1, 0,
3994 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VSUBiD,
3995 "vmvn", "$Vd, $Vm", "",
3996 [(set QPR:$Vd, (v4i32 (vnotq QPR:$Vm)))]>;
3997 def : Pat<(v2i32 (vnotd DPR:$src)), (VMVNd DPR:$src)>;
3998 def : Pat<(v4i32 (vnotq QPR:$src)), (VMVNq QPR:$src)>;
4000 // VBSL : Vector Bitwise Select
4001 def VBSLd : N3VX<1, 0, 0b01, 0b0001, 0, 1, (outs DPR:$Vd),
4002 (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4003 N3RegFrm, IIC_VCNTiD,
4004 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4006 (v2i32 (NEONvbsl DPR:$src1, DPR:$Vn, DPR:$Vm)))]>;
4008 def : Pat<(v2i32 (or (and DPR:$Vn, DPR:$Vd),
4009 (and DPR:$Vm, (vnotd DPR:$Vd)))),
4010 (VBSLd DPR:$Vd, DPR:$Vn, DPR:$Vm)>;
4012 def VBSLq : N3VX<1, 0, 0b01, 0b0001, 1, 1, (outs QPR:$Vd),
4013 (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4014 N3RegFrm, IIC_VCNTiQ,
4015 "vbsl", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4017 (v4i32 (NEONvbsl QPR:$src1, QPR:$Vn, QPR:$Vm)))]>;
4019 def : Pat<(v4i32 (or (and QPR:$Vn, QPR:$Vd),
4020 (and QPR:$Vm, (vnotq QPR:$Vd)))),
4021 (VBSLq QPR:$Vd, QPR:$Vn, QPR:$Vm)>;
4023 // VBIF : Vector Bitwise Insert if False
4024 // like VBSL but with: "vbif $dst, $src3, $src1", "$src2 = $dst",
4025 // FIXME: This instruction's encoding MAY NOT BE correct.
4026 def VBIFd : N3VX<1, 0, 0b11, 0b0001, 0, 1,
4027 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4028 N3RegFrm, IIC_VBINiD,
4029 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4031 def VBIFq : N3VX<1, 0, 0b11, 0b0001, 1, 1,
4032 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4033 N3RegFrm, IIC_VBINiQ,
4034 "vbif", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4037 // VBIT : Vector Bitwise Insert if True
4038 // like VBSL but with: "vbit $dst, $src2, $src1", "$src3 = $dst",
4039 // FIXME: This instruction's encoding MAY NOT BE correct.
4040 def VBITd : N3VX<1, 0, 0b10, 0b0001, 0, 1,
4041 (outs DPR:$Vd), (ins DPR:$src1, DPR:$Vn, DPR:$Vm),
4042 N3RegFrm, IIC_VBINiD,
4043 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4045 def VBITq : N3VX<1, 0, 0b10, 0b0001, 1, 1,
4046 (outs QPR:$Vd), (ins QPR:$src1, QPR:$Vn, QPR:$Vm),
4047 N3RegFrm, IIC_VBINiQ,
4048 "vbit", "$Vd, $Vn, $Vm", "$src1 = $Vd",
4051 // VBIT/VBIF are not yet implemented. The TwoAddress pass will not go looking
4052 // for equivalent operations with different register constraints; it just
4055 // Vector Absolute Differences.
4057 // VABD : Vector Absolute Difference
4058 defm VABDs : N3VInt_QHS<0, 0, 0b0111, 0, N3RegFrm,
4059 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4060 "vabd", "s", int_arm_neon_vabds, 1>;
4061 defm VABDu : N3VInt_QHS<1, 0, 0b0111, 0, N3RegFrm,
4062 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4063 "vabd", "u", int_arm_neon_vabdu, 1>;
4064 def VABDfd : N3VDInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBIND,
4065 "vabd", "f32", v2f32, v2f32, int_arm_neon_vabds, 1>;
4066 def VABDfq : N3VQInt<1, 0, 0b10, 0b1101, 0, N3RegFrm, IIC_VBINQ,
4067 "vabd", "f32", v4f32, v4f32, int_arm_neon_vabds, 1>;
4069 // VABDL : Vector Absolute Difference Long (Q = | D - D |)
4070 defm VABDLs : N3VLIntExt_QHS<0,1,0b0111,0, IIC_VSUBi4Q,
4071 "vabdl", "s", int_arm_neon_vabds, zext, 1>;
4072 defm VABDLu : N3VLIntExt_QHS<1,1,0b0111,0, IIC_VSUBi4Q,
4073 "vabdl", "u", int_arm_neon_vabdu, zext, 1>;
4075 // VABA : Vector Absolute Difference and Accumulate
4076 defm VABAs : N3VIntOp_QHS<0,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4077 "vaba", "s", int_arm_neon_vabds, add>;
4078 defm VABAu : N3VIntOp_QHS<1,0,0b0111,1, IIC_VABAD, IIC_VABAQ,
4079 "vaba", "u", int_arm_neon_vabdu, add>;
4081 // VABAL : Vector Absolute Difference and Accumulate Long (Q += | D - D |)
4082 defm VABALs : N3VLIntExtOp_QHS<0,1,0b0101,0, IIC_VABAD,
4083 "vabal", "s", int_arm_neon_vabds, zext, add>;
4084 defm VABALu : N3VLIntExtOp_QHS<1,1,0b0101,0, IIC_VABAD,
4085 "vabal", "u", int_arm_neon_vabdu, zext, add>;
4087 // Vector Maximum and Minimum.
4089 // VMAX : Vector Maximum
4090 defm VMAXs : N3VInt_QHS<0, 0, 0b0110, 0, N3RegFrm,
4091 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4092 "vmax", "s", int_arm_neon_vmaxs, 1>;
4093 defm VMAXu : N3VInt_QHS<1, 0, 0b0110, 0, N3RegFrm,
4094 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4095 "vmax", "u", int_arm_neon_vmaxu, 1>;
4096 def VMAXfd : N3VDInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBIND,
4098 v2f32, v2f32, int_arm_neon_vmaxs, 1>;
4099 def VMAXfq : N3VQInt<0, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4101 v4f32, v4f32, int_arm_neon_vmaxs, 1>;
4103 // VMIN : Vector Minimum
4104 defm VMINs : N3VInt_QHS<0, 0, 0b0110, 1, N3RegFrm,
4105 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4106 "vmin", "s", int_arm_neon_vmins, 1>;
4107 defm VMINu : N3VInt_QHS<1, 0, 0b0110, 1, N3RegFrm,
4108 IIC_VSUBi4D, IIC_VSUBi4D, IIC_VSUBi4Q, IIC_VSUBi4Q,
4109 "vmin", "u", int_arm_neon_vminu, 1>;
4110 def VMINfd : N3VDInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBIND,
4112 v2f32, v2f32, int_arm_neon_vmins, 1>;
4113 def VMINfq : N3VQInt<0, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VBINQ,
4115 v4f32, v4f32, int_arm_neon_vmins, 1>;
4117 // Vector Pairwise Operations.
4119 // VPADD : Vector Pairwise Add
4120 def VPADDi8 : N3VDInt<0, 0, 0b00, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4122 v8i8, v8i8, int_arm_neon_vpadd, 0>;
4123 def VPADDi16 : N3VDInt<0, 0, 0b01, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4125 v4i16, v4i16, int_arm_neon_vpadd, 0>;
4126 def VPADDi32 : N3VDInt<0, 0, 0b10, 0b1011, 1, N3RegFrm, IIC_VSHLiD,
4128 v2i32, v2i32, int_arm_neon_vpadd, 0>;
4129 def VPADDf : N3VDInt<1, 0, 0b00, 0b1101, 0, N3RegFrm,
4130 IIC_VPBIND, "vpadd", "f32",
4131 v2f32, v2f32, int_arm_neon_vpadd, 0>;
4133 // VPADDL : Vector Pairwise Add Long
4134 defm VPADDLs : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00100, 0, "vpaddl", "s",
4135 int_arm_neon_vpaddls>;
4136 defm VPADDLu : N2VPLInt_QHS<0b11, 0b11, 0b00, 0b00101, 0, "vpaddl", "u",
4137 int_arm_neon_vpaddlu>;
4139 // VPADAL : Vector Pairwise Add and Accumulate Long
4140 defm VPADALs : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01100, 0, "vpadal", "s",
4141 int_arm_neon_vpadals>;
4142 defm VPADALu : N2VPLInt2_QHS<0b11, 0b11, 0b00, 0b01101, 0, "vpadal", "u",
4143 int_arm_neon_vpadalu>;
4145 // VPMAX : Vector Pairwise Maximum
4146 def VPMAXs8 : N3VDInt<0, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4147 "s8", v8i8, v8i8, int_arm_neon_vpmaxs, 0>;
4148 def VPMAXs16 : N3VDInt<0, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4149 "s16", v4i16, v4i16, int_arm_neon_vpmaxs, 0>;
4150 def VPMAXs32 : N3VDInt<0, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4151 "s32", v2i32, v2i32, int_arm_neon_vpmaxs, 0>;
4152 def VPMAXu8 : N3VDInt<1, 0, 0b00, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4153 "u8", v8i8, v8i8, int_arm_neon_vpmaxu, 0>;
4154 def VPMAXu16 : N3VDInt<1, 0, 0b01, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4155 "u16", v4i16, v4i16, int_arm_neon_vpmaxu, 0>;
4156 def VPMAXu32 : N3VDInt<1, 0, 0b10, 0b1010, 0, N3RegFrm, IIC_VSUBi4D, "vpmax",
4157 "u32", v2i32, v2i32, int_arm_neon_vpmaxu, 0>;
4158 def VPMAXf : N3VDInt<1, 0, 0b00, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmax",
4159 "f32", v2f32, v2f32, int_arm_neon_vpmaxs, 0>;
4161 // VPMIN : Vector Pairwise Minimum
4162 def VPMINs8 : N3VDInt<0, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4163 "s8", v8i8, v8i8, int_arm_neon_vpmins, 0>;
4164 def VPMINs16 : N3VDInt<0, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4165 "s16", v4i16, v4i16, int_arm_neon_vpmins, 0>;
4166 def VPMINs32 : N3VDInt<0, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4167 "s32", v2i32, v2i32, int_arm_neon_vpmins, 0>;
4168 def VPMINu8 : N3VDInt<1, 0, 0b00, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4169 "u8", v8i8, v8i8, int_arm_neon_vpminu, 0>;
4170 def VPMINu16 : N3VDInt<1, 0, 0b01, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4171 "u16", v4i16, v4i16, int_arm_neon_vpminu, 0>;
4172 def VPMINu32 : N3VDInt<1, 0, 0b10, 0b1010, 1, N3RegFrm, IIC_VSUBi4D, "vpmin",
4173 "u32", v2i32, v2i32, int_arm_neon_vpminu, 0>;
4174 def VPMINf : N3VDInt<1, 0, 0b10, 0b1111, 0, N3RegFrm, IIC_VPBIND, "vpmin",
4175 "f32", v2f32, v2f32, int_arm_neon_vpmins, 0>;
4177 // Vector Reciprocal and Reciprocal Square Root Estimate and Step.
4179 // VRECPE : Vector Reciprocal Estimate
4180 def VRECPEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4181 IIC_VUNAD, "vrecpe", "u32",
4182 v2i32, v2i32, int_arm_neon_vrecpe>;
4183 def VRECPEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01000, 0,
4184 IIC_VUNAQ, "vrecpe", "u32",
4185 v4i32, v4i32, int_arm_neon_vrecpe>;
4186 def VRECPEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4187 IIC_VUNAD, "vrecpe", "f32",
4188 v2f32, v2f32, int_arm_neon_vrecpe>;
4189 def VRECPEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01010, 0,
4190 IIC_VUNAQ, "vrecpe", "f32",
4191 v4f32, v4f32, int_arm_neon_vrecpe>;
4193 // VRECPS : Vector Reciprocal Step
4194 def VRECPSfd : N3VDInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4195 IIC_VRECSD, "vrecps", "f32",
4196 v2f32, v2f32, int_arm_neon_vrecps, 1>;
4197 def VRECPSfq : N3VQInt<0, 0, 0b00, 0b1111, 1, N3RegFrm,
4198 IIC_VRECSQ, "vrecps", "f32",
4199 v4f32, v4f32, int_arm_neon_vrecps, 1>;
4201 // VRSQRTE : Vector Reciprocal Square Root Estimate
4202 def VRSQRTEd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4203 IIC_VUNAD, "vrsqrte", "u32",
4204 v2i32, v2i32, int_arm_neon_vrsqrte>;
4205 def VRSQRTEq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01001, 0,
4206 IIC_VUNAQ, "vrsqrte", "u32",
4207 v4i32, v4i32, int_arm_neon_vrsqrte>;
4208 def VRSQRTEfd : N2VDInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4209 IIC_VUNAD, "vrsqrte", "f32",
4210 v2f32, v2f32, int_arm_neon_vrsqrte>;
4211 def VRSQRTEfq : N2VQInt<0b11, 0b11, 0b10, 0b11, 0b01011, 0,
4212 IIC_VUNAQ, "vrsqrte", "f32",
4213 v4f32, v4f32, int_arm_neon_vrsqrte>;
4215 // VRSQRTS : Vector Reciprocal Square Root Step
4216 def VRSQRTSfd : N3VDInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4217 IIC_VRECSD, "vrsqrts", "f32",
4218 v2f32, v2f32, int_arm_neon_vrsqrts, 1>;
4219 def VRSQRTSfq : N3VQInt<0, 0, 0b10, 0b1111, 1, N3RegFrm,
4220 IIC_VRECSQ, "vrsqrts", "f32",
4221 v4f32, v4f32, int_arm_neon_vrsqrts, 1>;
4225 // VSHL : Vector Shift
4226 defm VSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 0, N3RegVShFrm,
4227 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4228 "vshl", "s", int_arm_neon_vshifts>;
4229 defm VSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 0, N3RegVShFrm,
4230 IIC_VSHLiD, IIC_VSHLiD, IIC_VSHLiQ, IIC_VSHLiQ,
4231 "vshl", "u", int_arm_neon_vshiftu>;
4233 // VSHL : Vector Shift Left (Immediate)
4234 defm VSHLi : N2VShL_QHSD<0, 1, 0b0101, 1, IIC_VSHLiD, "vshl", "i", NEONvshl>;
4236 // VSHR : Vector Shift Right (Immediate)
4237 defm VSHRs : N2VShR_QHSD<0, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "s",NEONvshrs>;
4238 defm VSHRu : N2VShR_QHSD<1, 1, 0b0000, 1, IIC_VSHLiD, "vshr", "u",NEONvshru>;
4240 // VSHLL : Vector Shift Left Long
4241 defm VSHLLs : N2VLSh_QHS<0, 1, 0b1010, 0, 0, 1, "vshll", "s", NEONvshlls>;
4242 defm VSHLLu : N2VLSh_QHS<1, 1, 0b1010, 0, 0, 1, "vshll", "u", NEONvshllu>;
4244 // VSHLL : Vector Shift Left Long (with maximum shift count)
4245 class N2VLShMax<bit op24, bit op23, bits<6> op21_16, bits<4> op11_8, bit op7,
4246 bit op6, bit op4, string OpcodeStr, string Dt, ValueType ResTy,
4247 ValueType OpTy, SDNode OpNode>
4248 : N2VLSh<op24, op23, op11_8, op7, op6, op4, OpcodeStr, Dt,
4249 ResTy, OpTy, OpNode> {
4250 let Inst{21-16} = op21_16;
4251 let DecoderMethod = "DecodeVSHLMaxInstruction";
4253 def VSHLLi8 : N2VLShMax<1, 1, 0b110010, 0b0011, 0, 0, 0, "vshll", "i8",
4254 v8i16, v8i8, NEONvshlli>;
4255 def VSHLLi16 : N2VLShMax<1, 1, 0b110110, 0b0011, 0, 0, 0, "vshll", "i16",
4256 v4i32, v4i16, NEONvshlli>;
4257 def VSHLLi32 : N2VLShMax<1, 1, 0b111010, 0b0011, 0, 0, 0, "vshll", "i32",
4258 v2i64, v2i32, NEONvshlli>;
4260 // VSHRN : Vector Shift Right and Narrow
4261 defm VSHRN : N2VNSh_HSD<0,1,0b1000,0,0,1, IIC_VSHLiD, "vshrn", "i",
4264 // VRSHL : Vector Rounding Shift
4265 defm VRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 0, N3RegVShFrm,
4266 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4267 "vrshl", "s", int_arm_neon_vrshifts>;
4268 defm VRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 0, N3RegVShFrm,
4269 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4270 "vrshl", "u", int_arm_neon_vrshiftu>;
4271 // VRSHR : Vector Rounding Shift Right
4272 defm VRSHRs : N2VShR_QHSD<0,1,0b0010,1, IIC_VSHLi4D, "vrshr", "s",NEONvrshrs>;
4273 defm VRSHRu : N2VShR_QHSD<1,1,0b0010,1, IIC_VSHLi4D, "vrshr", "u",NEONvrshru>;
4275 // VRSHRN : Vector Rounding Shift Right and Narrow
4276 defm VRSHRN : N2VNSh_HSD<0, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vrshrn", "i",
4279 // VQSHL : Vector Saturating Shift
4280 defm VQSHLs : N3VInt_QHSDSh<0, 0, 0b0100, 1, N3RegVShFrm,
4281 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4282 "vqshl", "s", int_arm_neon_vqshifts>;
4283 defm VQSHLu : N3VInt_QHSDSh<1, 0, 0b0100, 1, N3RegVShFrm,
4284 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4285 "vqshl", "u", int_arm_neon_vqshiftu>;
4286 // VQSHL : Vector Saturating Shift Left (Immediate)
4287 defm VQSHLsi : N2VShL_QHSD<0,1,0b0111,1, IIC_VSHLi4D, "vqshl", "s",NEONvqshls>;
4288 defm VQSHLui : N2VShL_QHSD<1,1,0b0111,1, IIC_VSHLi4D, "vqshl", "u",NEONvqshlu>;
4290 // VQSHLU : Vector Saturating Shift Left (Immediate, Unsigned)
4291 defm VQSHLsu : N2VShL_QHSD<1,1,0b0110,1, IIC_VSHLi4D,"vqshlu","s",NEONvqshlsu>;
4293 // VQSHRN : Vector Saturating Shift Right and Narrow
4294 defm VQSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "s",
4296 defm VQSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 0, 1, IIC_VSHLi4D, "vqshrn", "u",
4299 // VQSHRUN : Vector Saturating Shift Right and Narrow (Unsigned)
4300 defm VQSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 0, 1, IIC_VSHLi4D, "vqshrun", "s",
4303 // VQRSHL : Vector Saturating Rounding Shift
4304 defm VQRSHLs : N3VInt_QHSDSh<0, 0, 0b0101, 1, N3RegVShFrm,
4305 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4306 "vqrshl", "s", int_arm_neon_vqrshifts>;
4307 defm VQRSHLu : N3VInt_QHSDSh<1, 0, 0b0101, 1, N3RegVShFrm,
4308 IIC_VSHLi4D, IIC_VSHLi4D, IIC_VSHLi4Q, IIC_VSHLi4Q,
4309 "vqrshl", "u", int_arm_neon_vqrshiftu>;
4311 // VQRSHRN : Vector Saturating Rounding Shift Right and Narrow
4312 defm VQRSHRNs : N2VNSh_HSD<0, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "s",
4314 defm VQRSHRNu : N2VNSh_HSD<1, 1, 0b1001, 0, 1, 1, IIC_VSHLi4D, "vqrshrn", "u",
4317 // VQRSHRUN : Vector Saturating Rounding Shift Right and Narrow (Unsigned)
4318 defm VQRSHRUN : N2VNSh_HSD<1, 1, 0b1000, 0, 1, 1, IIC_VSHLi4D, "vqrshrun", "s",
4321 // VSRA : Vector Shift Right and Accumulate
4322 defm VSRAs : N2VShAdd_QHSD<0, 1, 0b0001, 1, "vsra", "s", NEONvshrs>;
4323 defm VSRAu : N2VShAdd_QHSD<1, 1, 0b0001, 1, "vsra", "u", NEONvshru>;
4324 // VRSRA : Vector Rounding Shift Right and Accumulate
4325 defm VRSRAs : N2VShAdd_QHSD<0, 1, 0b0011, 1, "vrsra", "s", NEONvrshrs>;
4326 defm VRSRAu : N2VShAdd_QHSD<1, 1, 0b0011, 1, "vrsra", "u", NEONvrshru>;
4328 // VSLI : Vector Shift Left and Insert
4329 defm VSLI : N2VShInsL_QHSD<1, 1, 0b0101, 1, "vsli">;
4331 // VSRI : Vector Shift Right and Insert
4332 defm VSRI : N2VShInsR_QHSD<1, 1, 0b0100, 1, "vsri">;
4334 // Vector Absolute and Saturating Absolute.
4336 // VABS : Vector Absolute Value
4337 defm VABS : N2VInt_QHS<0b11, 0b11, 0b01, 0b00110, 0,
4338 IIC_VUNAiD, IIC_VUNAiQ, "vabs", "s",
4340 def VABSfd : N2VDInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4341 IIC_VUNAD, "vabs", "f32",
4342 v2f32, v2f32, int_arm_neon_vabs>;
4343 def VABSfq : N2VQInt<0b11, 0b11, 0b10, 0b01, 0b01110, 0,
4344 IIC_VUNAQ, "vabs", "f32",
4345 v4f32, v4f32, int_arm_neon_vabs>;
4347 // VQABS : Vector Saturating Absolute Value
4348 defm VQABS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01110, 0,
4349 IIC_VQUNAiD, IIC_VQUNAiQ, "vqabs", "s",
4350 int_arm_neon_vqabs>;
4354 def vnegd : PatFrag<(ops node:$in),
4355 (sub (bitconvert (v2i32 NEONimmAllZerosV)), node:$in)>;
4356 def vnegq : PatFrag<(ops node:$in),
4357 (sub (bitconvert (v4i32 NEONimmAllZerosV)), node:$in)>;
4359 class VNEGD<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4360 : N2V<0b11, 0b11, size, 0b01, 0b00111, 0, 0, (outs DPR:$Vd), (ins DPR:$Vm),
4361 IIC_VSHLiD, OpcodeStr, Dt, "$Vd, $Vm", "",
4362 [(set DPR:$Vd, (Ty (vnegd DPR:$Vm)))]>;
4363 class VNEGQ<bits<2> size, string OpcodeStr, string Dt, ValueType Ty>
4364 : N2V<0b11, 0b11, size, 0b01, 0b00111, 1, 0, (outs QPR:$Vd), (ins QPR:$Vm),
4365 IIC_VSHLiQ, OpcodeStr, Dt, "$Vd, $Vm", "",
4366 [(set QPR:$Vd, (Ty (vnegq QPR:$Vm)))]>;
4368 // VNEG : Vector Negate (integer)
4369 def VNEGs8d : VNEGD<0b00, "vneg", "s8", v8i8>;
4370 def VNEGs16d : VNEGD<0b01, "vneg", "s16", v4i16>;
4371 def VNEGs32d : VNEGD<0b10, "vneg", "s32", v2i32>;
4372 def VNEGs8q : VNEGQ<0b00, "vneg", "s8", v16i8>;
4373 def VNEGs16q : VNEGQ<0b01, "vneg", "s16", v8i16>;
4374 def VNEGs32q : VNEGQ<0b10, "vneg", "s32", v4i32>;
4376 // VNEG : Vector Negate (floating-point)
4377 def VNEGfd : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 0, 0,
4378 (outs DPR:$Vd), (ins DPR:$Vm), IIC_VUNAD,
4379 "vneg", "f32", "$Vd, $Vm", "",
4380 [(set DPR:$Vd, (v2f32 (fneg DPR:$Vm)))]>;
4381 def VNEGf32q : N2V<0b11, 0b11, 0b10, 0b01, 0b01111, 1, 0,
4382 (outs QPR:$Vd), (ins QPR:$Vm), IIC_VUNAQ,
4383 "vneg", "f32", "$Vd, $Vm", "",
4384 [(set QPR:$Vd, (v4f32 (fneg QPR:$Vm)))]>;
4386 def : Pat<(v8i8 (vnegd DPR:$src)), (VNEGs8d DPR:$src)>;
4387 def : Pat<(v4i16 (vnegd DPR:$src)), (VNEGs16d DPR:$src)>;
4388 def : Pat<(v2i32 (vnegd DPR:$src)), (VNEGs32d DPR:$src)>;
4389 def : Pat<(v16i8 (vnegq QPR:$src)), (VNEGs8q QPR:$src)>;
4390 def : Pat<(v8i16 (vnegq QPR:$src)), (VNEGs16q QPR:$src)>;
4391 def : Pat<(v4i32 (vnegq QPR:$src)), (VNEGs32q QPR:$src)>;
4393 // VQNEG : Vector Saturating Negate
4394 defm VQNEG : N2VInt_QHS<0b11, 0b11, 0b00, 0b01111, 0,
4395 IIC_VQUNAiD, IIC_VQUNAiQ, "vqneg", "s",
4396 int_arm_neon_vqneg>;
4398 // Vector Bit Counting Operations.
4400 // VCLS : Vector Count Leading Sign Bits
4401 defm VCLS : N2VInt_QHS<0b11, 0b11, 0b00, 0b01000, 0,
4402 IIC_VCNTiD, IIC_VCNTiQ, "vcls", "s",
4404 // VCLZ : Vector Count Leading Zeros
4405 defm VCLZ : N2VInt_QHS<0b11, 0b11, 0b00, 0b01001, 0,
4406 IIC_VCNTiD, IIC_VCNTiQ, "vclz", "i",
4408 // VCNT : Vector Count One Bits
4409 def VCNTd : N2VDInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4410 IIC_VCNTiD, "vcnt", "8",
4411 v8i8, v8i8, int_arm_neon_vcnt>;
4412 def VCNTq : N2VQInt<0b11, 0b11, 0b00, 0b00, 0b01010, 0,
4413 IIC_VCNTiQ, "vcnt", "8",
4414 v16i8, v16i8, int_arm_neon_vcnt>;
4417 def VSWPd : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 0, 0,
4418 (outs DPR:$Vd), (ins DPR:$Vm), NoItinerary,
4419 "vswp", "$Vd, $Vm", "", []>;
4420 def VSWPq : N2VX<0b11, 0b11, 0b00, 0b10, 0b00000, 1, 0,
4421 (outs QPR:$Vd), (ins QPR:$Vm), NoItinerary,
4422 "vswp", "$Vd, $Vm", "", []>;
4424 // Vector Move Operations.
4426 // VMOV : Vector Move (Register)
4427 def : InstAlias<"vmov${p} $Vd, $Vm",
4428 (VORRd DPR:$Vd, DPR:$Vm, DPR:$Vm, pred:$p)>;
4429 def : InstAlias<"vmov${p} $Vd, $Vm",
4430 (VORRq QPR:$Vd, QPR:$Vm, QPR:$Vm, pred:$p)>;
4432 // VMOV : Vector Move (Immediate)
4434 let isReMaterializable = 1 in {
4435 def VMOVv8i8 : N1ModImm<1, 0b000, 0b1110, 0, 0, 0, 1, (outs DPR:$Vd),
4436 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4437 "vmov", "i8", "$Vd, $SIMM", "",
4438 [(set DPR:$Vd, (v8i8 (NEONvmovImm timm:$SIMM)))]>;
4439 def VMOVv16i8 : N1ModImm<1, 0b000, 0b1110, 0, 1, 0, 1, (outs QPR:$Vd),
4440 (ins nImmSplatI8:$SIMM), IIC_VMOVImm,
4441 "vmov", "i8", "$Vd, $SIMM", "",
4442 [(set QPR:$Vd, (v16i8 (NEONvmovImm timm:$SIMM)))]>;
4444 def VMOVv4i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 0, 0, 1, (outs DPR:$Vd),
4445 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4446 "vmov", "i16", "$Vd, $SIMM", "",
4447 [(set DPR:$Vd, (v4i16 (NEONvmovImm timm:$SIMM)))]> {
4448 let Inst{9} = SIMM{9};
4451 def VMOVv8i16 : N1ModImm<1, 0b000, {1,0,?,0}, 0, 1, 0, 1, (outs QPR:$Vd),
4452 (ins nImmSplatI16:$SIMM), IIC_VMOVImm,
4453 "vmov", "i16", "$Vd, $SIMM", "",
4454 [(set QPR:$Vd, (v8i16 (NEONvmovImm timm:$SIMM)))]> {
4455 let Inst{9} = SIMM{9};
4458 def VMOVv2i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 0, 0, 1, (outs DPR:$Vd),
4459 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4460 "vmov", "i32", "$Vd, $SIMM", "",
4461 [(set DPR:$Vd, (v2i32 (NEONvmovImm timm:$SIMM)))]> {
4462 let Inst{11-8} = SIMM{11-8};
4465 def VMOVv4i32 : N1ModImm<1, 0b000, {?,?,?,?}, 0, 1, 0, 1, (outs QPR:$Vd),
4466 (ins nImmVMOVI32:$SIMM), IIC_VMOVImm,
4467 "vmov", "i32", "$Vd, $SIMM", "",
4468 [(set QPR:$Vd, (v4i32 (NEONvmovImm timm:$SIMM)))]> {
4469 let Inst{11-8} = SIMM{11-8};
4472 def VMOVv1i64 : N1ModImm<1, 0b000, 0b1110, 0, 0, 1, 1, (outs DPR:$Vd),
4473 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4474 "vmov", "i64", "$Vd, $SIMM", "",
4475 [(set DPR:$Vd, (v1i64 (NEONvmovImm timm:$SIMM)))]>;
4476 def VMOVv2i64 : N1ModImm<1, 0b000, 0b1110, 0, 1, 1, 1, (outs QPR:$Vd),
4477 (ins nImmSplatI64:$SIMM), IIC_VMOVImm,
4478 "vmov", "i64", "$Vd, $SIMM", "",
4479 [(set QPR:$Vd, (v2i64 (NEONvmovImm timm:$SIMM)))]>;
4480 } // isReMaterializable
4482 // VMOV : Vector Get Lane (move scalar to ARM core register)
4484 def VGETLNs8 : NVGetLane<{1,1,1,0,0,1,?,1}, 0b1011, {?,?},
4485 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4486 IIC_VMOVSI, "vmov", "s8", "$R, $V$lane",
4487 [(set GPR:$R, (NEONvgetlanes (v8i8 DPR:$V),
4489 let Inst{21} = lane{2};
4490 let Inst{6-5} = lane{1-0};
4492 def VGETLNs16 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, {?,1},
4493 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4494 IIC_VMOVSI, "vmov", "s16", "$R, $V$lane",
4495 [(set GPR:$R, (NEONvgetlanes (v4i16 DPR:$V),
4497 let Inst{21} = lane{1};
4498 let Inst{6} = lane{0};
4500 def VGETLNu8 : NVGetLane<{1,1,1,0,1,1,?,1}, 0b1011, {?,?},
4501 (outs GPR:$R), (ins DPR:$V, VectorIndex8:$lane),
4502 IIC_VMOVSI, "vmov", "u8", "$R, $V$lane",
4503 [(set GPR:$R, (NEONvgetlaneu (v8i8 DPR:$V),
4505 let Inst{21} = lane{2};
4506 let Inst{6-5} = lane{1-0};
4508 def VGETLNu16 : NVGetLane<{1,1,1,0,1,0,?,1}, 0b1011, {?,1},
4509 (outs GPR:$R), (ins DPR:$V, VectorIndex16:$lane),
4510 IIC_VMOVSI, "vmov", "u16", "$R, $V$lane",
4511 [(set GPR:$R, (NEONvgetlaneu (v4i16 DPR:$V),
4513 let Inst{21} = lane{1};
4514 let Inst{6} = lane{0};
4516 def VGETLNi32 : NVGetLane<{1,1,1,0,0,0,?,1}, 0b1011, 0b00,
4517 (outs GPR:$R), (ins DPR:$V, VectorIndex32:$lane),
4518 IIC_VMOVSI, "vmov", "32", "$R, $V$lane",
4519 [(set GPR:$R, (extractelt (v2i32 DPR:$V),
4521 let Inst{21} = lane{0};
4523 // def VGETLNf32: see FMRDH and FMRDL in ARMInstrVFP.td
4524 def : Pat<(NEONvgetlanes (v16i8 QPR:$src), imm:$lane),
4525 (VGETLNs8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4526 (DSubReg_i8_reg imm:$lane))),
4527 (SubReg_i8_lane imm:$lane))>;
4528 def : Pat<(NEONvgetlanes (v8i16 QPR:$src), imm:$lane),
4529 (VGETLNs16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4530 (DSubReg_i16_reg imm:$lane))),
4531 (SubReg_i16_lane imm:$lane))>;
4532 def : Pat<(NEONvgetlaneu (v16i8 QPR:$src), imm:$lane),
4533 (VGETLNu8 (v8i8 (EXTRACT_SUBREG QPR:$src,
4534 (DSubReg_i8_reg imm:$lane))),
4535 (SubReg_i8_lane imm:$lane))>;
4536 def : Pat<(NEONvgetlaneu (v8i16 QPR:$src), imm:$lane),
4537 (VGETLNu16 (v4i16 (EXTRACT_SUBREG QPR:$src,
4538 (DSubReg_i16_reg imm:$lane))),
4539 (SubReg_i16_lane imm:$lane))>;
4540 def : Pat<(extractelt (v4i32 QPR:$src), imm:$lane),
4541 (VGETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src,
4542 (DSubReg_i32_reg imm:$lane))),
4543 (SubReg_i32_lane imm:$lane))>;
4544 def : Pat<(extractelt (v2f32 DPR:$src1), imm:$src2),
4545 (EXTRACT_SUBREG (v2f32 (COPY_TO_REGCLASS (v2f32 DPR:$src1),DPR_VFP2)),
4546 (SSubReg_f32_reg imm:$src2))>;
4547 def : Pat<(extractelt (v4f32 QPR:$src1), imm:$src2),
4548 (EXTRACT_SUBREG (v4f32 (COPY_TO_REGCLASS (v4f32 QPR:$src1),QPR_VFP2)),
4549 (SSubReg_f32_reg imm:$src2))>;
4550 //def : Pat<(extractelt (v2i64 QPR:$src1), imm:$src2),
4551 // (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4552 def : Pat<(extractelt (v2f64 QPR:$src1), imm:$src2),
4553 (EXTRACT_SUBREG QPR:$src1, (DSubReg_f64_reg imm:$src2))>;
4556 // VMOV : Vector Set Lane (move ARM core register to scalar)
4558 let Constraints = "$src1 = $V" in {
4559 def VSETLNi8 : NVSetLane<{1,1,1,0,0,1,?,0}, 0b1011, {?,?}, (outs DPR:$V),
4560 (ins DPR:$src1, GPR:$R, VectorIndex8:$lane),
4561 IIC_VMOVISL, "vmov", "8", "$V$lane, $R",
4562 [(set DPR:$V, (vector_insert (v8i8 DPR:$src1),
4563 GPR:$R, imm:$lane))]> {
4564 let Inst{21} = lane{2};
4565 let Inst{6-5} = lane{1-0};
4567 def VSETLNi16 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, {?,1}, (outs DPR:$V),
4568 (ins DPR:$src1, GPR:$R, VectorIndex16:$lane),
4569 IIC_VMOVISL, "vmov", "16", "$V$lane, $R",
4570 [(set DPR:$V, (vector_insert (v4i16 DPR:$src1),
4571 GPR:$R, imm:$lane))]> {
4572 let Inst{21} = lane{1};
4573 let Inst{6} = lane{0};
4575 def VSETLNi32 : NVSetLane<{1,1,1,0,0,0,?,0}, 0b1011, 0b00, (outs DPR:$V),
4576 (ins DPR:$src1, GPR:$R, VectorIndex32:$lane),
4577 IIC_VMOVISL, "vmov", "32", "$V$lane, $R",
4578 [(set DPR:$V, (insertelt (v2i32 DPR:$src1),
4579 GPR:$R, imm:$lane))]> {
4580 let Inst{21} = lane{0};
4583 def : Pat<(vector_insert (v16i8 QPR:$src1), GPR:$src2, imm:$lane),
4584 (v16i8 (INSERT_SUBREG QPR:$src1,
4585 (v8i8 (VSETLNi8 (v8i8 (EXTRACT_SUBREG QPR:$src1,
4586 (DSubReg_i8_reg imm:$lane))),
4587 GPR:$src2, (SubReg_i8_lane imm:$lane))),
4588 (DSubReg_i8_reg imm:$lane)))>;
4589 def : Pat<(vector_insert (v8i16 QPR:$src1), GPR:$src2, imm:$lane),
4590 (v8i16 (INSERT_SUBREG QPR:$src1,
4591 (v4i16 (VSETLNi16 (v4i16 (EXTRACT_SUBREG QPR:$src1,
4592 (DSubReg_i16_reg imm:$lane))),
4593 GPR:$src2, (SubReg_i16_lane imm:$lane))),
4594 (DSubReg_i16_reg imm:$lane)))>;
4595 def : Pat<(insertelt (v4i32 QPR:$src1), GPR:$src2, imm:$lane),
4596 (v4i32 (INSERT_SUBREG QPR:$src1,
4597 (v2i32 (VSETLNi32 (v2i32 (EXTRACT_SUBREG QPR:$src1,
4598 (DSubReg_i32_reg imm:$lane))),
4599 GPR:$src2, (SubReg_i32_lane imm:$lane))),
4600 (DSubReg_i32_reg imm:$lane)))>;
4602 def : Pat<(v2f32 (insertelt DPR:$src1, SPR:$src2, imm:$src3)),
4603 (INSERT_SUBREG (v2f32 (COPY_TO_REGCLASS DPR:$src1, DPR_VFP2)),
4604 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4605 def : Pat<(v4f32 (insertelt QPR:$src1, SPR:$src2, imm:$src3)),
4606 (INSERT_SUBREG (v4f32 (COPY_TO_REGCLASS QPR:$src1, QPR_VFP2)),
4607 SPR:$src2, (SSubReg_f32_reg imm:$src3))>;
4609 //def : Pat<(v2i64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4610 // (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4611 def : Pat<(v2f64 (insertelt QPR:$src1, DPR:$src2, imm:$src3)),
4612 (INSERT_SUBREG QPR:$src1, DPR:$src2, (DSubReg_f64_reg imm:$src3))>;
4614 def : Pat<(v2f32 (scalar_to_vector SPR:$src)),
4615 (INSERT_SUBREG (v2f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4616 def : Pat<(v2f64 (scalar_to_vector (f64 DPR:$src))),
4617 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), DPR:$src, dsub_0)>;
4618 def : Pat<(v4f32 (scalar_to_vector SPR:$src)),
4619 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), SPR:$src, ssub_0)>;
4621 def : Pat<(v8i8 (scalar_to_vector GPR:$src)),
4622 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4623 def : Pat<(v4i16 (scalar_to_vector GPR:$src)),
4624 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4625 def : Pat<(v2i32 (scalar_to_vector GPR:$src)),
4626 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0))>;
4628 def : Pat<(v16i8 (scalar_to_vector GPR:$src)),
4629 (INSERT_SUBREG (v16i8 (IMPLICIT_DEF)),
4630 (VSETLNi8 (v8i8 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4632 def : Pat<(v8i16 (scalar_to_vector GPR:$src)),
4633 (INSERT_SUBREG (v8i16 (IMPLICIT_DEF)),
4634 (VSETLNi16 (v4i16 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4636 def : Pat<(v4i32 (scalar_to_vector GPR:$src)),
4637 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
4638 (VSETLNi32 (v2i32 (IMPLICIT_DEF)), GPR:$src, (i32 0)),
4641 // VDUP : Vector Duplicate (from ARM core register to all elements)
4643 class VDUPD<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4644 : NVDup<opcod1, 0b1011, opcod3, (outs DPR:$V), (ins GPR:$R),
4645 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4646 [(set DPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4647 class VDUPQ<bits<8> opcod1, bits<2> opcod3, string Dt, ValueType Ty>
4648 : NVDup<opcod1, 0b1011, opcod3, (outs QPR:$V), (ins GPR:$R),
4649 IIC_VMOVIS, "vdup", Dt, "$V, $R",
4650 [(set QPR:$V, (Ty (NEONvdup (i32 GPR:$R))))]>;
4652 def VDUP8d : VDUPD<0b11101100, 0b00, "8", v8i8>;
4653 def VDUP16d : VDUPD<0b11101000, 0b01, "16", v4i16>;
4654 def VDUP32d : VDUPD<0b11101000, 0b00, "32", v2i32>;
4655 def VDUP8q : VDUPQ<0b11101110, 0b00, "8", v16i8>;
4656 def VDUP16q : VDUPQ<0b11101010, 0b01, "16", v8i16>;
4657 def VDUP32q : VDUPQ<0b11101010, 0b00, "32", v4i32>;
4659 def : Pat<(v2f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32d GPR:$R)>;
4660 def : Pat<(v4f32 (NEONvdup (f32 (bitconvert GPR:$R)))), (VDUP32q GPR:$R)>;
4662 // VDUP : Vector Duplicate Lane (from scalar to all elements)
4664 class VDUPLND<bits<4> op19_16, string OpcodeStr, string Dt,
4665 ValueType Ty, Operand IdxTy>
4666 : NVDupLane<op19_16, 0, (outs DPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4667 IIC_VMOVD, OpcodeStr, Dt, "$Vd, $Vm$lane",
4668 [(set DPR:$Vd, (Ty (NEONvduplane (Ty DPR:$Vm), imm:$lane)))]>;
4670 class VDUPLNQ<bits<4> op19_16, string OpcodeStr, string Dt,
4671 ValueType ResTy, ValueType OpTy, Operand IdxTy>
4672 : NVDupLane<op19_16, 1, (outs QPR:$Vd), (ins DPR:$Vm, IdxTy:$lane),
4673 IIC_VMOVQ, OpcodeStr, Dt, "$Vd, $Vm$lane",
4674 [(set QPR:$Vd, (ResTy (NEONvduplane (OpTy DPR:$Vm),
4675 VectorIndex32:$lane)))]>;
4677 // Inst{19-16} is partially specified depending on the element size.
4679 def VDUPLN8d : VDUPLND<{?,?,?,1}, "vdup", "8", v8i8, VectorIndex8> {
4681 let Inst{19-17} = lane{2-0};
4683 def VDUPLN16d : VDUPLND<{?,?,1,0}, "vdup", "16", v4i16, VectorIndex16> {
4685 let Inst{19-18} = lane{1-0};
4687 def VDUPLN32d : VDUPLND<{?,1,0,0}, "vdup", "32", v2i32, VectorIndex32> {
4689 let Inst{19} = lane{0};
4691 def VDUPLN8q : VDUPLNQ<{?,?,?,1}, "vdup", "8", v16i8, v8i8, VectorIndex8> {
4693 let Inst{19-17} = lane{2-0};
4695 def VDUPLN16q : VDUPLNQ<{?,?,1,0}, "vdup", "16", v8i16, v4i16, VectorIndex16> {
4697 let Inst{19-18} = lane{1-0};
4699 def VDUPLN32q : VDUPLNQ<{?,1,0,0}, "vdup", "32", v4i32, v2i32, VectorIndex32> {
4701 let Inst{19} = lane{0};
4704 def : Pat<(v2f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4705 (VDUPLN32d DPR:$Vm, imm:$lane)>;
4707 def : Pat<(v4f32 (NEONvduplane (v2f32 DPR:$Vm), imm:$lane)),
4708 (VDUPLN32q DPR:$Vm, imm:$lane)>;
4710 def : Pat<(v16i8 (NEONvduplane (v16i8 QPR:$src), imm:$lane)),
4711 (v16i8 (VDUPLN8q (v8i8 (EXTRACT_SUBREG QPR:$src,
4712 (DSubReg_i8_reg imm:$lane))),
4713 (SubReg_i8_lane imm:$lane)))>;
4714 def : Pat<(v8i16 (NEONvduplane (v8i16 QPR:$src), imm:$lane)),
4715 (v8i16 (VDUPLN16q (v4i16 (EXTRACT_SUBREG QPR:$src,
4716 (DSubReg_i16_reg imm:$lane))),
4717 (SubReg_i16_lane imm:$lane)))>;
4718 def : Pat<(v4i32 (NEONvduplane (v4i32 QPR:$src), imm:$lane)),
4719 (v4i32 (VDUPLN32q (v2i32 (EXTRACT_SUBREG QPR:$src,
4720 (DSubReg_i32_reg imm:$lane))),
4721 (SubReg_i32_lane imm:$lane)))>;
4722 def : Pat<(v4f32 (NEONvduplane (v4f32 QPR:$src), imm:$lane)),
4723 (v4f32 (VDUPLN32q (v2f32 (EXTRACT_SUBREG QPR:$src,
4724 (DSubReg_i32_reg imm:$lane))),
4725 (SubReg_i32_lane imm:$lane)))>;
4727 def VDUPfdf : PseudoNeonI<(outs DPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4728 [(set DPR:$dst, (v2f32 (NEONvdup (f32 SPR:$src))))]>;
4729 def VDUPfqf : PseudoNeonI<(outs QPR:$dst), (ins SPR:$src), IIC_VMOVD, "",
4730 [(set QPR:$dst, (v4f32 (NEONvdup (f32 SPR:$src))))]>;
4732 // VMOVN : Vector Narrowing Move
4733 defm VMOVN : N2VN_HSD<0b11,0b11,0b10,0b00100,0,0, IIC_VMOVN,
4734 "vmovn", "i", trunc>;
4735 // VQMOVN : Vector Saturating Narrowing Move
4736 defm VQMOVNs : N2VNInt_HSD<0b11,0b11,0b10,0b00101,0,0, IIC_VQUNAiD,
4737 "vqmovn", "s", int_arm_neon_vqmovns>;
4738 defm VQMOVNu : N2VNInt_HSD<0b11,0b11,0b10,0b00101,1,0, IIC_VQUNAiD,
4739 "vqmovn", "u", int_arm_neon_vqmovnu>;
4740 defm VQMOVNsu : N2VNInt_HSD<0b11,0b11,0b10,0b00100,1,0, IIC_VQUNAiD,
4741 "vqmovun", "s", int_arm_neon_vqmovnsu>;
4742 // VMOVL : Vector Lengthening Move
4743 defm VMOVLs : N2VL_QHS<0b01,0b10100,0,1, "vmovl", "s", sext>;
4744 defm VMOVLu : N2VL_QHS<0b11,0b10100,0,1, "vmovl", "u", zext>;
4746 // Vector Conversions.
4748 // VCVT : Vector Convert Between Floating-Point and Integers
4749 def VCVTf2sd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4750 v2i32, v2f32, fp_to_sint>;
4751 def VCVTf2ud : N2VD<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4752 v2i32, v2f32, fp_to_uint>;
4753 def VCVTs2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4754 v2f32, v2i32, sint_to_fp>;
4755 def VCVTu2fd : N2VD<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4756 v2f32, v2i32, uint_to_fp>;
4758 def VCVTf2sq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01110, 0, "vcvt", "s32.f32",
4759 v4i32, v4f32, fp_to_sint>;
4760 def VCVTf2uq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01111, 0, "vcvt", "u32.f32",
4761 v4i32, v4f32, fp_to_uint>;
4762 def VCVTs2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01100, 0, "vcvt", "f32.s32",
4763 v4f32, v4i32, sint_to_fp>;
4764 def VCVTu2fq : N2VQ<0b11, 0b11, 0b10, 0b11, 0b01101, 0, "vcvt", "f32.u32",
4765 v4f32, v4i32, uint_to_fp>;
4767 // VCVT : Vector Convert Between Floating-Point and Fixed-Point.
4768 def VCVTf2xsd : N2VCvtD<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4769 v2i32, v2f32, int_arm_neon_vcvtfp2fxs>;
4770 def VCVTf2xud : N2VCvtD<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4771 v2i32, v2f32, int_arm_neon_vcvtfp2fxu>;
4772 def VCVTxs2fd : N2VCvtD<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4773 v2f32, v2i32, int_arm_neon_vcvtfxs2fp>;
4774 def VCVTxu2fd : N2VCvtD<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4775 v2f32, v2i32, int_arm_neon_vcvtfxu2fp>;
4777 def VCVTf2xsq : N2VCvtQ<0, 1, 0b1111, 0, 1, "vcvt", "s32.f32",
4778 v4i32, v4f32, int_arm_neon_vcvtfp2fxs>;
4779 def VCVTf2xuq : N2VCvtQ<1, 1, 0b1111, 0, 1, "vcvt", "u32.f32",
4780 v4i32, v4f32, int_arm_neon_vcvtfp2fxu>;
4781 def VCVTxs2fq : N2VCvtQ<0, 1, 0b1110, 0, 1, "vcvt", "f32.s32",
4782 v4f32, v4i32, int_arm_neon_vcvtfxs2fp>;
4783 def VCVTxu2fq : N2VCvtQ<1, 1, 0b1110, 0, 1, "vcvt", "f32.u32",
4784 v4f32, v4i32, int_arm_neon_vcvtfxu2fp>;
4786 // VCVT : Vector Convert Between Half-Precision and Single-Precision.
4787 def VCVTf2h : N2VNInt<0b11, 0b11, 0b01, 0b10, 0b01100, 0, 0,
4788 IIC_VUNAQ, "vcvt", "f16.f32",
4789 v4i16, v4f32, int_arm_neon_vcvtfp2hf>,
4790 Requires<[HasNEON, HasFP16]>;
4791 def VCVTh2f : N2VLInt<0b11, 0b11, 0b01, 0b10, 0b01110, 0, 0,
4792 IIC_VUNAQ, "vcvt", "f32.f16",
4793 v4f32, v4i16, int_arm_neon_vcvthf2fp>,
4794 Requires<[HasNEON, HasFP16]>;
4798 // VREV64 : Vector Reverse elements within 64-bit doublewords
4800 class VREV64D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4801 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 0, 0, (outs DPR:$Vd),
4802 (ins DPR:$Vm), IIC_VMOVD,
4803 OpcodeStr, Dt, "$Vd, $Vm", "",
4804 [(set DPR:$Vd, (Ty (NEONvrev64 (Ty DPR:$Vm))))]>;
4805 class VREV64Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4806 : N2V<0b11, 0b11, op19_18, 0b00, 0b00000, 1, 0, (outs QPR:$Vd),
4807 (ins QPR:$Vm), IIC_VMOVQ,
4808 OpcodeStr, Dt, "$Vd, $Vm", "",
4809 [(set QPR:$Vd, (Ty (NEONvrev64 (Ty QPR:$Vm))))]>;
4811 def VREV64d8 : VREV64D<0b00, "vrev64", "8", v8i8>;
4812 def VREV64d16 : VREV64D<0b01, "vrev64", "16", v4i16>;
4813 def VREV64d32 : VREV64D<0b10, "vrev64", "32", v2i32>;
4814 def : Pat<(v2f32 (NEONvrev64 (v2f32 DPR:$Vm))), (VREV64d32 DPR:$Vm)>;
4816 def VREV64q8 : VREV64Q<0b00, "vrev64", "8", v16i8>;
4817 def VREV64q16 : VREV64Q<0b01, "vrev64", "16", v8i16>;
4818 def VREV64q32 : VREV64Q<0b10, "vrev64", "32", v4i32>;
4819 def : Pat<(v4f32 (NEONvrev64 (v4f32 QPR:$Vm))), (VREV64q32 QPR:$Vm)>;
4821 // VREV32 : Vector Reverse elements within 32-bit words
4823 class VREV32D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4824 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 0, 0, (outs DPR:$Vd),
4825 (ins DPR:$Vm), IIC_VMOVD,
4826 OpcodeStr, Dt, "$Vd, $Vm", "",
4827 [(set DPR:$Vd, (Ty (NEONvrev32 (Ty DPR:$Vm))))]>;
4828 class VREV32Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4829 : N2V<0b11, 0b11, op19_18, 0b00, 0b00001, 1, 0, (outs QPR:$Vd),
4830 (ins QPR:$Vm), IIC_VMOVQ,
4831 OpcodeStr, Dt, "$Vd, $Vm", "",
4832 [(set QPR:$Vd, (Ty (NEONvrev32 (Ty QPR:$Vm))))]>;
4834 def VREV32d8 : VREV32D<0b00, "vrev32", "8", v8i8>;
4835 def VREV32d16 : VREV32D<0b01, "vrev32", "16", v4i16>;
4837 def VREV32q8 : VREV32Q<0b00, "vrev32", "8", v16i8>;
4838 def VREV32q16 : VREV32Q<0b01, "vrev32", "16", v8i16>;
4840 // VREV16 : Vector Reverse elements within 16-bit halfwords
4842 class VREV16D<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4843 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 0, 0, (outs DPR:$Vd),
4844 (ins DPR:$Vm), IIC_VMOVD,
4845 OpcodeStr, Dt, "$Vd, $Vm", "",
4846 [(set DPR:$Vd, (Ty (NEONvrev16 (Ty DPR:$Vm))))]>;
4847 class VREV16Q<bits<2> op19_18, string OpcodeStr, string Dt, ValueType Ty>
4848 : N2V<0b11, 0b11, op19_18, 0b00, 0b00010, 1, 0, (outs QPR:$Vd),
4849 (ins QPR:$Vm), IIC_VMOVQ,
4850 OpcodeStr, Dt, "$Vd, $Vm", "",
4851 [(set QPR:$Vd, (Ty (NEONvrev16 (Ty QPR:$Vm))))]>;
4853 def VREV16d8 : VREV16D<0b00, "vrev16", "8", v8i8>;
4854 def VREV16q8 : VREV16Q<0b00, "vrev16", "8", v16i8>;
4856 // Other Vector Shuffles.
4858 // Aligned extractions: really just dropping registers
4860 class AlignedVEXTq<ValueType DestTy, ValueType SrcTy, SDNodeXForm LaneCVT>
4861 : Pat<(DestTy (vector_extract_subvec (SrcTy QPR:$src), (i32 imm:$start))),
4862 (EXTRACT_SUBREG (SrcTy QPR:$src), (LaneCVT imm:$start))>;
4864 def : AlignedVEXTq<v8i8, v16i8, DSubReg_i8_reg>;
4866 def : AlignedVEXTq<v4i16, v8i16, DSubReg_i16_reg>;
4868 def : AlignedVEXTq<v2i32, v4i32, DSubReg_i32_reg>;
4870 def : AlignedVEXTq<v1i64, v2i64, DSubReg_f64_reg>;
4872 def : AlignedVEXTq<v2f32, v4f32, DSubReg_i32_reg>;
4875 // VEXT : Vector Extract
4877 class VEXTd<string OpcodeStr, string Dt, ValueType Ty>
4878 : N3V<0,1,0b11,{?,?,?,?},0,0, (outs DPR:$Vd),
4879 (ins DPR:$Vn, DPR:$Vm, i32imm:$index), NVExtFrm,
4880 IIC_VEXTD, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4881 [(set DPR:$Vd, (Ty (NEONvext (Ty DPR:$Vn),
4882 (Ty DPR:$Vm), imm:$index)))]> {
4884 let Inst{11-8} = index{3-0};
4887 class VEXTq<string OpcodeStr, string Dt, ValueType Ty>
4888 : N3V<0,1,0b11,{?,?,?,?},1,0, (outs QPR:$Vd),
4889 (ins QPR:$Vn, QPR:$Vm, i32imm:$index), NVExtFrm,
4890 IIC_VEXTQ, OpcodeStr, Dt, "$Vd, $Vn, $Vm, $index", "",
4891 [(set QPR:$Vd, (Ty (NEONvext (Ty QPR:$Vn),
4892 (Ty QPR:$Vm), imm:$index)))]> {
4894 let Inst{11-8} = index{3-0};
4897 def VEXTd8 : VEXTd<"vext", "8", v8i8> {
4898 let Inst{11-8} = index{3-0};
4900 def VEXTd16 : VEXTd<"vext", "16", v4i16> {
4901 let Inst{11-9} = index{2-0};
4904 def VEXTd32 : VEXTd<"vext", "32", v2i32> {
4905 let Inst{11-10} = index{1-0};
4906 let Inst{9-8} = 0b00;
4908 def : Pat<(v2f32 (NEONvext (v2f32 DPR:$Vn),
4911 (VEXTd32 DPR:$Vn, DPR:$Vm, imm:$index)>;
4913 def VEXTq8 : VEXTq<"vext", "8", v16i8> {
4914 let Inst{11-8} = index{3-0};
4916 def VEXTq16 : VEXTq<"vext", "16", v8i16> {
4917 let Inst{11-9} = index{2-0};
4920 def VEXTq32 : VEXTq<"vext", "32", v4i32> {
4921 let Inst{11-10} = index{1-0};
4922 let Inst{9-8} = 0b00;
4924 def : Pat<(v4f32 (NEONvext (v4f32 QPR:$Vn),
4927 (VEXTq32 QPR:$Vn, QPR:$Vm, imm:$index)>;
4929 // VTRN : Vector Transpose
4931 def VTRNd8 : N2VDShuffle<0b00, 0b00001, "vtrn", "8">;
4932 def VTRNd16 : N2VDShuffle<0b01, 0b00001, "vtrn", "16">;
4933 def VTRNd32 : N2VDShuffle<0b10, 0b00001, "vtrn", "32">;
4935 def VTRNq8 : N2VQShuffle<0b00, 0b00001, IIC_VPERMQ, "vtrn", "8">;
4936 def VTRNq16 : N2VQShuffle<0b01, 0b00001, IIC_VPERMQ, "vtrn", "16">;
4937 def VTRNq32 : N2VQShuffle<0b10, 0b00001, IIC_VPERMQ, "vtrn", "32">;
4939 // VUZP : Vector Unzip (Deinterleave)
4941 def VUZPd8 : N2VDShuffle<0b00, 0b00010, "vuzp", "8">;
4942 def VUZPd16 : N2VDShuffle<0b01, 0b00010, "vuzp", "16">;
4943 def VUZPd32 : N2VDShuffle<0b10, 0b00010, "vuzp", "32">;
4945 def VUZPq8 : N2VQShuffle<0b00, 0b00010, IIC_VPERMQ3, "vuzp", "8">;
4946 def VUZPq16 : N2VQShuffle<0b01, 0b00010, IIC_VPERMQ3, "vuzp", "16">;
4947 def VUZPq32 : N2VQShuffle<0b10, 0b00010, IIC_VPERMQ3, "vuzp", "32">;
4949 // VZIP : Vector Zip (Interleave)
4951 def VZIPd8 : N2VDShuffle<0b00, 0b00011, "vzip", "8">;
4952 def VZIPd16 : N2VDShuffle<0b01, 0b00011, "vzip", "16">;
4953 def VZIPd32 : N2VDShuffle<0b10, 0b00011, "vzip", "32">;
4955 def VZIPq8 : N2VQShuffle<0b00, 0b00011, IIC_VPERMQ3, "vzip", "8">;
4956 def VZIPq16 : N2VQShuffle<0b01, 0b00011, IIC_VPERMQ3, "vzip", "16">;
4957 def VZIPq32 : N2VQShuffle<0b10, 0b00011, IIC_VPERMQ3, "vzip", "32">;
4959 // Vector Table Lookup and Table Extension.
4961 // VTBL : Vector Table Lookup
4962 let DecoderMethod = "DecodeTBLInstruction" in {
4964 : N3V<1,1,0b11,0b1000,0,0, (outs DPR:$Vd),
4965 (ins VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTB1,
4966 "vtbl", "8", "$Vd, $Vn, $Vm", "",
4967 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbl1 VecListOneD:$Vn, DPR:$Vm)))]>;
4968 let hasExtraSrcRegAllocReq = 1 in {
4970 : N3V<1,1,0b11,0b1001,0,0, (outs DPR:$Vd),
4971 (ins DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTB2,
4972 "vtbl", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "", []>;
4974 : N3V<1,1,0b11,0b1010,0,0, (outs DPR:$Vd),
4975 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm), NVTBLFrm, IIC_VTB3,
4976 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm", "", []>;
4978 : N3V<1,1,0b11,0b1011,0,0, (outs DPR:$Vd),
4979 (ins DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm),
4981 "vtbl", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm", "", []>;
4982 } // hasExtraSrcRegAllocReq = 1
4985 : PseudoNeonI<(outs DPR:$dst), (ins QPR:$tbl, DPR:$src), IIC_VTB2, "", []>;
4987 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB3, "", []>;
4989 : PseudoNeonI<(outs DPR:$dst), (ins QQPR:$tbl, DPR:$src), IIC_VTB4, "", []>;
4991 // VTBX : Vector Table Extension
4993 : N3V<1,1,0b11,0b1000,1,0, (outs DPR:$Vd),
4994 (ins DPR:$orig, VecListOneD:$Vn, DPR:$Vm), NVTBLFrm, IIC_VTBX1,
4995 "vtbx", "8", "$Vd, $Vn, $Vm", "$orig = $Vd",
4996 [(set DPR:$Vd, (v8i8 (int_arm_neon_vtbx1
4997 DPR:$orig, VecListOneD:$Vn, DPR:$Vm)))]>;
4998 let hasExtraSrcRegAllocReq = 1 in {
5000 : N3V<1,1,0b11,0b1001,1,0, (outs DPR:$Vd),
5001 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$Vm), NVTBLFrm, IIC_VTBX2,
5002 "vtbx", "8", "$Vd, \\{$Vn, $tbl2\\}, $Vm", "$orig = $Vd", []>;
5004 : N3V<1,1,0b11,0b1010,1,0, (outs DPR:$Vd),
5005 (ins DPR:$orig, DPR:$Vn, DPR:$tbl2, DPR:$tbl3, DPR:$Vm),
5006 NVTBLFrm, IIC_VTBX3,
5007 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3\\}, $Vm",
5010 : N3V<1,1,0b11,0b1011,1,0, (outs DPR:$Vd), (ins DPR:$orig, DPR:$Vn,
5011 DPR:$tbl2, DPR:$tbl3, DPR:$tbl4, DPR:$Vm), NVTBLFrm, IIC_VTBX4,
5012 "vtbx", "8", "$Vd, \\{$Vn, $tbl2, $tbl3, $tbl4\\}, $Vm",
5014 } // hasExtraSrcRegAllocReq = 1
5017 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QPR:$tbl, DPR:$src),
5018 IIC_VTBX2, "$orig = $dst", []>;
5020 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5021 IIC_VTBX3, "$orig = $dst", []>;
5023 : PseudoNeonI<(outs DPR:$dst), (ins DPR:$orig, QQPR:$tbl, DPR:$src),
5024 IIC_VTBX4, "$orig = $dst", []>;
5025 } // DecoderMethod = "DecodeTBLInstruction"
5027 //===----------------------------------------------------------------------===//
5028 // NEON instructions for single-precision FP math
5029 //===----------------------------------------------------------------------===//
5031 class N2VSPat<SDNode OpNode, NeonI Inst>
5032 : NEONFPPat<(f32 (OpNode SPR:$a)),
5034 (v2f32 (COPY_TO_REGCLASS (Inst
5036 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5037 SPR:$a, ssub_0)), DPR_VFP2)), ssub_0)>;
5039 class N3VSPat<SDNode OpNode, NeonI Inst>
5040 : NEONFPPat<(f32 (OpNode SPR:$a, SPR:$b)),
5042 (v2f32 (COPY_TO_REGCLASS (Inst
5044 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5047 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5048 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5050 class N3VSMulOpPat<SDNode MulNode, SDNode OpNode, NeonI Inst>
5051 : NEONFPPat<(f32 (OpNode SPR:$acc, (f32 (MulNode SPR:$a, SPR:$b)))),
5053 (v2f32 (COPY_TO_REGCLASS (Inst
5055 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5058 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5061 (v2f32 (COPY_TO_REGCLASS (v2f32 (IMPLICIT_DEF)), DPR_VFP2)),
5062 SPR:$b, ssub_0)), DPR_VFP2)), ssub_0)>;
5064 def : N3VSPat<fadd, VADDfd>;
5065 def : N3VSPat<fsub, VSUBfd>;
5066 def : N3VSPat<fmul, VMULfd>;
5067 def : N3VSMulOpPat<fmul, fadd, VMLAfd>,
5068 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5069 def : N3VSMulOpPat<fmul, fsub, VMLSfd>,
5070 Requires<[HasNEON, UseNEONForFP, UseFPVMLx]>;
5071 def : N2VSPat<fabs, VABSfd>;
5072 def : N2VSPat<fneg, VNEGfd>;
5073 def : N3VSPat<NEONfmax, VMAXfd>;
5074 def : N3VSPat<NEONfmin, VMINfd>;
5075 def : N2VSPat<arm_ftosi, VCVTf2sd>;
5076 def : N2VSPat<arm_ftoui, VCVTf2ud>;
5077 def : N2VSPat<arm_sitof, VCVTs2fd>;
5078 def : N2VSPat<arm_uitof, VCVTu2fd>;
5080 //===----------------------------------------------------------------------===//
5081 // Non-Instruction Patterns
5082 //===----------------------------------------------------------------------===//
5085 def : Pat<(v1i64 (bitconvert (v2i32 DPR:$src))), (v1i64 DPR:$src)>;
5086 def : Pat<(v1i64 (bitconvert (v4i16 DPR:$src))), (v1i64 DPR:$src)>;
5087 def : Pat<(v1i64 (bitconvert (v8i8 DPR:$src))), (v1i64 DPR:$src)>;
5088 def : Pat<(v1i64 (bitconvert (f64 DPR:$src))), (v1i64 DPR:$src)>;
5089 def : Pat<(v1i64 (bitconvert (v2f32 DPR:$src))), (v1i64 DPR:$src)>;
5090 def : Pat<(v2i32 (bitconvert (v1i64 DPR:$src))), (v2i32 DPR:$src)>;
5091 def : Pat<(v2i32 (bitconvert (v4i16 DPR:$src))), (v2i32 DPR:$src)>;
5092 def : Pat<(v2i32 (bitconvert (v8i8 DPR:$src))), (v2i32 DPR:$src)>;
5093 def : Pat<(v2i32 (bitconvert (f64 DPR:$src))), (v2i32 DPR:$src)>;
5094 def : Pat<(v2i32 (bitconvert (v2f32 DPR:$src))), (v2i32 DPR:$src)>;
5095 def : Pat<(v4i16 (bitconvert (v1i64 DPR:$src))), (v4i16 DPR:$src)>;
5096 def : Pat<(v4i16 (bitconvert (v2i32 DPR:$src))), (v4i16 DPR:$src)>;
5097 def : Pat<(v4i16 (bitconvert (v8i8 DPR:$src))), (v4i16 DPR:$src)>;
5098 def : Pat<(v4i16 (bitconvert (f64 DPR:$src))), (v4i16 DPR:$src)>;
5099 def : Pat<(v4i16 (bitconvert (v2f32 DPR:$src))), (v4i16 DPR:$src)>;
5100 def : Pat<(v8i8 (bitconvert (v1i64 DPR:$src))), (v8i8 DPR:$src)>;
5101 def : Pat<(v8i8 (bitconvert (v2i32 DPR:$src))), (v8i8 DPR:$src)>;
5102 def : Pat<(v8i8 (bitconvert (v4i16 DPR:$src))), (v8i8 DPR:$src)>;
5103 def : Pat<(v8i8 (bitconvert (f64 DPR:$src))), (v8i8 DPR:$src)>;
5104 def : Pat<(v8i8 (bitconvert (v2f32 DPR:$src))), (v8i8 DPR:$src)>;
5105 def : Pat<(f64 (bitconvert (v1i64 DPR:$src))), (f64 DPR:$src)>;
5106 def : Pat<(f64 (bitconvert (v2i32 DPR:$src))), (f64 DPR:$src)>;
5107 def : Pat<(f64 (bitconvert (v4i16 DPR:$src))), (f64 DPR:$src)>;
5108 def : Pat<(f64 (bitconvert (v8i8 DPR:$src))), (f64 DPR:$src)>;
5109 def : Pat<(f64 (bitconvert (v2f32 DPR:$src))), (f64 DPR:$src)>;
5110 def : Pat<(v2f32 (bitconvert (f64 DPR:$src))), (v2f32 DPR:$src)>;
5111 def : Pat<(v2f32 (bitconvert (v1i64 DPR:$src))), (v2f32 DPR:$src)>;
5112 def : Pat<(v2f32 (bitconvert (v2i32 DPR:$src))), (v2f32 DPR:$src)>;
5113 def : Pat<(v2f32 (bitconvert (v4i16 DPR:$src))), (v2f32 DPR:$src)>;
5114 def : Pat<(v2f32 (bitconvert (v8i8 DPR:$src))), (v2f32 DPR:$src)>;
5116 def : Pat<(v2i64 (bitconvert (v4i32 QPR:$src))), (v2i64 QPR:$src)>;
5117 def : Pat<(v2i64 (bitconvert (v8i16 QPR:$src))), (v2i64 QPR:$src)>;
5118 def : Pat<(v2i64 (bitconvert (v16i8 QPR:$src))), (v2i64 QPR:$src)>;
5119 def : Pat<(v2i64 (bitconvert (v2f64 QPR:$src))), (v2i64 QPR:$src)>;
5120 def : Pat<(v2i64 (bitconvert (v4f32 QPR:$src))), (v2i64 QPR:$src)>;
5121 def : Pat<(v4i32 (bitconvert (v2i64 QPR:$src))), (v4i32 QPR:$src)>;
5122 def : Pat<(v4i32 (bitconvert (v8i16 QPR:$src))), (v4i32 QPR:$src)>;
5123 def : Pat<(v4i32 (bitconvert (v16i8 QPR:$src))), (v4i32 QPR:$src)>;
5124 def : Pat<(v4i32 (bitconvert (v2f64 QPR:$src))), (v4i32 QPR:$src)>;
5125 def : Pat<(v4i32 (bitconvert (v4f32 QPR:$src))), (v4i32 QPR:$src)>;
5126 def : Pat<(v8i16 (bitconvert (v2i64 QPR:$src))), (v8i16 QPR:$src)>;
5127 def : Pat<(v8i16 (bitconvert (v4i32 QPR:$src))), (v8i16 QPR:$src)>;
5128 def : Pat<(v8i16 (bitconvert (v16i8 QPR:$src))), (v8i16 QPR:$src)>;
5129 def : Pat<(v8i16 (bitconvert (v2f64 QPR:$src))), (v8i16 QPR:$src)>;
5130 def : Pat<(v8i16 (bitconvert (v4f32 QPR:$src))), (v8i16 QPR:$src)>;
5131 def : Pat<(v16i8 (bitconvert (v2i64 QPR:$src))), (v16i8 QPR:$src)>;
5132 def : Pat<(v16i8 (bitconvert (v4i32 QPR:$src))), (v16i8 QPR:$src)>;
5133 def : Pat<(v16i8 (bitconvert (v8i16 QPR:$src))), (v16i8 QPR:$src)>;
5134 def : Pat<(v16i8 (bitconvert (v2f64 QPR:$src))), (v16i8 QPR:$src)>;
5135 def : Pat<(v16i8 (bitconvert (v4f32 QPR:$src))), (v16i8 QPR:$src)>;
5136 def : Pat<(v4f32 (bitconvert (v2i64 QPR:$src))), (v4f32 QPR:$src)>;
5137 def : Pat<(v4f32 (bitconvert (v4i32 QPR:$src))), (v4f32 QPR:$src)>;
5138 def : Pat<(v4f32 (bitconvert (v8i16 QPR:$src))), (v4f32 QPR:$src)>;
5139 def : Pat<(v4f32 (bitconvert (v16i8 QPR:$src))), (v4f32 QPR:$src)>;
5140 def : Pat<(v4f32 (bitconvert (v2f64 QPR:$src))), (v4f32 QPR:$src)>;
5141 def : Pat<(v2f64 (bitconvert (v2i64 QPR:$src))), (v2f64 QPR:$src)>;
5142 def : Pat<(v2f64 (bitconvert (v4i32 QPR:$src))), (v2f64 QPR:$src)>;
5143 def : Pat<(v2f64 (bitconvert (v8i16 QPR:$src))), (v2f64 QPR:$src)>;
5144 def : Pat<(v2f64 (bitconvert (v16i8 QPR:$src))), (v2f64 QPR:$src)>;
5145 def : Pat<(v2f64 (bitconvert (v4f32 QPR:$src))), (v2f64 QPR:$src)>;