1 //===- AArch64InstrFormats.td - AArch64 Instruction Formats --*- tblgen -*-===//
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
7 //===----------------------------------------------------------------------===//
9 //===----------------------------------------------------------------------===//
10 // Describe AArch64 instructions format here
13 // Format specifies the encoding used by the instruction. This is part of the
14 // ad-hoc solution used to emit machine instruction encodings by our machine
16 class Format<bits<2> val> {
20 def PseudoFrm : Format<0>;
21 def NormalFrm : Format<1>; // Do we need any others?
23 // AArch64 Instruction Format
24 class AArch64Inst<Format f, string cstr> : Instruction {
25 field bits<32> Inst; // Instruction encoding.
26 // Mask of bits that cause an encoding to be UNPREDICTABLE.
27 // If a bit is set, then if the corresponding bit in the
28 // target encoding differs from its value in the "Inst" field,
29 // the instruction is UNPREDICTABLE (SoftFail in abstract parlance).
30 field bits<32> Unpredictable = 0;
31 // SoftFail is the generic name for this field, but we alias it so
32 // as to make it more obvious what it means in ARM-land.
33 field bits<32> SoftFail = Unpredictable;
34 let Namespace = "AArch64";
36 bits<2> Form = F.Value;
38 let Constraints = cstr;
41 class InstSubst<string Asm, dag Result, bit EmitPriority = 0>
42 : InstAlias<Asm, Result, EmitPriority>, Requires<[UseNegativeImmediates]>;
44 // Pseudo instructions (don't have encoding information)
45 class Pseudo<dag oops, dag iops, list<dag> pattern, string cstr = "">
46 : AArch64Inst<PseudoFrm, cstr> {
47 dag OutOperandList = oops;
48 dag InOperandList = iops;
49 let Pattern = pattern;
50 let isCodeGenOnly = 1;
53 // Real instructions (have encoding information)
54 class EncodedI<string cstr, list<dag> pattern> : AArch64Inst<NormalFrm, cstr> {
55 let Pattern = pattern;
59 // Enum describing whether an instruction is
60 // destructive in its first source operand.
61 class DestructiveInstTypeEnum<bits<1> val> {
64 def NotDestructive : DestructiveInstTypeEnum<0>;
65 def Destructive : DestructiveInstTypeEnum<1>;
67 // Normal instructions
68 class I<dag oops, dag iops, string asm, string operands, string cstr,
70 : EncodedI<cstr, pattern> {
71 dag OutOperandList = oops;
72 dag InOperandList = iops;
73 let AsmString = !strconcat(asm, operands);
75 // Destructive operations (SVE)
76 DestructiveInstTypeEnum DestructiveInstType = NotDestructive;
77 ElementSizeEnum ElementSize = ElementSizeB;
79 let TSFlags{3} = DestructiveInstType.Value;
80 let TSFlags{2-0} = ElementSize.Value;
83 class TriOpFrag<dag res> : PatFrag<(ops node:$LHS, node:$MHS, node:$RHS), res>;
84 class BinOpFrag<dag res> : PatFrag<(ops node:$LHS, node:$RHS), res>;
85 class UnOpFrag<dag res> : PatFrag<(ops node:$LHS), res>;
87 // Helper fragment for an extract of the high portion of a 128-bit vector.
88 def extract_high_v16i8 :
89 UnOpFrag<(extract_subvector (v16i8 node:$LHS), (i64 8))>;
90 def extract_high_v8i16 :
91 UnOpFrag<(extract_subvector (v8i16 node:$LHS), (i64 4))>;
92 def extract_high_v4i32 :
93 UnOpFrag<(extract_subvector (v4i32 node:$LHS), (i64 2))>;
94 def extract_high_v2i64 :
95 UnOpFrag<(extract_subvector (v2i64 node:$LHS), (i64 1))>;
97 //===----------------------------------------------------------------------===//
98 // Asm Operand Classes.
101 // Shifter operand for arithmetic shifted encodings.
102 def ShifterOperand : AsmOperandClass {
103 let Name = "Shifter";
106 // Shifter operand for mov immediate encodings.
107 def MovImm32ShifterOperand : AsmOperandClass {
108 let SuperClasses = [ShifterOperand];
109 let Name = "MovImm32Shifter";
110 let RenderMethod = "addShifterOperands";
111 let DiagnosticType = "InvalidMovImm32Shift";
113 def MovImm64ShifterOperand : AsmOperandClass {
114 let SuperClasses = [ShifterOperand];
115 let Name = "MovImm64Shifter";
116 let RenderMethod = "addShifterOperands";
117 let DiagnosticType = "InvalidMovImm64Shift";
120 // Shifter operand for arithmetic register shifted encodings.
121 class ArithmeticShifterOperand<int width> : AsmOperandClass {
122 let SuperClasses = [ShifterOperand];
123 let Name = "ArithmeticShifter" # width;
124 let PredicateMethod = "isArithmeticShifter<" # width # ">";
125 let RenderMethod = "addShifterOperands";
126 let DiagnosticType = "AddSubRegShift" # width;
129 def ArithmeticShifterOperand32 : ArithmeticShifterOperand<32>;
130 def ArithmeticShifterOperand64 : ArithmeticShifterOperand<64>;
132 // Shifter operand for logical register shifted encodings.
133 class LogicalShifterOperand<int width> : AsmOperandClass {
134 let SuperClasses = [ShifterOperand];
135 let Name = "LogicalShifter" # width;
136 let PredicateMethod = "isLogicalShifter<" # width # ">";
137 let RenderMethod = "addShifterOperands";
138 let DiagnosticType = "AddSubRegShift" # width;
141 def LogicalShifterOperand32 : LogicalShifterOperand<32>;
142 def LogicalShifterOperand64 : LogicalShifterOperand<64>;
144 // Shifter operand for logical vector 128/64-bit shifted encodings.
145 def LogicalVecShifterOperand : AsmOperandClass {
146 let SuperClasses = [ShifterOperand];
147 let Name = "LogicalVecShifter";
148 let RenderMethod = "addShifterOperands";
150 def LogicalVecHalfWordShifterOperand : AsmOperandClass {
151 let SuperClasses = [LogicalVecShifterOperand];
152 let Name = "LogicalVecHalfWordShifter";
153 let RenderMethod = "addShifterOperands";
156 // The "MSL" shifter on the vector MOVI instruction.
157 def MoveVecShifterOperand : AsmOperandClass {
158 let SuperClasses = [ShifterOperand];
159 let Name = "MoveVecShifter";
160 let RenderMethod = "addShifterOperands";
163 // Extend operand for arithmetic encodings.
164 def ExtendOperand : AsmOperandClass {
166 let DiagnosticType = "AddSubRegExtendLarge";
168 def ExtendOperand64 : AsmOperandClass {
169 let SuperClasses = [ExtendOperand];
170 let Name = "Extend64";
171 let DiagnosticType = "AddSubRegExtendSmall";
173 // 'extend' that's a lsl of a 64-bit register.
174 def ExtendOperandLSL64 : AsmOperandClass {
175 let SuperClasses = [ExtendOperand];
176 let Name = "ExtendLSL64";
177 let RenderMethod = "addExtend64Operands";
178 let DiagnosticType = "AddSubRegExtendLarge";
181 // 8-bit floating-point immediate encodings.
182 def FPImmOperand : AsmOperandClass {
184 let ParserMethod = "tryParseFPImm<true>";
185 let DiagnosticType = "InvalidFPImm";
188 def CondCode : AsmOperandClass {
189 let Name = "CondCode";
190 let DiagnosticType = "InvalidCondCode";
193 // A 32-bit register pasrsed as 64-bit
194 def GPR32as64Operand : AsmOperandClass {
195 let Name = "GPR32as64";
197 "tryParseGPROperand<false, RegConstraintEqualityTy::EqualsSubReg>";
199 def GPR32as64 : RegisterOperand<GPR32> {
200 let ParserMatchClass = GPR32as64Operand;
203 // A 64-bit register pasrsed as 32-bit
204 def GPR64as32Operand : AsmOperandClass {
205 let Name = "GPR64as32";
207 "tryParseGPROperand<false, RegConstraintEqualityTy::EqualsSuperReg>";
209 def GPR64as32 : RegisterOperand<GPR64, "printGPR64as32"> {
210 let ParserMatchClass = GPR64as32Operand;
213 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
214 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
215 // are encoded as the eight bit value 'abcdefgh'.
216 def SIMDImmType10Operand : AsmOperandClass { let Name = "SIMDImmType10"; }
218 class UImmScaledMemoryIndexed<int Width, int Scale> : AsmOperandClass {
219 let Name = "UImm" # Width # "s" # Scale;
220 let DiagnosticType = "InvalidMemoryIndexed" # Scale # "UImm" # Width;
221 let RenderMethod = "addImmScaledOperands<" # Scale # ">";
222 let PredicateMethod = "isUImmScaled<" # Width # ", " # Scale # ">";
225 class SImmScaledMemoryIndexed<int Width, int Scale> : AsmOperandClass {
226 let Name = "SImm" # Width # "s" # Scale;
227 let DiagnosticType = "InvalidMemoryIndexed" # Scale # "SImm" # Width;
228 let RenderMethod = "addImmScaledOperands<" # Scale # ">";
229 let PredicateMethod = "isSImmScaled<" # Width # ", " # Scale # ">";
232 //===----------------------------------------------------------------------===//
233 // Operand Definitions.
236 // ADR[P] instruction labels.
237 def AdrpOperand : AsmOperandClass {
238 let Name = "AdrpLabel";
239 let ParserMethod = "tryParseAdrpLabel";
240 let DiagnosticType = "InvalidLabel";
242 def adrplabel : Operand<i64> {
243 let EncoderMethod = "getAdrLabelOpValue";
244 let PrintMethod = "printAdrpLabel";
245 let ParserMatchClass = AdrpOperand;
248 def AdrOperand : AsmOperandClass {
249 let Name = "AdrLabel";
250 let ParserMethod = "tryParseAdrLabel";
251 let DiagnosticType = "InvalidLabel";
253 def adrlabel : Operand<i64> {
254 let EncoderMethod = "getAdrLabelOpValue";
255 let ParserMatchClass = AdrOperand;
258 class SImmOperand<int width> : AsmOperandClass {
259 let Name = "SImm" # width;
260 let DiagnosticType = "InvalidMemoryIndexedSImm" # width;
261 let RenderMethod = "addImmOperands";
262 let PredicateMethod = "isSImm<" # width # ">";
266 class AsmImmRange<int Low, int High> : AsmOperandClass {
267 let Name = "Imm" # Low # "_" # High;
268 let DiagnosticType = "InvalidImm" # Low # "_" # High;
269 let RenderMethod = "addImmOperands";
270 let PredicateMethod = "isImmInRange<" # Low # "," # High # ">";
273 // Authenticated loads for v8.3 can have scaled 10-bit immediate offsets.
274 def SImm10s8Operand : SImmScaledMemoryIndexed<10, 8>;
275 def simm10Scaled : Operand<i64> {
276 let ParserMatchClass = SImm10s8Operand;
277 let DecoderMethod = "DecodeSImm<10>";
278 let PrintMethod = "printImmScale<8>";
281 def simm9s16 : Operand<i64> {
282 let ParserMatchClass = SImmScaledMemoryIndexed<9, 16>;
283 let DecoderMethod = "DecodeSImm<9>";
284 let PrintMethod = "printImmScale<16>";
287 // uimm6 predicate - True if the immediate is in the range [0, 63].
288 def UImm6Operand : AsmOperandClass {
290 let DiagnosticType = "InvalidImm0_63";
293 def uimm6 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= 0 && Imm < 64; }]> {
294 let ParserMatchClass = UImm6Operand;
297 def uimm16 : Operand<i16>, ImmLeaf<i16, [{return Imm >= 0 && Imm < 65536;}]>{
298 let ParserMatchClass = AsmImmRange<0, 65535>;
301 def SImm9Operand : SImmOperand<9>;
302 def simm9 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -256 && Imm < 256; }]> {
303 let ParserMatchClass = SImm9Operand;
304 let DecoderMethod = "DecodeSImm<9>";
307 def SImm8Operand : SImmOperand<8>;
308 def simm8 : Operand<i32>, ImmLeaf<i32, [{ return Imm >= -128 && Imm < 127; }]> {
309 let ParserMatchClass = SImm8Operand;
310 let DecoderMethod = "DecodeSImm<8>";
313 def SImm6Operand : SImmOperand<6>;
314 def simm6_32b : Operand<i32>, ImmLeaf<i32, [{ return Imm >= -32 && Imm < 32; }]> {
315 let ParserMatchClass = SImm6Operand;
316 let DecoderMethod = "DecodeSImm<6>";
319 def SImm5Operand : SImmOperand<5>;
320 def simm5_64b : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -16 && Imm < 16; }]> {
321 let ParserMatchClass = SImm5Operand;
322 let DecoderMethod = "DecodeSImm<5>";
325 def simm5_32b : Operand<i32>, ImmLeaf<i32, [{ return Imm >= -16 && Imm < 16; }]> {
326 let ParserMatchClass = SImm5Operand;
327 let DecoderMethod = "DecodeSImm<5>";
330 // simm7sN predicate - True if the immediate is a multiple of N in the range
331 // [-64 * N, 63 * N].
333 def SImm7s4Operand : SImmScaledMemoryIndexed<7, 4>;
334 def SImm7s8Operand : SImmScaledMemoryIndexed<7, 8>;
335 def SImm7s16Operand : SImmScaledMemoryIndexed<7, 16>;
337 def simm7s4 : Operand<i32> {
338 let ParserMatchClass = SImm7s4Operand;
339 let PrintMethod = "printImmScale<4>";
342 def simm7s8 : Operand<i32> {
343 let ParserMatchClass = SImm7s8Operand;
344 let PrintMethod = "printImmScale<8>";
347 def simm7s16 : Operand<i32> {
348 let ParserMatchClass = SImm7s16Operand;
349 let PrintMethod = "printImmScale<16>";
352 def am_indexed7s8 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S8", []>;
353 def am_indexed7s16 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S16", []>;
354 def am_indexed7s32 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S32", []>;
355 def am_indexed7s64 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S64", []>;
356 def am_indexed7s128 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S128", []>;
358 def am_indexedu6s128 : ComplexPattern<i64, 2, "SelectAddrModeIndexedU6S128", []>;
359 def am_indexeds9s128 : ComplexPattern<i64, 2, "SelectAddrModeIndexedS9S128", []>;
361 def UImmS2XForm : SDNodeXForm<imm, [{
362 return CurDAG->getTargetConstant(N->getZExtValue() / 2, SDLoc(N), MVT::i64);
364 def UImmS4XForm : SDNodeXForm<imm, [{
365 return CurDAG->getTargetConstant(N->getZExtValue() / 4, SDLoc(N), MVT::i64);
367 def UImmS8XForm : SDNodeXForm<imm, [{
368 return CurDAG->getTargetConstant(N->getZExtValue() / 8, SDLoc(N), MVT::i64);
371 // uimm5sN predicate - True if the immediate is a multiple of N in the range
373 def UImm5s2Operand : UImmScaledMemoryIndexed<5, 2>;
374 def UImm5s4Operand : UImmScaledMemoryIndexed<5, 4>;
375 def UImm5s8Operand : UImmScaledMemoryIndexed<5, 8>;
377 def uimm5s2 : Operand<i64>, ImmLeaf<i64,
378 [{ return Imm >= 0 && Imm < (32*2) && ((Imm % 2) == 0); }],
380 let ParserMatchClass = UImm5s2Operand;
381 let PrintMethod = "printImmScale<2>";
383 def uimm5s4 : Operand<i64>, ImmLeaf<i64,
384 [{ return Imm >= 0 && Imm < (32*4) && ((Imm % 4) == 0); }],
386 let ParserMatchClass = UImm5s4Operand;
387 let PrintMethod = "printImmScale<4>";
389 def uimm5s8 : Operand<i64>, ImmLeaf<i64,
390 [{ return Imm >= 0 && Imm < (32*8) && ((Imm % 8) == 0); }],
392 let ParserMatchClass = UImm5s8Operand;
393 let PrintMethod = "printImmScale<8>";
396 // tuimm5sN predicate - similiar to uimm5sN, but use TImmLeaf (TargetConstant)
397 // instead of ImmLeaf (Constant)
398 def tuimm5s2 : Operand<i64>, TImmLeaf<i64,
399 [{ return Imm >= 0 && Imm < (32*2) && ((Imm % 2) == 0); }],
401 let ParserMatchClass = UImm5s2Operand;
402 let PrintMethod = "printImmScale<2>";
404 def tuimm5s4 : Operand<i64>, TImmLeaf<i64,
405 [{ return Imm >= 0 && Imm < (32*4) && ((Imm % 4) == 0); }],
407 let ParserMatchClass = UImm5s4Operand;
408 let PrintMethod = "printImmScale<4>";
410 def tuimm5s8 : Operand<i64>, TImmLeaf<i64,
411 [{ return Imm >= 0 && Imm < (32*8) && ((Imm % 8) == 0); }],
413 let ParserMatchClass = UImm5s8Operand;
414 let PrintMethod = "printImmScale<8>";
417 // uimm6sN predicate - True if the immediate is a multiple of N in the range
419 def UImm6s1Operand : UImmScaledMemoryIndexed<6, 1>;
420 def UImm6s2Operand : UImmScaledMemoryIndexed<6, 2>;
421 def UImm6s4Operand : UImmScaledMemoryIndexed<6, 4>;
422 def UImm6s8Operand : UImmScaledMemoryIndexed<6, 8>;
423 def UImm6s16Operand : UImmScaledMemoryIndexed<6, 16>;
425 def uimm6s1 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= 0 && Imm < 64; }]> {
426 let ParserMatchClass = UImm6s1Operand;
428 def uimm6s2 : Operand<i64>, ImmLeaf<i64,
429 [{ return Imm >= 0 && Imm < (64*2) && ((Imm % 2) == 0); }]> {
430 let PrintMethod = "printImmScale<2>";
431 let ParserMatchClass = UImm6s2Operand;
433 def uimm6s4 : Operand<i64>, ImmLeaf<i64,
434 [{ return Imm >= 0 && Imm < (64*4) && ((Imm % 4) == 0); }]> {
435 let PrintMethod = "printImmScale<4>";
436 let ParserMatchClass = UImm6s4Operand;
438 def uimm6s8 : Operand<i64>, ImmLeaf<i64,
439 [{ return Imm >= 0 && Imm < (64*8) && ((Imm % 8) == 0); }]> {
440 let PrintMethod = "printImmScale<8>";
441 let ParserMatchClass = UImm6s8Operand;
443 def uimm6s16 : Operand<i64>, ImmLeaf<i64,
444 [{ return Imm >= 0 && Imm < (64*16) && ((Imm % 16) == 0); }]> {
445 let PrintMethod = "printImmScale<16>";
446 let ParserMatchClass = UImm6s16Operand;
449 // simm6sN predicate - True if the immediate is a multiple of N in the range
450 // [-32 * N, 31 * N].
451 def SImm6s1Operand : SImmScaledMemoryIndexed<6, 1>;
452 def simm6s1 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -32 && Imm < 32; }]> {
453 let ParserMatchClass = SImm6s1Operand;
454 let DecoderMethod = "DecodeSImm<6>";
457 // simm4sN predicate - True if the immediate is a multiple of N in the range
459 def SImm4s1Operand : SImmScaledMemoryIndexed<4, 1>;
460 def SImm4s2Operand : SImmScaledMemoryIndexed<4, 2>;
461 def SImm4s3Operand : SImmScaledMemoryIndexed<4, 3>;
462 def SImm4s4Operand : SImmScaledMemoryIndexed<4, 4>;
463 def SImm4s16Operand : SImmScaledMemoryIndexed<4, 16>;
465 def simm4s1 : Operand<i64>, ImmLeaf<i64,
466 [{ return Imm >=-8 && Imm <= 7; }]> {
467 let ParserMatchClass = SImm4s1Operand;
468 let DecoderMethod = "DecodeSImm<4>";
471 def simm4s2 : Operand<i64>, ImmLeaf<i64,
472 [{ return Imm >=-16 && Imm <= 14 && (Imm % 2) == 0x0; }]> {
473 let PrintMethod = "printImmScale<2>";
474 let ParserMatchClass = SImm4s2Operand;
475 let DecoderMethod = "DecodeSImm<4>";
478 def simm4s3 : Operand<i64>, ImmLeaf<i64,
479 [{ return Imm >=-24 && Imm <= 21 && (Imm % 3) == 0x0; }]> {
480 let PrintMethod = "printImmScale<3>";
481 let ParserMatchClass = SImm4s3Operand;
482 let DecoderMethod = "DecodeSImm<4>";
485 def simm4s4 : Operand<i64>, ImmLeaf<i64,
486 [{ return Imm >=-32 && Imm <= 28 && (Imm % 4) == 0x0; }]> {
487 let PrintMethod = "printImmScale<4>";
488 let ParserMatchClass = SImm4s4Operand;
489 let DecoderMethod = "DecodeSImm<4>";
491 def simm4s16 : Operand<i64>, ImmLeaf<i64,
492 [{ return Imm >=-128 && Imm <= 112 && (Imm % 16) == 0x0; }]> {
493 let PrintMethod = "printImmScale<16>";
494 let ParserMatchClass = SImm4s16Operand;
495 let DecoderMethod = "DecodeSImm<4>";
498 def Imm1_8Operand : AsmImmRange<1, 8>;
499 def Imm1_16Operand : AsmImmRange<1, 16>;
500 def Imm1_32Operand : AsmImmRange<1, 32>;
501 def Imm1_64Operand : AsmImmRange<1, 64>;
503 class BranchTarget<int N> : AsmOperandClass {
504 let Name = "BranchTarget" # N;
505 let DiagnosticType = "InvalidLabel";
506 let PredicateMethod = "isBranchTarget<" # N # ">";
509 class PCRelLabel<int N> : BranchTarget<N> {
510 let Name = "PCRelLabel" # N;
513 def BranchTarget14Operand : BranchTarget<14>;
514 def BranchTarget26Operand : BranchTarget<26>;
515 def PCRelLabel19Operand : PCRelLabel<19>;
517 def MovWSymbolG3AsmOperand : AsmOperandClass {
518 let Name = "MovWSymbolG3";
519 let RenderMethod = "addImmOperands";
522 def movw_symbol_g3 : Operand<i32> {
523 let ParserMatchClass = MovWSymbolG3AsmOperand;
526 def MovWSymbolG2AsmOperand : AsmOperandClass {
527 let Name = "MovWSymbolG2";
528 let RenderMethod = "addImmOperands";
531 def movw_symbol_g2 : Operand<i32> {
532 let ParserMatchClass = MovWSymbolG2AsmOperand;
535 def MovWSymbolG1AsmOperand : AsmOperandClass {
536 let Name = "MovWSymbolG1";
537 let RenderMethod = "addImmOperands";
540 def movw_symbol_g1 : Operand<i32> {
541 let ParserMatchClass = MovWSymbolG1AsmOperand;
544 def MovWSymbolG0AsmOperand : AsmOperandClass {
545 let Name = "MovWSymbolG0";
546 let RenderMethod = "addImmOperands";
549 def movw_symbol_g0 : Operand<i32> {
550 let ParserMatchClass = MovWSymbolG0AsmOperand;
553 class fixedpoint_i32<ValueType FloatVT>
555 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<32>", [fpimm, ld]> {
556 let EncoderMethod = "getFixedPointScaleOpValue";
557 let DecoderMethod = "DecodeFixedPointScaleImm32";
558 let ParserMatchClass = Imm1_32Operand;
561 class fixedpoint_i64<ValueType FloatVT>
563 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<64>", [fpimm, ld]> {
564 let EncoderMethod = "getFixedPointScaleOpValue";
565 let DecoderMethod = "DecodeFixedPointScaleImm64";
566 let ParserMatchClass = Imm1_64Operand;
569 def fixedpoint_f16_i32 : fixedpoint_i32<f16>;
570 def fixedpoint_f32_i32 : fixedpoint_i32<f32>;
571 def fixedpoint_f64_i32 : fixedpoint_i32<f64>;
573 def fixedpoint_f16_i64 : fixedpoint_i64<f16>;
574 def fixedpoint_f32_i64 : fixedpoint_i64<f32>;
575 def fixedpoint_f64_i64 : fixedpoint_i64<f64>;
577 def vecshiftR8 : Operand<i32>, ImmLeaf<i32, [{
578 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
580 let EncoderMethod = "getVecShiftR8OpValue";
581 let DecoderMethod = "DecodeVecShiftR8Imm";
582 let ParserMatchClass = Imm1_8Operand;
584 def vecshiftR16 : Operand<i32>, ImmLeaf<i32, [{
585 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
587 let EncoderMethod = "getVecShiftR16OpValue";
588 let DecoderMethod = "DecodeVecShiftR16Imm";
589 let ParserMatchClass = Imm1_16Operand;
591 def vecshiftR16Narrow : Operand<i32>, ImmLeaf<i32, [{
592 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
594 let EncoderMethod = "getVecShiftR16OpValue";
595 let DecoderMethod = "DecodeVecShiftR16ImmNarrow";
596 let ParserMatchClass = Imm1_8Operand;
598 def vecshiftR32 : Operand<i32>, ImmLeaf<i32, [{
599 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
601 let EncoderMethod = "getVecShiftR32OpValue";
602 let DecoderMethod = "DecodeVecShiftR32Imm";
603 let ParserMatchClass = Imm1_32Operand;
605 def vecshiftR32Narrow : Operand<i32>, ImmLeaf<i32, [{
606 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
608 let EncoderMethod = "getVecShiftR32OpValue";
609 let DecoderMethod = "DecodeVecShiftR32ImmNarrow";
610 let ParserMatchClass = Imm1_16Operand;
612 def vecshiftR64 : Operand<i32>, ImmLeaf<i32, [{
613 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 65);
615 let EncoderMethod = "getVecShiftR64OpValue";
616 let DecoderMethod = "DecodeVecShiftR64Imm";
617 let ParserMatchClass = Imm1_64Operand;
619 def vecshiftR64Narrow : Operand<i32>, ImmLeaf<i32, [{
620 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
622 let EncoderMethod = "getVecShiftR64OpValue";
623 let DecoderMethod = "DecodeVecShiftR64ImmNarrow";
624 let ParserMatchClass = Imm1_32Operand;
627 // Same as vecshiftR#N, but use TargetConstant (TimmLeaf) instead of Constant
629 def tvecshiftR8 : Operand<i32>, TImmLeaf<i32, [{
630 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
632 let EncoderMethod = "getVecShiftR8OpValue";
633 let DecoderMethod = "DecodeVecShiftR8Imm";
634 let ParserMatchClass = Imm1_8Operand;
636 def tvecshiftR16 : Operand<i32>, TImmLeaf<i32, [{
637 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
639 let EncoderMethod = "getVecShiftR16OpValue";
640 let DecoderMethod = "DecodeVecShiftR16Imm";
641 let ParserMatchClass = Imm1_16Operand;
643 def tvecshiftR32 : Operand<i32>, TImmLeaf<i32, [{
644 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
646 let EncoderMethod = "getVecShiftR32OpValue";
647 let DecoderMethod = "DecodeVecShiftR32Imm";
648 let ParserMatchClass = Imm1_32Operand;
651 def Imm0_1Operand : AsmImmRange<0, 1>;
652 def Imm0_7Operand : AsmImmRange<0, 7>;
653 def Imm0_15Operand : AsmImmRange<0, 15>;
654 def Imm0_31Operand : AsmImmRange<0, 31>;
655 def Imm0_63Operand : AsmImmRange<0, 63>;
657 def vecshiftL8 : Operand<i32>, ImmLeaf<i32, [{
658 return (((uint32_t)Imm) < 8);
660 let EncoderMethod = "getVecShiftL8OpValue";
661 let DecoderMethod = "DecodeVecShiftL8Imm";
662 let ParserMatchClass = Imm0_7Operand;
664 def vecshiftL16 : Operand<i32>, ImmLeaf<i32, [{
665 return (((uint32_t)Imm) < 16);
667 let EncoderMethod = "getVecShiftL16OpValue";
668 let DecoderMethod = "DecodeVecShiftL16Imm";
669 let ParserMatchClass = Imm0_15Operand;
671 def vecshiftL32 : Operand<i32>, ImmLeaf<i32, [{
672 return (((uint32_t)Imm) < 32);
674 let EncoderMethod = "getVecShiftL32OpValue";
675 let DecoderMethod = "DecodeVecShiftL32Imm";
676 let ParserMatchClass = Imm0_31Operand;
678 def vecshiftL64 : Operand<i32>, ImmLeaf<i32, [{
679 return (((uint32_t)Imm) < 64);
681 let EncoderMethod = "getVecShiftL64OpValue";
682 let DecoderMethod = "DecodeVecShiftL64Imm";
683 let ParserMatchClass = Imm0_63Operand;
687 // Crazy immediate formats used by 32-bit and 64-bit logical immediate
688 // instructions for splatting repeating bit patterns across the immediate.
689 def logical_imm32_XFORM : SDNodeXForm<imm, [{
690 uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 32);
691 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
693 def logical_imm64_XFORM : SDNodeXForm<imm, [{
694 uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 64);
695 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
698 def gi_logical_imm32_XFORM : GICustomOperandRenderer<"renderLogicalImm32">,
699 GISDNodeXFormEquiv<logical_imm32_XFORM>;
700 def gi_logical_imm64_XFORM : GICustomOperandRenderer<"renderLogicalImm64">,
701 GISDNodeXFormEquiv<logical_imm64_XFORM>;
703 let DiagnosticType = "LogicalSecondSource" in {
704 def LogicalImm32Operand : AsmOperandClass {
705 let Name = "LogicalImm32";
706 let PredicateMethod = "isLogicalImm<int32_t>";
707 let RenderMethod = "addLogicalImmOperands<int32_t>";
709 def LogicalImm64Operand : AsmOperandClass {
710 let Name = "LogicalImm64";
711 let PredicateMethod = "isLogicalImm<int64_t>";
712 let RenderMethod = "addLogicalImmOperands<int64_t>";
714 def LogicalImm32NotOperand : AsmOperandClass {
715 let Name = "LogicalImm32Not";
716 let PredicateMethod = "isLogicalImm<int32_t>";
717 let RenderMethod = "addLogicalImmNotOperands<int32_t>";
719 def LogicalImm64NotOperand : AsmOperandClass {
720 let Name = "LogicalImm64Not";
721 let PredicateMethod = "isLogicalImm<int64_t>";
722 let RenderMethod = "addLogicalImmNotOperands<int64_t>";
725 def logical_imm32 : Operand<i32>, IntImmLeaf<i32, [{
726 return AArch64_AM::isLogicalImmediate(Imm.getZExtValue(), 32);
727 }], logical_imm32_XFORM> {
728 let PrintMethod = "printLogicalImm<int32_t>";
729 let ParserMatchClass = LogicalImm32Operand;
731 def logical_imm64 : Operand<i64>, IntImmLeaf<i64, [{
732 return AArch64_AM::isLogicalImmediate(Imm.getZExtValue(), 64);
733 }], logical_imm64_XFORM> {
734 let PrintMethod = "printLogicalImm<int64_t>";
735 let ParserMatchClass = LogicalImm64Operand;
737 def logical_imm32_not : Operand<i32> {
738 let ParserMatchClass = LogicalImm32NotOperand;
740 def logical_imm64_not : Operand<i64> {
741 let ParserMatchClass = LogicalImm64NotOperand;
744 // iXX_imm0_65535 predicates - True if the immediate is in the range [0,65535].
745 let ParserMatchClass = AsmImmRange<0, 65535>, PrintMethod = "printImmHex" in {
746 def i32_imm0_65535 : Operand<i32>, TImmLeaf<i32, [{
747 return ((uint32_t)Imm) < 65536;
750 def i64_imm0_65535 : Operand<i64>, TImmLeaf<i64, [{
751 return ((uint64_t)Imm) < 65536;
755 // imm0_255 predicate - True if the immediate is in the range [0,255].
756 def Imm0_255Operand : AsmImmRange<0,255>;
758 def imm0_255 : Operand<i32>, ImmLeaf<i32, [{
759 return ((uint32_t)Imm) < 256;
761 let ParserMatchClass = Imm0_255Operand;
762 let PrintMethod = "printImm";
765 // imm0_127 predicate - True if the immediate is in the range [0,127]
766 def Imm0_127Operand : AsmImmRange<0, 127>;
767 def imm0_127 : Operand<i32>, ImmLeaf<i32, [{
768 return ((uint32_t)Imm) < 128;
770 let ParserMatchClass = Imm0_127Operand;
771 let PrintMethod = "printImm";
774 def imm0_127_64b : Operand<i64>, ImmLeaf<i64, [{
775 return ((uint64_t)Imm) < 128;
777 let ParserMatchClass = Imm0_127Operand;
778 let PrintMethod = "printImm";
781 // NOTE: These imm0_N operands have to be of type i64 because i64 is the size
782 // for all shift-amounts.
784 // imm0_63 predicate - True if the immediate is in the range [0,63]
785 def imm0_63 : Operand<i64>, ImmLeaf<i64, [{
786 return ((uint64_t)Imm) < 64;
788 let ParserMatchClass = Imm0_63Operand;
791 // imm0_31 predicate - True if the immediate is in the range [0,31]
792 def imm0_31 : Operand<i64>, ImmLeaf<i64, [{
793 return ((uint64_t)Imm) < 32;
795 let ParserMatchClass = Imm0_31Operand;
798 // timm0_31 predicate - same ass imm0_31, but use TargetConstant (TimmLeaf)
799 // instead of Contant (ImmLeaf)
800 def timm0_31 : Operand<i64>, TImmLeaf<i64, [{
801 return ((uint64_t)Imm) < 32;
803 let ParserMatchClass = Imm0_31Operand;
806 // True if the 32-bit immediate is in the range [0,31]
807 def imm32_0_31 : Operand<i32>, ImmLeaf<i32, [{
808 return ((uint64_t)Imm) < 32;
810 let ParserMatchClass = Imm0_31Operand;
813 // imm0_1 predicate - True if the immediate is in the range [0,1]
814 def imm0_1 : Operand<i64>, ImmLeaf<i64, [{
815 return ((uint64_t)Imm) < 2;
817 let ParserMatchClass = Imm0_1Operand;
820 // imm0_15 predicate - True if the immediate is in the range [0,15]
821 def imm0_15 : Operand<i64>, ImmLeaf<i64, [{
822 return ((uint64_t)Imm) < 16;
824 let ParserMatchClass = Imm0_15Operand;
827 // imm0_7 predicate - True if the immediate is in the range [0,7]
828 def imm0_7 : Operand<i64>, ImmLeaf<i64, [{
829 return ((uint64_t)Imm) < 8;
831 let ParserMatchClass = Imm0_7Operand;
834 // imm32_0_7 predicate - True if the 32-bit immediate is in the range [0,7]
835 def imm32_0_7 : Operand<i32>, ImmLeaf<i32, [{
836 return ((uint32_t)Imm) < 8;
838 let ParserMatchClass = Imm0_7Operand;
841 // imm32_0_15 predicate - True if the 32-bit immediate is in the range [0,15]
842 def imm32_0_15 : Operand<i32>, ImmLeaf<i32, [{
843 return ((uint32_t)Imm) < 16;
845 let ParserMatchClass = Imm0_15Operand;
848 // An arithmetic shifter operand:
849 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr
851 class arith_shift<ValueType Ty, int width> : Operand<Ty> {
852 let PrintMethod = "printShifter";
853 let ParserMatchClass = !cast<AsmOperandClass>(
854 "ArithmeticShifterOperand" # width);
857 def arith_shift32 : arith_shift<i32, 32>;
858 def arith_shift64 : arith_shift<i64, 64>;
860 class arith_shifted_reg<ValueType Ty, RegisterClass regclass, int width>
862 ComplexPattern<Ty, 2, "SelectArithShiftedRegister", []> {
863 let PrintMethod = "printShiftedRegister";
864 let MIOperandInfo = (ops regclass, !cast<Operand>("arith_shift" # width));
867 def arith_shifted_reg32 : arith_shifted_reg<i32, GPR32, 32>;
868 def arith_shifted_reg64 : arith_shifted_reg<i64, GPR64, 64>;
870 def gi_arith_shifted_reg32 :
871 GIComplexOperandMatcher<s32, "selectArithShiftedRegister">,
872 GIComplexPatternEquiv<arith_shifted_reg32>;
874 def gi_arith_shifted_reg64 :
875 GIComplexOperandMatcher<s64, "selectArithShiftedRegister">,
876 GIComplexPatternEquiv<arith_shifted_reg64>;
878 // An arithmetic shifter operand:
879 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr, 11 = ror
881 class logical_shift<int width> : Operand<i32> {
882 let PrintMethod = "printShifter";
883 let ParserMatchClass = !cast<AsmOperandClass>(
884 "LogicalShifterOperand" # width);
887 def logical_shift32 : logical_shift<32>;
888 def logical_shift64 : logical_shift<64>;
890 class logical_shifted_reg<ValueType Ty, RegisterClass regclass, Operand shiftop>
892 ComplexPattern<Ty, 2, "SelectLogicalShiftedRegister", []> {
893 let PrintMethod = "printShiftedRegister";
894 let MIOperandInfo = (ops regclass, shiftop);
897 def logical_shifted_reg32 : logical_shifted_reg<i32, GPR32, logical_shift32>;
898 def logical_shifted_reg64 : logical_shifted_reg<i64, GPR64, logical_shift64>;
900 def gi_logical_shifted_reg32 :
901 GIComplexOperandMatcher<s32, "selectLogicalShiftedRegister">,
902 GIComplexPatternEquiv<logical_shifted_reg32>;
904 def gi_logical_shifted_reg64 :
905 GIComplexOperandMatcher<s64, "selectLogicalShiftedRegister">,
906 GIComplexPatternEquiv<logical_shifted_reg64>;
908 // A logical vector shifter operand:
909 // {7-6} - shift type: 00 = lsl
910 // {5-0} - imm6: #0, #8, #16, or #24
911 def logical_vec_shift : Operand<i32> {
912 let PrintMethod = "printShifter";
913 let EncoderMethod = "getVecShifterOpValue";
914 let ParserMatchClass = LogicalVecShifterOperand;
917 // A logical vector half-word shifter operand:
918 // {7-6} - shift type: 00 = lsl
919 // {5-0} - imm6: #0 or #8
920 def logical_vec_hw_shift : Operand<i32> {
921 let PrintMethod = "printShifter";
922 let EncoderMethod = "getVecShifterOpValue";
923 let ParserMatchClass = LogicalVecHalfWordShifterOperand;
926 // A vector move shifter operand:
927 // {0} - imm1: #8 or #16
928 def move_vec_shift : Operand<i32> {
929 let PrintMethod = "printShifter";
930 let EncoderMethod = "getMoveVecShifterOpValue";
931 let ParserMatchClass = MoveVecShifterOperand;
934 let DiagnosticType = "AddSubSecondSource" in {
935 def AddSubImmOperand : AsmOperandClass {
936 let Name = "AddSubImm";
937 let ParserMethod = "tryParseImmWithOptionalShift";
938 let RenderMethod = "addImmWithOptionalShiftOperands<12>";
940 def AddSubImmNegOperand : AsmOperandClass {
941 let Name = "AddSubImmNeg";
942 let ParserMethod = "tryParseImmWithOptionalShift";
943 let RenderMethod = "addImmNegWithOptionalShiftOperands<12>";
946 // An ADD/SUB immediate shifter operand:
948 // {7-6} - shift type: 00 = lsl
949 // {5-0} - imm6: #0 or #12
950 class addsub_shifted_imm<ValueType Ty>
951 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectArithImmed", [imm]> {
952 let PrintMethod = "printAddSubImm";
953 let EncoderMethod = "getAddSubImmOpValue";
954 let ParserMatchClass = AddSubImmOperand;
955 let MIOperandInfo = (ops i32imm, i32imm);
958 class addsub_shifted_imm_neg<ValueType Ty>
960 let EncoderMethod = "getAddSubImmOpValue";
961 let ParserMatchClass = AddSubImmNegOperand;
962 let MIOperandInfo = (ops i32imm, i32imm);
965 def addsub_shifted_imm32 : addsub_shifted_imm<i32>;
966 def addsub_shifted_imm64 : addsub_shifted_imm<i64>;
967 def addsub_shifted_imm32_neg : addsub_shifted_imm_neg<i32>;
968 def addsub_shifted_imm64_neg : addsub_shifted_imm_neg<i64>;
970 def gi_addsub_shifted_imm32 :
971 GIComplexOperandMatcher<s32, "selectArithImmed">,
972 GIComplexPatternEquiv<addsub_shifted_imm32>;
974 def gi_addsub_shifted_imm64 :
975 GIComplexOperandMatcher<s64, "selectArithImmed">,
976 GIComplexPatternEquiv<addsub_shifted_imm64>;
978 class neg_addsub_shifted_imm<ValueType Ty>
979 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectNegArithImmed", [imm]> {
980 let PrintMethod = "printAddSubImm";
981 let EncoderMethod = "getAddSubImmOpValue";
982 let ParserMatchClass = AddSubImmOperand;
983 let MIOperandInfo = (ops i32imm, i32imm);
986 def neg_addsub_shifted_imm32 : neg_addsub_shifted_imm<i32>;
987 def neg_addsub_shifted_imm64 : neg_addsub_shifted_imm<i64>;
989 def gi_neg_addsub_shifted_imm32 :
990 GIComplexOperandMatcher<s32, "selectNegArithImmed">,
991 GIComplexPatternEquiv<neg_addsub_shifted_imm32>;
993 def gi_neg_addsub_shifted_imm64 :
994 GIComplexOperandMatcher<s64, "selectNegArithImmed">,
995 GIComplexPatternEquiv<neg_addsub_shifted_imm64>;
997 // An extend operand:
998 // {5-3} - extend type
1000 def arith_extend : Operand<i32> {
1001 let PrintMethod = "printArithExtend";
1002 let ParserMatchClass = ExtendOperand;
1004 def arith_extend64 : Operand<i32> {
1005 let PrintMethod = "printArithExtend";
1006 let ParserMatchClass = ExtendOperand64;
1009 // 'extend' that's a lsl of a 64-bit register.
1010 def arith_extendlsl64 : Operand<i32> {
1011 let PrintMethod = "printArithExtend";
1012 let ParserMatchClass = ExtendOperandLSL64;
1015 class arith_extended_reg32<ValueType Ty> : Operand<Ty>,
1016 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
1017 let PrintMethod = "printExtendedRegister";
1018 let MIOperandInfo = (ops GPR32, arith_extend);
1021 class arith_extended_reg32to64<ValueType Ty> : Operand<Ty>,
1022 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
1023 let PrintMethod = "printExtendedRegister";
1024 let MIOperandInfo = (ops GPR32, arith_extend64);
1027 def arith_extended_reg32_i32 : arith_extended_reg32<i32>;
1028 def gi_arith_extended_reg32_i32 :
1029 GIComplexOperandMatcher<s32, "selectArithExtendedRegister">,
1030 GIComplexPatternEquiv<arith_extended_reg32_i32>;
1032 def arith_extended_reg32_i64 : arith_extended_reg32<i64>;
1033 def gi_arith_extended_reg32_i64 :
1034 GIComplexOperandMatcher<s64, "selectArithExtendedRegister">,
1035 GIComplexPatternEquiv<arith_extended_reg32_i64>;
1037 def arith_extended_reg32to64_i64 : arith_extended_reg32to64<i64>;
1038 def gi_arith_extended_reg32to64_i64 :
1039 GIComplexOperandMatcher<s64, "selectArithExtendedRegister">,
1040 GIComplexPatternEquiv<arith_extended_reg32to64_i64>;
1042 // Floating-point immediate.
1043 def fpimm16 : Operand<f16>,
1045 return AArch64_AM::getFP16Imm(Imm) != -1;
1046 }], SDNodeXForm<fpimm, [{
1047 APFloat InVal = N->getValueAPF();
1048 uint32_t enc = AArch64_AM::getFP16Imm(InVal);
1049 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
1051 let ParserMatchClass = FPImmOperand;
1052 let PrintMethod = "printFPImmOperand";
1054 def fpimm32 : Operand<f32>,
1056 return AArch64_AM::getFP32Imm(Imm) != -1;
1057 }], SDNodeXForm<fpimm, [{
1058 APFloat InVal = N->getValueAPF();
1059 uint32_t enc = AArch64_AM::getFP32Imm(InVal);
1060 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
1062 let ParserMatchClass = FPImmOperand;
1063 let PrintMethod = "printFPImmOperand";
1065 def fpimm64 : Operand<f64>,
1067 return AArch64_AM::getFP64Imm(Imm) != -1;
1068 }], SDNodeXForm<fpimm, [{
1069 APFloat InVal = N->getValueAPF();
1070 uint32_t enc = AArch64_AM::getFP64Imm(InVal);
1071 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
1073 let ParserMatchClass = FPImmOperand;
1074 let PrintMethod = "printFPImmOperand";
1077 def fpimm8 : Operand<i32> {
1078 let ParserMatchClass = FPImmOperand;
1079 let PrintMethod = "printFPImmOperand";
1082 def fpimm0 : FPImmLeaf<fAny, [{
1083 return Imm.isExactlyValue(+0.0);
1086 // Vector lane operands
1087 class AsmVectorIndex<int Min, int Max, string NamePrefix=""> : AsmOperandClass {
1088 let Name = NamePrefix # "IndexRange" # Min # "_" # Max;
1089 let DiagnosticType = "Invalid" # Name;
1090 let PredicateMethod = "isVectorIndex<" # Min # ", " # Max # ">";
1091 let RenderMethod = "addVectorIndexOperands";
1094 class AsmVectorIndexOpnd<ValueType ty, AsmOperandClass mc, code pred>
1095 : Operand<ty>, ImmLeaf<ty, pred> {
1096 let ParserMatchClass = mc;
1097 let PrintMethod = "printVectorIndex";
1100 def VectorIndex1Operand : AsmVectorIndex<1, 1>;
1101 def VectorIndexBOperand : AsmVectorIndex<0, 15>;
1102 def VectorIndexHOperand : AsmVectorIndex<0, 7>;
1103 def VectorIndexSOperand : AsmVectorIndex<0, 3>;
1104 def VectorIndexDOperand : AsmVectorIndex<0, 1>;
1106 def VectorIndex1 : AsmVectorIndexOpnd<i64, VectorIndex1Operand, [{ return ((uint64_t)Imm) == 1; }]>;
1107 def VectorIndexB : AsmVectorIndexOpnd<i64, VectorIndexBOperand, [{ return ((uint64_t)Imm) < 16; }]>;
1108 def VectorIndexH : AsmVectorIndexOpnd<i64, VectorIndexHOperand, [{ return ((uint64_t)Imm) < 8; }]>;
1109 def VectorIndexS : AsmVectorIndexOpnd<i64, VectorIndexSOperand, [{ return ((uint64_t)Imm) < 4; }]>;
1110 def VectorIndexD : AsmVectorIndexOpnd<i64, VectorIndexDOperand, [{ return ((uint64_t)Imm) < 2; }]>;
1112 def VectorIndex132b : AsmVectorIndexOpnd<i32, VectorIndex1Operand, [{ return ((uint64_t)Imm) == 1; }]>;
1113 def VectorIndexB32b : AsmVectorIndexOpnd<i32, VectorIndexBOperand, [{ return ((uint64_t)Imm) < 16; }]>;
1114 def VectorIndexH32b : AsmVectorIndexOpnd<i32, VectorIndexHOperand, [{ return ((uint64_t)Imm) < 8; }]>;
1115 def VectorIndexS32b : AsmVectorIndexOpnd<i32, VectorIndexSOperand, [{ return ((uint64_t)Imm) < 4; }]>;
1116 def VectorIndexD32b : AsmVectorIndexOpnd<i32, VectorIndexDOperand, [{ return ((uint64_t)Imm) < 2; }]>;
1118 def SVEVectorIndexExtDupBOperand : AsmVectorIndex<0, 63, "SVE">;
1119 def SVEVectorIndexExtDupHOperand : AsmVectorIndex<0, 31, "SVE">;
1120 def SVEVectorIndexExtDupSOperand : AsmVectorIndex<0, 15, "SVE">;
1121 def SVEVectorIndexExtDupDOperand : AsmVectorIndex<0, 7, "SVE">;
1122 def SVEVectorIndexExtDupQOperand : AsmVectorIndex<0, 3, "SVE">;
1124 def sve_elm_idx_extdup_b
1125 : AsmVectorIndexOpnd<i64, SVEVectorIndexExtDupBOperand, [{ return ((uint64_t)Imm) < 64; }]>;
1126 def sve_elm_idx_extdup_h
1127 : AsmVectorIndexOpnd<i64, SVEVectorIndexExtDupHOperand, [{ return ((uint64_t)Imm) < 32; }]>;
1128 def sve_elm_idx_extdup_s
1129 : AsmVectorIndexOpnd<i64, SVEVectorIndexExtDupSOperand, [{ return ((uint64_t)Imm) < 16; }]>;
1130 def sve_elm_idx_extdup_d
1131 : AsmVectorIndexOpnd<i64, SVEVectorIndexExtDupDOperand, [{ return ((uint64_t)Imm) < 8; }]>;
1132 def sve_elm_idx_extdup_q
1133 : AsmVectorIndexOpnd<i64, SVEVectorIndexExtDupQOperand, [{ return ((uint64_t)Imm) < 4; }]>;
1135 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
1136 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
1137 // are encoded as the eight bit value 'abcdefgh'.
1138 def simdimmtype10 : Operand<i32>,
1140 return AArch64_AM::isAdvSIMDModImmType10(
1141 Imm.bitcastToAPInt().getZExtValue());
1142 }], SDNodeXForm<fpimm, [{
1143 APFloat InVal = N->getValueAPF();
1144 uint32_t enc = AArch64_AM::encodeAdvSIMDModImmType10(N->getValueAPF()
1147 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
1149 let ParserMatchClass = SIMDImmType10Operand;
1150 let PrintMethod = "printSIMDType10Operand";
1155 // System management
1158 // Base encoding for system instruction operands.
1159 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
1160 class BaseSystemI<bit L, dag oops, dag iops, string asm, string operands,
1161 list<dag> pattern = []>
1162 : I<oops, iops, asm, operands, "", pattern> {
1163 let Inst{31-22} = 0b1101010100;
1167 // System instructions which do not have an Rt register.
1168 class SimpleSystemI<bit L, dag iops, string asm, string operands,
1169 list<dag> pattern = []>
1170 : BaseSystemI<L, (outs), iops, asm, operands, pattern> {
1171 let Inst{4-0} = 0b11111;
1174 // System instructions which have an Rt register.
1175 class RtSystemI<bit L, dag oops, dag iops, string asm, string operands>
1176 : BaseSystemI<L, oops, iops, asm, operands>,
1182 // System instructions for transactional memory extension
1183 class TMBaseSystemI<bit L, bits<4> CRm, bits<3> op2, dag oops, dag iops,
1184 string asm, string operands, list<dag> pattern>
1185 : BaseSystemI<L, oops, iops, asm, operands, pattern>,
1187 let Inst{20-12} = 0b000110011;
1188 let Inst{11-8} = CRm;
1189 let Inst{7-5} = op2;
1190 let DecoderMethod = "";
1196 // System instructions for transactional memory - single input operand
1197 class TMSystemI<bits<4> CRm, string asm, list<dag> pattern>
1198 : TMBaseSystemI<0b1, CRm, 0b011,
1199 (outs GPR64:$Rt), (ins), asm, "\t$Rt", pattern> {
1204 // System instructions for transactional memory - no operand
1205 class TMSystemINoOperand<bits<4> CRm, string asm, list<dag> pattern>
1206 : TMBaseSystemI<0b0, CRm, 0b011, (outs), (ins), asm, "", pattern> {
1207 let Inst{4-0} = 0b11111;
1210 // System instructions for exit from transactions
1211 class TMSystemException<bits<3> op1, string asm, list<dag> pattern>
1212 : I<(outs), (ins i64_imm0_65535:$imm), asm, "\t$imm", "", pattern>,
1215 let Inst{31-24} = 0b11010100;
1216 let Inst{23-21} = op1;
1217 let Inst{20-5} = imm;
1218 let Inst{4-0} = 0b00000;
1221 // Hint instructions that take both a CRm and a 3-bit immediate.
1222 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
1223 // model patterns with sufficiently fine granularity
1224 let mayStore = 1, mayLoad = 1, hasSideEffects = 1 in
1225 class HintI<string mnemonic>
1226 : SimpleSystemI<0, (ins imm0_127:$imm), mnemonic#"\t$imm", "",
1227 [(int_aarch64_hint imm0_127:$imm)]>,
1228 Sched<[WriteHint]> {
1230 let Inst{20-12} = 0b000110010;
1231 let Inst{11-5} = imm;
1234 // System instructions taking a single literal operand which encodes into
1235 // CRm. op2 differentiates the opcodes.
1236 def BarrierAsmOperand : AsmOperandClass {
1237 let Name = "Barrier";
1238 let ParserMethod = "tryParseBarrierOperand";
1240 def barrier_op : Operand<i32> {
1241 let PrintMethod = "printBarrierOption";
1242 let ParserMatchClass = BarrierAsmOperand;
1244 class CRmSystemI<Operand crmtype, bits<3> opc, string asm,
1245 list<dag> pattern = []>
1246 : SimpleSystemI<0, (ins crmtype:$CRm), asm, "\t$CRm", pattern>,
1247 Sched<[WriteBarrier]> {
1249 let Inst{20-12} = 0b000110011;
1250 let Inst{11-8} = CRm;
1251 let Inst{7-5} = opc;
1254 class SystemNoOperands<bits<3> op2, string asm, list<dag> pattern = []>
1255 : SimpleSystemI<0, (ins), asm, "", pattern>,
1259 let Inst{31-12} = 0b11010101000000110010;
1260 let Inst{11-8} = CRm;
1261 let Inst{7-5} = op2;
1262 let Inst{4-0} = 0b11111;
1265 // MRS/MSR system instructions. These have different operand classes because
1266 // a different subset of registers can be accessed through each instruction.
1267 def MRSSystemRegisterOperand : AsmOperandClass {
1268 let Name = "MRSSystemRegister";
1269 let ParserMethod = "tryParseSysReg";
1270 let DiagnosticType = "MRS";
1272 // concatenation of op0, op1, CRn, CRm, op2. 16-bit immediate.
1273 def mrs_sysreg_op : Operand<i32> {
1274 let ParserMatchClass = MRSSystemRegisterOperand;
1275 let DecoderMethod = "DecodeMRSSystemRegister";
1276 let PrintMethod = "printMRSSystemRegister";
1279 def MSRSystemRegisterOperand : AsmOperandClass {
1280 let Name = "MSRSystemRegister";
1281 let ParserMethod = "tryParseSysReg";
1282 let DiagnosticType = "MSR";
1284 def msr_sysreg_op : Operand<i32> {
1285 let ParserMatchClass = MSRSystemRegisterOperand;
1286 let DecoderMethod = "DecodeMSRSystemRegister";
1287 let PrintMethod = "printMSRSystemRegister";
1290 def PSBHintOperand : AsmOperandClass {
1291 let Name = "PSBHint";
1292 let ParserMethod = "tryParsePSBHint";
1294 def psbhint_op : Operand<i32> {
1295 let ParserMatchClass = PSBHintOperand;
1296 let PrintMethod = "printPSBHintOp";
1297 let MCOperandPredicate = [{
1298 // Check, if operand is valid, to fix exhaustive aliasing in disassembly.
1299 // "psb" is an alias to "hint" only for certain values of CRm:Op2 fields.
1302 return AArch64PSBHint::lookupPSBByEncoding(MCOp.getImm()) != nullptr;
1306 def BTIHintOperand : AsmOperandClass {
1307 let Name = "BTIHint";
1308 let ParserMethod = "tryParseBTIHint";
1310 def btihint_op : Operand<i32> {
1311 let ParserMatchClass = BTIHintOperand;
1312 let PrintMethod = "printBTIHintOp";
1313 let MCOperandPredicate = [{
1314 // "bti" is an alias to "hint" only for certain values of CRm:Op2 fields.
1317 return AArch64BTIHint::lookupBTIByEncoding((MCOp.getImm() ^ 32) >> 1) != nullptr;
1321 class MRSI : RtSystemI<1, (outs GPR64:$Rt), (ins mrs_sysreg_op:$systemreg),
1322 "mrs", "\t$Rt, $systemreg"> {
1324 let Inst{20-5} = systemreg;
1327 // FIXME: Some of these def NZCV, others don't. Best way to model that?
1328 // Explicitly modeling each of the system register as a register class
1329 // would do it, but feels like overkill at this point.
1330 class MSRI : RtSystemI<0, (outs), (ins msr_sysreg_op:$systemreg, GPR64:$Rt),
1331 "msr", "\t$systemreg, $Rt"> {
1333 let Inst{20-5} = systemreg;
1336 def SystemPStateFieldWithImm0_15Operand : AsmOperandClass {
1337 let Name = "SystemPStateFieldWithImm0_15";
1338 let ParserMethod = "tryParseSysReg";
1340 def pstatefield4_op : Operand<i32> {
1341 let ParserMatchClass = SystemPStateFieldWithImm0_15Operand;
1342 let PrintMethod = "printSystemPStateField";
1345 // Instructions to modify PSTATE, no input reg
1346 let Defs = [NZCV] in
1347 class PstateWriteSimple<dag iops, string asm, string operands>
1348 : SimpleSystemI<0, iops, asm, operands> {
1350 let Inst{20-19} = 0b00;
1351 let Inst{15-12} = 0b0100;
1354 class MSRpstateImm0_15
1355 : PstateWriteSimple<(ins pstatefield4_op:$pstatefield, imm0_15:$imm), "msr",
1356 "\t$pstatefield, $imm">,
1359 bits<6> pstatefield;
1361 let Inst{18-16} = pstatefield{5-3};
1362 let Inst{11-8} = imm;
1363 let Inst{7-5} = pstatefield{2-0};
1365 let DecoderMethod = "DecodeSystemPStateInstruction";
1366 // MSRpstateI aliases with MSRI. When the MSRpstateI decoder method returns
1367 // Fail the decoder should attempt to decode the instruction as MSRI.
1368 let hasCompleteDecoder = 0;
1371 def SystemPStateFieldWithImm0_1Operand : AsmOperandClass {
1372 let Name = "SystemPStateFieldWithImm0_1";
1373 let ParserMethod = "tryParseSysReg";
1375 def pstatefield1_op : Operand<i32> {
1376 let ParserMatchClass = SystemPStateFieldWithImm0_1Operand;
1377 let PrintMethod = "printSystemPStateField";
1380 class MSRpstateImm0_1
1381 : PstateWriteSimple<(ins pstatefield1_op:$pstatefield, imm0_1:$imm), "msr",
1382 "\t$pstatefield, $imm">,
1385 bits<6> pstatefield;
1387 let Inst{18-16} = pstatefield{5-3};
1388 let Inst{11-9} = 0b000;
1390 let Inst{7-5} = pstatefield{2-0};
1392 let DecoderMethod = "DecodeSystemPStateInstruction";
1393 // MSRpstateI aliases with MSRI. When the MSRpstateI decoder method returns
1394 // Fail the decoder should attempt to decode the instruction as MSRI.
1395 let hasCompleteDecoder = 0;
1398 // SYS and SYSL generic system instructions.
1399 def SysCRAsmOperand : AsmOperandClass {
1401 let ParserMethod = "tryParseSysCROperand";
1404 def sys_cr_op : Operand<i32> {
1405 let PrintMethod = "printSysCROperand";
1406 let ParserMatchClass = SysCRAsmOperand;
1409 class SystemXtI<bit L, string asm>
1410 : RtSystemI<L, (outs),
1411 (ins imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, GPR64:$Rt),
1412 asm, "\t$op1, $Cn, $Cm, $op2, $Rt"> {
1417 let Inst{20-19} = 0b01;
1418 let Inst{18-16} = op1;
1419 let Inst{15-12} = Cn;
1420 let Inst{11-8} = Cm;
1421 let Inst{7-5} = op2;
1424 class SystemLXtI<bit L, string asm>
1425 : RtSystemI<L, (outs),
1426 (ins GPR64:$Rt, imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2),
1427 asm, "\t$Rt, $op1, $Cn, $Cm, $op2"> {
1432 let Inst{20-19} = 0b01;
1433 let Inst{18-16} = op1;
1434 let Inst{15-12} = Cn;
1435 let Inst{11-8} = Cm;
1436 let Inst{7-5} = op2;
1440 // Branch (register) instructions:
1448 // otherwise UNDEFINED
1449 class BaseBranchReg<bits<4> opc, dag oops, dag iops, string asm,
1450 string operands, list<dag> pattern>
1451 : I<oops, iops, asm, operands, "", pattern>, Sched<[WriteBrReg]> {
1452 let Inst{31-25} = 0b1101011;
1453 let Inst{24-21} = opc;
1454 let Inst{20-16} = 0b11111;
1455 let Inst{15-10} = 0b000000;
1456 let Inst{4-0} = 0b00000;
1459 class BranchReg<bits<4> opc, string asm, list<dag> pattern>
1460 : BaseBranchReg<opc, (outs), (ins GPR64:$Rn), asm, "\t$Rn", pattern> {
1465 let mayLoad = 0, mayStore = 0, hasSideEffects = 1, isReturn = 1 in
1466 class SpecialReturn<bits<4> opc, string asm>
1467 : BaseBranchReg<opc, (outs), (ins), asm, "", []> {
1468 let Inst{9-5} = 0b11111;
1472 class RCPCLoad<bits<2> sz, string asm, RegisterClass RC>
1473 : I<(outs RC:$Rt), (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]", "", []>,
1477 let Inst{31-30} = sz;
1478 let Inst{29-10} = 0b11100010111111110000;
1483 class AuthBase<bits<1> M, dag oops, dag iops, string asm, string operands,
1485 : I<oops, iops, asm, operands, "", pattern>, Sched<[]> {
1486 let isAuthenticated = 1;
1487 let Inst{31-25} = 0b1101011;
1488 let Inst{20-11} = 0b1111100001;
1490 let Inst{4-0} = 0b11111;
1493 class AuthBranchTwoOperands<bits<1> op, bits<1> M, string asm>
1494 : AuthBase<M, (outs), (ins GPR64:$Rn, GPR64sp:$Rm), asm, "\t$Rn, $Rm", []> {
1497 let Inst{24-22} = 0b100;
1503 class AuthOneOperand<bits<3> opc, bits<1> M, string asm>
1504 : AuthBase<M, (outs), (ins GPR64:$Rn), asm, "\t$Rn", []> {
1507 let Inst{23-21} = opc;
1511 let Uses = [LR,SP] in
1512 class AuthReturn<bits<3> op, bits<1> M, string asm>
1513 : AuthBase<M, (outs), (ins), asm, "", []> {
1515 let Inst{23-21} = op;
1516 let Inst{9-0} = 0b1111111111;
1520 class BaseAuthLoad<bit M, bit W, dag oops, dag iops, string asm,
1521 string operands, string cstr, Operand opr>
1522 : I<oops, iops, asm, operands, cstr, []>, Sched<[]> {
1526 let isAuthenticated = 1;
1527 let Inst{31-24} = 0b11111000;
1529 let Inst{22} = offset{9};
1531 let Inst{20-12} = offset{8-0};
1538 multiclass AuthLoad<bit M, string asm, Operand opr> {
1539 def indexed : BaseAuthLoad<M, 0, (outs GPR64:$Rt),
1540 (ins GPR64sp:$Rn, opr:$offset),
1541 asm, "\t$Rt, [$Rn, $offset]", "", opr>;
1542 def writeback : BaseAuthLoad<M, 1, (outs GPR64sp:$wback, GPR64:$Rt),
1543 (ins GPR64sp:$Rn, opr:$offset),
1544 asm, "\t$Rt, [$Rn, $offset]!",
1545 "$Rn = $wback,@earlyclobber $wback", opr>;
1547 def : InstAlias<asm # "\t$Rt, [$Rn]",
1548 (!cast<Instruction>(NAME # "indexed") GPR64:$Rt, GPR64sp:$Rn, 0)>;
1550 def : InstAlias<asm # "\t$Rt, [$wback]!",
1551 (!cast<Instruction>(NAME # "writeback") GPR64sp:$wback, GPR64:$Rt, 0), 0>;
1555 // Conditional branch instruction.
1559 // 4-bit immediate. Pretty-printed as <cc>
1560 def ccode : Operand<i32> {
1561 let PrintMethod = "printCondCode";
1562 let ParserMatchClass = CondCode;
1564 def inv_ccode : Operand<i32> {
1565 // AL and NV are invalid in the aliases which use inv_ccode
1566 let PrintMethod = "printInverseCondCode";
1567 let ParserMatchClass = CondCode;
1568 let MCOperandPredicate = [{
1569 return MCOp.isImm() &&
1570 MCOp.getImm() != AArch64CC::AL &&
1571 MCOp.getImm() != AArch64CC::NV;
1575 // Conditional branch target. 19-bit immediate. The low two bits of the target
1576 // offset are implied zero and so are not part of the immediate.
1577 def am_brcond : Operand<OtherVT> {
1578 let EncoderMethod = "getCondBranchTargetOpValue";
1579 let DecoderMethod = "DecodePCRelLabel19";
1580 let PrintMethod = "printAlignedLabel";
1581 let ParserMatchClass = PCRelLabel19Operand;
1582 let OperandType = "OPERAND_PCREL";
1585 class BranchCond : I<(outs), (ins ccode:$cond, am_brcond:$target),
1586 "b", ".$cond\t$target", "",
1587 [(AArch64brcond bb:$target, imm:$cond, NZCV)]>,
1590 let isTerminator = 1;
1595 let Inst{31-24} = 0b01010100;
1596 let Inst{23-5} = target;
1598 let Inst{3-0} = cond;
1602 // Compare-and-branch instructions.
1604 class BaseCmpBranch<RegisterClass regtype, bit op, string asm, SDNode node>
1605 : I<(outs), (ins regtype:$Rt, am_brcond:$target),
1606 asm, "\t$Rt, $target", "",
1607 [(node regtype:$Rt, bb:$target)]>,
1610 let isTerminator = 1;
1614 let Inst{30-25} = 0b011010;
1616 let Inst{23-5} = target;
1620 multiclass CmpBranch<bit op, string asm, SDNode node> {
1621 def W : BaseCmpBranch<GPR32, op, asm, node> {
1624 def X : BaseCmpBranch<GPR64, op, asm, node> {
1630 // Test-bit-and-branch instructions.
1632 // Test-and-branch target. 14-bit sign-extended immediate. The low two bits of
1633 // the target offset are implied zero and so are not part of the immediate.
1634 def am_tbrcond : Operand<OtherVT> {
1635 let EncoderMethod = "getTestBranchTargetOpValue";
1636 let PrintMethod = "printAlignedLabel";
1637 let ParserMatchClass = BranchTarget14Operand;
1638 let OperandType = "OPERAND_PCREL";
1641 // AsmOperand classes to emit (or not) special diagnostics
1642 def TBZImm0_31Operand : AsmOperandClass {
1643 let Name = "TBZImm0_31";
1644 let PredicateMethod = "isImmInRange<0,31>";
1645 let RenderMethod = "addImmOperands";
1647 def TBZImm32_63Operand : AsmOperandClass {
1648 let Name = "Imm32_63";
1649 let PredicateMethod = "isImmInRange<32,63>";
1650 let DiagnosticType = "InvalidImm0_63";
1651 let RenderMethod = "addImmOperands";
1654 class tbz_imm0_31<AsmOperandClass matcher> : Operand<i64>, ImmLeaf<i64, [{
1655 return (((uint32_t)Imm) < 32);
1657 let ParserMatchClass = matcher;
1660 def tbz_imm0_31_diag : tbz_imm0_31<Imm0_31Operand>;
1661 def tbz_imm0_31_nodiag : tbz_imm0_31<TBZImm0_31Operand>;
1663 def tbz_imm32_63 : Operand<i64>, ImmLeaf<i64, [{
1664 return (((uint32_t)Imm) > 31) && (((uint32_t)Imm) < 64);
1666 let ParserMatchClass = TBZImm32_63Operand;
1669 class BaseTestBranch<RegisterClass regtype, Operand immtype,
1670 bit op, string asm, SDNode node>
1671 : I<(outs), (ins regtype:$Rt, immtype:$bit_off, am_tbrcond:$target),
1672 asm, "\t$Rt, $bit_off, $target", "",
1673 [(node regtype:$Rt, immtype:$bit_off, bb:$target)]>,
1676 let isTerminator = 1;
1682 let Inst{30-25} = 0b011011;
1684 let Inst{23-19} = bit_off{4-0};
1685 let Inst{18-5} = target;
1688 let DecoderMethod = "DecodeTestAndBranch";
1691 multiclass TestBranch<bit op, string asm, SDNode node> {
1692 def W : BaseTestBranch<GPR32, tbz_imm0_31_diag, op, asm, node> {
1696 def X : BaseTestBranch<GPR64, tbz_imm32_63, op, asm, node> {
1700 // Alias X-reg with 0-31 imm to W-Reg.
1701 def : InstAlias<asm # "\t$Rd, $imm, $target",
1702 (!cast<Instruction>(NAME#"W") GPR32as64:$Rd,
1703 tbz_imm0_31_nodiag:$imm, am_tbrcond:$target), 0>;
1704 def : Pat<(node GPR64:$Rn, tbz_imm0_31_diag:$imm, bb:$target),
1705 (!cast<Instruction>(NAME#"W") (EXTRACT_SUBREG GPR64:$Rn, sub_32),
1706 tbz_imm0_31_diag:$imm, bb:$target)>;
1710 // Unconditional branch (immediate) instructions.
1712 def am_b_target : Operand<OtherVT> {
1713 let EncoderMethod = "getBranchTargetOpValue";
1714 let PrintMethod = "printAlignedLabel";
1715 let ParserMatchClass = BranchTarget26Operand;
1716 let OperandType = "OPERAND_PCREL";
1718 def am_bl_target : Operand<i64> {
1719 let EncoderMethod = "getBranchTargetOpValue";
1720 let PrintMethod = "printAlignedLabel";
1721 let ParserMatchClass = BranchTarget26Operand;
1722 let OperandType = "OPERAND_PCREL";
1725 class BImm<bit op, dag iops, string asm, list<dag> pattern>
1726 : I<(outs), iops, asm, "\t$addr", "", pattern>, Sched<[WriteBr]> {
1729 let Inst{30-26} = 0b00101;
1730 let Inst{25-0} = addr;
1732 let DecoderMethod = "DecodeUnconditionalBranch";
1735 class BranchImm<bit op, string asm, list<dag> pattern>
1736 : BImm<op, (ins am_b_target:$addr), asm, pattern>;
1737 class CallImm<bit op, string asm, list<dag> pattern>
1738 : BImm<op, (ins am_bl_target:$addr), asm, pattern>;
1741 // Basic one-operand data processing instructions.
1744 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1745 class BaseOneOperandData<bits<3> opc, RegisterClass regtype, string asm,
1746 SDPatternOperator node>
1747 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
1748 [(set regtype:$Rd, (node regtype:$Rn))]>,
1749 Sched<[WriteI, ReadI]> {
1753 let Inst{30-13} = 0b101101011000000000;
1754 let Inst{12-10} = opc;
1759 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1760 multiclass OneOperandData<bits<3> opc, string asm,
1761 SDPatternOperator node = null_frag> {
1762 def Wr : BaseOneOperandData<opc, GPR32, asm, node> {
1766 def Xr : BaseOneOperandData<opc, GPR64, asm, node> {
1771 class OneWRegData<bits<3> opc, string asm, SDPatternOperator node>
1772 : BaseOneOperandData<opc, GPR32, asm, node> {
1776 class OneXRegData<bits<3> opc, string asm, SDPatternOperator node>
1777 : BaseOneOperandData<opc, GPR64, asm, node> {
1781 class SignAuthOneData<bits<3> opcode_prefix, bits<2> opcode, string asm>
1782 : I<(outs GPR64:$Rd), (ins GPR64sp:$Rn), asm, "\t$Rd, $Rn", "",
1784 Sched<[WriteI, ReadI]> {
1787 let Inst{31-15} = 0b11011010110000010;
1788 let Inst{14-12} = opcode_prefix;
1789 let Inst{11-10} = opcode;
1794 class SignAuthZero<bits<3> opcode_prefix, bits<2> opcode, string asm>
1795 : I<(outs GPR64:$Rd), (ins), asm, "\t$Rd", "", []>, Sched<[]> {
1797 let Inst{31-15} = 0b11011010110000010;
1798 let Inst{14-12} = opcode_prefix;
1799 let Inst{11-10} = opcode;
1800 let Inst{9-5} = 0b11111;
1804 class SignAuthTwoOperand<bits<4> opc, string asm,
1805 SDPatternOperator OpNode>
1806 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64sp:$Rm),
1807 asm, "\t$Rd, $Rn, $Rm", "",
1808 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64sp:$Rm))]>,
1809 Sched<[WriteI, ReadI, ReadI]> {
1813 let Inst{31-21} = 0b10011010110;
1814 let Inst{20-16} = Rm;
1815 let Inst{15-14} = 0b00;
1816 let Inst{13-10} = opc;
1821 // Base class for the Armv8.4-A 8 and 16-bit flag manipulation instructions
1822 class BaseFlagManipulation<bit sf, bit sz, dag iops, string asm, string ops>
1823 : I<(outs), iops, asm, ops, "", []>,
1824 Sched<[WriteI, ReadI, ReadI]> {
1828 let Inst{30-15} = 0b0111010000000000;
1830 let Inst{13-10} = 0b0010;
1832 let Inst{4-0} = 0b01101;
1835 class FlagRotate<dag iops, string asm, string ops>
1836 : BaseFlagManipulation<0b1, 0b0, iops, asm, ops> {
1839 let Inst{20-15} = imm;
1840 let Inst{13-10} = 0b0001;
1842 let Inst{3-0} = mask;
1846 // Basic two-operand data processing instructions.
1848 class BaseBaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1850 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1851 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
1852 Sched<[WriteI, ReadI, ReadI]> {
1857 let Inst{30} = isSub;
1858 let Inst{28-21} = 0b11010000;
1859 let Inst{20-16} = Rm;
1860 let Inst{15-10} = 0;
1865 class BaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1867 : BaseBaseAddSubCarry<isSub, regtype, asm,
1868 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, NZCV))]>;
1870 class BaseAddSubCarrySetFlags<bit isSub, RegisterClass regtype, string asm,
1872 : BaseBaseAddSubCarry<isSub, regtype, asm,
1873 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, NZCV)),
1878 multiclass AddSubCarry<bit isSub, string asm, string asm_setflags,
1879 SDNode OpNode, SDNode OpNode_setflags> {
1880 def Wr : BaseAddSubCarry<isSub, GPR32, asm, OpNode> {
1884 def Xr : BaseAddSubCarry<isSub, GPR64, asm, OpNode> {
1890 def SWr : BaseAddSubCarrySetFlags<isSub, GPR32, asm_setflags,
1895 def SXr : BaseAddSubCarrySetFlags<isSub, GPR64, asm_setflags,
1902 class BaseTwoOperand<bits<4> opc, RegisterClass regtype, string asm,
1903 SDPatternOperator OpNode,
1904 RegisterClass in1regtype = regtype,
1905 RegisterClass in2regtype = regtype>
1906 : I<(outs regtype:$Rd), (ins in1regtype:$Rn, in2regtype:$Rm),
1907 asm, "\t$Rd, $Rn, $Rm", "",
1908 [(set regtype:$Rd, (OpNode in1regtype:$Rn, in2regtype:$Rm))]> {
1912 let Inst{30-21} = 0b0011010110;
1913 let Inst{20-16} = Rm;
1914 let Inst{15-14} = 0b00;
1915 let Inst{13-10} = opc;
1920 class BaseDiv<bit isSigned, RegisterClass regtype, string asm,
1921 SDPatternOperator OpNode>
1922 : BaseTwoOperand<{0,0,1,?}, regtype, asm, OpNode> {
1923 let Inst{10} = isSigned;
1926 multiclass Div<bit isSigned, string asm, SDPatternOperator OpNode> {
1927 def Wr : BaseDiv<isSigned, GPR32, asm, OpNode>,
1928 Sched<[WriteID32, ReadID, ReadID]> {
1931 def Xr : BaseDiv<isSigned, GPR64, asm, OpNode>,
1932 Sched<[WriteID64, ReadID, ReadID]> {
1937 class BaseShift<bits<2> shift_type, RegisterClass regtype, string asm,
1938 SDPatternOperator OpNode = null_frag>
1939 : BaseTwoOperand<{1,0,?,?}, regtype, asm, OpNode>,
1940 Sched<[WriteIS, ReadI]> {
1941 let Inst{11-10} = shift_type;
1944 multiclass Shift<bits<2> shift_type, string asm, SDNode OpNode> {
1945 def Wr : BaseShift<shift_type, GPR32, asm> {
1949 def Xr : BaseShift<shift_type, GPR64, asm, OpNode> {
1953 def : Pat<(i32 (OpNode GPR32:$Rn, i64:$Rm)),
1954 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn,
1955 (EXTRACT_SUBREG i64:$Rm, sub_32))>;
1957 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (zext GPR32:$Rm)))),
1958 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1960 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (anyext GPR32:$Rm)))),
1961 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1963 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (sext GPR32:$Rm)))),
1964 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1966 def : Pat<(i64 (OpNode GPR64:$Rn, (i64 (sext GPR32:$Rm)))),
1967 (!cast<Instruction>(NAME # "Xr") GPR64:$Rn,
1968 (SUBREG_TO_REG (i32 0), GPR32:$Rm, sub_32))>;
1970 def : Pat<(i64 (OpNode GPR64:$Rn, (i64 (zext GPR32:$Rm)))),
1971 (!cast<Instruction>(NAME # "Xr") GPR64:$Rn,
1972 (SUBREG_TO_REG (i32 0), GPR32:$Rm, sub_32))>;
1975 class ShiftAlias<string asm, Instruction inst, RegisterClass regtype>
1976 : InstAlias<asm#"\t$dst, $src1, $src2",
1977 (inst regtype:$dst, regtype:$src1, regtype:$src2), 0>;
1979 class BaseMulAccum<bit isSub, bits<3> opc, RegisterClass multype,
1980 RegisterClass addtype, string asm,
1982 : I<(outs addtype:$Rd), (ins multype:$Rn, multype:$Rm, addtype:$Ra),
1983 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pattern> {
1988 let Inst{30-24} = 0b0011011;
1989 let Inst{23-21} = opc;
1990 let Inst{20-16} = Rm;
1991 let Inst{15} = isSub;
1992 let Inst{14-10} = Ra;
1997 multiclass MulAccum<bit isSub, string asm, SDNode AccNode> {
1998 // MADD/MSUB generation is decided by MachineCombiner.cpp
1999 def Wrrr : BaseMulAccum<isSub, 0b000, GPR32, GPR32, asm,
2000 [/*(set GPR32:$Rd, (AccNode GPR32:$Ra, (mul GPR32:$Rn, GPR32:$Rm)))*/]>,
2001 Sched<[WriteIM32, ReadIM, ReadIM, ReadIMA]> {
2005 def Xrrr : BaseMulAccum<isSub, 0b000, GPR64, GPR64, asm,
2006 [/*(set GPR64:$Rd, (AccNode GPR64:$Ra, (mul GPR64:$Rn, GPR64:$Rm)))*/]>,
2007 Sched<[WriteIM64, ReadIM, ReadIM, ReadIMA]> {
2012 class WideMulAccum<bit isSub, bits<3> opc, string asm,
2013 SDNode AccNode, SDNode ExtNode>
2014 : BaseMulAccum<isSub, opc, GPR32, GPR64, asm,
2015 [(set GPR64:$Rd, (AccNode GPR64:$Ra,
2016 (mul (ExtNode GPR32:$Rn), (ExtNode GPR32:$Rm))))]>,
2017 Sched<[WriteIM32, ReadIM, ReadIM, ReadIMA]> {
2021 class MulHi<bits<3> opc, string asm, SDNode OpNode>
2022 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64:$Rm),
2023 asm, "\t$Rd, $Rn, $Rm", "",
2024 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64:$Rm))]>,
2025 Sched<[WriteIM64, ReadIM, ReadIM]> {
2029 let Inst{31-24} = 0b10011011;
2030 let Inst{23-21} = opc;
2031 let Inst{20-16} = Rm;
2036 // The Ra field of SMULH and UMULH is unused: it should be assembled as 31
2037 // (i.e. all bits 1) but is ignored by the processor.
2038 let PostEncoderMethod = "fixMulHigh";
2041 class MulAccumWAlias<string asm, Instruction inst>
2042 : InstAlias<asm#"\t$dst, $src1, $src2",
2043 (inst GPR32:$dst, GPR32:$src1, GPR32:$src2, WZR)>;
2044 class MulAccumXAlias<string asm, Instruction inst>
2045 : InstAlias<asm#"\t$dst, $src1, $src2",
2046 (inst GPR64:$dst, GPR64:$src1, GPR64:$src2, XZR)>;
2047 class WideMulAccumAlias<string asm, Instruction inst>
2048 : InstAlias<asm#"\t$dst, $src1, $src2",
2049 (inst GPR64:$dst, GPR32:$src1, GPR32:$src2, XZR)>;
2051 class BaseCRC32<bit sf, bits<2> sz, bit C, RegisterClass StreamReg,
2052 SDPatternOperator OpNode, string asm>
2053 : I<(outs GPR32:$Rd), (ins GPR32:$Rn, StreamReg:$Rm),
2054 asm, "\t$Rd, $Rn, $Rm", "",
2055 [(set GPR32:$Rd, (OpNode GPR32:$Rn, StreamReg:$Rm))]>,
2056 Sched<[WriteISReg, ReadI, ReadISReg]> {
2062 let Inst{30-21} = 0b0011010110;
2063 let Inst{20-16} = Rm;
2064 let Inst{15-13} = 0b010;
2066 let Inst{11-10} = sz;
2069 let Predicates = [HasCRC];
2073 // Address generation.
2076 class ADRI<bit page, string asm, Operand adr, list<dag> pattern>
2077 : I<(outs GPR64:$Xd), (ins adr:$label), asm, "\t$Xd, $label", "",
2082 let Inst{31} = page;
2083 let Inst{30-29} = label{1-0};
2084 let Inst{28-24} = 0b10000;
2085 let Inst{23-5} = label{20-2};
2088 let DecoderMethod = "DecodeAdrInstruction";
2095 def movimm32_imm : Operand<i32> {
2096 let ParserMatchClass = AsmImmRange<0, 65535>;
2097 let EncoderMethod = "getMoveWideImmOpValue";
2098 let PrintMethod = "printImm";
2100 def movimm32_shift : Operand<i32> {
2101 let PrintMethod = "printShifter";
2102 let ParserMatchClass = MovImm32ShifterOperand;
2104 def movimm64_shift : Operand<i32> {
2105 let PrintMethod = "printShifter";
2106 let ParserMatchClass = MovImm64ShifterOperand;
2109 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2110 class BaseMoveImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
2112 : I<(outs regtype:$Rd), (ins movimm32_imm:$imm, shifter:$shift),
2113 asm, "\t$Rd, $imm$shift", "", []>,
2118 let Inst{30-29} = opc;
2119 let Inst{28-23} = 0b100101;
2120 let Inst{22-21} = shift{5-4};
2121 let Inst{20-5} = imm;
2124 let DecoderMethod = "DecodeMoveImmInstruction";
2127 multiclass MoveImmediate<bits<2> opc, string asm> {
2128 def Wi : BaseMoveImmediate<opc, GPR32, movimm32_shift, asm> {
2132 def Xi : BaseMoveImmediate<opc, GPR64, movimm64_shift, asm> {
2137 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2138 class BaseInsertImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
2140 : I<(outs regtype:$Rd),
2141 (ins regtype:$src, movimm32_imm:$imm, shifter:$shift),
2142 asm, "\t$Rd, $imm$shift", "$src = $Rd", []>,
2143 Sched<[WriteI, ReadI]> {
2147 let Inst{30-29} = opc;
2148 let Inst{28-23} = 0b100101;
2149 let Inst{22-21} = shift{5-4};
2150 let Inst{20-5} = imm;
2153 let DecoderMethod = "DecodeMoveImmInstruction";
2156 multiclass InsertImmediate<bits<2> opc, string asm> {
2157 def Wi : BaseInsertImmediate<opc, GPR32, movimm32_shift, asm> {
2161 def Xi : BaseInsertImmediate<opc, GPR64, movimm64_shift, asm> {
2170 class BaseAddSubImm<bit isSub, bit setFlags, RegisterClass dstRegtype,
2171 string asm_inst, string asm_ops,
2172 dag inputs, dag pattern>
2173 : I<(outs dstRegtype:$Rd), inputs, asm_inst, asm_ops, "", [pattern]>,
2174 Sched<[WriteI, ReadI]> {
2177 let Inst{30} = isSub;
2178 let Inst{29} = setFlags;
2179 let Inst{28-24} = 0b10001;
2184 class AddSubImmShift<bit isSub, bit setFlags, RegisterClass dstRegtype,
2185 RegisterClass srcRegtype, addsub_shifted_imm immtype,
2186 string asm_inst, SDPatternOperator OpNode>
2187 : BaseAddSubImm<isSub, setFlags, dstRegtype, asm_inst, "\t$Rd, $Rn, $imm",
2188 (ins srcRegtype:$Rn, immtype:$imm),
2189 (set dstRegtype:$Rd, (OpNode srcRegtype:$Rn, immtype:$imm))> {
2191 let Inst{23-22} = imm{13-12}; // '00' => lsl #0, '01' => lsl #12
2192 let Inst{21-10} = imm{11-0};
2193 let DecoderMethod = "DecodeAddSubImmShift";
2196 class BaseAddSubRegPseudo<RegisterClass regtype,
2197 SDPatternOperator OpNode>
2198 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
2199 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
2200 Sched<[WriteI, ReadI, ReadI]>;
2202 class BaseAddSubSReg<bit isSub, bit setFlags, RegisterClass regtype,
2203 arith_shifted_reg shifted_regtype, string asm,
2204 SDPatternOperator OpNode>
2205 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
2206 asm, "\t$Rd, $Rn, $Rm", "",
2207 [(set regtype:$Rd, (OpNode regtype:$Rn, shifted_regtype:$Rm))]>,
2208 Sched<[WriteISReg, ReadI, ReadISReg]> {
2209 // The operands are in order to match the 'addr' MI operands, so we
2210 // don't need an encoder method and by-name matching. Just use the default
2211 // in-order handling. Since we're using by-order, make sure the names
2217 let Inst{30} = isSub;
2218 let Inst{29} = setFlags;
2219 let Inst{28-24} = 0b01011;
2220 let Inst{23-22} = shift{7-6};
2222 let Inst{20-16} = src2;
2223 let Inst{15-10} = shift{5-0};
2224 let Inst{9-5} = src1;
2225 let Inst{4-0} = dst;
2227 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
2230 class BaseAddSubEReg<bit isSub, bit setFlags, RegisterClass dstRegtype,
2231 RegisterClass src1Regtype, Operand src2Regtype,
2232 string asm, SDPatternOperator OpNode>
2233 : I<(outs dstRegtype:$R1),
2234 (ins src1Regtype:$R2, src2Regtype:$R3),
2235 asm, "\t$R1, $R2, $R3", "",
2236 [(set dstRegtype:$R1, (OpNode src1Regtype:$R2, src2Regtype:$R3))]>,
2237 Sched<[WriteIEReg, ReadI, ReadIEReg]> {
2242 let Inst{30} = isSub;
2243 let Inst{29} = setFlags;
2244 let Inst{28-24} = 0b01011;
2245 let Inst{23-21} = 0b001;
2246 let Inst{20-16} = Rm;
2247 let Inst{15-13} = ext{5-3};
2248 let Inst{12-10} = ext{2-0};
2252 let DecoderMethod = "DecodeAddSubERegInstruction";
2255 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2256 class BaseAddSubEReg64<bit isSub, bit setFlags, RegisterClass dstRegtype,
2257 RegisterClass src1Regtype, RegisterClass src2Regtype,
2258 Operand ext_op, string asm>
2259 : I<(outs dstRegtype:$Rd),
2260 (ins src1Regtype:$Rn, src2Regtype:$Rm, ext_op:$ext),
2261 asm, "\t$Rd, $Rn, $Rm$ext", "", []>,
2262 Sched<[WriteIEReg, ReadI, ReadIEReg]> {
2267 let Inst{30} = isSub;
2268 let Inst{29} = setFlags;
2269 let Inst{28-24} = 0b01011;
2270 let Inst{23-21} = 0b001;
2271 let Inst{20-16} = Rm;
2272 let Inst{15} = ext{5};
2273 let Inst{12-10} = ext{2-0};
2277 let DecoderMethod = "DecodeAddSubERegInstruction";
2280 // Aliases for register+register add/subtract.
2281 class AddSubRegAlias<string asm, Instruction inst, RegisterClass dstRegtype,
2282 RegisterClass src1Regtype, RegisterClass src2Regtype,
2284 : InstAlias<asm#"\t$dst, $src1, $src2",
2285 (inst dstRegtype:$dst, src1Regtype:$src1, src2Regtype:$src2,
2288 multiclass AddSub<bit isSub, string mnemonic, string alias,
2289 SDPatternOperator OpNode = null_frag> {
2290 let hasSideEffects = 0, isReMaterializable = 1, isAsCheapAsAMove = 1 in {
2291 // Add/Subtract immediate
2292 // Increase the weight of the immediate variant to try to match it before
2293 // the extended register variant.
2294 // We used to match the register variant before the immediate when the
2295 // register argument could be implicitly zero-extended.
2296 let AddedComplexity = 6 in
2297 def Wri : AddSubImmShift<isSub, 0, GPR32sp, GPR32sp, addsub_shifted_imm32,
2301 let AddedComplexity = 6 in
2302 def Xri : AddSubImmShift<isSub, 0, GPR64sp, GPR64sp, addsub_shifted_imm64,
2307 // Add/Subtract register - Only used for CodeGen
2308 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
2309 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
2311 // Add/Subtract shifted register
2312 def Wrs : BaseAddSubSReg<isSub, 0, GPR32, arith_shifted_reg32, mnemonic,
2316 def Xrs : BaseAddSubSReg<isSub, 0, GPR64, arith_shifted_reg64, mnemonic,
2322 // Add/Subtract extended register
2323 let AddedComplexity = 1, hasSideEffects = 0 in {
2324 def Wrx : BaseAddSubEReg<isSub, 0, GPR32sp, GPR32sp,
2325 arith_extended_reg32_i32, mnemonic, OpNode> {
2328 def Xrx : BaseAddSubEReg<isSub, 0, GPR64sp, GPR64sp,
2329 arith_extended_reg32to64_i64, mnemonic, OpNode> {
2334 def Xrx64 : BaseAddSubEReg64<isSub, 0, GPR64sp, GPR64sp, GPR64,
2335 arith_extendlsl64, mnemonic> {
2336 // UXTX and SXTX only.
2337 let Inst{14-13} = 0b11;
2341 // add Rd, Rb, -imm -> sub Rd, Rn, imm
2342 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2343 (!cast<Instruction>(NAME # "Wri") GPR32sp:$Rd, GPR32sp:$Rn,
2344 addsub_shifted_imm32_neg:$imm), 0>;
2345 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2346 (!cast<Instruction>(NAME # "Xri") GPR64sp:$Rd, GPR64sp:$Rn,
2347 addsub_shifted_imm64_neg:$imm), 0>;
2349 // Register/register aliases with no shift when SP is not used.
2350 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
2351 GPR32, GPR32, GPR32, 0>;
2352 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
2353 GPR64, GPR64, GPR64, 0>;
2355 // Register/register aliases with no shift when either the destination or
2356 // first source register is SP.
2357 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2358 GPR32sponly, GPR32sp, GPR32, 16>; // UXTW #0
2359 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2360 GPR32sp, GPR32sponly, GPR32, 16>; // UXTW #0
2361 def : AddSubRegAlias<mnemonic,
2362 !cast<Instruction>(NAME#"Xrx64"),
2363 GPR64sponly, GPR64sp, GPR64, 24>; // UXTX #0
2364 def : AddSubRegAlias<mnemonic,
2365 !cast<Instruction>(NAME#"Xrx64"),
2366 GPR64sp, GPR64sponly, GPR64, 24>; // UXTX #0
2369 multiclass AddSubS<bit isSub, string mnemonic, SDNode OpNode, string cmp,
2370 string alias, string cmpAlias> {
2371 let isCompare = 1, Defs = [NZCV] in {
2372 // Add/Subtract immediate
2373 def Wri : AddSubImmShift<isSub, 1, GPR32, GPR32sp, addsub_shifted_imm32,
2377 def Xri : AddSubImmShift<isSub, 1, GPR64, GPR64sp, addsub_shifted_imm64,
2382 // Add/Subtract register
2383 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
2384 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
2386 // Add/Subtract shifted register
2387 def Wrs : BaseAddSubSReg<isSub, 1, GPR32, arith_shifted_reg32, mnemonic,
2391 def Xrs : BaseAddSubSReg<isSub, 1, GPR64, arith_shifted_reg64, mnemonic,
2396 // Add/Subtract extended register
2397 let AddedComplexity = 1 in {
2398 def Wrx : BaseAddSubEReg<isSub, 1, GPR32, GPR32sp,
2399 arith_extended_reg32_i32, mnemonic, OpNode> {
2402 def Xrx : BaseAddSubEReg<isSub, 1, GPR64, GPR64sp,
2403 arith_extended_reg32_i64, mnemonic, OpNode> {
2408 def Xrx64 : BaseAddSubEReg64<isSub, 1, GPR64, GPR64sp, GPR64,
2409 arith_extendlsl64, mnemonic> {
2410 // UXTX and SXTX only.
2411 let Inst{14-13} = 0b11;
2416 // Support negative immediates, e.g. adds Rd, Rn, -imm -> subs Rd, Rn, imm
2417 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2418 (!cast<Instruction>(NAME # "Wri") GPR32:$Rd, GPR32sp:$Rn,
2419 addsub_shifted_imm32_neg:$imm), 0>;
2420 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2421 (!cast<Instruction>(NAME # "Xri") GPR64:$Rd, GPR64sp:$Rn,
2422 addsub_shifted_imm64_neg:$imm), 0>;
2425 def : InstAlias<cmp#"\t$src, $imm", (!cast<Instruction>(NAME#"Wri")
2426 WZR, GPR32sp:$src, addsub_shifted_imm32:$imm), 5>;
2427 def : InstAlias<cmp#"\t$src, $imm", (!cast<Instruction>(NAME#"Xri")
2428 XZR, GPR64sp:$src, addsub_shifted_imm64:$imm), 5>;
2429 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Wrx")
2430 WZR, GPR32sp:$src1, GPR32:$src2, arith_extend:$sh), 4>;
2431 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrx")
2432 XZR, GPR64sp:$src1, GPR32:$src2, arith_extend:$sh), 4>;
2433 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrx64")
2434 XZR, GPR64sp:$src1, GPR64:$src2, arith_extendlsl64:$sh), 4>;
2435 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Wrs")
2436 WZR, GPR32:$src1, GPR32:$src2, arith_shift32:$sh), 4>;
2437 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrs")
2438 XZR, GPR64:$src1, GPR64:$src2, arith_shift64:$sh), 4>;
2440 // Support negative immediates, e.g. cmp Rn, -imm -> cmn Rn, imm
2441 def : InstSubst<cmpAlias#"\t$src, $imm", (!cast<Instruction>(NAME#"Wri")
2442 WZR, GPR32sp:$src, addsub_shifted_imm32_neg:$imm), 0>;
2443 def : InstSubst<cmpAlias#"\t$src, $imm", (!cast<Instruction>(NAME#"Xri")
2444 XZR, GPR64sp:$src, addsub_shifted_imm64_neg:$imm), 0>;
2446 // Compare shorthands
2447 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Wrs")
2448 WZR, GPR32:$src1, GPR32:$src2, 0), 5>;
2449 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Xrs")
2450 XZR, GPR64:$src1, GPR64:$src2, 0), 5>;
2451 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Wrx")
2452 WZR, GPR32sponly:$src1, GPR32:$src2, 16), 5>;
2453 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Xrx64")
2454 XZR, GPR64sponly:$src1, GPR64:$src2, 24), 5>;
2456 // Register/register aliases with no shift when SP is not used.
2457 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
2458 GPR32, GPR32, GPR32, 0>;
2459 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
2460 GPR64, GPR64, GPR64, 0>;
2462 // Register/register aliases with no shift when the first source register
2464 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2465 GPR32, GPR32sponly, GPR32, 16>; // UXTW #0
2466 def : AddSubRegAlias<mnemonic,
2467 !cast<Instruction>(NAME#"Xrx64"),
2468 GPR64, GPR64sponly, GPR64, 24>; // UXTX #0
2471 class AddSubG<bit isSub, string asm_inst, SDPatternOperator OpNode>
2473 isSub, 0, GPR64sp, asm_inst, "\t$Rd, $Rn, $imm6, $imm4",
2474 (ins GPR64sp:$Rn, uimm6s16:$imm6, imm0_15:$imm4),
2475 (set GPR64sp:$Rd, (OpNode GPR64sp:$Rn, imm0_63:$imm6, imm0_15:$imm4))> {
2479 let Inst{23-22} = 0b10;
2480 let Inst{21-16} = imm6;
2481 let Inst{15-14} = 0b00;
2482 let Inst{13-10} = imm4;
2483 let Unpredictable{15-14} = 0b11;
2486 class SUBP<bit setsFlags, string asm_instr, SDPatternOperator OpNode>
2487 : BaseTwoOperand<0b0000, GPR64, asm_instr, OpNode, GPR64sp, GPR64sp> {
2489 let Inst{29} = setsFlags;
2495 def SDTA64EXTR : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>,
2497 def AArch64Extr : SDNode<"AArch64ISD::EXTR", SDTA64EXTR>;
2499 class BaseExtractImm<RegisterClass regtype, Operand imm_type, string asm,
2501 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, imm_type:$imm),
2502 asm, "\t$Rd, $Rn, $Rm, $imm", "", patterns>,
2503 Sched<[WriteExtr, ReadExtrHi]> {
2509 let Inst{30-23} = 0b00100111;
2511 let Inst{20-16} = Rm;
2512 let Inst{15-10} = imm;
2517 multiclass ExtractImm<string asm> {
2518 def Wrri : BaseExtractImm<GPR32, imm0_31, asm,
2520 (AArch64Extr GPR32:$Rn, GPR32:$Rm, imm0_31:$imm))]> {
2523 // imm<5> must be zero.
2526 def Xrri : BaseExtractImm<GPR64, imm0_63, asm,
2528 (AArch64Extr GPR64:$Rn, GPR64:$Rm, imm0_63:$imm))]> {
2539 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2540 class BaseBitfieldImm<bits<2> opc,
2541 RegisterClass regtype, Operand imm_type, string asm>
2542 : I<(outs regtype:$Rd), (ins regtype:$Rn, imm_type:$immr, imm_type:$imms),
2543 asm, "\t$Rd, $Rn, $immr, $imms", "", []>,
2544 Sched<[WriteIS, ReadI]> {
2550 let Inst{30-29} = opc;
2551 let Inst{28-23} = 0b100110;
2552 let Inst{21-16} = immr;
2553 let Inst{15-10} = imms;
2558 multiclass BitfieldImm<bits<2> opc, string asm> {
2559 def Wri : BaseBitfieldImm<opc, GPR32, imm0_31, asm> {
2562 // imms<5> and immr<5> must be zero, else ReservedValue().
2566 def Xri : BaseBitfieldImm<opc, GPR64, imm0_63, asm> {
2572 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2573 class BaseBitfieldImmWith2RegArgs<bits<2> opc,
2574 RegisterClass regtype, Operand imm_type, string asm>
2575 : I<(outs regtype:$Rd), (ins regtype:$src, regtype:$Rn, imm_type:$immr,
2577 asm, "\t$Rd, $Rn, $immr, $imms", "$src = $Rd", []>,
2578 Sched<[WriteIS, ReadI]> {
2584 let Inst{30-29} = opc;
2585 let Inst{28-23} = 0b100110;
2586 let Inst{21-16} = immr;
2587 let Inst{15-10} = imms;
2592 multiclass BitfieldImmWith2RegArgs<bits<2> opc, string asm> {
2593 def Wri : BaseBitfieldImmWith2RegArgs<opc, GPR32, imm0_31, asm> {
2596 // imms<5> and immr<5> must be zero, else ReservedValue().
2600 def Xri : BaseBitfieldImmWith2RegArgs<opc, GPR64, imm0_63, asm> {
2610 // Logical (immediate)
2611 class BaseLogicalImm<bits<2> opc, RegisterClass dregtype,
2612 RegisterClass sregtype, Operand imm_type, string asm,
2614 : I<(outs dregtype:$Rd), (ins sregtype:$Rn, imm_type:$imm),
2615 asm, "\t$Rd, $Rn, $imm", "", pattern>,
2616 Sched<[WriteI, ReadI]> {
2620 let Inst{30-29} = opc;
2621 let Inst{28-23} = 0b100100;
2622 let Inst{22} = imm{12};
2623 let Inst{21-16} = imm{11-6};
2624 let Inst{15-10} = imm{5-0};
2628 let DecoderMethod = "DecodeLogicalImmInstruction";
2631 // Logical (shifted register)
2632 class BaseLogicalSReg<bits<2> opc, bit N, RegisterClass regtype,
2633 logical_shifted_reg shifted_regtype, string asm,
2635 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
2636 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
2637 Sched<[WriteISReg, ReadI, ReadISReg]> {
2638 // The operands are in order to match the 'addr' MI operands, so we
2639 // don't need an encoder method and by-name matching. Just use the default
2640 // in-order handling. Since we're using by-order, make sure the names
2646 let Inst{30-29} = opc;
2647 let Inst{28-24} = 0b01010;
2648 let Inst{23-22} = shift{7-6};
2650 let Inst{20-16} = src2;
2651 let Inst{15-10} = shift{5-0};
2652 let Inst{9-5} = src1;
2653 let Inst{4-0} = dst;
2655 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
2658 // Aliases for register+register logical instructions.
2659 class LogicalRegAlias<string asm, Instruction inst, RegisterClass regtype>
2660 : InstAlias<asm#"\t$dst, $src1, $src2",
2661 (inst regtype:$dst, regtype:$src1, regtype:$src2, 0)>;
2663 multiclass LogicalImm<bits<2> opc, string mnemonic, SDNode OpNode,
2665 let AddedComplexity = 6, isReMaterializable = 1, isAsCheapAsAMove = 1 in
2666 def Wri : BaseLogicalImm<opc, GPR32sp, GPR32, logical_imm32, mnemonic,
2667 [(set GPR32sp:$Rd, (OpNode GPR32:$Rn,
2668 logical_imm32:$imm))]> {
2670 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
2672 let AddedComplexity = 6, isReMaterializable = 1, isAsCheapAsAMove = 1 in
2673 def Xri : BaseLogicalImm<opc, GPR64sp, GPR64, logical_imm64, mnemonic,
2674 [(set GPR64sp:$Rd, (OpNode GPR64:$Rn,
2675 logical_imm64:$imm))]> {
2679 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2680 (!cast<Instruction>(NAME # "Wri") GPR32sp:$Rd, GPR32:$Rn,
2681 logical_imm32_not:$imm), 0>;
2682 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2683 (!cast<Instruction>(NAME # "Xri") GPR64sp:$Rd, GPR64:$Rn,
2684 logical_imm64_not:$imm), 0>;
2687 multiclass LogicalImmS<bits<2> opc, string mnemonic, SDNode OpNode,
2689 let isCompare = 1, Defs = [NZCV] in {
2690 def Wri : BaseLogicalImm<opc, GPR32, GPR32, logical_imm32, mnemonic,
2691 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_imm32:$imm))]> {
2693 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
2695 def Xri : BaseLogicalImm<opc, GPR64, GPR64, logical_imm64, mnemonic,
2696 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_imm64:$imm))]> {
2699 } // end Defs = [NZCV]
2701 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2702 (!cast<Instruction>(NAME # "Wri") GPR32:$Rd, GPR32:$Rn,
2703 logical_imm32_not:$imm), 0>;
2704 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2705 (!cast<Instruction>(NAME # "Xri") GPR64:$Rd, GPR64:$Rn,
2706 logical_imm64_not:$imm), 0>;
2709 class BaseLogicalRegPseudo<RegisterClass regtype, SDPatternOperator OpNode>
2710 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
2711 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
2712 Sched<[WriteI, ReadI, ReadI]>;
2714 // Split from LogicalImm as not all instructions have both.
2715 multiclass LogicalReg<bits<2> opc, bit N, string mnemonic,
2716 SDPatternOperator OpNode> {
2717 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
2718 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
2719 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
2722 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
2723 [(set GPR32:$Rd, (OpNode GPR32:$Rn,
2724 logical_shifted_reg32:$Rm))]> {
2727 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
2728 [(set GPR64:$Rd, (OpNode GPR64:$Rn,
2729 logical_shifted_reg64:$Rm))]> {
2733 def : LogicalRegAlias<mnemonic,
2734 !cast<Instruction>(NAME#"Wrs"), GPR32>;
2735 def : LogicalRegAlias<mnemonic,
2736 !cast<Instruction>(NAME#"Xrs"), GPR64>;
2739 // Split from LogicalReg to allow setting NZCV Defs
2740 multiclass LogicalRegS<bits<2> opc, bit N, string mnemonic,
2741 SDPatternOperator OpNode = null_frag> {
2742 let Defs = [NZCV], mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
2743 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
2744 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
2746 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
2747 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_shifted_reg32:$Rm))]> {
2750 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
2751 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_shifted_reg64:$Rm))]> {
2756 def : LogicalRegAlias<mnemonic,
2757 !cast<Instruction>(NAME#"Wrs"), GPR32>;
2758 def : LogicalRegAlias<mnemonic,
2759 !cast<Instruction>(NAME#"Xrs"), GPR64>;
2763 // Conditionally set flags
2766 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2767 class BaseCondComparisonImm<bit op, RegisterClass regtype, ImmLeaf immtype,
2768 string mnemonic, SDNode OpNode>
2769 : I<(outs), (ins regtype:$Rn, immtype:$imm, imm32_0_15:$nzcv, ccode:$cond),
2770 mnemonic, "\t$Rn, $imm, $nzcv, $cond", "",
2771 [(set NZCV, (OpNode regtype:$Rn, immtype:$imm, (i32 imm:$nzcv),
2772 (i32 imm:$cond), NZCV))]>,
2773 Sched<[WriteI, ReadI]> {
2783 let Inst{29-21} = 0b111010010;
2784 let Inst{20-16} = imm;
2785 let Inst{15-12} = cond;
2786 let Inst{11-10} = 0b10;
2789 let Inst{3-0} = nzcv;
2792 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2793 class BaseCondComparisonReg<bit op, RegisterClass regtype, string mnemonic,
2795 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm32_0_15:$nzcv, ccode:$cond),
2796 mnemonic, "\t$Rn, $Rm, $nzcv, $cond", "",
2797 [(set NZCV, (OpNode regtype:$Rn, regtype:$Rm, (i32 imm:$nzcv),
2798 (i32 imm:$cond), NZCV))]>,
2799 Sched<[WriteI, ReadI, ReadI]> {
2809 let Inst{29-21} = 0b111010010;
2810 let Inst{20-16} = Rm;
2811 let Inst{15-12} = cond;
2812 let Inst{11-10} = 0b00;
2815 let Inst{3-0} = nzcv;
2818 multiclass CondComparison<bit op, string mnemonic, SDNode OpNode> {
2819 // immediate operand variants
2820 def Wi : BaseCondComparisonImm<op, GPR32, imm32_0_31, mnemonic, OpNode> {
2823 def Xi : BaseCondComparisonImm<op, GPR64, imm0_31, mnemonic, OpNode> {
2826 // register operand variants
2827 def Wr : BaseCondComparisonReg<op, GPR32, mnemonic, OpNode> {
2830 def Xr : BaseCondComparisonReg<op, GPR64, mnemonic, OpNode> {
2836 // Conditional select
2839 class BaseCondSelect<bit op, bits<2> op2, RegisterClass regtype, string asm>
2840 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
2841 asm, "\t$Rd, $Rn, $Rm, $cond", "",
2843 (AArch64csel regtype:$Rn, regtype:$Rm, (i32 imm:$cond), NZCV))]>,
2844 Sched<[WriteI, ReadI, ReadI]> {
2853 let Inst{29-21} = 0b011010100;
2854 let Inst{20-16} = Rm;
2855 let Inst{15-12} = cond;
2856 let Inst{11-10} = op2;
2861 multiclass CondSelect<bit op, bits<2> op2, string asm> {
2862 def Wr : BaseCondSelect<op, op2, GPR32, asm> {
2865 def Xr : BaseCondSelect<op, op2, GPR64, asm> {
2870 class BaseCondSelectOp<bit op, bits<2> op2, RegisterClass regtype, string asm,
2872 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
2873 asm, "\t$Rd, $Rn, $Rm, $cond", "",
2875 (AArch64csel regtype:$Rn, (frag regtype:$Rm),
2876 (i32 imm:$cond), NZCV))]>,
2877 Sched<[WriteI, ReadI, ReadI]> {
2886 let Inst{29-21} = 0b011010100;
2887 let Inst{20-16} = Rm;
2888 let Inst{15-12} = cond;
2889 let Inst{11-10} = op2;
2894 def inv_cond_XFORM : SDNodeXForm<imm, [{
2895 AArch64CC::CondCode CC = static_cast<AArch64CC::CondCode>(N->getZExtValue());
2896 return CurDAG->getTargetConstant(AArch64CC::getInvertedCondCode(CC), SDLoc(N),
2900 multiclass CondSelectOp<bit op, bits<2> op2, string asm, PatFrag frag> {
2901 def Wr : BaseCondSelectOp<op, op2, GPR32, asm, frag> {
2904 def Xr : BaseCondSelectOp<op, op2, GPR64, asm, frag> {
2908 def : Pat<(AArch64csel (frag GPR32:$Rm), GPR32:$Rn, (i32 imm:$cond), NZCV),
2909 (!cast<Instruction>(NAME # Wr) GPR32:$Rn, GPR32:$Rm,
2910 (inv_cond_XFORM imm:$cond))>;
2912 def : Pat<(AArch64csel (frag GPR64:$Rm), GPR64:$Rn, (i32 imm:$cond), NZCV),
2913 (!cast<Instruction>(NAME # Xr) GPR64:$Rn, GPR64:$Rm,
2914 (inv_cond_XFORM imm:$cond))>;
2918 // Special Mask Value
2920 def maski8_or_more : Operand<i32>,
2921 ImmLeaf<i32, [{ return (Imm & 0xff) == 0xff; }]> {
2923 def maski16_or_more : Operand<i32>,
2924 ImmLeaf<i32, [{ return (Imm & 0xffff) == 0xffff; }]> {
2932 // (unsigned immediate)
2933 // Indexed for 8-bit registers. offset is in range [0,4095].
2934 def am_indexed8 : ComplexPattern<i64, 2, "SelectAddrModeIndexed8", []>;
2935 def am_indexed16 : ComplexPattern<i64, 2, "SelectAddrModeIndexed16", []>;
2936 def am_indexed32 : ComplexPattern<i64, 2, "SelectAddrModeIndexed32", []>;
2937 def am_indexed64 : ComplexPattern<i64, 2, "SelectAddrModeIndexed64", []>;
2938 def am_indexed128 : ComplexPattern<i64, 2, "SelectAddrModeIndexed128", []>;
2940 def gi_am_indexed8 :
2941 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<8>">,
2942 GIComplexPatternEquiv<am_indexed8>;
2943 def gi_am_indexed16 :
2944 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<16>">,
2945 GIComplexPatternEquiv<am_indexed16>;
2946 def gi_am_indexed32 :
2947 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<32>">,
2948 GIComplexPatternEquiv<am_indexed32>;
2949 def gi_am_indexed64 :
2950 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<64>">,
2951 GIComplexPatternEquiv<am_indexed64>;
2952 def gi_am_indexed128 :
2953 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<128>">,
2954 GIComplexPatternEquiv<am_indexed128>;
2956 class UImm12OffsetOperand<int Scale> : AsmOperandClass {
2957 let Name = "UImm12Offset" # Scale;
2958 let RenderMethod = "addUImm12OffsetOperands<" # Scale # ">";
2959 let PredicateMethod = "isUImm12Offset<" # Scale # ">";
2960 let DiagnosticType = "InvalidMemoryIndexed" # Scale;
2963 def UImm12OffsetScale1Operand : UImm12OffsetOperand<1>;
2964 def UImm12OffsetScale2Operand : UImm12OffsetOperand<2>;
2965 def UImm12OffsetScale4Operand : UImm12OffsetOperand<4>;
2966 def UImm12OffsetScale8Operand : UImm12OffsetOperand<8>;
2967 def UImm12OffsetScale16Operand : UImm12OffsetOperand<16>;
2969 class uimm12_scaled<int Scale> : Operand<i64> {
2970 let ParserMatchClass
2971 = !cast<AsmOperandClass>("UImm12OffsetScale" # Scale # "Operand");
2973 = "getLdStUImm12OpValue<AArch64::fixup_aarch64_ldst_imm12_scale" # Scale # ">";
2974 let PrintMethod = "printUImm12Offset<" # Scale # ">";
2977 def uimm12s1 : uimm12_scaled<1>;
2978 def uimm12s2 : uimm12_scaled<2>;
2979 def uimm12s4 : uimm12_scaled<4>;
2980 def uimm12s8 : uimm12_scaled<8>;
2981 def uimm12s16 : uimm12_scaled<16>;
2983 class BaseLoadStoreUI<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2984 string asm, list<dag> pattern>
2985 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", pattern> {
2991 let Inst{31-30} = sz;
2992 let Inst{29-27} = 0b111;
2994 let Inst{25-24} = 0b01;
2995 let Inst{23-22} = opc;
2996 let Inst{21-10} = offset;
3000 let DecoderMethod = "DecodeUnsignedLdStInstruction";
3003 multiclass LoadUI<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3004 Operand indextype, string asm, list<dag> pattern> {
3005 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3006 def ui : BaseLoadStoreUI<sz, V, opc, (outs regtype:$Rt),
3007 (ins GPR64sp:$Rn, indextype:$offset),
3011 def : InstAlias<asm # "\t$Rt, [$Rn]",
3012 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
3015 multiclass StoreUI<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3016 Operand indextype, string asm, list<dag> pattern> {
3017 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3018 def ui : BaseLoadStoreUI<sz, V, opc, (outs),
3019 (ins regtype:$Rt, GPR64sp:$Rn, indextype:$offset),
3023 def : InstAlias<asm # "\t$Rt, [$Rn]",
3024 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
3027 // Same as StoreUI, but take a RegisterOperand. This is used by GlobalISel to
3028 // substitute zero-registers automatically.
3030 // TODO: Roll out zero-register subtitution to GPR32/GPR64 and fold this back
3032 multiclass StoreUIz<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3033 Operand indextype, string asm, list<dag> pattern> {
3034 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3035 def ui : BaseLoadStoreUI<sz, V, opc, (outs),
3036 (ins regtype:$Rt, GPR64sp:$Rn, indextype:$offset),
3040 def : InstAlias<asm # "\t$Rt, [$Rn]",
3041 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
3044 def PrefetchOperand : AsmOperandClass {
3045 let Name = "Prefetch";
3046 let ParserMethod = "tryParsePrefetch";
3048 def prfop : Operand<i32> {
3049 let PrintMethod = "printPrefetchOp";
3050 let ParserMatchClass = PrefetchOperand;
3053 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3054 class PrefetchUI<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
3055 : BaseLoadStoreUI<sz, V, opc,
3056 (outs), (ins prfop:$Rt, GPR64sp:$Rn, uimm12s8:$offset),
3064 // Load literal address: 19-bit immediate. The low two bits of the target
3065 // offset are implied zero and so are not part of the immediate.
3066 def am_ldrlit : Operand<iPTR> {
3067 let EncoderMethod = "getLoadLiteralOpValue";
3068 let DecoderMethod = "DecodePCRelLabel19";
3069 let PrintMethod = "printAlignedLabel";
3070 let ParserMatchClass = PCRelLabel19Operand;
3071 let OperandType = "OPERAND_PCREL";
3074 let mayLoad = 1, mayStore = 0, hasSideEffects = 0, AddedComplexity = 20 in
3075 class LoadLiteral<bits<2> opc, bit V, RegisterOperand regtype, string asm, list<dag> pat>
3076 : I<(outs regtype:$Rt), (ins am_ldrlit:$label),
3077 asm, "\t$Rt, $label", "", pat>,
3081 let Inst{31-30} = opc;
3082 let Inst{29-27} = 0b011;
3084 let Inst{25-24} = 0b00;
3085 let Inst{23-5} = label;
3089 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3090 class PrefetchLiteral<bits<2> opc, bit V, string asm, list<dag> pat>
3091 : I<(outs), (ins prfop:$Rt, am_ldrlit:$label),
3092 asm, "\t$Rt, $label", "", pat>,
3096 let Inst{31-30} = opc;
3097 let Inst{29-27} = 0b011;
3099 let Inst{25-24} = 0b00;
3100 let Inst{23-5} = label;
3105 // Load/store register offset
3108 def ro_Xindexed8 : ComplexPattern<i64, 4, "SelectAddrModeXRO<8>", []>;
3109 def ro_Xindexed16 : ComplexPattern<i64, 4, "SelectAddrModeXRO<16>", []>;
3110 def ro_Xindexed32 : ComplexPattern<i64, 4, "SelectAddrModeXRO<32>", []>;
3111 def ro_Xindexed64 : ComplexPattern<i64, 4, "SelectAddrModeXRO<64>", []>;
3112 def ro_Xindexed128 : ComplexPattern<i64, 4, "SelectAddrModeXRO<128>", []>;
3114 def gi_ro_Xindexed8 :
3115 GIComplexOperandMatcher<s64, "selectAddrModeXRO<8>">,
3116 GIComplexPatternEquiv<ro_Xindexed8>;
3117 def gi_ro_Xindexed16 :
3118 GIComplexOperandMatcher<s64, "selectAddrModeXRO<16>">,
3119 GIComplexPatternEquiv<ro_Xindexed16>;
3120 def gi_ro_Xindexed32 :
3121 GIComplexOperandMatcher<s64, "selectAddrModeXRO<32>">,
3122 GIComplexPatternEquiv<ro_Xindexed32>;
3123 def gi_ro_Xindexed64 :
3124 GIComplexOperandMatcher<s64, "selectAddrModeXRO<64>">,
3125 GIComplexPatternEquiv<ro_Xindexed64>;
3126 def gi_ro_Xindexed128 :
3127 GIComplexOperandMatcher<s64, "selectAddrModeXRO<128>">,
3128 GIComplexPatternEquiv<ro_Xindexed128>;
3130 def ro_Windexed8 : ComplexPattern<i64, 4, "SelectAddrModeWRO<8>", []>;
3131 def ro_Windexed16 : ComplexPattern<i64, 4, "SelectAddrModeWRO<16>", []>;
3132 def ro_Windexed32 : ComplexPattern<i64, 4, "SelectAddrModeWRO<32>", []>;
3133 def ro_Windexed64 : ComplexPattern<i64, 4, "SelectAddrModeWRO<64>", []>;
3134 def ro_Windexed128 : ComplexPattern<i64, 4, "SelectAddrModeWRO<128>", []>;
3136 def gi_ro_Windexed8 :
3137 GIComplexOperandMatcher<s64, "selectAddrModeWRO<8>">,
3138 GIComplexPatternEquiv<ro_Windexed8>;
3139 def gi_ro_Windexed16 :
3140 GIComplexOperandMatcher<s64, "selectAddrModeWRO<16>">,
3141 GIComplexPatternEquiv<ro_Windexed16>;
3142 def gi_ro_Windexed32 :
3143 GIComplexOperandMatcher<s64, "selectAddrModeWRO<32>">,
3144 GIComplexPatternEquiv<ro_Windexed32>;
3145 def gi_ro_Windexed64 :
3146 GIComplexOperandMatcher<s64, "selectAddrModeWRO<64>">,
3147 GIComplexPatternEquiv<ro_Windexed64>;
3148 def gi_ro_Windexed128 :
3149 GIComplexOperandMatcher<s64, "selectAddrModeWRO<128>">,
3150 GIComplexPatternEquiv<ro_Windexed128>;
3152 class MemExtendOperand<string Reg, int Width> : AsmOperandClass {
3153 let Name = "Mem" # Reg # "Extend" # Width;
3154 let PredicateMethod = "isMem" # Reg # "Extend<" # Width # ">";
3155 let RenderMethod = "addMemExtendOperands";
3156 let DiagnosticType = "InvalidMemory" # Reg # "Extend" # Width;
3159 def MemWExtend8Operand : MemExtendOperand<"W", 8> {
3160 // The address "[x0, x1, lsl #0]" actually maps to the variant which performs
3161 // the trivial shift.
3162 let RenderMethod = "addMemExtend8Operands";
3164 def MemWExtend16Operand : MemExtendOperand<"W", 16>;
3165 def MemWExtend32Operand : MemExtendOperand<"W", 32>;
3166 def MemWExtend64Operand : MemExtendOperand<"W", 64>;
3167 def MemWExtend128Operand : MemExtendOperand<"W", 128>;
3169 def MemXExtend8Operand : MemExtendOperand<"X", 8> {
3170 // The address "[x0, x1, lsl #0]" actually maps to the variant which performs
3171 // the trivial shift.
3172 let RenderMethod = "addMemExtend8Operands";
3174 def MemXExtend16Operand : MemExtendOperand<"X", 16>;
3175 def MemXExtend32Operand : MemExtendOperand<"X", 32>;
3176 def MemXExtend64Operand : MemExtendOperand<"X", 64>;
3177 def MemXExtend128Operand : MemExtendOperand<"X", 128>;
3179 class ro_extend<AsmOperandClass ParserClass, string Reg, int Width>
3181 let ParserMatchClass = ParserClass;
3182 let PrintMethod = "printMemExtend<'" # Reg # "', " # Width # ">";
3183 let DecoderMethod = "DecodeMemExtend";
3184 let EncoderMethod = "getMemExtendOpValue";
3185 let MIOperandInfo = (ops i32imm:$signed, i32imm:$doshift);
3188 def ro_Wextend8 : ro_extend<MemWExtend8Operand, "w", 8>;
3189 def ro_Wextend16 : ro_extend<MemWExtend16Operand, "w", 16>;
3190 def ro_Wextend32 : ro_extend<MemWExtend32Operand, "w", 32>;
3191 def ro_Wextend64 : ro_extend<MemWExtend64Operand, "w", 64>;
3192 def ro_Wextend128 : ro_extend<MemWExtend128Operand, "w", 128>;
3194 def ro_Xextend8 : ro_extend<MemXExtend8Operand, "x", 8>;
3195 def ro_Xextend16 : ro_extend<MemXExtend16Operand, "x", 16>;
3196 def ro_Xextend32 : ro_extend<MemXExtend32Operand, "x", 32>;
3197 def ro_Xextend64 : ro_extend<MemXExtend64Operand, "x", 64>;
3198 def ro_Xextend128 : ro_extend<MemXExtend128Operand, "x", 128>;
3200 class ROAddrMode<ComplexPattern windex, ComplexPattern xindex,
3201 Operand wextend, Operand xextend> {
3202 // CodeGen-level pattern covering the entire addressing mode.
3203 ComplexPattern Wpat = windex;
3204 ComplexPattern Xpat = xindex;
3206 // Asm-level Operand covering the valid "uxtw #3" style syntax.
3207 Operand Wext = wextend;
3208 Operand Xext = xextend;
3211 def ro8 : ROAddrMode<ro_Windexed8, ro_Xindexed8, ro_Wextend8, ro_Xextend8>;
3212 def ro16 : ROAddrMode<ro_Windexed16, ro_Xindexed16, ro_Wextend16, ro_Xextend16>;
3213 def ro32 : ROAddrMode<ro_Windexed32, ro_Xindexed32, ro_Wextend32, ro_Xextend32>;
3214 def ro64 : ROAddrMode<ro_Windexed64, ro_Xindexed64, ro_Wextend64, ro_Xextend64>;
3215 def ro128 : ROAddrMode<ro_Windexed128, ro_Xindexed128, ro_Wextend128,
3218 class LoadStore8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3219 string asm, dag ins, dag outs, list<dag> pat>
3220 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3225 let Inst{31-30} = sz;
3226 let Inst{29-27} = 0b111;
3228 let Inst{25-24} = 0b00;
3229 let Inst{23-22} = opc;
3231 let Inst{20-16} = Rm;
3232 let Inst{15} = extend{1}; // sign extend Rm?
3234 let Inst{12} = extend{0}; // do shift?
3235 let Inst{11-10} = 0b10;
3240 class ROInstAlias<string asm, RegisterOperand regtype, Instruction INST>
3241 : InstAlias<asm # "\t$Rt, [$Rn, $Rm]",
3242 (INST regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, 0, 0)>;
3244 multiclass Load8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3245 string asm, ValueType Ty, SDPatternOperator loadop> {
3246 let AddedComplexity = 10 in
3247 def roW : LoadStore8RO<sz, V, opc, regtype, asm,
3249 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$extend),
3250 [(set (Ty regtype:$Rt),
3251 (loadop (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm,
3252 ro_Wextend8:$extend)))]>,
3253 Sched<[WriteLDIdx, ReadAdrBase]> {
3257 let AddedComplexity = 10 in
3258 def roX : LoadStore8RO<sz, V, opc, regtype, asm,
3260 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend8:$extend),
3261 [(set (Ty regtype:$Rt),
3262 (loadop (ro_Xindexed8 GPR64sp:$Rn, GPR64:$Rm,
3263 ro_Xextend8:$extend)))]>,
3264 Sched<[WriteLDIdx, ReadAdrBase]> {
3268 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3271 multiclass Store8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3272 string asm, ValueType Ty, SDPatternOperator storeop> {
3273 let AddedComplexity = 10 in
3274 def roW : LoadStore8RO<sz, V, opc, regtype, asm, (outs),
3275 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$extend),
3276 [(storeop (Ty regtype:$Rt),
3277 (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm,
3278 ro_Wextend8:$extend))]>,
3279 Sched<[WriteSTIdx, ReadAdrBase]> {
3283 let AddedComplexity = 10 in
3284 def roX : LoadStore8RO<sz, V, opc, regtype, asm, (outs),
3285 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend8:$extend),
3286 [(storeop (Ty regtype:$Rt),
3287 (ro_Xindexed8 GPR64sp:$Rn, GPR64:$Rm,
3288 ro_Xextend8:$extend))]>,
3289 Sched<[WriteSTIdx, ReadAdrBase]> {
3293 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3296 class LoadStore16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3297 string asm, dag ins, dag outs, list<dag> pat>
3298 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3303 let Inst{31-30} = sz;
3304 let Inst{29-27} = 0b111;
3306 let Inst{25-24} = 0b00;
3307 let Inst{23-22} = opc;
3309 let Inst{20-16} = Rm;
3310 let Inst{15} = extend{1}; // sign extend Rm?
3312 let Inst{12} = extend{0}; // do shift?
3313 let Inst{11-10} = 0b10;
3318 multiclass Load16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3319 string asm, ValueType Ty, SDPatternOperator loadop> {
3320 let AddedComplexity = 10 in
3321 def roW : LoadStore16RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3322 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend16:$extend),
3323 [(set (Ty regtype:$Rt),
3324 (loadop (ro_Windexed16 GPR64sp:$Rn, GPR32:$Rm,
3325 ro_Wextend16:$extend)))]>,
3326 Sched<[WriteLDIdx, ReadAdrBase]> {
3330 let AddedComplexity = 10 in
3331 def roX : LoadStore16RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3332 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend16:$extend),
3333 [(set (Ty regtype:$Rt),
3334 (loadop (ro_Xindexed16 GPR64sp:$Rn, GPR64:$Rm,
3335 ro_Xextend16:$extend)))]>,
3336 Sched<[WriteLDIdx, ReadAdrBase]> {
3340 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3343 multiclass Store16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3344 string asm, ValueType Ty, SDPatternOperator storeop> {
3345 let AddedComplexity = 10 in
3346 def roW : LoadStore16RO<sz, V, opc, regtype, asm, (outs),
3347 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend16:$extend),
3348 [(storeop (Ty regtype:$Rt),
3349 (ro_Windexed16 GPR64sp:$Rn, GPR32:$Rm,
3350 ro_Wextend16:$extend))]>,
3351 Sched<[WriteSTIdx, ReadAdrBase]> {
3355 let AddedComplexity = 10 in
3356 def roX : LoadStore16RO<sz, V, opc, regtype, asm, (outs),
3357 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend16:$extend),
3358 [(storeop (Ty regtype:$Rt),
3359 (ro_Xindexed16 GPR64sp:$Rn, GPR64:$Rm,
3360 ro_Xextend16:$extend))]>,
3361 Sched<[WriteSTIdx, ReadAdrBase]> {
3365 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3368 class LoadStore32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3369 string asm, dag ins, dag outs, list<dag> pat>
3370 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3375 let Inst{31-30} = sz;
3376 let Inst{29-27} = 0b111;
3378 let Inst{25-24} = 0b00;
3379 let Inst{23-22} = opc;
3381 let Inst{20-16} = Rm;
3382 let Inst{15} = extend{1}; // sign extend Rm?
3384 let Inst{12} = extend{0}; // do shift?
3385 let Inst{11-10} = 0b10;
3390 multiclass Load32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3391 string asm, ValueType Ty, SDPatternOperator loadop> {
3392 let AddedComplexity = 10 in
3393 def roW : LoadStore32RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3394 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend32:$extend),
3395 [(set (Ty regtype:$Rt),
3396 (loadop (ro_Windexed32 GPR64sp:$Rn, GPR32:$Rm,
3397 ro_Wextend32:$extend)))]>,
3398 Sched<[WriteLDIdx, ReadAdrBase]> {
3402 let AddedComplexity = 10 in
3403 def roX : LoadStore32RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3404 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend32:$extend),
3405 [(set (Ty regtype:$Rt),
3406 (loadop (ro_Xindexed32 GPR64sp:$Rn, GPR64:$Rm,
3407 ro_Xextend32:$extend)))]>,
3408 Sched<[WriteLDIdx, ReadAdrBase]> {
3412 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3415 multiclass Store32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3416 string asm, ValueType Ty, SDPatternOperator storeop> {
3417 let AddedComplexity = 10 in
3418 def roW : LoadStore32RO<sz, V, opc, regtype, asm, (outs),
3419 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend32:$extend),
3420 [(storeop (Ty regtype:$Rt),
3421 (ro_Windexed32 GPR64sp:$Rn, GPR32:$Rm,
3422 ro_Wextend32:$extend))]>,
3423 Sched<[WriteSTIdx, ReadAdrBase]> {
3427 let AddedComplexity = 10 in
3428 def roX : LoadStore32RO<sz, V, opc, regtype, asm, (outs),
3429 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend32:$extend),
3430 [(storeop (Ty regtype:$Rt),
3431 (ro_Xindexed32 GPR64sp:$Rn, GPR64:$Rm,
3432 ro_Xextend32:$extend))]>,
3433 Sched<[WriteSTIdx, ReadAdrBase]> {
3437 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3440 class LoadStore64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3441 string asm, dag ins, dag outs, list<dag> pat>
3442 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3447 let Inst{31-30} = sz;
3448 let Inst{29-27} = 0b111;
3450 let Inst{25-24} = 0b00;
3451 let Inst{23-22} = opc;
3453 let Inst{20-16} = Rm;
3454 let Inst{15} = extend{1}; // sign extend Rm?
3456 let Inst{12} = extend{0}; // do shift?
3457 let Inst{11-10} = 0b10;
3462 multiclass Load64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3463 string asm, ValueType Ty, SDPatternOperator loadop> {
3464 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3465 def roW : LoadStore64RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3466 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3467 [(set (Ty regtype:$Rt),
3468 (loadop (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3469 ro_Wextend64:$extend)))]>,
3470 Sched<[WriteLDIdx, ReadAdrBase]> {
3474 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3475 def roX : LoadStore64RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3476 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3477 [(set (Ty regtype:$Rt),
3478 (loadop (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3479 ro_Xextend64:$extend)))]>,
3480 Sched<[WriteLDIdx, ReadAdrBase]> {
3484 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3487 multiclass Store64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3488 string asm, ValueType Ty, SDPatternOperator storeop> {
3489 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3490 def roW : LoadStore64RO<sz, V, opc, regtype, asm, (outs),
3491 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3492 [(storeop (Ty regtype:$Rt),
3493 (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3494 ro_Wextend64:$extend))]>,
3495 Sched<[WriteSTIdx, ReadAdrBase]> {
3499 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3500 def roX : LoadStore64RO<sz, V, opc, regtype, asm, (outs),
3501 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3502 [(storeop (Ty regtype:$Rt),
3503 (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3504 ro_Xextend64:$extend))]>,
3505 Sched<[WriteSTIdx, ReadAdrBase]> {
3509 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3512 class LoadStore128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3513 string asm, dag ins, dag outs, list<dag> pat>
3514 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3519 let Inst{31-30} = sz;
3520 let Inst{29-27} = 0b111;
3522 let Inst{25-24} = 0b00;
3523 let Inst{23-22} = opc;
3525 let Inst{20-16} = Rm;
3526 let Inst{15} = extend{1}; // sign extend Rm?
3528 let Inst{12} = extend{0}; // do shift?
3529 let Inst{11-10} = 0b10;
3534 multiclass Load128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3535 string asm, ValueType Ty, SDPatternOperator loadop> {
3536 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3537 def roW : LoadStore128RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3538 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend),
3539 [(set (Ty regtype:$Rt),
3540 (loadop (ro_Windexed128 GPR64sp:$Rn, GPR32:$Rm,
3541 ro_Wextend128:$extend)))]>,
3542 Sched<[WriteLDIdx, ReadAdrBase]> {
3546 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3547 def roX : LoadStore128RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3548 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend128:$extend),
3549 [(set (Ty regtype:$Rt),
3550 (loadop (ro_Xindexed128 GPR64sp:$Rn, GPR64:$Rm,
3551 ro_Xextend128:$extend)))]>,
3552 Sched<[WriteLDIdx, ReadAdrBase]> {
3556 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3559 multiclass Store128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3560 string asm, ValueType Ty, SDPatternOperator storeop> {
3561 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3562 def roW : LoadStore128RO<sz, V, opc, regtype, asm, (outs),
3563 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend),
3565 Sched<[WriteSTIdx, ReadAdrBase]> {
3569 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3570 def roX : LoadStore128RO<sz, V, opc, regtype, asm, (outs),
3571 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend128:$extend),
3573 Sched<[WriteSTIdx, ReadAdrBase]> {
3577 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3580 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3581 class BasePrefetchRO<bits<2> sz, bit V, bits<2> opc, dag outs, dag ins,
3582 string asm, list<dag> pat>
3583 : I<outs, ins, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat>,
3589 let Inst{31-30} = sz;
3590 let Inst{29-27} = 0b111;
3592 let Inst{25-24} = 0b00;
3593 let Inst{23-22} = opc;
3595 let Inst{20-16} = Rm;
3596 let Inst{15} = extend{1}; // sign extend Rm?
3598 let Inst{12} = extend{0}; // do shift?
3599 let Inst{11-10} = 0b10;
3604 multiclass PrefetchRO<bits<2> sz, bit V, bits<2> opc, string asm> {
3605 def roW : BasePrefetchRO<sz, V, opc, (outs),
3606 (ins prfop:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3607 asm, [(AArch64Prefetch imm:$Rt,
3608 (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3609 ro_Wextend64:$extend))]> {
3613 def roX : BasePrefetchRO<sz, V, opc, (outs),
3614 (ins prfop:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3615 asm, [(AArch64Prefetch imm:$Rt,
3616 (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3617 ro_Xextend64:$extend))]> {
3621 def : InstAlias<"prfm $Rt, [$Rn, $Rm]",
3622 (!cast<Instruction>(NAME # "roX") prfop:$Rt,
3623 GPR64sp:$Rn, GPR64:$Rm, 0, 0)>;
3627 // Load/store unscaled immediate
3630 def am_unscaled8 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled8", []>;
3631 def am_unscaled16 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled16", []>;
3632 def am_unscaled32 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled32", []>;
3633 def am_unscaled64 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled64", []>;
3634 def am_unscaled128 :ComplexPattern<i64, 2, "SelectAddrModeUnscaled128", []>;
3636 def gi_am_unscaled8 :
3637 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled8">,
3638 GIComplexPatternEquiv<am_unscaled8>;
3639 def gi_am_unscaled16 :
3640 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled16">,
3641 GIComplexPatternEquiv<am_unscaled16>;
3642 def gi_am_unscaled32 :
3643 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled32">,
3644 GIComplexPatternEquiv<am_unscaled32>;
3645 def gi_am_unscaled64 :
3646 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled64">,
3647 GIComplexPatternEquiv<am_unscaled64>;
3648 def gi_am_unscaled128 :
3649 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled128">,
3650 GIComplexPatternEquiv<am_unscaled128>;
3653 class BaseLoadStoreUnscale<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3654 string asm, list<dag> pattern>
3655 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", pattern> {
3659 let Inst{31-30} = sz;
3660 let Inst{29-27} = 0b111;
3662 let Inst{25-24} = 0b00;
3663 let Inst{23-22} = opc;
3665 let Inst{20-12} = offset;
3666 let Inst{11-10} = 0b00;
3670 let DecoderMethod = "DecodeSignedLdStInstruction";
3673 // Armv8.4 LDAPR & STLR with Immediate Offset instruction
3674 multiclass BaseLoadUnscaleV84<string asm, bits<2> sz, bits<2> opc,
3675 RegisterOperand regtype > {
3676 def i : BaseLoadStoreUnscale<sz, 0, opc, (outs regtype:$Rt),
3677 (ins GPR64sp:$Rn, simm9:$offset), asm, []>,
3682 def : InstAlias<asm # "\t$Rt, [$Rn]",
3683 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3686 multiclass BaseStoreUnscaleV84<string asm, bits<2> sz, bits<2> opc,
3687 RegisterOperand regtype > {
3688 def i : BaseLoadStoreUnscale<sz, 0, opc, (outs),
3689 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3695 def : InstAlias<asm # "\t$Rt, [$Rn]",
3696 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3699 multiclass LoadUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3700 string asm, list<dag> pattern> {
3701 let AddedComplexity = 1 in // try this before LoadUI
3702 def i : BaseLoadStoreUnscale<sz, V, opc, (outs regtype:$Rt),
3703 (ins GPR64sp:$Rn, simm9:$offset), asm, pattern>,
3706 def : InstAlias<asm # "\t$Rt, [$Rn]",
3707 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3710 multiclass StoreUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3711 string asm, list<dag> pattern> {
3712 let AddedComplexity = 1 in // try this before StoreUI
3713 def i : BaseLoadStoreUnscale<sz, V, opc, (outs),
3714 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3718 def : InstAlias<asm # "\t$Rt, [$Rn]",
3719 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3722 multiclass PrefetchUnscaled<bits<2> sz, bit V, bits<2> opc, string asm,
3724 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3725 def i : BaseLoadStoreUnscale<sz, V, opc, (outs),
3726 (ins prfop:$Rt, GPR64sp:$Rn, simm9:$offset),
3730 def : InstAlias<asm # "\t$Rt, [$Rn]",
3731 (!cast<Instruction>(NAME # "i") prfop:$Rt, GPR64sp:$Rn, 0)>;
3735 // Load/store unscaled immediate, unprivileged
3738 class BaseLoadStoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
3739 dag oops, dag iops, string asm>
3740 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", []> {
3744 let Inst{31-30} = sz;
3745 let Inst{29-27} = 0b111;
3747 let Inst{25-24} = 0b00;
3748 let Inst{23-22} = opc;
3750 let Inst{20-12} = offset;
3751 let Inst{11-10} = 0b10;
3755 let DecoderMethod = "DecodeSignedLdStInstruction";
3758 multiclass LoadUnprivileged<bits<2> sz, bit V, bits<2> opc,
3759 RegisterClass regtype, string asm> {
3760 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in
3761 def i : BaseLoadStoreUnprivileged<sz, V, opc, (outs regtype:$Rt),
3762 (ins GPR64sp:$Rn, simm9:$offset), asm>,
3765 def : InstAlias<asm # "\t$Rt, [$Rn]",
3766 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3769 multiclass StoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
3770 RegisterClass regtype, string asm> {
3771 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
3772 def i : BaseLoadStoreUnprivileged<sz, V, opc, (outs),
3773 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3777 def : InstAlias<asm # "\t$Rt, [$Rn]",
3778 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3782 // Load/store pre-indexed
3785 class BaseLoadStorePreIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3786 string asm, string cstr, list<dag> pat>
3787 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]!", cstr, pat> {
3791 let Inst{31-30} = sz;
3792 let Inst{29-27} = 0b111;
3794 let Inst{25-24} = 0;
3795 let Inst{23-22} = opc;
3797 let Inst{20-12} = offset;
3798 let Inst{11-10} = 0b11;
3802 let DecoderMethod = "DecodeSignedLdStInstruction";
3805 let hasSideEffects = 0 in {
3806 let mayStore = 0, mayLoad = 1 in
3807 class LoadPreIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3809 : BaseLoadStorePreIdx<sz, V, opc,
3810 (outs GPR64sp:$wback, regtype:$Rt),
3811 (ins GPR64sp:$Rn, simm9:$offset), asm,
3812 "$Rn = $wback,@earlyclobber $wback", []>,
3813 Sched<[WriteLD, WriteAdr]>;
3815 let mayStore = 1, mayLoad = 0 in
3816 class StorePreIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3817 string asm, SDPatternOperator storeop, ValueType Ty>
3818 : BaseLoadStorePreIdx<sz, V, opc,
3819 (outs GPR64sp:$wback),
3820 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3821 asm, "$Rn = $wback,@earlyclobber $wback",
3822 [(set GPR64sp:$wback,
3823 (storeop (Ty regtype:$Rt), GPR64sp:$Rn, simm9:$offset))]>,
3824 Sched<[WriteAdr, WriteST]>;
3825 } // hasSideEffects = 0
3828 // Load/store post-indexed
3831 class BaseLoadStorePostIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3832 string asm, string cstr, list<dag> pat>
3833 : I<oops, iops, asm, "\t$Rt, [$Rn], $offset", cstr, pat> {
3837 let Inst{31-30} = sz;
3838 let Inst{29-27} = 0b111;
3840 let Inst{25-24} = 0b00;
3841 let Inst{23-22} = opc;
3843 let Inst{20-12} = offset;
3844 let Inst{11-10} = 0b01;
3848 let DecoderMethod = "DecodeSignedLdStInstruction";
3851 let hasSideEffects = 0 in {
3852 let mayStore = 0, mayLoad = 1 in
3853 class LoadPostIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3855 : BaseLoadStorePostIdx<sz, V, opc,
3856 (outs GPR64sp:$wback, regtype:$Rt),
3857 (ins GPR64sp:$Rn, simm9:$offset),
3858 asm, "$Rn = $wback,@earlyclobber $wback", []>,
3859 Sched<[WriteLD, WriteAdr]>;
3861 let mayStore = 1, mayLoad = 0 in
3862 class StorePostIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3863 string asm, SDPatternOperator storeop, ValueType Ty>
3864 : BaseLoadStorePostIdx<sz, V, opc,
3865 (outs GPR64sp:$wback),
3866 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3867 asm, "$Rn = $wback,@earlyclobber $wback",
3868 [(set GPR64sp:$wback,
3869 (storeop (Ty regtype:$Rt), GPR64sp:$Rn, simm9:$offset))]>,
3870 Sched<[WriteAdr, WriteST]>;
3871 } // hasSideEffects = 0
3878 // (indexed, offset)
3880 class BaseLoadStorePairOffset<bits<2> opc, bit V, bit L, dag oops, dag iops,
3882 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]", "", []> {
3887 let Inst{31-30} = opc;
3888 let Inst{29-27} = 0b101;
3890 let Inst{25-23} = 0b010;
3892 let Inst{21-15} = offset;
3893 let Inst{14-10} = Rt2;
3897 let DecoderMethod = "DecodePairLdStInstruction";
3900 multiclass LoadPairOffset<bits<2> opc, bit V, RegisterOperand regtype,
3901 Operand indextype, string asm> {
3902 let hasSideEffects = 0, mayStore = 0, mayLoad = 1 in
3903 def i : BaseLoadStorePairOffset<opc, V, 1,
3904 (outs regtype:$Rt, regtype:$Rt2),
3905 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3906 Sched<[WriteLD, WriteLDHi]>;
3908 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3909 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3914 multiclass StorePairOffset<bits<2> opc, bit V, RegisterOperand regtype,
3915 Operand indextype, string asm> {
3916 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in
3917 def i : BaseLoadStorePairOffset<opc, V, 0, (outs),
3918 (ins regtype:$Rt, regtype:$Rt2,
3919 GPR64sp:$Rn, indextype:$offset),
3923 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3924 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3929 class BaseLoadStorePairPreIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
3931 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]!", "$Rn = $wback,@earlyclobber $wback", []> {
3936 let Inst{31-30} = opc;
3937 let Inst{29-27} = 0b101;
3939 let Inst{25-23} = 0b011;
3941 let Inst{21-15} = offset;
3942 let Inst{14-10} = Rt2;
3946 let DecoderMethod = "DecodePairLdStInstruction";
3949 let hasSideEffects = 0 in {
3950 let mayStore = 0, mayLoad = 1 in
3951 class LoadPairPreIdx<bits<2> opc, bit V, RegisterOperand regtype,
3952 Operand indextype, string asm>
3953 : BaseLoadStorePairPreIdx<opc, V, 1,
3954 (outs GPR64sp:$wback, regtype:$Rt, regtype:$Rt2),
3955 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3956 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
3958 let mayStore = 1, mayLoad = 0 in
3959 class StorePairPreIdx<bits<2> opc, bit V, RegisterOperand regtype,
3960 Operand indextype, string asm>
3961 : BaseLoadStorePairPreIdx<opc, V, 0, (outs GPR64sp:$wback),
3962 (ins regtype:$Rt, regtype:$Rt2,
3963 GPR64sp:$Rn, indextype:$offset),
3965 Sched<[WriteAdr, WriteSTP]>;
3966 } // hasSideEffects = 0
3970 class BaseLoadStorePairPostIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
3972 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn], $offset", "$Rn = $wback,@earlyclobber $wback", []> {
3977 let Inst{31-30} = opc;
3978 let Inst{29-27} = 0b101;
3980 let Inst{25-23} = 0b001;
3982 let Inst{21-15} = offset;
3983 let Inst{14-10} = Rt2;
3987 let DecoderMethod = "DecodePairLdStInstruction";
3990 let hasSideEffects = 0 in {
3991 let mayStore = 0, mayLoad = 1 in
3992 class LoadPairPostIdx<bits<2> opc, bit V, RegisterOperand regtype,
3993 Operand idxtype, string asm>
3994 : BaseLoadStorePairPostIdx<opc, V, 1,
3995 (outs GPR64sp:$wback, regtype:$Rt, regtype:$Rt2),
3996 (ins GPR64sp:$Rn, idxtype:$offset), asm>,
3997 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
3999 let mayStore = 1, mayLoad = 0 in
4000 class StorePairPostIdx<bits<2> opc, bit V, RegisterOperand regtype,
4001 Operand idxtype, string asm>
4002 : BaseLoadStorePairPostIdx<opc, V, 0, (outs GPR64sp:$wback),
4003 (ins regtype:$Rt, regtype:$Rt2,
4004 GPR64sp:$Rn, idxtype:$offset),
4006 Sched<[WriteAdr, WriteSTP]>;
4007 } // hasSideEffects = 0
4011 class BaseLoadStorePairNoAlloc<bits<2> opc, bit V, bit L, dag oops, dag iops,
4013 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]", "", []> {
4018 let Inst{31-30} = opc;
4019 let Inst{29-27} = 0b101;
4021 let Inst{25-23} = 0b000;
4023 let Inst{21-15} = offset;
4024 let Inst{14-10} = Rt2;
4028 let DecoderMethod = "DecodePairLdStInstruction";
4031 multiclass LoadPairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
4032 Operand indextype, string asm> {
4033 let hasSideEffects = 0, mayStore = 0, mayLoad = 1 in
4034 def i : BaseLoadStorePairNoAlloc<opc, V, 1,
4035 (outs regtype:$Rt, regtype:$Rt2),
4036 (ins GPR64sp:$Rn, indextype:$offset), asm>,
4037 Sched<[WriteLD, WriteLDHi]>;
4040 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
4041 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
4045 multiclass StorePairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
4046 Operand indextype, string asm> {
4047 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in
4048 def i : BaseLoadStorePairNoAlloc<opc, V, 0, (outs),
4049 (ins regtype:$Rt, regtype:$Rt2,
4050 GPR64sp:$Rn, indextype:$offset),
4054 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
4055 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
4060 // Load/store exclusive
4063 // True exclusive operations write to and/or read from the system's exclusive
4064 // monitors, which as far as a compiler is concerned can be modelled as a
4065 // random shared memory address. Hence LoadExclusive mayStore.
4067 // Since these instructions have the undefined register bits set to 1 in
4068 // their canonical form, we need a post encoder method to set those bits
4069 // to 1 when encoding these instructions. We do this using the
4070 // fixLoadStoreExclusive function. This function has template parameters:
4072 // fixLoadStoreExclusive<int hasRs, int hasRt2>
4074 // hasRs indicates that the instruction uses the Rs field, so we won't set
4075 // it to 1 (and the same for Rt2). We don't need template parameters for
4076 // the other register fields since Rt and Rn are always used.
4078 let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
4079 class BaseLoadStoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4080 dag oops, dag iops, string asm, string operands>
4081 : I<oops, iops, asm, operands, "", []> {
4082 let Inst{31-30} = sz;
4083 let Inst{29-24} = 0b001000;
4089 let DecoderMethod = "DecodeExclusiveLdStInstruction";
4092 // Neither Rs nor Rt2 operands.
4093 class LoadStoreExclusiveSimple<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4094 dag oops, dag iops, string asm, string operands>
4095 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, oops, iops, asm, operands> {
4098 let Inst{20-16} = 0b11111;
4099 let Unpredictable{20-16} = 0b11111;
4100 let Inst{14-10} = 0b11111;
4101 let Unpredictable{14-10} = 0b11111;
4105 let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
4108 // Simple load acquires don't set the exclusive monitor
4109 let mayLoad = 1, mayStore = 0 in
4110 class LoadAcquire<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4111 RegisterClass regtype, string asm>
4112 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
4113 (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]">,
4116 class LoadExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4117 RegisterClass regtype, string asm>
4118 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
4119 (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]">,
4122 class LoadExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4123 RegisterClass regtype, string asm>
4124 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
4125 (outs regtype:$Rt, regtype:$Rt2),
4126 (ins GPR64sp0:$Rn), asm,
4127 "\t$Rt, $Rt2, [$Rn]">,
4128 Sched<[WriteLD, WriteLDHi]> {
4132 let Inst{14-10} = Rt2;
4136 let PostEncoderMethod = "fixLoadStoreExclusive<0,1>";
4139 // Simple store release operations do not check the exclusive monitor.
4140 let mayLoad = 0, mayStore = 1 in
4141 class StoreRelease<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4142 RegisterClass regtype, string asm>
4143 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs),
4144 (ins regtype:$Rt, GPR64sp0:$Rn),
4145 asm, "\t$Rt, [$Rn]">,
4148 let mayLoad = 1, mayStore = 1 in
4149 class StoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4150 RegisterClass regtype, string asm>
4151 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, (outs GPR32:$Ws),
4152 (ins regtype:$Rt, GPR64sp0:$Rn),
4153 asm, "\t$Ws, $Rt, [$Rn]">,
4158 let Inst{20-16} = Ws;
4162 let Constraints = "@earlyclobber $Ws";
4163 let PostEncoderMethod = "fixLoadStoreExclusive<1,0>";
4166 class StoreExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
4167 RegisterClass regtype, string asm>
4168 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
4170 (ins regtype:$Rt, regtype:$Rt2, GPR64sp0:$Rn),
4171 asm, "\t$Ws, $Rt, $Rt2, [$Rn]">,
4177 let Inst{20-16} = Ws;
4178 let Inst{14-10} = Rt2;
4182 let Constraints = "@earlyclobber $Ws";
4185 // Armv8.5-A Memory Tagging Extension
4186 class BaseMemTag<bits<2> opc1, bits<2> opc2, string asm_insn,
4187 string asm_opnds, string cstr, dag oops, dag iops>
4188 : I<oops, iops, asm_insn, asm_opnds, cstr, []>,
4192 let Inst{31-24} = 0b11011001;
4193 let Inst{23-22} = opc1;
4195 // Inst{20-12} defined by subclass
4196 let Inst{11-10} = opc2;
4198 // Inst{4-0} defined by subclass
4201 class MemTagVector<bit Load, string asm_insn, string asm_opnds,
4203 : BaseMemTag<{0b1, Load}, 0b00, asm_insn, asm_opnds,
4207 let Inst{20-12} = 0b000000000;
4213 class MemTagLoad<string asm_insn, string asm_opnds>
4214 : BaseMemTag<0b01, 0b00, asm_insn, asm_opnds, "$Rt = $wback",
4215 (outs GPR64:$wback),
4216 (ins GPR64:$Rt, GPR64sp:$Rn, simm9s16:$offset)> {
4220 let Inst{20-12} = offset;
4226 class BaseMemTagStore<bits<2> opc1, bits<2> opc2, string asm_insn,
4227 string asm_opnds, string cstr, dag oops, dag iops>
4228 : BaseMemTag<opc1, opc2, asm_insn, asm_opnds, cstr, oops, iops> {
4232 let Inst{20-12} = offset;
4238 multiclass MemTagStore<bits<2> opc1, string insn> {
4240 BaseMemTagStore<opc1, 0b10, insn, "\t$Rt, [$Rn, $offset]", "",
4241 (outs), (ins GPR64sp:$Rt, GPR64sp:$Rn, simm9s16:$offset)>;
4243 BaseMemTagStore<opc1, 0b11, insn, "\t$Rt, [$Rn, $offset]!",
4245 (outs GPR64sp:$wback),
4246 (ins GPR64sp:$Rt, GPR64sp:$Rn, simm9s16:$offset)>;
4248 BaseMemTagStore<opc1, 0b01, insn, "\t$Rt, [$Rn], $offset",
4250 (outs GPR64sp:$wback),
4251 (ins GPR64sp:$Rt, GPR64sp:$Rn, simm9s16:$offset)>;
4253 def : InstAlias<insn # "\t$Rt, [$Rn]",
4254 (!cast<Instruction>(NAME # "Offset") GPR64sp:$Rt, GPR64sp:$Rn, 0)>;
4258 // Exception generation
4261 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
4262 class ExceptionGeneration<bits<3> op1, bits<2> ll, string asm>
4263 : I<(outs), (ins i32_imm0_65535:$imm), asm, "\t$imm", "", []>,
4266 let Inst{31-24} = 0b11010100;
4267 let Inst{23-21} = op1;
4268 let Inst{20-5} = imm;
4269 let Inst{4-2} = 0b000;
4274 // UDF : Permanently UNDEFINED instructions. Format: Opc = 0x0000, 16 bit imm.
4276 let hasSideEffects = 1, isTrap = 1, mayLoad = 0, mayStore = 0 in {
4277 class UDFType<bits<16> opc, string asm>
4278 : I<(outs), (ins uimm16:$imm),
4279 asm, "\t$imm", "", []>,
4282 let Inst{31-16} = opc;
4283 let Inst{15-0} = imm;
4286 let Predicates = [HasFPARMv8] in {
4289 // Floating point to integer conversion
4292 class BaseFPToIntegerUnscaled<bits<2> type, bits<2> rmode, bits<3> opcode,
4293 RegisterClass srcType, RegisterClass dstType,
4294 string asm, list<dag> pattern>
4295 : I<(outs dstType:$Rd), (ins srcType:$Rn),
4296 asm, "\t$Rd, $Rn", "", pattern>,
4297 Sched<[WriteFCvt]> {
4300 let Inst{30-29} = 0b00;
4301 let Inst{28-24} = 0b11110;
4302 let Inst{23-22} = type;
4304 let Inst{20-19} = rmode;
4305 let Inst{18-16} = opcode;
4306 let Inst{15-10} = 0;
4311 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4312 class BaseFPToInteger<bits<2> type, bits<2> rmode, bits<3> opcode,
4313 RegisterClass srcType, RegisterClass dstType,
4314 Operand immType, string asm, list<dag> pattern>
4315 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
4316 asm, "\t$Rd, $Rn, $scale", "", pattern>,
4317 Sched<[WriteFCvt]> {
4321 let Inst{30-29} = 0b00;
4322 let Inst{28-24} = 0b11110;
4323 let Inst{23-22} = type;
4325 let Inst{20-19} = rmode;
4326 let Inst{18-16} = opcode;
4327 let Inst{15-10} = scale;
4332 multiclass FPToIntegerUnscaled<bits<2> rmode, bits<3> opcode, string asm,
4333 SDPatternOperator OpN> {
4334 // Unscaled half-precision to 32-bit
4335 def UWHr : BaseFPToIntegerUnscaled<0b11, rmode, opcode, FPR16, GPR32, asm,
4336 [(set GPR32:$Rd, (OpN FPR16:$Rn))]> {
4337 let Inst{31} = 0; // 32-bit GPR flag
4338 let Predicates = [HasFullFP16];
4341 // Unscaled half-precision to 64-bit
4342 def UXHr : BaseFPToIntegerUnscaled<0b11, rmode, opcode, FPR16, GPR64, asm,
4343 [(set GPR64:$Rd, (OpN FPR16:$Rn))]> {
4344 let Inst{31} = 1; // 64-bit GPR flag
4345 let Predicates = [HasFullFP16];
4348 // Unscaled single-precision to 32-bit
4349 def UWSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR32, asm,
4350 [(set GPR32:$Rd, (OpN FPR32:$Rn))]> {
4351 let Inst{31} = 0; // 32-bit GPR flag
4354 // Unscaled single-precision to 64-bit
4355 def UXSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR64, asm,
4356 [(set GPR64:$Rd, (OpN FPR32:$Rn))]> {
4357 let Inst{31} = 1; // 64-bit GPR flag
4360 // Unscaled double-precision to 32-bit
4361 def UWDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR32, asm,
4362 [(set GPR32:$Rd, (OpN (f64 FPR64:$Rn)))]> {
4363 let Inst{31} = 0; // 32-bit GPR flag
4366 // Unscaled double-precision to 64-bit
4367 def UXDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR64, asm,
4368 [(set GPR64:$Rd, (OpN (f64 FPR64:$Rn)))]> {
4369 let Inst{31} = 1; // 64-bit GPR flag
4373 multiclass FPToIntegerScaled<bits<2> rmode, bits<3> opcode, string asm,
4374 SDPatternOperator OpN> {
4375 // Scaled half-precision to 32-bit
4376 def SWHri : BaseFPToInteger<0b11, rmode, opcode, FPR16, GPR32,
4377 fixedpoint_f16_i32, asm,
4378 [(set GPR32:$Rd, (OpN (fmul FPR16:$Rn,
4379 fixedpoint_f16_i32:$scale)))]> {
4380 let Inst{31} = 0; // 32-bit GPR flag
4382 let Predicates = [HasFullFP16];
4385 // Scaled half-precision to 64-bit
4386 def SXHri : BaseFPToInteger<0b11, rmode, opcode, FPR16, GPR64,
4387 fixedpoint_f16_i64, asm,
4388 [(set GPR64:$Rd, (OpN (fmul FPR16:$Rn,
4389 fixedpoint_f16_i64:$scale)))]> {
4390 let Inst{31} = 1; // 64-bit GPR flag
4391 let Predicates = [HasFullFP16];
4394 // Scaled single-precision to 32-bit
4395 def SWSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR32,
4396 fixedpoint_f32_i32, asm,
4397 [(set GPR32:$Rd, (OpN (fmul FPR32:$Rn,
4398 fixedpoint_f32_i32:$scale)))]> {
4399 let Inst{31} = 0; // 32-bit GPR flag
4403 // Scaled single-precision to 64-bit
4404 def SXSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR64,
4405 fixedpoint_f32_i64, asm,
4406 [(set GPR64:$Rd, (OpN (fmul FPR32:$Rn,
4407 fixedpoint_f32_i64:$scale)))]> {
4408 let Inst{31} = 1; // 64-bit GPR flag
4411 // Scaled double-precision to 32-bit
4412 def SWDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR32,
4413 fixedpoint_f64_i32, asm,
4414 [(set GPR32:$Rd, (OpN (fmul FPR64:$Rn,
4415 fixedpoint_f64_i32:$scale)))]> {
4416 let Inst{31} = 0; // 32-bit GPR flag
4420 // Scaled double-precision to 64-bit
4421 def SXDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR64,
4422 fixedpoint_f64_i64, asm,
4423 [(set GPR64:$Rd, (OpN (fmul FPR64:$Rn,
4424 fixedpoint_f64_i64:$scale)))]> {
4425 let Inst{31} = 1; // 64-bit GPR flag
4430 // Integer to floating point conversion
4433 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
4434 class BaseIntegerToFP<bit isUnsigned,
4435 RegisterClass srcType, RegisterClass dstType,
4436 Operand immType, string asm, list<dag> pattern>
4437 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
4438 asm, "\t$Rd, $Rn, $scale", "", pattern>,
4439 Sched<[WriteFCvt]> {
4443 let Inst{30-24} = 0b0011110;
4444 let Inst{21-17} = 0b00001;
4445 let Inst{16} = isUnsigned;
4446 let Inst{15-10} = scale;
4451 class BaseIntegerToFPUnscaled<bit isUnsigned,
4452 RegisterClass srcType, RegisterClass dstType,
4453 ValueType dvt, string asm, SDNode node>
4454 : I<(outs dstType:$Rd), (ins srcType:$Rn),
4455 asm, "\t$Rd, $Rn", "", [(set (dvt dstType:$Rd), (node srcType:$Rn))]>,
4456 Sched<[WriteFCvt]> {
4460 let Inst{30-24} = 0b0011110;
4461 let Inst{21-17} = 0b10001;
4462 let Inst{16} = isUnsigned;
4463 let Inst{15-10} = 0b000000;
4468 multiclass IntegerToFP<bit isUnsigned, string asm, SDNode node> {
4470 def UWHri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR16, f16, asm, node> {
4471 let Inst{31} = 0; // 32-bit GPR flag
4472 let Inst{23-22} = 0b11; // 16-bit FPR flag
4473 let Predicates = [HasFullFP16];
4476 def UWSri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR32, f32, asm, node> {
4477 let Inst{31} = 0; // 32-bit GPR flag
4478 let Inst{23-22} = 0b00; // 32-bit FPR flag
4481 def UWDri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR64, f64, asm, node> {
4482 let Inst{31} = 0; // 32-bit GPR flag
4483 let Inst{23-22} = 0b01; // 64-bit FPR flag
4486 def UXHri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR16, f16, asm, node> {
4487 let Inst{31} = 1; // 64-bit GPR flag
4488 let Inst{23-22} = 0b11; // 16-bit FPR flag
4489 let Predicates = [HasFullFP16];
4492 def UXSri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR32, f32, asm, node> {
4493 let Inst{31} = 1; // 64-bit GPR flag
4494 let Inst{23-22} = 0b00; // 32-bit FPR flag
4497 def UXDri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR64, f64, asm, node> {
4498 let Inst{31} = 1; // 64-bit GPR flag
4499 let Inst{23-22} = 0b01; // 64-bit FPR flag
4503 def SWHri: BaseIntegerToFP<isUnsigned, GPR32, FPR16, fixedpoint_f16_i32, asm,
4505 (fdiv (node GPR32:$Rn),
4506 fixedpoint_f16_i32:$scale))]> {
4507 let Inst{31} = 0; // 32-bit GPR flag
4508 let Inst{23-22} = 0b11; // 16-bit FPR flag
4510 let Predicates = [HasFullFP16];
4513 def SWSri: BaseIntegerToFP<isUnsigned, GPR32, FPR32, fixedpoint_f32_i32, asm,
4515 (fdiv (node GPR32:$Rn),
4516 fixedpoint_f32_i32:$scale))]> {
4517 let Inst{31} = 0; // 32-bit GPR flag
4518 let Inst{23-22} = 0b00; // 32-bit FPR flag
4522 def SWDri: BaseIntegerToFP<isUnsigned, GPR32, FPR64, fixedpoint_f64_i32, asm,
4524 (fdiv (node GPR32:$Rn),
4525 fixedpoint_f64_i32:$scale))]> {
4526 let Inst{31} = 0; // 32-bit GPR flag
4527 let Inst{23-22} = 0b01; // 64-bit FPR flag
4531 def SXHri: BaseIntegerToFP<isUnsigned, GPR64, FPR16, fixedpoint_f16_i64, asm,
4533 (fdiv (node GPR64:$Rn),
4534 fixedpoint_f16_i64:$scale))]> {
4535 let Inst{31} = 1; // 64-bit GPR flag
4536 let Inst{23-22} = 0b11; // 16-bit FPR flag
4537 let Predicates = [HasFullFP16];
4540 def SXSri: BaseIntegerToFP<isUnsigned, GPR64, FPR32, fixedpoint_f32_i64, asm,
4542 (fdiv (node GPR64:$Rn),
4543 fixedpoint_f32_i64:$scale))]> {
4544 let Inst{31} = 1; // 64-bit GPR flag
4545 let Inst{23-22} = 0b00; // 32-bit FPR flag
4548 def SXDri: BaseIntegerToFP<isUnsigned, GPR64, FPR64, fixedpoint_f64_i64, asm,
4550 (fdiv (node GPR64:$Rn),
4551 fixedpoint_f64_i64:$scale))]> {
4552 let Inst{31} = 1; // 64-bit GPR flag
4553 let Inst{23-22} = 0b01; // 64-bit FPR flag
4558 // Unscaled integer <-> floating point conversion (i.e. FMOV)
4561 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4562 class BaseUnscaledConversion<bits<2> rmode, bits<3> opcode,
4563 RegisterClass srcType, RegisterClass dstType,
4565 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "",
4566 // We use COPY_TO_REGCLASS for these bitconvert operations.
4567 // copyPhysReg() expands the resultant COPY instructions after
4568 // regalloc is done. This gives greater freedom for the allocator
4569 // and related passes (coalescing, copy propagation, et. al.) to
4570 // be more effective.
4571 [/*(set (dvt dstType:$Rd), (bitconvert (svt srcType:$Rn)))*/]>,
4572 Sched<[WriteFCopy]> {
4575 let Inst{30-24} = 0b0011110;
4577 let Inst{20-19} = rmode;
4578 let Inst{18-16} = opcode;
4579 let Inst{15-10} = 0b000000;
4584 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4585 class BaseUnscaledConversionToHigh<bits<2> rmode, bits<3> opcode,
4586 RegisterClass srcType, RegisterOperand dstType, string asm,
4588 : I<(outs dstType:$Rd), (ins srcType:$Rn, VectorIndex1:$idx), asm,
4589 "{\t$Rd"#kind#"$idx, $Rn|"#kind#"\t$Rd$idx, $Rn}", "", []>,
4590 Sched<[WriteFCopy]> {
4593 let Inst{30-23} = 0b00111101;
4595 let Inst{20-19} = rmode;
4596 let Inst{18-16} = opcode;
4597 let Inst{15-10} = 0b000000;
4601 let DecoderMethod = "DecodeFMOVLaneInstruction";
4604 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4605 class BaseUnscaledConversionFromHigh<bits<2> rmode, bits<3> opcode,
4606 RegisterOperand srcType, RegisterClass dstType, string asm,
4608 : I<(outs dstType:$Rd), (ins srcType:$Rn, VectorIndex1:$idx), asm,
4609 "{\t$Rd, $Rn"#kind#"$idx|"#kind#"\t$Rd, $Rn$idx}", "", []>,
4610 Sched<[WriteFCopy]> {
4613 let Inst{30-23} = 0b00111101;
4615 let Inst{20-19} = rmode;
4616 let Inst{18-16} = opcode;
4617 let Inst{15-10} = 0b000000;
4621 let DecoderMethod = "DecodeFMOVLaneInstruction";
4625 multiclass UnscaledConversion<string asm> {
4626 def WHr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR16, asm> {
4627 let Inst{31} = 0; // 32-bit GPR flag
4628 let Inst{23-22} = 0b11; // 16-bit FPR flag
4629 let Predicates = [HasFullFP16];
4632 def XHr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR16, asm> {
4633 let Inst{31} = 1; // 64-bit GPR flag
4634 let Inst{23-22} = 0b11; // 16-bit FPR flag
4635 let Predicates = [HasFullFP16];
4638 def WSr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR32, asm> {
4639 let Inst{31} = 0; // 32-bit GPR flag
4640 let Inst{23-22} = 0b00; // 32-bit FPR flag
4643 def XDr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR64, asm> {
4644 let Inst{31} = 1; // 64-bit GPR flag
4645 let Inst{23-22} = 0b01; // 64-bit FPR flag
4648 def HWr : BaseUnscaledConversion<0b00, 0b110, FPR16, GPR32, asm> {
4649 let Inst{31} = 0; // 32-bit GPR flag
4650 let Inst{23-22} = 0b11; // 16-bit FPR flag
4651 let Predicates = [HasFullFP16];
4654 def HXr : BaseUnscaledConversion<0b00, 0b110, FPR16, GPR64, asm> {
4655 let Inst{31} = 1; // 64-bit GPR flag
4656 let Inst{23-22} = 0b11; // 16-bit FPR flag
4657 let Predicates = [HasFullFP16];
4660 def SWr : BaseUnscaledConversion<0b00, 0b110, FPR32, GPR32, asm> {
4661 let Inst{31} = 0; // 32-bit GPR flag
4662 let Inst{23-22} = 0b00; // 32-bit FPR flag
4665 def DXr : BaseUnscaledConversion<0b00, 0b110, FPR64, GPR64, asm> {
4666 let Inst{31} = 1; // 64-bit GPR flag
4667 let Inst{23-22} = 0b01; // 64-bit FPR flag
4670 def XDHighr : BaseUnscaledConversionToHigh<0b01, 0b111, GPR64, V128,
4676 def DXHighr : BaseUnscaledConversionFromHigh<0b01, 0b110, V128, GPR64,
4684 // Floating point conversion
4687 class BaseFPConversion<bits<2> type, bits<2> opcode, RegisterClass dstType,
4688 RegisterClass srcType, string asm, list<dag> pattern>
4689 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "", pattern>,
4690 Sched<[WriteFCvt]> {
4693 let Inst{31-24} = 0b00011110;
4694 let Inst{23-22} = type;
4695 let Inst{21-17} = 0b10001;
4696 let Inst{16-15} = opcode;
4697 let Inst{14-10} = 0b10000;
4702 multiclass FPConversion<string asm> {
4703 // Double-precision to Half-precision
4704 def HDr : BaseFPConversion<0b01, 0b11, FPR16, FPR64, asm,
4705 [(set FPR16:$Rd, (fpround FPR64:$Rn))]>;
4707 // Double-precision to Single-precision
4708 def SDr : BaseFPConversion<0b01, 0b00, FPR32, FPR64, asm,
4709 [(set FPR32:$Rd, (fpround FPR64:$Rn))]>;
4711 // Half-precision to Double-precision
4712 def DHr : BaseFPConversion<0b11, 0b01, FPR64, FPR16, asm,
4713 [(set FPR64:$Rd, (fpextend FPR16:$Rn))]>;
4715 // Half-precision to Single-precision
4716 def SHr : BaseFPConversion<0b11, 0b00, FPR32, FPR16, asm,
4717 [(set FPR32:$Rd, (fpextend FPR16:$Rn))]>;
4719 // Single-precision to Double-precision
4720 def DSr : BaseFPConversion<0b00, 0b01, FPR64, FPR32, asm,
4721 [(set FPR64:$Rd, (fpextend FPR32:$Rn))]>;
4723 // Single-precision to Half-precision
4724 def HSr : BaseFPConversion<0b00, 0b11, FPR16, FPR32, asm,
4725 [(set FPR16:$Rd, (fpround FPR32:$Rn))]>;
4729 // Single operand floating point data processing
4732 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4733 class BaseSingleOperandFPData<bits<6> opcode, RegisterClass regtype,
4734 ValueType vt, string asm, SDPatternOperator node>
4735 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
4736 [(set (vt regtype:$Rd), (node (vt regtype:$Rn)))]>,
4740 let Inst{31-24} = 0b00011110;
4742 let Inst{20-15} = opcode;
4743 let Inst{14-10} = 0b10000;
4748 multiclass SingleOperandFPData<bits<4> opcode, string asm,
4749 SDPatternOperator node = null_frag> {
4751 def Hr : BaseSingleOperandFPData<{0b00,opcode}, FPR16, f16, asm, node> {
4752 let Inst{23-22} = 0b11; // 16-bit size flag
4753 let Predicates = [HasFullFP16];
4756 def Sr : BaseSingleOperandFPData<{0b00,opcode}, FPR32, f32, asm, node> {
4757 let Inst{23-22} = 0b00; // 32-bit size flag
4760 def Dr : BaseSingleOperandFPData<{0b00,opcode}, FPR64, f64, asm, node> {
4761 let Inst{23-22} = 0b01; // 64-bit size flag
4765 multiclass SingleOperandFPNo16<bits<6> opcode, string asm,
4766 SDPatternOperator node = null_frag>{
4768 def Sr : BaseSingleOperandFPData<opcode, FPR32, f32, asm, node> {
4769 let Inst{23-22} = 0b00; // 32-bit registers
4772 def Dr : BaseSingleOperandFPData<opcode, FPR64, f64, asm, node> {
4773 let Inst{23-22} = 0b01; // 64-bit registers
4777 // FRInt[32|64][Z|N] instructions
4778 multiclass FRIntNNT<bits<2> opcode, string asm, SDPatternOperator node = null_frag> :
4779 SingleOperandFPNo16<{0b0100,opcode}, asm, node>;
4782 // Two operand floating point data processing
4785 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4786 class BaseTwoOperandFPData<bits<4> opcode, RegisterClass regtype,
4787 string asm, list<dag> pat>
4788 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
4789 asm, "\t$Rd, $Rn, $Rm", "", pat>,
4794 let Inst{31-24} = 0b00011110;
4796 let Inst{20-16} = Rm;
4797 let Inst{15-12} = opcode;
4798 let Inst{11-10} = 0b10;
4803 multiclass TwoOperandFPData<bits<4> opcode, string asm,
4804 SDPatternOperator node = null_frag> {
4805 def Hrr : BaseTwoOperandFPData<opcode, FPR16, asm,
4806 [(set (f16 FPR16:$Rd),
4807 (node (f16 FPR16:$Rn), (f16 FPR16:$Rm)))]> {
4808 let Inst{23-22} = 0b11; // 16-bit size flag
4809 let Predicates = [HasFullFP16];
4812 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
4813 [(set (f32 FPR32:$Rd),
4814 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]> {
4815 let Inst{23-22} = 0b00; // 32-bit size flag
4818 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
4819 [(set (f64 FPR64:$Rd),
4820 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]> {
4821 let Inst{23-22} = 0b01; // 64-bit size flag
4825 multiclass TwoOperandFPDataNeg<bits<4> opcode, string asm, SDNode node> {
4826 def Hrr : BaseTwoOperandFPData<opcode, FPR16, asm,
4827 [(set FPR16:$Rd, (fneg (node FPR16:$Rn, (f16 FPR16:$Rm))))]> {
4828 let Inst{23-22} = 0b11; // 16-bit size flag
4829 let Predicates = [HasFullFP16];
4832 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
4833 [(set FPR32:$Rd, (fneg (node FPR32:$Rn, (f32 FPR32:$Rm))))]> {
4834 let Inst{23-22} = 0b00; // 32-bit size flag
4837 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
4838 [(set FPR64:$Rd, (fneg (node FPR64:$Rn, (f64 FPR64:$Rm))))]> {
4839 let Inst{23-22} = 0b01; // 64-bit size flag
4845 // Three operand floating point data processing
4848 class BaseThreeOperandFPData<bit isNegated, bit isSub,
4849 RegisterClass regtype, string asm, list<dag> pat>
4850 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, regtype: $Ra),
4851 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pat>,
4852 Sched<[WriteFMul]> {
4857 let Inst{31-24} = 0b00011111;
4858 let Inst{21} = isNegated;
4859 let Inst{20-16} = Rm;
4860 let Inst{15} = isSub;
4861 let Inst{14-10} = Ra;
4866 multiclass ThreeOperandFPData<bit isNegated, bit isSub,string asm,
4867 SDPatternOperator node> {
4868 def Hrrr : BaseThreeOperandFPData<isNegated, isSub, FPR16, asm,
4870 (node (f16 FPR16:$Rn), (f16 FPR16:$Rm), (f16 FPR16:$Ra)))]> {
4871 let Inst{23-22} = 0b11; // 16-bit size flag
4872 let Predicates = [HasFullFP16];
4875 def Srrr : BaseThreeOperandFPData<isNegated, isSub, FPR32, asm,
4877 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm), (f32 FPR32:$Ra)))]> {
4878 let Inst{23-22} = 0b00; // 32-bit size flag
4881 def Drrr : BaseThreeOperandFPData<isNegated, isSub, FPR64, asm,
4883 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm), (f64 FPR64:$Ra)))]> {
4884 let Inst{23-22} = 0b01; // 64-bit size flag
4889 // Floating point data comparisons
4892 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4893 class BaseOneOperandFPComparison<bit signalAllNans,
4894 RegisterClass regtype, string asm,
4896 : I<(outs), (ins regtype:$Rn), asm, "\t$Rn, #0.0", "", pat>,
4897 Sched<[WriteFCmp]> {
4899 let Inst{31-24} = 0b00011110;
4902 let Inst{15-10} = 0b001000;
4904 let Inst{4} = signalAllNans;
4905 let Inst{3-0} = 0b1000;
4907 // Rm should be 0b00000 canonically, but we need to accept any value.
4908 let PostEncoderMethod = "fixOneOperandFPComparison";
4911 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4912 class BaseTwoOperandFPComparison<bit signalAllNans, RegisterClass regtype,
4913 string asm, list<dag> pat>
4914 : I<(outs), (ins regtype:$Rn, regtype:$Rm), asm, "\t$Rn, $Rm", "", pat>,
4915 Sched<[WriteFCmp]> {
4918 let Inst{31-24} = 0b00011110;
4920 let Inst{20-16} = Rm;
4921 let Inst{15-10} = 0b001000;
4923 let Inst{4} = signalAllNans;
4924 let Inst{3-0} = 0b0000;
4927 multiclass FPComparison<bit signalAllNans, string asm,
4928 SDPatternOperator OpNode = null_frag> {
4929 let Defs = [NZCV] in {
4930 def Hrr : BaseTwoOperandFPComparison<signalAllNans, FPR16, asm,
4931 [(OpNode FPR16:$Rn, (f16 FPR16:$Rm)), (implicit NZCV)]> {
4932 let Inst{23-22} = 0b11;
4933 let Predicates = [HasFullFP16];
4936 def Hri : BaseOneOperandFPComparison<signalAllNans, FPR16, asm,
4937 [(OpNode (f16 FPR16:$Rn), fpimm0), (implicit NZCV)]> {
4938 let Inst{23-22} = 0b11;
4939 let Predicates = [HasFullFP16];
4942 def Srr : BaseTwoOperandFPComparison<signalAllNans, FPR32, asm,
4943 [(OpNode FPR32:$Rn, (f32 FPR32:$Rm)), (implicit NZCV)]> {
4944 let Inst{23-22} = 0b00;
4947 def Sri : BaseOneOperandFPComparison<signalAllNans, FPR32, asm,
4948 [(OpNode (f32 FPR32:$Rn), fpimm0), (implicit NZCV)]> {
4949 let Inst{23-22} = 0b00;
4952 def Drr : BaseTwoOperandFPComparison<signalAllNans, FPR64, asm,
4953 [(OpNode FPR64:$Rn, (f64 FPR64:$Rm)), (implicit NZCV)]> {
4954 let Inst{23-22} = 0b01;
4957 def Dri : BaseOneOperandFPComparison<signalAllNans, FPR64, asm,
4958 [(OpNode (f64 FPR64:$Rn), fpimm0), (implicit NZCV)]> {
4959 let Inst{23-22} = 0b01;
4965 // Floating point conditional comparisons
4968 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4969 class BaseFPCondComparison<bit signalAllNans, RegisterClass regtype,
4970 string mnemonic, list<dag> pat>
4971 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm32_0_15:$nzcv, ccode:$cond),
4972 mnemonic, "\t$Rn, $Rm, $nzcv, $cond", "", pat>,
4973 Sched<[WriteFCmp]> {
4982 let Inst{31-24} = 0b00011110;
4984 let Inst{20-16} = Rm;
4985 let Inst{15-12} = cond;
4986 let Inst{11-10} = 0b01;
4988 let Inst{4} = signalAllNans;
4989 let Inst{3-0} = nzcv;
4992 multiclass FPCondComparison<bit signalAllNans, string mnemonic,
4993 SDPatternOperator OpNode = null_frag> {
4994 def Hrr : BaseFPCondComparison<signalAllNans, FPR16, mnemonic,
4995 [(set NZCV, (OpNode (f16 FPR16:$Rn), (f16 FPR16:$Rm), (i32 imm:$nzcv),
4996 (i32 imm:$cond), NZCV))]> {
4997 let Inst{23-22} = 0b11;
4998 let Predicates = [HasFullFP16];
5001 def Srr : BaseFPCondComparison<signalAllNans, FPR32, mnemonic,
5002 [(set NZCV, (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm), (i32 imm:$nzcv),
5003 (i32 imm:$cond), NZCV))]> {
5004 let Inst{23-22} = 0b00;
5007 def Drr : BaseFPCondComparison<signalAllNans, FPR64, mnemonic,
5008 [(set NZCV, (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm), (i32 imm:$nzcv),
5009 (i32 imm:$cond), NZCV))]> {
5010 let Inst{23-22} = 0b01;
5015 // Floating point conditional select
5018 class BaseFPCondSelect<RegisterClass regtype, ValueType vt, string asm>
5019 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
5020 asm, "\t$Rd, $Rn, $Rm, $cond", "",
5022 (AArch64csel (vt regtype:$Rn), regtype:$Rm,
5023 (i32 imm:$cond), NZCV))]>,
5030 let Inst{31-24} = 0b00011110;
5032 let Inst{20-16} = Rm;
5033 let Inst{15-12} = cond;
5034 let Inst{11-10} = 0b11;
5039 multiclass FPCondSelect<string asm> {
5040 let Uses = [NZCV] in {
5041 def Hrrr : BaseFPCondSelect<FPR16, f16, asm> {
5042 let Inst{23-22} = 0b11;
5043 let Predicates = [HasFullFP16];
5046 def Srrr : BaseFPCondSelect<FPR32, f32, asm> {
5047 let Inst{23-22} = 0b00;
5050 def Drrr : BaseFPCondSelect<FPR64, f64, asm> {
5051 let Inst{23-22} = 0b01;
5057 // Floating move immediate
5060 class BaseFPMoveImmediate<RegisterClass regtype, Operand fpimmtype, string asm>
5061 : I<(outs regtype:$Rd), (ins fpimmtype:$imm), asm, "\t$Rd, $imm", "",
5062 [(set regtype:$Rd, fpimmtype:$imm)]>,
5063 Sched<[WriteFImm]> {
5066 let Inst{31-24} = 0b00011110;
5068 let Inst{20-13} = imm;
5069 let Inst{12-5} = 0b10000000;
5073 multiclass FPMoveImmediate<string asm> {
5074 def Hi : BaseFPMoveImmediate<FPR16, fpimm16, asm> {
5075 let Inst{23-22} = 0b11;
5076 let Predicates = [HasFullFP16];
5079 def Si : BaseFPMoveImmediate<FPR32, fpimm32, asm> {
5080 let Inst{23-22} = 0b00;
5083 def Di : BaseFPMoveImmediate<FPR64, fpimm64, asm> {
5084 let Inst{23-22} = 0b01;
5087 } // end of 'let Predicates = [HasFPARMv8]'
5089 //----------------------------------------------------------------------------
5091 //----------------------------------------------------------------------------
5093 let Predicates = [HasNEON] in {
5095 //----------------------------------------------------------------------------
5096 // AdvSIMD three register vector instructions
5097 //----------------------------------------------------------------------------
5099 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5100 class BaseSIMDThreeSameVector<bit Q, bit U, bits<3> size, bits<5> opcode,
5101 RegisterOperand regtype, string asm, string kind,
5103 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
5104 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
5105 "|" # kind # "\t$Rd, $Rn, $Rm|}", "", pattern>,
5113 let Inst{28-24} = 0b01110;
5114 let Inst{23-21} = size;
5115 let Inst{20-16} = Rm;
5116 let Inst{15-11} = opcode;
5122 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5123 class BaseSIMDThreeSameVectorTied<bit Q, bit U, bits<3> size, bits<5> opcode,
5124 RegisterOperand regtype, string asm, string kind,
5126 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn, regtype:$Rm), asm,
5127 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
5128 "|" # kind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
5136 let Inst{28-24} = 0b01110;
5137 let Inst{23-21} = size;
5138 let Inst{20-16} = Rm;
5139 let Inst{15-11} = opcode;
5145 // All operand sizes distinguished in the encoding.
5146 multiclass SIMDThreeSameVector<bit U, bits<5> opc, string asm,
5147 SDPatternOperator OpNode> {
5148 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
5150 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5151 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
5153 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
5154 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
5156 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5157 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
5159 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
5160 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
5162 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5163 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
5165 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
5166 def v2i64 : BaseSIMDThreeSameVector<1, U, 0b111, opc, V128,
5168 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
5171 multiclass SIMDThreeSameVectorExtraPatterns<string inst, SDPatternOperator OpNode> {
5172 def : Pat<(v8i8 (OpNode V64:$LHS, V64:$RHS)),
5173 (!cast<Instruction>(inst#"v8i8") V64:$LHS, V64:$RHS)>;
5174 def : Pat<(v4i16 (OpNode V64:$LHS, V64:$RHS)),
5175 (!cast<Instruction>(inst#"v4i16") V64:$LHS, V64:$RHS)>;
5176 def : Pat<(v2i32 (OpNode V64:$LHS, V64:$RHS)),
5177 (!cast<Instruction>(inst#"v2i32") V64:$LHS, V64:$RHS)>;
5179 def : Pat<(v16i8 (OpNode V128:$LHS, V128:$RHS)),
5180 (!cast<Instruction>(inst#"v16i8") V128:$LHS, V128:$RHS)>;
5181 def : Pat<(v8i16 (OpNode V128:$LHS, V128:$RHS)),
5182 (!cast<Instruction>(inst#"v8i16") V128:$LHS, V128:$RHS)>;
5183 def : Pat<(v4i32 (OpNode V128:$LHS, V128:$RHS)),
5184 (!cast<Instruction>(inst#"v4i32") V128:$LHS, V128:$RHS)>;
5185 def : Pat<(v2i64 (OpNode V128:$LHS, V128:$RHS)),
5186 (!cast<Instruction>(inst#"v2i64") V128:$LHS, V128:$RHS)>;
5189 // As above, but D sized elements unsupported.
5190 multiclass SIMDThreeSameVectorBHS<bit U, bits<5> opc, string asm,
5191 SDPatternOperator OpNode> {
5192 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
5194 [(set V64:$Rd, (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))]>;
5195 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
5197 [(set V128:$Rd, (v16i8 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm))))]>;
5198 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
5200 [(set V64:$Rd, (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))]>;
5201 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
5203 [(set V128:$Rd, (v8i16 (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm))))]>;
5204 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
5206 [(set V64:$Rd, (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))]>;
5207 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
5209 [(set V128:$Rd, (v4i32 (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm))))]>;
5212 multiclass SIMDThreeSameVectorBHSTied<bit U, bits<5> opc, string asm,
5213 SDPatternOperator OpNode> {
5214 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, 0b001, opc, V64,
5216 [(set (v8i8 V64:$dst),
5217 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5218 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, 0b001, opc, V128,
5220 [(set (v16i8 V128:$dst),
5221 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
5222 def v4i16 : BaseSIMDThreeSameVectorTied<0, U, 0b011, opc, V64,
5224 [(set (v4i16 V64:$dst),
5225 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5226 def v8i16 : BaseSIMDThreeSameVectorTied<1, U, 0b011, opc, V128,
5228 [(set (v8i16 V128:$dst),
5229 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
5230 def v2i32 : BaseSIMDThreeSameVectorTied<0, U, 0b101, opc, V64,
5232 [(set (v2i32 V64:$dst),
5233 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5234 def v4i32 : BaseSIMDThreeSameVectorTied<1, U, 0b101, opc, V128,
5236 [(set (v4i32 V128:$dst),
5237 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
5240 // As above, but only B sized elements supported.
5241 multiclass SIMDThreeSameVectorB<bit U, bits<5> opc, string asm,
5242 SDPatternOperator OpNode> {
5243 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
5245 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5246 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
5248 [(set (v16i8 V128:$Rd),
5249 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
5252 // As above, but only floating point elements supported.
5253 multiclass SIMDThreeSameVectorFP<bit U, bit S, bits<3> opc,
5254 string asm, SDPatternOperator OpNode> {
5255 let Predicates = [HasNEON, HasFullFP16] in {
5256 def v4f16 : BaseSIMDThreeSameVector<0, U, {S,0b10}, {0b00,opc}, V64,
5258 [(set (v4f16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
5259 def v8f16 : BaseSIMDThreeSameVector<1, U, {S,0b10}, {0b00,opc}, V128,
5261 [(set (v8f16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
5262 } // Predicates = [HasNEON, HasFullFP16]
5263 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0b01}, {0b11,opc}, V64,
5265 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
5266 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0b01}, {0b11,opc}, V128,
5268 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
5269 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,0b11}, {0b11,opc}, V128,
5271 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
5274 multiclass SIMDThreeSameVectorFPCmp<bit U, bit S, bits<3> opc,
5276 SDPatternOperator OpNode> {
5277 let Predicates = [HasNEON, HasFullFP16] in {
5278 def v4f16 : BaseSIMDThreeSameVector<0, U, {S,0b10}, {0b00,opc}, V64,
5280 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
5281 def v8f16 : BaseSIMDThreeSameVector<1, U, {S,0b10}, {0b00,opc}, V128,
5283 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
5284 } // Predicates = [HasNEON, HasFullFP16]
5285 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0b01}, {0b11,opc}, V64,
5287 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
5288 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0b01}, {0b11,opc}, V128,
5290 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
5291 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,0b11}, {0b11,opc}, V128,
5293 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
5296 multiclass SIMDThreeSameVectorFPTied<bit U, bit S, bits<3> opc,
5297 string asm, SDPatternOperator OpNode> {
5298 let Predicates = [HasNEON, HasFullFP16] in {
5299 def v4f16 : BaseSIMDThreeSameVectorTied<0, U, {S,0b10}, {0b00,opc}, V64,
5301 [(set (v4f16 V64:$dst),
5302 (OpNode (v4f16 V64:$Rd), (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
5303 def v8f16 : BaseSIMDThreeSameVectorTied<1, U, {S,0b10}, {0b00,opc}, V128,
5305 [(set (v8f16 V128:$dst),
5306 (OpNode (v8f16 V128:$Rd), (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
5307 } // Predicates = [HasNEON, HasFullFP16]
5308 def v2f32 : BaseSIMDThreeSameVectorTied<0, U, {S,0b01}, {0b11,opc}, V64,
5310 [(set (v2f32 V64:$dst),
5311 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
5312 def v4f32 : BaseSIMDThreeSameVectorTied<1, U, {S,0b01}, {0b11,opc}, V128,
5314 [(set (v4f32 V128:$dst),
5315 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
5316 def v2f64 : BaseSIMDThreeSameVectorTied<1, U, {S,0b11}, {0b11,opc}, V128,
5318 [(set (v2f64 V128:$dst),
5319 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
5322 // As above, but D and B sized elements unsupported.
5323 multiclass SIMDThreeSameVectorHS<bit U, bits<5> opc, string asm,
5324 SDPatternOperator OpNode> {
5325 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
5327 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5328 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
5330 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
5331 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
5333 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5334 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
5336 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
5339 // Logical three vector ops share opcode bits, and only use B sized elements.
5340 multiclass SIMDLogicalThreeVector<bit U, bits<2> size, string asm,
5341 SDPatternOperator OpNode = null_frag> {
5342 def v8i8 : BaseSIMDThreeSameVector<0, U, {size,1}, 0b00011, V64,
5344 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn, V64:$Rm))]>;
5345 def v16i8 : BaseSIMDThreeSameVector<1, U, {size,1}, 0b00011, V128,
5347 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn, V128:$Rm))]>;
5349 def : Pat<(v4i16 (OpNode V64:$LHS, V64:$RHS)),
5350 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
5351 def : Pat<(v2i32 (OpNode V64:$LHS, V64:$RHS)),
5352 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
5353 def : Pat<(v1i64 (OpNode V64:$LHS, V64:$RHS)),
5354 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
5356 def : Pat<(v8i16 (OpNode V128:$LHS, V128:$RHS)),
5357 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
5358 def : Pat<(v4i32 (OpNode V128:$LHS, V128:$RHS)),
5359 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
5360 def : Pat<(v2i64 (OpNode V128:$LHS, V128:$RHS)),
5361 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
5364 multiclass SIMDLogicalThreeVectorTied<bit U, bits<2> size,
5365 string asm, SDPatternOperator OpNode> {
5366 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, {size,1}, 0b00011, V64,
5368 [(set (v8i8 V64:$dst),
5369 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5370 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, {size,1}, 0b00011, V128,
5372 [(set (v16i8 V128:$dst),
5373 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
5374 (v16i8 V128:$Rm)))]>;
5376 def : Pat<(v4i16 (OpNode (v4i16 V64:$LHS), (v4i16 V64:$MHS),
5378 (!cast<Instruction>(NAME#"v8i8")
5379 V64:$LHS, V64:$MHS, V64:$RHS)>;
5380 def : Pat<(v2i32 (OpNode (v2i32 V64:$LHS), (v2i32 V64:$MHS),
5382 (!cast<Instruction>(NAME#"v8i8")
5383 V64:$LHS, V64:$MHS, V64:$RHS)>;
5384 def : Pat<(v1i64 (OpNode (v1i64 V64:$LHS), (v1i64 V64:$MHS),
5386 (!cast<Instruction>(NAME#"v8i8")
5387 V64:$LHS, V64:$MHS, V64:$RHS)>;
5389 def : Pat<(v8i16 (OpNode (v8i16 V128:$LHS), (v8i16 V128:$MHS),
5390 (v8i16 V128:$RHS))),
5391 (!cast<Instruction>(NAME#"v16i8")
5392 V128:$LHS, V128:$MHS, V128:$RHS)>;
5393 def : Pat<(v4i32 (OpNode (v4i32 V128:$LHS), (v4i32 V128:$MHS),
5394 (v4i32 V128:$RHS))),
5395 (!cast<Instruction>(NAME#"v16i8")
5396 V128:$LHS, V128:$MHS, V128:$RHS)>;
5397 def : Pat<(v2i64 (OpNode (v2i64 V128:$LHS), (v2i64 V128:$MHS),
5398 (v2i64 V128:$RHS))),
5399 (!cast<Instruction>(NAME#"v16i8")
5400 V128:$LHS, V128:$MHS, V128:$RHS)>;
5403 // ARMv8.2-A Dot Product Instructions (Vector): These instructions extract
5404 // bytes from S-sized elements.
5405 class BaseSIMDThreeSameVectorDot<bit Q, bit U, string asm, string kind1,
5406 string kind2, RegisterOperand RegType,
5407 ValueType AccumType, ValueType InputType,
5408 SDPatternOperator OpNode> :
5409 BaseSIMDThreeSameVectorTied<Q, U, 0b100, 0b10010, RegType, asm, kind1,
5410 [(set (AccumType RegType:$dst),
5411 (OpNode (AccumType RegType:$Rd),
5412 (InputType RegType:$Rn),
5413 (InputType RegType:$Rm)))]> {
5414 let AsmString = !strconcat(asm, "{\t$Rd" # kind1 # ", $Rn" # kind2 # ", $Rm" # kind2 # "}");
5417 multiclass SIMDThreeSameVectorDot<bit U, string asm, SDPatternOperator OpNode> {
5418 def v8i8 : BaseSIMDThreeSameVectorDot<0, U, asm, ".2s", ".8b", V64,
5419 v2i32, v8i8, OpNode>;
5420 def v16i8 : BaseSIMDThreeSameVectorDot<1, U, asm, ".4s", ".16b", V128,
5421 v4i32, v16i8, OpNode>;
5424 // ARMv8.2-A Fused Multiply Add-Long Instructions (Vector): These instructions
5425 // select inputs from 4H vectors and accumulate outputs to a 2S vector (or from
5426 // 8H to 4S, when Q=1).
5427 class BaseSIMDThreeSameVectorFML<bit Q, bit U, bit b13, bits<3> size, string asm, string kind1,
5428 string kind2, RegisterOperand RegType,
5429 ValueType AccumType, ValueType InputType,
5430 SDPatternOperator OpNode> :
5431 BaseSIMDThreeSameVectorTied<Q, U, size, 0b11101, RegType, asm, kind1,
5432 [(set (AccumType RegType:$dst),
5433 (OpNode (AccumType RegType:$Rd),
5434 (InputType RegType:$Rn),
5435 (InputType RegType:$Rm)))]> {
5436 let AsmString = !strconcat(asm, "{\t$Rd" # kind1 # ", $Rn" # kind2 # ", $Rm" # kind2 # "}");
5440 multiclass SIMDThreeSameVectorFML<bit U, bit b13, bits<3> size, string asm,
5441 SDPatternOperator OpNode> {
5442 def v4f16 : BaseSIMDThreeSameVectorFML<0, U, b13, size, asm, ".2s", ".2h", V64,
5443 v2f32, v4f16, OpNode>;
5444 def v8f16 : BaseSIMDThreeSameVectorFML<1, U, b13, size, asm, ".4s", ".4h", V128,
5445 v4f32, v8f16, OpNode>;
5449 //----------------------------------------------------------------------------
5450 // AdvSIMD two register vector instructions.
5451 //----------------------------------------------------------------------------
5453 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5454 class BaseSIMDTwoSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5455 bits<2> size2, RegisterOperand regtype, string asm,
5456 string dstkind, string srckind, list<dag> pattern>
5457 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5458 "{\t$Rd" # dstkind # ", $Rn" # srckind #
5459 "|" # dstkind # "\t$Rd, $Rn}", "", pattern>,
5466 let Inst{28-24} = 0b01110;
5467 let Inst{23-22} = size;
5469 let Inst{20-19} = size2;
5470 let Inst{18-17} = 0b00;
5471 let Inst{16-12} = opcode;
5472 let Inst{11-10} = 0b10;
5477 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5478 class BaseSIMDTwoSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5479 bits<2> size2, RegisterOperand regtype,
5480 string asm, string dstkind, string srckind,
5482 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn), asm,
5483 "{\t$Rd" # dstkind # ", $Rn" # srckind #
5484 "|" # dstkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
5491 let Inst{28-24} = 0b01110;
5492 let Inst{23-22} = size;
5494 let Inst{20-19} = size2;
5495 let Inst{18-17} = 0b00;
5496 let Inst{16-12} = opcode;
5497 let Inst{11-10} = 0b10;
5502 // Supports B, H, and S element sizes.
5503 multiclass SIMDTwoVectorBHS<bit U, bits<5> opc, string asm,
5504 SDPatternOperator OpNode> {
5505 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5507 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5508 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5509 asm, ".16b", ".16b",
5510 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5511 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5513 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5514 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5516 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5517 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5519 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5520 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5522 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5525 class BaseSIMDVectorLShiftLongBySize<bit Q, bits<2> size,
5526 RegisterOperand regtype, string asm, string dstkind,
5527 string srckind, string amount>
5528 : I<(outs V128:$Rd), (ins regtype:$Rn), asm,
5529 "{\t$Rd" # dstkind # ", $Rn" # srckind # ", #" # amount #
5530 "|" # dstkind # "\t$Rd, $Rn, #" # amount # "}", "", []>,
5536 let Inst{29-24} = 0b101110;
5537 let Inst{23-22} = size;
5538 let Inst{21-10} = 0b100001001110;
5543 multiclass SIMDVectorLShiftLongBySizeBHS {
5544 let hasSideEffects = 0 in {
5545 def v8i8 : BaseSIMDVectorLShiftLongBySize<0, 0b00, V64,
5546 "shll", ".8h", ".8b", "8">;
5547 def v16i8 : BaseSIMDVectorLShiftLongBySize<1, 0b00, V128,
5548 "shll2", ".8h", ".16b", "8">;
5549 def v4i16 : BaseSIMDVectorLShiftLongBySize<0, 0b01, V64,
5550 "shll", ".4s", ".4h", "16">;
5551 def v8i16 : BaseSIMDVectorLShiftLongBySize<1, 0b01, V128,
5552 "shll2", ".4s", ".8h", "16">;
5553 def v2i32 : BaseSIMDVectorLShiftLongBySize<0, 0b10, V64,
5554 "shll", ".2d", ".2s", "32">;
5555 def v4i32 : BaseSIMDVectorLShiftLongBySize<1, 0b10, V128,
5556 "shll2", ".2d", ".4s", "32">;
5560 // Supports all element sizes.
5561 multiclass SIMDLongTwoVector<bit U, bits<5> opc, string asm,
5562 SDPatternOperator OpNode> {
5563 def v8i8_v4i16 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5565 [(set (v4i16 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5566 def v16i8_v8i16 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5568 [(set (v8i16 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5569 def v4i16_v2i32 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5571 [(set (v2i32 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5572 def v8i16_v4i32 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5574 [(set (v4i32 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5575 def v2i32_v1i64 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5577 [(set (v1i64 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5578 def v4i32_v2i64 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5580 [(set (v2i64 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5583 multiclass SIMDLongTwoVectorTied<bit U, bits<5> opc, string asm,
5584 SDPatternOperator OpNode> {
5585 def v8i8_v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, 0b00, V64,
5587 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd),
5589 def v16i8_v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, 0b00, V128,
5591 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd),
5592 (v16i8 V128:$Rn)))]>;
5593 def v4i16_v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, 0b00, V64,
5595 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd),
5596 (v4i16 V64:$Rn)))]>;
5597 def v8i16_v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, 0b00, V128,
5599 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd),
5600 (v8i16 V128:$Rn)))]>;
5601 def v2i32_v1i64 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, 0b00, V64,
5603 [(set (v1i64 V64:$dst), (OpNode (v1i64 V64:$Rd),
5604 (v2i32 V64:$Rn)))]>;
5605 def v4i32_v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, 0b00, V128,
5607 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd),
5608 (v4i32 V128:$Rn)))]>;
5611 // Supports all element sizes, except 1xD.
5612 multiclass SIMDTwoVectorBHSDTied<bit U, bits<5> opc, string asm,
5613 SDPatternOperator OpNode> {
5614 def v8i8 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, 0b00, V64,
5616 [(set (v8i8 V64:$dst), (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn)))]>;
5617 def v16i8 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, 0b00, V128,
5618 asm, ".16b", ".16b",
5619 [(set (v16i8 V128:$dst), (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
5620 def v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, 0b00, V64,
5622 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn)))]>;
5623 def v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, 0b00, V128,
5625 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn)))]>;
5626 def v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, 0b00, V64,
5628 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn)))]>;
5629 def v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, 0b00, V128,
5631 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
5632 def v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b11, opc, 0b00, V128,
5634 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn)))]>;
5637 multiclass SIMDTwoVectorBHSD<bit U, bits<5> opc, string asm,
5638 SDPatternOperator OpNode = null_frag> {
5639 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5641 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5642 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5643 asm, ".16b", ".16b",
5644 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5645 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5647 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5648 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5650 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5651 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5653 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5654 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5656 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5657 def v2i64 : BaseSIMDTwoSameVector<1, U, 0b11, opc, 0b00, V128,
5659 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5663 // Supports only B element sizes.
5664 multiclass SIMDTwoVectorB<bit U, bits<2> size, bits<5> opc, string asm,
5665 SDPatternOperator OpNode> {
5666 def v8i8 : BaseSIMDTwoSameVector<0, U, size, opc, 0b00, V64,
5668 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5669 def v16i8 : BaseSIMDTwoSameVector<1, U, size, opc, 0b00, V128,
5670 asm, ".16b", ".16b",
5671 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5675 // Supports only B and H element sizes.
5676 multiclass SIMDTwoVectorBH<bit U, bits<5> opc, string asm,
5677 SDPatternOperator OpNode> {
5678 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5680 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn))]>;
5681 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5682 asm, ".16b", ".16b",
5683 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn))]>;
5684 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5686 [(set (v4i16 V64:$Rd), (OpNode V64:$Rn))]>;
5687 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5689 [(set (v8i16 V128:$Rd), (OpNode V128:$Rn))]>;
5692 // Supports H, S and D element sizes, uses high bit of the size field
5693 // as an extra opcode bit.
5694 multiclass SIMDTwoVectorFP<bit U, bit S, bits<5> opc, string asm,
5695 SDPatternOperator OpNode> {
5696 let Predicates = [HasNEON, HasFullFP16] in {
5697 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5699 [(set (v4f16 V64:$Rd), (OpNode (v4f16 V64:$Rn)))]>;
5700 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5702 [(set (v8f16 V128:$Rd), (OpNode (v8f16 V128:$Rn)))]>;
5703 } // Predicates = [HasNEON, HasFullFP16]
5704 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5706 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5707 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5709 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5710 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5712 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5715 // Supports only S and D element sizes
5716 multiclass SIMDTwoVectorSD<bit U, bits<5> opc, string asm,
5717 SDPatternOperator OpNode = null_frag> {
5719 def v2f32 : BaseSIMDTwoSameVector<0, U, 00, opc, 0b00, V64,
5721 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5722 def v4f32 : BaseSIMDTwoSameVector<1, U, 00, opc, 0b00, V128,
5724 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5725 def v2f64 : BaseSIMDTwoSameVector<1, U, 01, opc, 0b00, V128,
5727 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5730 multiclass FRIntNNTVector<bit U, bit op, string asm,
5731 SDPatternOperator OpNode = null_frag> :
5732 SIMDTwoVectorSD<U, {0b1111,op}, asm, OpNode>;
5734 // Supports only S element size.
5735 multiclass SIMDTwoVectorS<bit U, bit S, bits<5> opc, string asm,
5736 SDPatternOperator OpNode> {
5737 def v2i32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5739 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5740 def v4i32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5742 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5746 multiclass SIMDTwoVectorFPToInt<bit U, bit S, bits<5> opc, string asm,
5747 SDPatternOperator OpNode> {
5748 let Predicates = [HasNEON, HasFullFP16] in {
5749 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5751 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn)))]>;
5752 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5754 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn)))]>;
5755 } // Predicates = [HasNEON, HasFullFP16]
5756 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5758 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5759 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5761 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5762 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5764 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5767 multiclass SIMDTwoVectorIntToFP<bit U, bit S, bits<5> opc, string asm,
5768 SDPatternOperator OpNode> {
5769 let Predicates = [HasNEON, HasFullFP16] in {
5770 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5772 [(set (v4f16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5773 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5775 [(set (v8f16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5776 } // Predicates = [HasNEON, HasFullFP16]
5777 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5779 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5780 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5782 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5783 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5785 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5789 class BaseSIMDMixedTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5790 RegisterOperand inreg, RegisterOperand outreg,
5791 string asm, string outkind, string inkind,
5793 : I<(outs outreg:$Rd), (ins inreg:$Rn), asm,
5794 "{\t$Rd" # outkind # ", $Rn" # inkind #
5795 "|" # outkind # "\t$Rd, $Rn}", "", pattern>,
5802 let Inst{28-24} = 0b01110;
5803 let Inst{23-22} = size;
5804 let Inst{21-17} = 0b10000;
5805 let Inst{16-12} = opcode;
5806 let Inst{11-10} = 0b10;
5811 class BaseSIMDMixedTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5812 RegisterOperand inreg, RegisterOperand outreg,
5813 string asm, string outkind, string inkind,
5815 : I<(outs outreg:$dst), (ins outreg:$Rd, inreg:$Rn), asm,
5816 "{\t$Rd" # outkind # ", $Rn" # inkind #
5817 "|" # outkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
5824 let Inst{28-24} = 0b01110;
5825 let Inst{23-22} = size;
5826 let Inst{21-17} = 0b10000;
5827 let Inst{16-12} = opcode;
5828 let Inst{11-10} = 0b10;
5833 multiclass SIMDMixedTwoVector<bit U, bits<5> opc, string asm,
5834 SDPatternOperator OpNode> {
5835 def v8i8 : BaseSIMDMixedTwoVector<0, U, 0b00, opc, V128, V64,
5837 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5838 def v16i8 : BaseSIMDMixedTwoVectorTied<1, U, 0b00, opc, V128, V128,
5839 asm#"2", ".16b", ".8h", []>;
5840 def v4i16 : BaseSIMDMixedTwoVector<0, U, 0b01, opc, V128, V64,
5842 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5843 def v8i16 : BaseSIMDMixedTwoVectorTied<1, U, 0b01, opc, V128, V128,
5844 asm#"2", ".8h", ".4s", []>;
5845 def v2i32 : BaseSIMDMixedTwoVector<0, U, 0b10, opc, V128, V64,
5847 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5848 def v4i32 : BaseSIMDMixedTwoVectorTied<1, U, 0b10, opc, V128, V128,
5849 asm#"2", ".4s", ".2d", []>;
5851 def : Pat<(concat_vectors (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn))),
5852 (!cast<Instruction>(NAME # "v16i8")
5853 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5854 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn))),
5855 (!cast<Instruction>(NAME # "v8i16")
5856 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5857 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn))),
5858 (!cast<Instruction>(NAME # "v4i32")
5859 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5862 class BaseSIMDCmpTwoVector<bit Q, bit U, bits<2> size, bits<2> size2,
5863 bits<5> opcode, RegisterOperand regtype, string asm,
5864 string kind, string zero, ValueType dty,
5865 ValueType sty, SDNode OpNode>
5866 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5867 "{\t$Rd" # kind # ", $Rn" # kind # ", #" # zero #
5868 "|" # kind # "\t$Rd, $Rn, #" # zero # "}", "",
5869 [(set (dty regtype:$Rd), (OpNode (sty regtype:$Rn)))]>,
5876 let Inst{28-24} = 0b01110;
5877 let Inst{23-22} = size;
5879 let Inst{20-19} = size2;
5880 let Inst{18-17} = 0b00;
5881 let Inst{16-12} = opcode;
5882 let Inst{11-10} = 0b10;
5887 // Comparisons support all element sizes, except 1xD.
5888 multiclass SIMDCmpTwoVector<bit U, bits<5> opc, string asm,
5890 def v8i8rz : BaseSIMDCmpTwoVector<0, U, 0b00, 0b00, opc, V64,
5892 v8i8, v8i8, OpNode>;
5893 def v16i8rz : BaseSIMDCmpTwoVector<1, U, 0b00, 0b00, opc, V128,
5895 v16i8, v16i8, OpNode>;
5896 def v4i16rz : BaseSIMDCmpTwoVector<0, U, 0b01, 0b00, opc, V64,
5898 v4i16, v4i16, OpNode>;
5899 def v8i16rz : BaseSIMDCmpTwoVector<1, U, 0b01, 0b00, opc, V128,
5901 v8i16, v8i16, OpNode>;
5902 def v2i32rz : BaseSIMDCmpTwoVector<0, U, 0b10, 0b00, opc, V64,
5904 v2i32, v2i32, OpNode>;
5905 def v4i32rz : BaseSIMDCmpTwoVector<1, U, 0b10, 0b00, opc, V128,
5907 v4i32, v4i32, OpNode>;
5908 def v2i64rz : BaseSIMDCmpTwoVector<1, U, 0b11, 0b00, opc, V128,
5910 v2i64, v2i64, OpNode>;
5913 // FP Comparisons support only S and D element sizes (and H for v8.2a).
5914 multiclass SIMDFPCmpTwoVector<bit U, bit S, bits<5> opc,
5915 string asm, SDNode OpNode> {
5917 let Predicates = [HasNEON, HasFullFP16] in {
5918 def v4i16rz : BaseSIMDCmpTwoVector<0, U, {S,1}, 0b11, opc, V64,
5920 v4i16, v4f16, OpNode>;
5921 def v8i16rz : BaseSIMDCmpTwoVector<1, U, {S,1}, 0b11, opc, V128,
5923 v8i16, v8f16, OpNode>;
5924 } // Predicates = [HasNEON, HasFullFP16]
5925 def v2i32rz : BaseSIMDCmpTwoVector<0, U, {S,0}, 0b00, opc, V64,
5927 v2i32, v2f32, OpNode>;
5928 def v4i32rz : BaseSIMDCmpTwoVector<1, U, {S,0}, 0b00, opc, V128,
5930 v4i32, v4f32, OpNode>;
5931 def v2i64rz : BaseSIMDCmpTwoVector<1, U, {S,1}, 0b00, opc, V128,
5933 v2i64, v2f64, OpNode>;
5935 let Predicates = [HasNEON, HasFullFP16] in {
5936 def : InstAlias<asm # "\t$Vd.4h, $Vn.4h, #0",
5937 (!cast<Instruction>(NAME # v4i16rz) V64:$Vd, V64:$Vn), 0>;
5938 def : InstAlias<asm # "\t$Vd.8h, $Vn.8h, #0",
5939 (!cast<Instruction>(NAME # v8i16rz) V128:$Vd, V128:$Vn), 0>;
5941 def : InstAlias<asm # "\t$Vd.2s, $Vn.2s, #0",
5942 (!cast<Instruction>(NAME # v2i32rz) V64:$Vd, V64:$Vn), 0>;
5943 def : InstAlias<asm # "\t$Vd.4s, $Vn.4s, #0",
5944 (!cast<Instruction>(NAME # v4i32rz) V128:$Vd, V128:$Vn), 0>;
5945 def : InstAlias<asm # "\t$Vd.2d, $Vn.2d, #0",
5946 (!cast<Instruction>(NAME # v2i64rz) V128:$Vd, V128:$Vn), 0>;
5947 let Predicates = [HasNEON, HasFullFP16] in {
5948 def : InstAlias<asm # ".4h\t$Vd, $Vn, #0",
5949 (!cast<Instruction>(NAME # v4i16rz) V64:$Vd, V64:$Vn), 0>;
5950 def : InstAlias<asm # ".8h\t$Vd, $Vn, #0",
5951 (!cast<Instruction>(NAME # v8i16rz) V128:$Vd, V128:$Vn), 0>;
5953 def : InstAlias<asm # ".2s\t$Vd, $Vn, #0",
5954 (!cast<Instruction>(NAME # v2i32rz) V64:$Vd, V64:$Vn), 0>;
5955 def : InstAlias<asm # ".4s\t$Vd, $Vn, #0",
5956 (!cast<Instruction>(NAME # v4i32rz) V128:$Vd, V128:$Vn), 0>;
5957 def : InstAlias<asm # ".2d\t$Vd, $Vn, #0",
5958 (!cast<Instruction>(NAME # v2i64rz) V128:$Vd, V128:$Vn), 0>;
5961 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5962 class BaseSIMDFPCvtTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5963 RegisterOperand outtype, RegisterOperand intype,
5964 string asm, string VdTy, string VnTy,
5966 : I<(outs outtype:$Rd), (ins intype:$Rn), asm,
5967 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "", pattern>,
5974 let Inst{28-24} = 0b01110;
5975 let Inst{23-22} = size;
5976 let Inst{21-17} = 0b10000;
5977 let Inst{16-12} = opcode;
5978 let Inst{11-10} = 0b10;
5983 class BaseSIMDFPCvtTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5984 RegisterOperand outtype, RegisterOperand intype,
5985 string asm, string VdTy, string VnTy,
5987 : I<(outs outtype:$dst), (ins outtype:$Rd, intype:$Rn), asm,
5988 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "$Rd = $dst", pattern>,
5995 let Inst{28-24} = 0b01110;
5996 let Inst{23-22} = size;
5997 let Inst{21-17} = 0b10000;
5998 let Inst{16-12} = opcode;
5999 let Inst{11-10} = 0b10;
6004 multiclass SIMDFPWidenTwoVector<bit U, bit S, bits<5> opc, string asm> {
6005 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V128, V64,
6006 asm, ".4s", ".4h", []>;
6007 def v8i16 : BaseSIMDFPCvtTwoVector<1, U, {S,0}, opc, V128, V128,
6008 asm#"2", ".4s", ".8h", []>;
6009 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V128, V64,
6010 asm, ".2d", ".2s", []>;
6011 def v4i32 : BaseSIMDFPCvtTwoVector<1, U, {S,1}, opc, V128, V128,
6012 asm#"2", ".2d", ".4s", []>;
6015 multiclass SIMDFPNarrowTwoVector<bit U, bit S, bits<5> opc, string asm> {
6016 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V64, V128,
6017 asm, ".4h", ".4s", []>;
6018 def v8i16 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,0}, opc, V128, V128,
6019 asm#"2", ".8h", ".4s", []>;
6020 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
6021 asm, ".2s", ".2d", []>;
6022 def v4i32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
6023 asm#"2", ".4s", ".2d", []>;
6026 multiclass SIMDFPInexactCvtTwoVector<bit U, bit S, bits<5> opc, string asm,
6028 def v2f32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
6030 [(set (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
6031 def v4f32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
6032 asm#"2", ".4s", ".2d", []>;
6034 def : Pat<(concat_vectors (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn))),
6035 (!cast<Instruction>(NAME # "v4f32")
6036 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
6039 //----------------------------------------------------------------------------
6040 // AdvSIMD three register different-size vector instructions.
6041 //----------------------------------------------------------------------------
6043 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6044 class BaseSIMDDifferentThreeVector<bit U, bits<3> size, bits<4> opcode,
6045 RegisterOperand outtype, RegisterOperand intype1,
6046 RegisterOperand intype2, string asm,
6047 string outkind, string inkind1, string inkind2,
6049 : I<(outs outtype:$Rd), (ins intype1:$Rn, intype2:$Rm), asm,
6050 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
6051 "|" # outkind # "\t$Rd, $Rn, $Rm}", "", pattern>,
6057 let Inst{30} = size{0};
6059 let Inst{28-24} = 0b01110;
6060 let Inst{23-22} = size{2-1};
6062 let Inst{20-16} = Rm;
6063 let Inst{15-12} = opcode;
6064 let Inst{11-10} = 0b00;
6069 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6070 class BaseSIMDDifferentThreeVectorTied<bit U, bits<3> size, bits<4> opcode,
6071 RegisterOperand outtype, RegisterOperand intype1,
6072 RegisterOperand intype2, string asm,
6073 string outkind, string inkind1, string inkind2,
6075 : I<(outs outtype:$dst), (ins outtype:$Rd, intype1:$Rn, intype2:$Rm), asm,
6076 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
6077 "|" # outkind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
6083 let Inst{30} = size{0};
6085 let Inst{28-24} = 0b01110;
6086 let Inst{23-22} = size{2-1};
6088 let Inst{20-16} = Rm;
6089 let Inst{15-12} = opcode;
6090 let Inst{11-10} = 0b00;
6095 // FIXME: TableGen doesn't know how to deal with expanded types that also
6096 // change the element count (in this case, placing the results in
6097 // the high elements of the result register rather than the low
6098 // elements). Until that's fixed, we can't code-gen those.
6099 multiclass SIMDNarrowThreeVectorBHS<bit U, bits<4> opc, string asm,
6101 def v8i16_v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6103 asm, ".8b", ".8h", ".8h",
6104 [(set (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
6105 def v8i16_v16i8 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
6107 asm#"2", ".16b", ".8h", ".8h",
6109 def v4i32_v4i16 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6111 asm, ".4h", ".4s", ".4s",
6112 [(set (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
6113 def v4i32_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
6115 asm#"2", ".8h", ".4s", ".4s",
6117 def v2i64_v2i32 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6119 asm, ".2s", ".2d", ".2d",
6120 [(set (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
6121 def v2i64_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
6123 asm#"2", ".4s", ".2d", ".2d",
6127 // Patterns for the '2' variants involve INSERT_SUBREG, which you can't put in
6128 // a version attached to an instruction.
6129 def : Pat<(concat_vectors (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn),
6131 (!cast<Instruction>(NAME # "v8i16_v16i8")
6132 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6133 V128:$Rn, V128:$Rm)>;
6134 def : Pat<(concat_vectors (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn),
6136 (!cast<Instruction>(NAME # "v4i32_v8i16")
6137 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6138 V128:$Rn, V128:$Rm)>;
6139 def : Pat<(concat_vectors (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn),
6141 (!cast<Instruction>(NAME # "v2i64_v4i32")
6142 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6143 V128:$Rn, V128:$Rm)>;
6146 multiclass SIMDDifferentThreeVectorBD<bit U, bits<4> opc, string asm,
6148 def v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6150 asm, ".8h", ".8b", ".8b",
6151 [(set (v8i16 V128:$Rd), (IntOp (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
6152 def v16i8 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
6154 asm#"2", ".8h", ".16b", ".16b", []>;
6155 let Predicates = [HasAES] in {
6156 def v1i64 : BaseSIMDDifferentThreeVector<U, 0b110, opc,
6158 asm, ".1q", ".1d", ".1d", []>;
6159 def v2i64 : BaseSIMDDifferentThreeVector<U, 0b111, opc,
6161 asm#"2", ".1q", ".2d", ".2d", []>;
6164 def : Pat<(v8i16 (IntOp (v8i8 (extract_high_v16i8 V128:$Rn)),
6165 (v8i8 (extract_high_v16i8 V128:$Rm)))),
6166 (!cast<Instruction>(NAME#"v16i8") V128:$Rn, V128:$Rm)>;
6169 multiclass SIMDLongThreeVectorHS<bit U, bits<4> opc, string asm,
6170 SDPatternOperator OpNode> {
6171 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6173 asm, ".4s", ".4h", ".4h",
6174 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
6175 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
6177 asm#"2", ".4s", ".8h", ".8h",
6178 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
6179 (extract_high_v8i16 V128:$Rm)))]>;
6180 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6182 asm, ".2d", ".2s", ".2s",
6183 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
6184 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
6186 asm#"2", ".2d", ".4s", ".4s",
6187 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
6188 (extract_high_v4i32 V128:$Rm)))]>;
6191 multiclass SIMDLongThreeVectorBHSabdl<bit U, bits<4> opc, string asm,
6192 SDPatternOperator OpNode = null_frag> {
6193 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6195 asm, ".8h", ".8b", ".8b",
6196 [(set (v8i16 V128:$Rd),
6197 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))))]>;
6198 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
6200 asm#"2", ".8h", ".16b", ".16b",
6201 [(set (v8i16 V128:$Rd),
6202 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
6203 (extract_high_v16i8 V128:$Rm)))))]>;
6204 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6206 asm, ".4s", ".4h", ".4h",
6207 [(set (v4i32 V128:$Rd),
6208 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))))]>;
6209 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
6211 asm#"2", ".4s", ".8h", ".8h",
6212 [(set (v4i32 V128:$Rd),
6213 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
6214 (extract_high_v8i16 V128:$Rm)))))]>;
6215 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6217 asm, ".2d", ".2s", ".2s",
6218 [(set (v2i64 V128:$Rd),
6219 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))))]>;
6220 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
6222 asm#"2", ".2d", ".4s", ".4s",
6223 [(set (v2i64 V128:$Rd),
6224 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
6225 (extract_high_v4i32 V128:$Rm)))))]>;
6228 multiclass SIMDLongThreeVectorTiedBHSabal<bit U, bits<4> opc,
6230 SDPatternOperator OpNode> {
6231 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
6233 asm, ".8h", ".8b", ".8b",
6234 [(set (v8i16 V128:$dst),
6235 (add (v8i16 V128:$Rd),
6236 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))))]>;
6237 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
6239 asm#"2", ".8h", ".16b", ".16b",
6240 [(set (v8i16 V128:$dst),
6241 (add (v8i16 V128:$Rd),
6242 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
6243 (extract_high_v16i8 V128:$Rm))))))]>;
6244 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
6246 asm, ".4s", ".4h", ".4h",
6247 [(set (v4i32 V128:$dst),
6248 (add (v4i32 V128:$Rd),
6249 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))))]>;
6250 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
6252 asm#"2", ".4s", ".8h", ".8h",
6253 [(set (v4i32 V128:$dst),
6254 (add (v4i32 V128:$Rd),
6255 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
6256 (extract_high_v8i16 V128:$Rm))))))]>;
6257 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
6259 asm, ".2d", ".2s", ".2s",
6260 [(set (v2i64 V128:$dst),
6261 (add (v2i64 V128:$Rd),
6262 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))))]>;
6263 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
6265 asm#"2", ".2d", ".4s", ".4s",
6266 [(set (v2i64 V128:$dst),
6267 (add (v2i64 V128:$Rd),
6268 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
6269 (extract_high_v4i32 V128:$Rm))))))]>;
6272 multiclass SIMDLongThreeVectorBHS<bit U, bits<4> opc, string asm,
6273 SDPatternOperator OpNode = null_frag> {
6274 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6276 asm, ".8h", ".8b", ".8b",
6277 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
6278 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
6280 asm#"2", ".8h", ".16b", ".16b",
6281 [(set (v8i16 V128:$Rd), (OpNode (extract_high_v16i8 V128:$Rn),
6282 (extract_high_v16i8 V128:$Rm)))]>;
6283 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6285 asm, ".4s", ".4h", ".4h",
6286 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
6287 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
6289 asm#"2", ".4s", ".8h", ".8h",
6290 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
6291 (extract_high_v8i16 V128:$Rm)))]>;
6292 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6294 asm, ".2d", ".2s", ".2s",
6295 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
6296 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
6298 asm#"2", ".2d", ".4s", ".4s",
6299 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
6300 (extract_high_v4i32 V128:$Rm)))]>;
6303 multiclass SIMDLongThreeVectorTiedBHS<bit U, bits<4> opc,
6305 SDPatternOperator OpNode> {
6306 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
6308 asm, ".8h", ".8b", ".8b",
6309 [(set (v8i16 V128:$dst),
6310 (OpNode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
6311 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
6313 asm#"2", ".8h", ".16b", ".16b",
6314 [(set (v8i16 V128:$dst),
6315 (OpNode (v8i16 V128:$Rd),
6316 (extract_high_v16i8 V128:$Rn),
6317 (extract_high_v16i8 V128:$Rm)))]>;
6318 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
6320 asm, ".4s", ".4h", ".4h",
6321 [(set (v4i32 V128:$dst),
6322 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
6323 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
6325 asm#"2", ".4s", ".8h", ".8h",
6326 [(set (v4i32 V128:$dst),
6327 (OpNode (v4i32 V128:$Rd),
6328 (extract_high_v8i16 V128:$Rn),
6329 (extract_high_v8i16 V128:$Rm)))]>;
6330 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
6332 asm, ".2d", ".2s", ".2s",
6333 [(set (v2i64 V128:$dst),
6334 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
6335 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
6337 asm#"2", ".2d", ".4s", ".4s",
6338 [(set (v2i64 V128:$dst),
6339 (OpNode (v2i64 V128:$Rd),
6340 (extract_high_v4i32 V128:$Rn),
6341 (extract_high_v4i32 V128:$Rm)))]>;
6344 multiclass SIMDLongThreeVectorSQDMLXTiedHS<bit U, bits<4> opc, string asm,
6345 SDPatternOperator Accum> {
6346 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
6348 asm, ".4s", ".4h", ".4h",
6349 [(set (v4i32 V128:$dst),
6350 (Accum (v4i32 V128:$Rd),
6351 (v4i32 (int_aarch64_neon_sqdmull (v4i16 V64:$Rn),
6352 (v4i16 V64:$Rm)))))]>;
6353 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
6355 asm#"2", ".4s", ".8h", ".8h",
6356 [(set (v4i32 V128:$dst),
6357 (Accum (v4i32 V128:$Rd),
6358 (v4i32 (int_aarch64_neon_sqdmull (extract_high_v8i16 V128:$Rn),
6359 (extract_high_v8i16 V128:$Rm)))))]>;
6360 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
6362 asm, ".2d", ".2s", ".2s",
6363 [(set (v2i64 V128:$dst),
6364 (Accum (v2i64 V128:$Rd),
6365 (v2i64 (int_aarch64_neon_sqdmull (v2i32 V64:$Rn),
6366 (v2i32 V64:$Rm)))))]>;
6367 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
6369 asm#"2", ".2d", ".4s", ".4s",
6370 [(set (v2i64 V128:$dst),
6371 (Accum (v2i64 V128:$Rd),
6372 (v2i64 (int_aarch64_neon_sqdmull (extract_high_v4i32 V128:$Rn),
6373 (extract_high_v4i32 V128:$Rm)))))]>;
6376 multiclass SIMDWideThreeVectorBHS<bit U, bits<4> opc, string asm,
6377 SDPatternOperator OpNode> {
6378 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6380 asm, ".8h", ".8h", ".8b",
6381 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i8 V64:$Rm)))]>;
6382 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
6384 asm#"2", ".8h", ".8h", ".16b",
6385 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
6386 (extract_high_v16i8 V128:$Rm)))]>;
6387 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6389 asm, ".4s", ".4s", ".4h",
6390 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i16 V64:$Rm)))]>;
6391 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
6393 asm#"2", ".4s", ".4s", ".8h",
6394 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
6395 (extract_high_v8i16 V128:$Rm)))]>;
6396 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6398 asm, ".2d", ".2d", ".2s",
6399 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i32 V64:$Rm)))]>;
6400 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
6402 asm#"2", ".2d", ".2d", ".4s",
6403 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
6404 (extract_high_v4i32 V128:$Rm)))]>;
6407 //----------------------------------------------------------------------------
6408 // AdvSIMD bitwise extract from vector
6409 //----------------------------------------------------------------------------
6411 class BaseSIMDBitwiseExtract<bit size, RegisterOperand regtype, ValueType vty,
6412 string asm, string kind>
6413 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, i32imm:$imm), asm,
6414 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $imm" #
6415 "|" # kind # "\t$Rd, $Rn, $Rm, $imm}", "",
6416 [(set (vty regtype:$Rd),
6417 (AArch64ext regtype:$Rn, regtype:$Rm, (i32 imm:$imm)))]>,
6424 let Inst{30} = size;
6425 let Inst{29-21} = 0b101110000;
6426 let Inst{20-16} = Rm;
6428 let Inst{14-11} = imm;
6435 multiclass SIMDBitwiseExtract<string asm> {
6436 def v8i8 : BaseSIMDBitwiseExtract<0, V64, v8i8, asm, ".8b"> {
6439 def v16i8 : BaseSIMDBitwiseExtract<1, V128, v16i8, asm, ".16b">;
6442 //----------------------------------------------------------------------------
6443 // AdvSIMD zip vector
6444 //----------------------------------------------------------------------------
6446 class BaseSIMDZipVector<bits<3> size, bits<3> opc, RegisterOperand regtype,
6447 string asm, string kind, SDNode OpNode, ValueType valty>
6448 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
6449 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
6450 "|" # kind # "\t$Rd, $Rn, $Rm}", "",
6451 [(set (valty regtype:$Rd), (OpNode regtype:$Rn, regtype:$Rm))]>,
6457 let Inst{30} = size{0};
6458 let Inst{29-24} = 0b001110;
6459 let Inst{23-22} = size{2-1};
6461 let Inst{20-16} = Rm;
6463 let Inst{14-12} = opc;
6464 let Inst{11-10} = 0b10;
6469 multiclass SIMDZipVector<bits<3>opc, string asm,
6471 def v8i8 : BaseSIMDZipVector<0b000, opc, V64,
6472 asm, ".8b", OpNode, v8i8>;
6473 def v16i8 : BaseSIMDZipVector<0b001, opc, V128,
6474 asm, ".16b", OpNode, v16i8>;
6475 def v4i16 : BaseSIMDZipVector<0b010, opc, V64,
6476 asm, ".4h", OpNode, v4i16>;
6477 def v8i16 : BaseSIMDZipVector<0b011, opc, V128,
6478 asm, ".8h", OpNode, v8i16>;
6479 def v2i32 : BaseSIMDZipVector<0b100, opc, V64,
6480 asm, ".2s", OpNode, v2i32>;
6481 def v4i32 : BaseSIMDZipVector<0b101, opc, V128,
6482 asm, ".4s", OpNode, v4i32>;
6483 def v2i64 : BaseSIMDZipVector<0b111, opc, V128,
6484 asm, ".2d", OpNode, v2i64>;
6486 def : Pat<(v4f16 (OpNode V64:$Rn, V64:$Rm)),
6487 (!cast<Instruction>(NAME#"v4i16") V64:$Rn, V64:$Rm)>;
6488 def : Pat<(v8f16 (OpNode V128:$Rn, V128:$Rm)),
6489 (!cast<Instruction>(NAME#"v8i16") V128:$Rn, V128:$Rm)>;
6490 def : Pat<(v2f32 (OpNode V64:$Rn, V64:$Rm)),
6491 (!cast<Instruction>(NAME#"v2i32") V64:$Rn, V64:$Rm)>;
6492 def : Pat<(v4f32 (OpNode V128:$Rn, V128:$Rm)),
6493 (!cast<Instruction>(NAME#"v4i32") V128:$Rn, V128:$Rm)>;
6494 def : Pat<(v2f64 (OpNode V128:$Rn, V128:$Rm)),
6495 (!cast<Instruction>(NAME#"v2i64") V128:$Rn, V128:$Rm)>;
6498 //----------------------------------------------------------------------------
6499 // AdvSIMD three register scalar instructions
6500 //----------------------------------------------------------------------------
6502 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6503 class BaseSIMDThreeScalar<bit U, bits<3> size, bits<5> opcode,
6504 RegisterClass regtype, string asm,
6506 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
6507 "\t$Rd, $Rn, $Rm", "", pattern>,
6512 let Inst{31-30} = 0b01;
6514 let Inst{28-24} = 0b11110;
6515 let Inst{23-21} = size;
6516 let Inst{20-16} = Rm;
6517 let Inst{15-11} = opcode;
6523 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6524 class BaseSIMDThreeScalarTied<bit U, bits<2> size, bit R, bits<5> opcode,
6525 dag oops, dag iops, string asm,
6527 : I<oops, iops, asm, "\t$Rd, $Rn, $Rm", "$Rd = $dst", pattern>,
6532 let Inst{31-30} = 0b01;
6534 let Inst{28-24} = 0b11110;
6535 let Inst{23-22} = size;
6537 let Inst{20-16} = Rm;
6538 let Inst{15-11} = opcode;
6544 multiclass SIMDThreeScalarD<bit U, bits<5> opc, string asm,
6545 SDPatternOperator OpNode> {
6546 def v1i64 : BaseSIMDThreeScalar<U, 0b111, opc, FPR64, asm,
6547 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
6550 multiclass SIMDThreeScalarBHSD<bit U, bits<5> opc, string asm,
6551 SDPatternOperator OpNode> {
6552 def v1i64 : BaseSIMDThreeScalar<U, 0b111, opc, FPR64, asm,
6553 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
6554 def v1i32 : BaseSIMDThreeScalar<U, 0b101, opc, FPR32, asm, []>;
6555 def v1i16 : BaseSIMDThreeScalar<U, 0b011, opc, FPR16, asm, []>;
6556 def v1i8 : BaseSIMDThreeScalar<U, 0b001, opc, FPR8 , asm, []>;
6558 def : Pat<(i64 (OpNode (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6559 (!cast<Instruction>(NAME#"v1i64") FPR64:$Rn, FPR64:$Rm)>;
6560 def : Pat<(i32 (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm))),
6561 (!cast<Instruction>(NAME#"v1i32") FPR32:$Rn, FPR32:$Rm)>;
6564 multiclass SIMDThreeScalarHS<bit U, bits<5> opc, string asm,
6565 SDPatternOperator OpNode> {
6566 def v1i32 : BaseSIMDThreeScalar<U, 0b101, opc, FPR32, asm,
6567 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
6568 def v1i16 : BaseSIMDThreeScalar<U, 0b011, opc, FPR16, asm, []>;
6571 multiclass SIMDThreeScalarHSTied<bit U, bit R, bits<5> opc, string asm,
6572 SDPatternOperator OpNode = null_frag> {
6573 def v1i32: BaseSIMDThreeScalarTied<U, 0b10, R, opc, (outs FPR32:$dst),
6574 (ins FPR32:$Rd, FPR32:$Rn, FPR32:$Rm),
6576 def v1i16: BaseSIMDThreeScalarTied<U, 0b01, R, opc, (outs FPR16:$dst),
6577 (ins FPR16:$Rd, FPR16:$Rn, FPR16:$Rm),
6581 multiclass SIMDFPThreeScalar<bit U, bit S, bits<3> opc, string asm,
6582 SDPatternOperator OpNode = null_frag> {
6583 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6584 def #NAME#64 : BaseSIMDThreeScalar<U, {S,0b11}, {0b11,opc}, FPR64, asm,
6585 [(set (f64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
6586 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0b01}, {0b11,opc}, FPR32, asm,
6587 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
6588 let Predicates = [HasNEON, HasFullFP16] in {
6589 def #NAME#16 : BaseSIMDThreeScalar<U, {S,0b10}, {0b00,opc}, FPR16, asm,
6590 [(set FPR16:$Rd, (OpNode FPR16:$Rn, FPR16:$Rm))]>;
6591 } // Predicates = [HasNEON, HasFullFP16]
6594 def : Pat<(v1f64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
6595 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
6598 multiclass SIMDThreeScalarFPCmp<bit U, bit S, bits<3> opc, string asm,
6599 SDPatternOperator OpNode = null_frag> {
6600 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6601 def #NAME#64 : BaseSIMDThreeScalar<U, {S,0b11}, {0b11,opc}, FPR64, asm,
6602 [(set (i64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
6603 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0b01}, {0b11,opc}, FPR32, asm,
6604 [(set (i32 FPR32:$Rd), (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]>;
6605 let Predicates = [HasNEON, HasFullFP16] in {
6606 def #NAME#16 : BaseSIMDThreeScalar<U, {S,0b10}, {0b00,opc}, FPR16, asm,
6608 } // Predicates = [HasNEON, HasFullFP16]
6611 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
6612 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
6615 class BaseSIMDThreeScalarMixed<bit U, bits<2> size, bits<5> opcode,
6616 dag oops, dag iops, string asm, string cstr, list<dag> pat>
6617 : I<oops, iops, asm,
6618 "\t$Rd, $Rn, $Rm", cstr, pat>,
6623 let Inst{31-30} = 0b01;
6625 let Inst{28-24} = 0b11110;
6626 let Inst{23-22} = size;
6628 let Inst{20-16} = Rm;
6629 let Inst{15-11} = opcode;
6635 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6636 multiclass SIMDThreeScalarMixedHS<bit U, bits<5> opc, string asm,
6637 SDPatternOperator OpNode = null_frag> {
6638 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
6640 (ins FPR16:$Rn, FPR16:$Rm), asm, "", []>;
6641 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
6643 (ins FPR32:$Rn, FPR32:$Rm), asm, "",
6644 [(set (i64 FPR64:$Rd), (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
6647 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6648 multiclass SIMDThreeScalarMixedTiedHS<bit U, bits<5> opc, string asm,
6649 SDPatternOperator OpNode = null_frag> {
6650 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
6652 (ins FPR32:$Rd, FPR16:$Rn, FPR16:$Rm),
6653 asm, "$Rd = $dst", []>;
6654 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
6656 (ins FPR64:$Rd, FPR32:$Rn, FPR32:$Rm),
6658 [(set (i64 FPR64:$dst),
6659 (OpNode (i64 FPR64:$Rd), (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
6662 //----------------------------------------------------------------------------
6663 // AdvSIMD two register scalar instructions
6664 //----------------------------------------------------------------------------
6666 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6667 class BaseSIMDTwoScalar<bit U, bits<2> size, bits<2> size2, bits<5> opcode,
6668 RegisterClass regtype, RegisterClass regtype2,
6669 string asm, list<dag> pat>
6670 : I<(outs regtype:$Rd), (ins regtype2:$Rn), asm,
6671 "\t$Rd, $Rn", "", pat>,
6675 let Inst{31-30} = 0b01;
6677 let Inst{28-24} = 0b11110;
6678 let Inst{23-22} = size;
6680 let Inst{20-19} = size2;
6681 let Inst{18-17} = 0b00;
6682 let Inst{16-12} = opcode;
6683 let Inst{11-10} = 0b10;
6688 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6689 class BaseSIMDTwoScalarTied<bit U, bits<2> size, bits<5> opcode,
6690 RegisterClass regtype, RegisterClass regtype2,
6691 string asm, list<dag> pat>
6692 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype2:$Rn), asm,
6693 "\t$Rd, $Rn", "$Rd = $dst", pat>,
6697 let Inst{31-30} = 0b01;
6699 let Inst{28-24} = 0b11110;
6700 let Inst{23-22} = size;
6701 let Inst{21-17} = 0b10000;
6702 let Inst{16-12} = opcode;
6703 let Inst{11-10} = 0b10;
6709 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6710 class BaseSIMDCmpTwoScalar<bit U, bits<2> size, bits<2> size2, bits<5> opcode,
6711 RegisterClass regtype, string asm, string zero>
6712 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
6713 "\t$Rd, $Rn, #" # zero, "", []>,
6717 let Inst{31-30} = 0b01;
6719 let Inst{28-24} = 0b11110;
6720 let Inst{23-22} = size;
6722 let Inst{20-19} = size2;
6723 let Inst{18-17} = 0b00;
6724 let Inst{16-12} = opcode;
6725 let Inst{11-10} = 0b10;
6730 class SIMDInexactCvtTwoScalar<bits<5> opcode, string asm>
6731 : I<(outs FPR32:$Rd), (ins FPR64:$Rn), asm, "\t$Rd, $Rn", "",
6732 [(set (f32 FPR32:$Rd), (int_aarch64_sisd_fcvtxn (f64 FPR64:$Rn)))]>,
6736 let Inst{31-17} = 0b011111100110000;
6737 let Inst{16-12} = opcode;
6738 let Inst{11-10} = 0b10;
6743 multiclass SIMDCmpTwoScalarD<bit U, bits<5> opc, string asm,
6744 SDPatternOperator OpNode> {
6745 def v1i64rz : BaseSIMDCmpTwoScalar<U, 0b11, 0b00, opc, FPR64, asm, "0">;
6747 def : Pat<(v1i64 (OpNode FPR64:$Rn)),
6748 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
6751 multiclass SIMDFPCmpTwoScalar<bit U, bit S, bits<5> opc, string asm,
6752 SDPatternOperator OpNode> {
6753 def v1i64rz : BaseSIMDCmpTwoScalar<U, {S,1}, 0b00, opc, FPR64, asm, "0.0">;
6754 def v1i32rz : BaseSIMDCmpTwoScalar<U, {S,0}, 0b00, opc, FPR32, asm, "0.0">;
6755 let Predicates = [HasNEON, HasFullFP16] in {
6756 def v1i16rz : BaseSIMDCmpTwoScalar<U, {S,1}, 0b11, opc, FPR16, asm, "0.0">;
6759 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6760 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rd, FPR64:$Rn), 0>;
6761 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6762 (!cast<Instruction>(NAME # v1i32rz) FPR32:$Rd, FPR32:$Rn), 0>;
6763 let Predicates = [HasNEON, HasFullFP16] in {
6764 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6765 (!cast<Instruction>(NAME # v1i16rz) FPR16:$Rd, FPR16:$Rn), 0>;
6768 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn))),
6769 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
6772 multiclass SIMDTwoScalarD<bit U, bits<5> opc, string asm,
6773 SDPatternOperator OpNode = null_frag> {
6774 def v1i64 : BaseSIMDTwoScalar<U, 0b11, 0b00, opc, FPR64, FPR64, asm,
6775 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn)))]>;
6777 def : Pat<(i64 (OpNode (i64 FPR64:$Rn))),
6778 (!cast<Instruction>(NAME # "v1i64") FPR64:$Rn)>;
6781 multiclass SIMDFPTwoScalar<bit U, bit S, bits<5> opc, string asm> {
6782 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, 0b00, opc, FPR64, FPR64, asm,[]>;
6783 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, 0b00, opc, FPR32, FPR32, asm,[]>;
6784 let Predicates = [HasNEON, HasFullFP16] in {
6785 def v1f16 : BaseSIMDTwoScalar<U, {S,1}, 0b11, opc, FPR16, FPR16, asm,[]>;
6789 multiclass SIMDFPTwoScalarCVT<bit U, bit S, bits<5> opc, string asm,
6790 SDPatternOperator OpNode> {
6791 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, 0b00, opc, FPR64, FPR64, asm,
6792 [(set FPR64:$Rd, (OpNode (f64 FPR64:$Rn)))]>;
6793 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, 0b00, opc, FPR32, FPR32, asm,
6794 [(set FPR32:$Rd, (OpNode (f32 FPR32:$Rn)))]>;
6795 let Predicates = [HasNEON, HasFullFP16] in {
6796 def v1i16 : BaseSIMDTwoScalar<U, {S,1}, 0b11, opc, FPR16, FPR16, asm,
6797 [(set FPR16:$Rd, (OpNode (f16 FPR16:$Rn)))]>;
6801 multiclass SIMDTwoScalarBHSD<bit U, bits<5> opc, string asm,
6802 SDPatternOperator OpNode = null_frag> {
6803 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6804 def v1i64 : BaseSIMDTwoScalar<U, 0b11, 0b00, opc, FPR64, FPR64, asm,
6805 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
6806 def v1i32 : BaseSIMDTwoScalar<U, 0b10, 0b00, opc, FPR32, FPR32, asm,
6807 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
6808 def v1i16 : BaseSIMDTwoScalar<U, 0b01, 0b00, opc, FPR16, FPR16, asm, []>;
6809 def v1i8 : BaseSIMDTwoScalar<U, 0b00, 0b00, opc, FPR8 , FPR8 , asm, []>;
6812 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn))),
6813 (!cast<Instruction>(NAME # v1i64) FPR64:$Rn)>;
6816 multiclass SIMDTwoScalarBHSDTied<bit U, bits<5> opc, string asm,
6818 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6819 def v1i64 : BaseSIMDTwoScalarTied<U, 0b11, opc, FPR64, FPR64, asm,
6820 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn)))]>;
6821 def v1i32 : BaseSIMDTwoScalarTied<U, 0b10, opc, FPR32, FPR32, asm,
6822 [(set (i32 FPR32:$dst), (OpNode (i32 FPR32:$Rd), (i32 FPR32:$Rn)))]>;
6823 def v1i16 : BaseSIMDTwoScalarTied<U, 0b01, opc, FPR16, FPR16, asm, []>;
6824 def v1i8 : BaseSIMDTwoScalarTied<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
6827 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn))),
6828 (!cast<Instruction>(NAME # v1i64) FPR64:$Rd, FPR64:$Rn)>;
6833 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6834 multiclass SIMDTwoScalarMixedBHS<bit U, bits<5> opc, string asm,
6835 SDPatternOperator OpNode = null_frag> {
6836 def v1i32 : BaseSIMDTwoScalar<U, 0b10, 0b00, opc, FPR32, FPR64, asm,
6837 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
6838 def v1i16 : BaseSIMDTwoScalar<U, 0b01, 0b00, opc, FPR16, FPR32, asm, []>;
6839 def v1i8 : BaseSIMDTwoScalar<U, 0b00, 0b00, opc, FPR8 , FPR16, asm, []>;
6842 //----------------------------------------------------------------------------
6843 // AdvSIMD scalar pairwise instructions
6844 //----------------------------------------------------------------------------
6846 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6847 class BaseSIMDPairwiseScalar<bit U, bits<2> size, bits<5> opcode,
6848 RegisterOperand regtype, RegisterOperand vectype,
6849 string asm, string kind>
6850 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
6851 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", []>,
6855 let Inst{31-30} = 0b01;
6857 let Inst{28-24} = 0b11110;
6858 let Inst{23-22} = size;
6859 let Inst{21-17} = 0b11000;
6860 let Inst{16-12} = opcode;
6861 let Inst{11-10} = 0b10;
6866 multiclass SIMDPairwiseScalarD<bit U, bits<5> opc, string asm> {
6867 def v2i64p : BaseSIMDPairwiseScalar<U, 0b11, opc, FPR64Op, V128,
6871 multiclass SIMDFPPairwiseScalar<bit S, bits<5> opc, string asm> {
6872 let Predicates = [HasNEON, HasFullFP16] in {
6873 def v2i16p : BaseSIMDPairwiseScalar<0, {S,0}, opc, FPR16Op, V64,
6876 def v2i32p : BaseSIMDPairwiseScalar<1, {S,0}, opc, FPR32Op, V64,
6878 def v2i64p : BaseSIMDPairwiseScalar<1, {S,1}, opc, FPR64Op, V128,
6882 //----------------------------------------------------------------------------
6883 // AdvSIMD across lanes instructions
6884 //----------------------------------------------------------------------------
6886 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6887 class BaseSIMDAcrossLanes<bit Q, bit U, bits<2> size, bits<5> opcode,
6888 RegisterClass regtype, RegisterOperand vectype,
6889 string asm, string kind, list<dag> pattern>
6890 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
6891 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", pattern>,
6898 let Inst{28-24} = 0b01110;
6899 let Inst{23-22} = size;
6900 let Inst{21-17} = 0b11000;
6901 let Inst{16-12} = opcode;
6902 let Inst{11-10} = 0b10;
6907 multiclass SIMDAcrossLanesBHS<bit U, bits<5> opcode,
6909 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR8, V64,
6911 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR8, V128,
6913 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR16, V64,
6915 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR16, V128,
6917 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR32, V128,
6921 multiclass SIMDAcrossLanesHSD<bit U, bits<5> opcode, string asm> {
6922 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR16, V64,
6924 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR16, V128,
6926 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR32, V64,
6928 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR32, V128,
6930 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR64, V128,
6934 multiclass SIMDFPAcrossLanes<bits<5> opcode, bit sz1, string asm,
6936 let Predicates = [HasNEON, HasFullFP16] in {
6937 def v4i16v : BaseSIMDAcrossLanes<0, 0, {sz1, 0}, opcode, FPR16, V64,
6939 [(set FPR16:$Rd, (intOp (v4f16 V64:$Rn)))]>;
6940 def v8i16v : BaseSIMDAcrossLanes<1, 0, {sz1, 0}, opcode, FPR16, V128,
6942 [(set FPR16:$Rd, (intOp (v8f16 V128:$Rn)))]>;
6943 } // Predicates = [HasNEON, HasFullFP16]
6944 def v4i32v : BaseSIMDAcrossLanes<1, 1, {sz1, 0}, opcode, FPR32, V128,
6946 [(set FPR32:$Rd, (intOp (v4f32 V128:$Rn)))]>;
6949 //----------------------------------------------------------------------------
6950 // AdvSIMD INS/DUP instructions
6951 //----------------------------------------------------------------------------
6953 // FIXME: There has got to be a better way to factor these. ugh.
6955 class BaseSIMDInsDup<bit Q, bit op, dag outs, dag ins, string asm,
6956 string operands, string constraints, list<dag> pattern>
6957 : I<outs, ins, asm, operands, constraints, pattern>,
6964 let Inst{28-21} = 0b01110000;
6971 class SIMDDupFromMain<bit Q, bits<5> imm5, string size, ValueType vectype,
6972 RegisterOperand vecreg, RegisterClass regtype>
6973 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins regtype:$Rn), "dup",
6974 "{\t$Rd" # size # ", $Rn" #
6975 "|" # size # "\t$Rd, $Rn}", "",
6976 [(set (vectype vecreg:$Rd), (AArch64dup regtype:$Rn))]> {
6977 let Inst{20-16} = imm5;
6978 let Inst{14-11} = 0b0001;
6981 class SIMDDupFromElement<bit Q, string dstkind, string srckind,
6982 ValueType vectype, ValueType insreg,
6983 RegisterOperand vecreg, Operand idxtype,
6984 ValueType elttype, SDNode OpNode>
6985 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins V128:$Rn, idxtype:$idx), "dup",
6986 "{\t$Rd" # dstkind # ", $Rn" # srckind # "$idx" #
6987 "|" # dstkind # "\t$Rd, $Rn$idx}", "",
6988 [(set (vectype vecreg:$Rd),
6989 (OpNode (insreg V128:$Rn), idxtype:$idx))]> {
6990 let Inst{14-11} = 0b0000;
6993 class SIMDDup64FromElement
6994 : SIMDDupFromElement<1, ".2d", ".d", v2i64, v2i64, V128,
6995 VectorIndexD, i64, AArch64duplane64> {
6998 let Inst{19-16} = 0b1000;
7001 class SIMDDup32FromElement<bit Q, string size, ValueType vectype,
7002 RegisterOperand vecreg>
7003 : SIMDDupFromElement<Q, size, ".s", vectype, v4i32, vecreg,
7004 VectorIndexS, i64, AArch64duplane32> {
7006 let Inst{20-19} = idx;
7007 let Inst{18-16} = 0b100;
7010 class SIMDDup16FromElement<bit Q, string size, ValueType vectype,
7011 RegisterOperand vecreg>
7012 : SIMDDupFromElement<Q, size, ".h", vectype, v8i16, vecreg,
7013 VectorIndexH, i64, AArch64duplane16> {
7015 let Inst{20-18} = idx;
7016 let Inst{17-16} = 0b10;
7019 class SIMDDup8FromElement<bit Q, string size, ValueType vectype,
7020 RegisterOperand vecreg>
7021 : SIMDDupFromElement<Q, size, ".b", vectype, v16i8, vecreg,
7022 VectorIndexB, i64, AArch64duplane8> {
7024 let Inst{20-17} = idx;
7028 class BaseSIMDMov<bit Q, string size, bits<4> imm4, RegisterClass regtype,
7029 Operand idxtype, string asm, list<dag> pattern>
7030 : BaseSIMDInsDup<Q, 0, (outs regtype:$Rd), (ins V128:$Rn, idxtype:$idx), asm,
7031 "{\t$Rd, $Rn" # size # "$idx" #
7032 "|" # size # "\t$Rd, $Rn$idx}", "", pattern> {
7033 let Inst{14-11} = imm4;
7036 class SIMDSMov<bit Q, string size, RegisterClass regtype,
7038 : BaseSIMDMov<Q, size, 0b0101, regtype, idxtype, "smov", []>;
7039 class SIMDUMov<bit Q, string size, ValueType vectype, RegisterClass regtype,
7041 : BaseSIMDMov<Q, size, 0b0111, regtype, idxtype, "umov",
7042 [(set regtype:$Rd, (vector_extract (vectype V128:$Rn), idxtype:$idx))]>;
7044 class SIMDMovAlias<string asm, string size, Instruction inst,
7045 RegisterClass regtype, Operand idxtype>
7046 : InstAlias<asm#"{\t$dst, $src"#size#"$idx" #
7047 "|" # size # "\t$dst, $src$idx}",
7048 (inst regtype:$dst, V128:$src, idxtype:$idx)>;
7051 def vi8to32 : SIMDSMov<0, ".b", GPR32, VectorIndexB> {
7053 let Inst{20-17} = idx;
7056 def vi8to64 : SIMDSMov<1, ".b", GPR64, VectorIndexB> {
7058 let Inst{20-17} = idx;
7061 def vi16to32 : SIMDSMov<0, ".h", GPR32, VectorIndexH> {
7063 let Inst{20-18} = idx;
7064 let Inst{17-16} = 0b10;
7066 def vi16to64 : SIMDSMov<1, ".h", GPR64, VectorIndexH> {
7068 let Inst{20-18} = idx;
7069 let Inst{17-16} = 0b10;
7071 def vi32to64 : SIMDSMov<1, ".s", GPR64, VectorIndexS> {
7073 let Inst{20-19} = idx;
7074 let Inst{18-16} = 0b100;
7079 def vi8 : SIMDUMov<0, ".b", v16i8, GPR32, VectorIndexB> {
7081 let Inst{20-17} = idx;
7084 def vi16 : SIMDUMov<0, ".h", v8i16, GPR32, VectorIndexH> {
7086 let Inst{20-18} = idx;
7087 let Inst{17-16} = 0b10;
7089 def vi32 : SIMDUMov<0, ".s", v4i32, GPR32, VectorIndexS> {
7091 let Inst{20-19} = idx;
7092 let Inst{18-16} = 0b100;
7094 def vi64 : SIMDUMov<1, ".d", v2i64, GPR64, VectorIndexD> {
7097 let Inst{19-16} = 0b1000;
7099 def : SIMDMovAlias<"mov", ".s",
7100 !cast<Instruction>(NAME#"vi32"),
7101 GPR32, VectorIndexS>;
7102 def : SIMDMovAlias<"mov", ".d",
7103 !cast<Instruction>(NAME#"vi64"),
7104 GPR64, VectorIndexD>;
7107 class SIMDInsFromMain<string size, ValueType vectype,
7108 RegisterClass regtype, Operand idxtype>
7109 : BaseSIMDInsDup<1, 0, (outs V128:$dst),
7110 (ins V128:$Rd, idxtype:$idx, regtype:$Rn), "ins",
7111 "{\t$Rd" # size # "$idx, $Rn" #
7112 "|" # size # "\t$Rd$idx, $Rn}",
7115 (vector_insert (vectype V128:$Rd), regtype:$Rn, idxtype:$idx))]> {
7116 let Inst{14-11} = 0b0011;
7119 class SIMDInsFromElement<string size, ValueType vectype,
7120 ValueType elttype, Operand idxtype>
7121 : BaseSIMDInsDup<1, 1, (outs V128:$dst),
7122 (ins V128:$Rd, idxtype:$idx, V128:$Rn, idxtype:$idx2), "ins",
7123 "{\t$Rd" # size # "$idx, $Rn" # size # "$idx2" #
7124 "|" # size # "\t$Rd$idx, $Rn$idx2}",
7129 (elttype (vector_extract (vectype V128:$Rn), idxtype:$idx2)),
7132 class SIMDInsMainMovAlias<string size, Instruction inst,
7133 RegisterClass regtype, Operand idxtype>
7134 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" #
7135 "|" # size #"\t$dst$idx, $src}",
7136 (inst V128:$dst, idxtype:$idx, regtype:$src)>;
7137 class SIMDInsElementMovAlias<string size, Instruction inst,
7139 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" # size # "$idx2" #
7140 # "|" # size #"\t$dst$idx, $src$idx2}",
7141 (inst V128:$dst, idxtype:$idx, V128:$src, idxtype:$idx2)>;
7144 multiclass SIMDIns {
7145 def vi8gpr : SIMDInsFromMain<".b", v16i8, GPR32, VectorIndexB> {
7147 let Inst{20-17} = idx;
7150 def vi16gpr : SIMDInsFromMain<".h", v8i16, GPR32, VectorIndexH> {
7152 let Inst{20-18} = idx;
7153 let Inst{17-16} = 0b10;
7155 def vi32gpr : SIMDInsFromMain<".s", v4i32, GPR32, VectorIndexS> {
7157 let Inst{20-19} = idx;
7158 let Inst{18-16} = 0b100;
7160 def vi64gpr : SIMDInsFromMain<".d", v2i64, GPR64, VectorIndexD> {
7163 let Inst{19-16} = 0b1000;
7166 def vi8lane : SIMDInsFromElement<".b", v16i8, i32, VectorIndexB> {
7169 let Inst{20-17} = idx;
7171 let Inst{14-11} = idx2;
7173 def vi16lane : SIMDInsFromElement<".h", v8i16, i32, VectorIndexH> {
7176 let Inst{20-18} = idx;
7177 let Inst{17-16} = 0b10;
7178 let Inst{14-12} = idx2;
7181 def vi32lane : SIMDInsFromElement<".s", v4i32, i32, VectorIndexS> {
7184 let Inst{20-19} = idx;
7185 let Inst{18-16} = 0b100;
7186 let Inst{14-13} = idx2;
7187 let Inst{12-11} = {?,?};
7189 def vi64lane : SIMDInsFromElement<".d", v2i64, i64, VectorIndexD> {
7193 let Inst{19-16} = 0b1000;
7194 let Inst{14} = idx2;
7195 let Inst{13-11} = {?,?,?};
7198 // For all forms of the INS instruction, the "mov" mnemonic is the
7199 // preferred alias. Why they didn't just call the instruction "mov" in
7200 // the first place is a very good question indeed...
7201 def : SIMDInsMainMovAlias<".b", !cast<Instruction>(NAME#"vi8gpr"),
7202 GPR32, VectorIndexB>;
7203 def : SIMDInsMainMovAlias<".h", !cast<Instruction>(NAME#"vi16gpr"),
7204 GPR32, VectorIndexH>;
7205 def : SIMDInsMainMovAlias<".s", !cast<Instruction>(NAME#"vi32gpr"),
7206 GPR32, VectorIndexS>;
7207 def : SIMDInsMainMovAlias<".d", !cast<Instruction>(NAME#"vi64gpr"),
7208 GPR64, VectorIndexD>;
7210 def : SIMDInsElementMovAlias<".b", !cast<Instruction>(NAME#"vi8lane"),
7212 def : SIMDInsElementMovAlias<".h", !cast<Instruction>(NAME#"vi16lane"),
7214 def : SIMDInsElementMovAlias<".s", !cast<Instruction>(NAME#"vi32lane"),
7216 def : SIMDInsElementMovAlias<".d", !cast<Instruction>(NAME#"vi64lane"),
7220 //----------------------------------------------------------------------------
7222 //----------------------------------------------------------------------------
7224 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
7225 class BaseSIMDTableLookup<bit Q, bits<2> len, bit op, RegisterOperand vectype,
7226 RegisterOperand listtype, string asm, string kind>
7227 : I<(outs vectype:$Vd), (ins listtype:$Vn, vectype:$Vm), asm,
7228 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "", []>,
7235 let Inst{29-21} = 0b001110000;
7236 let Inst{20-16} = Vm;
7238 let Inst{14-13} = len;
7240 let Inst{11-10} = 0b00;
7245 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
7246 class BaseSIMDTableLookupTied<bit Q, bits<2> len, bit op, RegisterOperand vectype,
7247 RegisterOperand listtype, string asm, string kind>
7248 : I<(outs vectype:$dst), (ins vectype:$Vd, listtype:$Vn, vectype:$Vm), asm,
7249 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "$Vd = $dst", []>,
7256 let Inst{29-21} = 0b001110000;
7257 let Inst{20-16} = Vm;
7259 let Inst{14-13} = len;
7261 let Inst{11-10} = 0b00;
7266 class SIMDTableLookupAlias<string asm, Instruction inst,
7267 RegisterOperand vectype, RegisterOperand listtype>
7268 : InstAlias<!strconcat(asm, "\t$dst, $lst, $index"),
7269 (inst vectype:$dst, listtype:$lst, vectype:$index), 0>;
7271 multiclass SIMDTableLookup<bit op, string asm> {
7272 def v8i8One : BaseSIMDTableLookup<0, 0b00, op, V64, VecListOne16b,
7274 def v8i8Two : BaseSIMDTableLookup<0, 0b01, op, V64, VecListTwo16b,
7276 def v8i8Three : BaseSIMDTableLookup<0, 0b10, op, V64, VecListThree16b,
7278 def v8i8Four : BaseSIMDTableLookup<0, 0b11, op, V64, VecListFour16b,
7280 def v16i8One : BaseSIMDTableLookup<1, 0b00, op, V128, VecListOne16b,
7282 def v16i8Two : BaseSIMDTableLookup<1, 0b01, op, V128, VecListTwo16b,
7284 def v16i8Three: BaseSIMDTableLookup<1, 0b10, op, V128, VecListThree16b,
7286 def v16i8Four : BaseSIMDTableLookup<1, 0b11, op, V128, VecListFour16b,
7289 def : SIMDTableLookupAlias<asm # ".8b",
7290 !cast<Instruction>(NAME#"v8i8One"),
7291 V64, VecListOne128>;
7292 def : SIMDTableLookupAlias<asm # ".8b",
7293 !cast<Instruction>(NAME#"v8i8Two"),
7294 V64, VecListTwo128>;
7295 def : SIMDTableLookupAlias<asm # ".8b",
7296 !cast<Instruction>(NAME#"v8i8Three"),
7297 V64, VecListThree128>;
7298 def : SIMDTableLookupAlias<asm # ".8b",
7299 !cast<Instruction>(NAME#"v8i8Four"),
7300 V64, VecListFour128>;
7301 def : SIMDTableLookupAlias<asm # ".16b",
7302 !cast<Instruction>(NAME#"v16i8One"),
7303 V128, VecListOne128>;
7304 def : SIMDTableLookupAlias<asm # ".16b",
7305 !cast<Instruction>(NAME#"v16i8Two"),
7306 V128, VecListTwo128>;
7307 def : SIMDTableLookupAlias<asm # ".16b",
7308 !cast<Instruction>(NAME#"v16i8Three"),
7309 V128, VecListThree128>;
7310 def : SIMDTableLookupAlias<asm # ".16b",
7311 !cast<Instruction>(NAME#"v16i8Four"),
7312 V128, VecListFour128>;
7315 multiclass SIMDTableLookupTied<bit op, string asm> {
7316 def v8i8One : BaseSIMDTableLookupTied<0, 0b00, op, V64, VecListOne16b,
7318 def v8i8Two : BaseSIMDTableLookupTied<0, 0b01, op, V64, VecListTwo16b,
7320 def v8i8Three : BaseSIMDTableLookupTied<0, 0b10, op, V64, VecListThree16b,
7322 def v8i8Four : BaseSIMDTableLookupTied<0, 0b11, op, V64, VecListFour16b,
7324 def v16i8One : BaseSIMDTableLookupTied<1, 0b00, op, V128, VecListOne16b,
7326 def v16i8Two : BaseSIMDTableLookupTied<1, 0b01, op, V128, VecListTwo16b,
7328 def v16i8Three: BaseSIMDTableLookupTied<1, 0b10, op, V128, VecListThree16b,
7330 def v16i8Four : BaseSIMDTableLookupTied<1, 0b11, op, V128, VecListFour16b,
7333 def : SIMDTableLookupAlias<asm # ".8b",
7334 !cast<Instruction>(NAME#"v8i8One"),
7335 V64, VecListOne128>;
7336 def : SIMDTableLookupAlias<asm # ".8b",
7337 !cast<Instruction>(NAME#"v8i8Two"),
7338 V64, VecListTwo128>;
7339 def : SIMDTableLookupAlias<asm # ".8b",
7340 !cast<Instruction>(NAME#"v8i8Three"),
7341 V64, VecListThree128>;
7342 def : SIMDTableLookupAlias<asm # ".8b",
7343 !cast<Instruction>(NAME#"v8i8Four"),
7344 V64, VecListFour128>;
7345 def : SIMDTableLookupAlias<asm # ".16b",
7346 !cast<Instruction>(NAME#"v16i8One"),
7347 V128, VecListOne128>;
7348 def : SIMDTableLookupAlias<asm # ".16b",
7349 !cast<Instruction>(NAME#"v16i8Two"),
7350 V128, VecListTwo128>;
7351 def : SIMDTableLookupAlias<asm # ".16b",
7352 !cast<Instruction>(NAME#"v16i8Three"),
7353 V128, VecListThree128>;
7354 def : SIMDTableLookupAlias<asm # ".16b",
7355 !cast<Instruction>(NAME#"v16i8Four"),
7356 V128, VecListFour128>;
7360 //----------------------------------------------------------------------------
7361 // AdvSIMD scalar CPY
7362 //----------------------------------------------------------------------------
7363 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7364 class BaseSIMDScalarCPY<RegisterClass regtype, RegisterOperand vectype,
7365 string kind, Operand idxtype>
7366 : I<(outs regtype:$dst), (ins vectype:$src, idxtype:$idx), "mov",
7367 "{\t$dst, $src" # kind # "$idx" #
7368 "|\t$dst, $src$idx}", "", []>,
7372 let Inst{31-21} = 0b01011110000;
7373 let Inst{15-10} = 0b000001;
7374 let Inst{9-5} = src;
7375 let Inst{4-0} = dst;
7378 class SIMDScalarCPYAlias<string asm, string size, Instruction inst,
7379 RegisterClass regtype, RegisterOperand vectype, Operand idxtype>
7380 : InstAlias<asm # "{\t$dst, $src" # size # "$index" #
7381 # "|\t$dst, $src$index}",
7382 (inst regtype:$dst, vectype:$src, idxtype:$index), 0>;
7385 multiclass SIMDScalarCPY<string asm> {
7386 def i8 : BaseSIMDScalarCPY<FPR8, V128, ".b", VectorIndexB> {
7388 let Inst{20-17} = idx;
7391 def i16 : BaseSIMDScalarCPY<FPR16, V128, ".h", VectorIndexH> {
7393 let Inst{20-18} = idx;
7394 let Inst{17-16} = 0b10;
7396 def i32 : BaseSIMDScalarCPY<FPR32, V128, ".s", VectorIndexS> {
7398 let Inst{20-19} = idx;
7399 let Inst{18-16} = 0b100;
7401 def i64 : BaseSIMDScalarCPY<FPR64, V128, ".d", VectorIndexD> {
7404 let Inst{19-16} = 0b1000;
7407 def : Pat<(v1i64 (scalar_to_vector (i64 (vector_extract (v2i64 V128:$src),
7408 VectorIndexD:$idx)))),
7409 (!cast<Instruction>(NAME # i64) V128:$src, VectorIndexD:$idx)>;
7411 // 'DUP' mnemonic aliases.
7412 def : SIMDScalarCPYAlias<"dup", ".b",
7413 !cast<Instruction>(NAME#"i8"),
7414 FPR8, V128, VectorIndexB>;
7415 def : SIMDScalarCPYAlias<"dup", ".h",
7416 !cast<Instruction>(NAME#"i16"),
7417 FPR16, V128, VectorIndexH>;
7418 def : SIMDScalarCPYAlias<"dup", ".s",
7419 !cast<Instruction>(NAME#"i32"),
7420 FPR32, V128, VectorIndexS>;
7421 def : SIMDScalarCPYAlias<"dup", ".d",
7422 !cast<Instruction>(NAME#"i64"),
7423 FPR64, V128, VectorIndexD>;
7426 //----------------------------------------------------------------------------
7427 // AdvSIMD modified immediate instructions
7428 //----------------------------------------------------------------------------
7430 class BaseSIMDModifiedImm<bit Q, bit op, bit op2, dag oops, dag iops,
7431 string asm, string op_string,
7432 string cstr, list<dag> pattern>
7433 : I<oops, iops, asm, op_string, cstr, pattern>,
7440 let Inst{28-19} = 0b0111100000;
7441 let Inst{18-16} = imm8{7-5};
7444 let Inst{9-5} = imm8{4-0};
7448 class BaseSIMDModifiedImmVector<bit Q, bit op, bit op2, RegisterOperand vectype,
7449 Operand immtype, dag opt_shift_iop,
7450 string opt_shift, string asm, string kind,
7452 : BaseSIMDModifiedImm<Q, op, op2, (outs vectype:$Rd),
7453 !con((ins immtype:$imm8), opt_shift_iop), asm,
7454 "{\t$Rd" # kind # ", $imm8" # opt_shift #
7455 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
7457 let DecoderMethod = "DecodeModImmInstruction";
7460 class BaseSIMDModifiedImmVectorTied<bit Q, bit op, RegisterOperand vectype,
7461 Operand immtype, dag opt_shift_iop,
7462 string opt_shift, string asm, string kind,
7464 : BaseSIMDModifiedImm<Q, op, 0, (outs vectype:$dst),
7465 !con((ins vectype:$Rd, immtype:$imm8), opt_shift_iop),
7466 asm, "{\t$Rd" # kind # ", $imm8" # opt_shift #
7467 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
7468 "$Rd = $dst", pattern> {
7469 let DecoderMethod = "DecodeModImmTiedInstruction";
7472 class BaseSIMDModifiedImmVectorShift<bit Q, bit op, bits<2> b15_b12,
7473 RegisterOperand vectype, string asm,
7474 string kind, list<dag> pattern>
7475 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7476 (ins logical_vec_shift:$shift),
7477 "$shift", asm, kind, pattern> {
7479 let Inst{15} = b15_b12{1};
7480 let Inst{14-13} = shift;
7481 let Inst{12} = b15_b12{0};
7484 class BaseSIMDModifiedImmVectorShiftTied<bit Q, bit op, bits<2> b15_b12,
7485 RegisterOperand vectype, string asm,
7486 string kind, list<dag> pattern>
7487 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
7488 (ins logical_vec_shift:$shift),
7489 "$shift", asm, kind, pattern> {
7491 let Inst{15} = b15_b12{1};
7492 let Inst{14-13} = shift;
7493 let Inst{12} = b15_b12{0};
7497 class BaseSIMDModifiedImmVectorShiftHalf<bit Q, bit op, bits<2> b15_b12,
7498 RegisterOperand vectype, string asm,
7499 string kind, list<dag> pattern>
7500 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7501 (ins logical_vec_hw_shift:$shift),
7502 "$shift", asm, kind, pattern> {
7504 let Inst{15} = b15_b12{1};
7506 let Inst{13} = shift{0};
7507 let Inst{12} = b15_b12{0};
7510 class BaseSIMDModifiedImmVectorShiftHalfTied<bit Q, bit op, bits<2> b15_b12,
7511 RegisterOperand vectype, string asm,
7512 string kind, list<dag> pattern>
7513 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
7514 (ins logical_vec_hw_shift:$shift),
7515 "$shift", asm, kind, pattern> {
7517 let Inst{15} = b15_b12{1};
7519 let Inst{13} = shift{0};
7520 let Inst{12} = b15_b12{0};
7523 multiclass SIMDModifiedImmVectorShift<bit op, bits<2> hw_cmode, bits<2> w_cmode,
7525 def v4i16 : BaseSIMDModifiedImmVectorShiftHalf<0, op, hw_cmode, V64,
7527 def v8i16 : BaseSIMDModifiedImmVectorShiftHalf<1, op, hw_cmode, V128,
7530 def v2i32 : BaseSIMDModifiedImmVectorShift<0, op, w_cmode, V64,
7532 def v4i32 : BaseSIMDModifiedImmVectorShift<1, op, w_cmode, V128,
7536 multiclass SIMDModifiedImmVectorShiftTied<bit op, bits<2> hw_cmode,
7537 bits<2> w_cmode, string asm,
7539 def v4i16 : BaseSIMDModifiedImmVectorShiftHalfTied<0, op, hw_cmode, V64,
7541 [(set (v4i16 V64:$dst), (OpNode V64:$Rd,
7543 (i32 imm:$shift)))]>;
7544 def v8i16 : BaseSIMDModifiedImmVectorShiftHalfTied<1, op, hw_cmode, V128,
7546 [(set (v8i16 V128:$dst), (OpNode V128:$Rd,
7548 (i32 imm:$shift)))]>;
7550 def v2i32 : BaseSIMDModifiedImmVectorShiftTied<0, op, w_cmode, V64,
7552 [(set (v2i32 V64:$dst), (OpNode V64:$Rd,
7554 (i32 imm:$shift)))]>;
7555 def v4i32 : BaseSIMDModifiedImmVectorShiftTied<1, op, w_cmode, V128,
7557 [(set (v4i32 V128:$dst), (OpNode V128:$Rd,
7559 (i32 imm:$shift)))]>;
7562 class SIMDModifiedImmMoveMSL<bit Q, bit op, bits<4> cmode,
7563 RegisterOperand vectype, string asm,
7564 string kind, list<dag> pattern>
7565 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7566 (ins move_vec_shift:$shift),
7567 "$shift", asm, kind, pattern> {
7569 let Inst{15-13} = cmode{3-1};
7570 let Inst{12} = shift;
7573 class SIMDModifiedImmVectorNoShift<bit Q, bit op, bit op2, bits<4> cmode,
7574 RegisterOperand vectype,
7575 Operand imm_type, string asm,
7576 string kind, list<dag> pattern>
7577 : BaseSIMDModifiedImmVector<Q, op, op2, vectype, imm_type, (ins), "",
7578 asm, kind, pattern> {
7579 let Inst{15-12} = cmode;
7582 class SIMDModifiedImmScalarNoShift<bit Q, bit op, bits<4> cmode, string asm,
7584 : BaseSIMDModifiedImm<Q, op, 0, (outs FPR64:$Rd), (ins simdimmtype10:$imm8), asm,
7585 "\t$Rd, $imm8", "", pattern> {
7586 let Inst{15-12} = cmode;
7587 let DecoderMethod = "DecodeModImmInstruction";
7590 //----------------------------------------------------------------------------
7591 // AdvSIMD indexed element
7592 //----------------------------------------------------------------------------
7594 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7595 class BaseSIMDIndexed<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
7596 RegisterOperand dst_reg, RegisterOperand lhs_reg,
7597 RegisterOperand rhs_reg, Operand vec_idx, string asm,
7598 string apple_kind, string dst_kind, string lhs_kind,
7599 string rhs_kind, list<dag> pattern>
7600 : I<(outs dst_reg:$Rd), (ins lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx),
7602 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
7603 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "", pattern>,
7612 let Inst{28} = Scalar;
7613 let Inst{27-24} = 0b1111;
7614 let Inst{23-22} = size;
7615 // Bit 21 must be set by the derived class.
7616 let Inst{20-16} = Rm;
7617 let Inst{15-12} = opc;
7618 // Bit 11 must be set by the derived class.
7624 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7625 class BaseSIMDIndexedTied<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
7626 RegisterOperand dst_reg, RegisterOperand lhs_reg,
7627 RegisterOperand rhs_reg, Operand vec_idx, string asm,
7628 string apple_kind, string dst_kind, string lhs_kind,
7629 string rhs_kind, list<dag> pattern>
7630 : I<(outs dst_reg:$dst),
7631 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx), asm,
7632 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
7633 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "$Rd = $dst", pattern>,
7642 let Inst{28} = Scalar;
7643 let Inst{27-24} = 0b1111;
7644 let Inst{23-22} = size;
7645 // Bit 21 must be set by the derived class.
7646 let Inst{20-16} = Rm;
7647 let Inst{15-12} = opc;
7648 // Bit 11 must be set by the derived class.
7654 // ARMv8.2-A Dot Product Instructions (Indexed)
7655 class BaseSIMDThreeSameVectorDotIndex<bit Q, bit U, string asm, string dst_kind,
7656 string lhs_kind, string rhs_kind,
7657 RegisterOperand RegType,
7658 ValueType AccumType, ValueType InputType,
7659 SDPatternOperator OpNode> :
7660 BaseSIMDIndexedTied<Q, U, 0b0, 0b10, 0b1110, RegType, RegType, V128,
7661 VectorIndexS, asm, "", dst_kind, lhs_kind, rhs_kind,
7662 [(set (AccumType RegType:$dst),
7663 (AccumType (OpNode (AccumType RegType:$Rd),
7664 (InputType RegType:$Rn),
7665 (InputType (bitconvert (AccumType
7666 (AArch64duplane32 (v4i32 V128:$Rm),
7667 VectorIndexS:$idx)))))))]> {
7669 let Inst{21} = idx{0}; // L
7670 let Inst{11} = idx{1}; // H
7673 multiclass SIMDThreeSameVectorDotIndex<bit U, string asm,
7674 SDPatternOperator OpNode> {
7675 def v8i8 : BaseSIMDThreeSameVectorDotIndex<0, U, asm, ".2s", ".8b", ".4b",
7676 V64, v2i32, v8i8, OpNode>;
7677 def v16i8 : BaseSIMDThreeSameVectorDotIndex<1, U, asm, ".4s", ".16b", ".4b",
7678 V128, v4i32, v16i8, OpNode>;
7681 // ARMv8.2-A Fused Multiply Add-Long Instructions (Indexed)
7682 class BaseSIMDThreeSameVectorFMLIndex<bit Q, bit U, bits<4> opc, string asm,
7683 string dst_kind, string lhs_kind,
7684 string rhs_kind, RegisterOperand RegType,
7685 ValueType AccumType, ValueType InputType,
7686 SDPatternOperator OpNode> :
7687 BaseSIMDIndexedTied<Q, U, 0, 0b10, opc, RegType, RegType, V128,
7688 VectorIndexH, asm, "", dst_kind, lhs_kind, rhs_kind,
7689 [(set (AccumType RegType:$dst),
7690 (AccumType (OpNode (AccumType RegType:$Rd),
7691 (InputType RegType:$Rn),
7692 (InputType (AArch64duplane16 (v8f16 V128:$Rm),
7693 VectorIndexH:$idx)))))]> {
7696 let Inst{11} = idx{2}; // H
7697 let Inst{21} = idx{1}; // L
7698 let Inst{20} = idx{0}; // M
7701 multiclass SIMDThreeSameVectorFMLIndex<bit U, bits<4> opc, string asm,
7702 SDPatternOperator OpNode> {
7703 def v4f16 : BaseSIMDThreeSameVectorFMLIndex<0, U, opc, asm, ".2s", ".2h", ".h",
7704 V64, v2f32, v4f16, OpNode>;
7705 def v8f16 : BaseSIMDThreeSameVectorFMLIndex<1, U, opc, asm, ".4s", ".4h", ".h",
7706 V128, v4f32, v8f16, OpNode>;
7709 multiclass SIMDFPIndexed<bit U, bits<4> opc, string asm,
7710 SDPatternOperator OpNode> {
7711 let Predicates = [HasNEON, HasFullFP16] in {
7712 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b00, opc,
7714 V128_lo, VectorIndexH,
7715 asm, ".4h", ".4h", ".4h", ".h",
7716 [(set (v4f16 V64:$Rd),
7717 (OpNode (v4f16 V64:$Rn),
7718 (v4f16 (AArch64duplane16 (v8f16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7720 let Inst{11} = idx{2};
7721 let Inst{21} = idx{1};
7722 let Inst{20} = idx{0};
7725 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b00, opc,
7727 V128_lo, VectorIndexH,
7728 asm, ".8h", ".8h", ".8h", ".h",
7729 [(set (v8f16 V128:$Rd),
7730 (OpNode (v8f16 V128:$Rn),
7731 (v8f16 (AArch64duplane16 (v8f16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7733 let Inst{11} = idx{2};
7734 let Inst{21} = idx{1};
7735 let Inst{20} = idx{0};
7737 } // Predicates = [HasNEON, HasFullFP16]
7739 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7742 asm, ".2s", ".2s", ".2s", ".s",
7743 [(set (v2f32 V64:$Rd),
7744 (OpNode (v2f32 V64:$Rn),
7745 (v2f32 (AArch64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
7747 let Inst{11} = idx{1};
7748 let Inst{21} = idx{0};
7751 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7754 asm, ".4s", ".4s", ".4s", ".s",
7755 [(set (v4f32 V128:$Rd),
7756 (OpNode (v4f32 V128:$Rn),
7757 (v4f32 (AArch64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
7759 let Inst{11} = idx{1};
7760 let Inst{21} = idx{0};
7763 def v2i64_indexed : BaseSIMDIndexed<1, U, 0, 0b11, opc,
7766 asm, ".2d", ".2d", ".2d", ".d",
7767 [(set (v2f64 V128:$Rd),
7768 (OpNode (v2f64 V128:$Rn),
7769 (v2f64 (AArch64duplane64 (v2f64 V128:$Rm), VectorIndexD:$idx))))]> {
7771 let Inst{11} = idx{0};
7775 let Predicates = [HasNEON, HasFullFP16] in {
7776 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b00, opc,
7777 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7778 asm, ".h", "", "", ".h",
7779 [(set (f16 FPR16Op:$Rd),
7780 (OpNode (f16 FPR16Op:$Rn),
7781 (f16 (vector_extract (v8f16 V128_lo:$Rm),
7782 VectorIndexH:$idx))))]> {
7784 let Inst{11} = idx{2};
7785 let Inst{21} = idx{1};
7786 let Inst{20} = idx{0};
7788 } // Predicates = [HasNEON, HasFullFP16]
7790 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
7791 FPR32Op, FPR32Op, V128, VectorIndexS,
7792 asm, ".s", "", "", ".s",
7793 [(set (f32 FPR32Op:$Rd),
7794 (OpNode (f32 FPR32Op:$Rn),
7795 (f32 (vector_extract (v4f32 V128:$Rm),
7796 VectorIndexS:$idx))))]> {
7798 let Inst{11} = idx{1};
7799 let Inst{21} = idx{0};
7802 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b11, opc,
7803 FPR64Op, FPR64Op, V128, VectorIndexD,
7804 asm, ".d", "", "", ".d",
7805 [(set (f64 FPR64Op:$Rd),
7806 (OpNode (f64 FPR64Op:$Rn),
7807 (f64 (vector_extract (v2f64 V128:$Rm),
7808 VectorIndexD:$idx))))]> {
7810 let Inst{11} = idx{0};
7815 multiclass SIMDFPIndexedTiedPatterns<string INST, SDPatternOperator OpNode> {
7816 // 2 variants for the .2s version: DUPLANE from 128-bit and DUP scalar.
7817 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7818 (AArch64duplane32 (v4f32 V128:$Rm),
7819 VectorIndexS:$idx))),
7820 (!cast<Instruction>(INST # v2i32_indexed)
7821 V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7822 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7823 (AArch64dup (f32 FPR32Op:$Rm)))),
7824 (!cast<Instruction>(INST # "v2i32_indexed") V64:$Rd, V64:$Rn,
7825 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7828 // 2 variants for the .4s version: DUPLANE from 128-bit and DUP scalar.
7829 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7830 (AArch64duplane32 (v4f32 V128:$Rm),
7831 VectorIndexS:$idx))),
7832 (!cast<Instruction>(INST # "v4i32_indexed")
7833 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7834 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7835 (AArch64dup (f32 FPR32Op:$Rm)))),
7836 (!cast<Instruction>(INST # "v4i32_indexed") V128:$Rd, V128:$Rn,
7837 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7839 // 2 variants for the .2d version: DUPLANE from 128-bit and DUP scalar.
7840 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7841 (AArch64duplane64 (v2f64 V128:$Rm),
7842 VectorIndexD:$idx))),
7843 (!cast<Instruction>(INST # "v2i64_indexed")
7844 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7845 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7846 (AArch64dup (f64 FPR64Op:$Rm)))),
7847 (!cast<Instruction>(INST # "v2i64_indexed") V128:$Rd, V128:$Rn,
7848 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
7850 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
7851 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7852 (vector_extract (v4f32 V128:$Rm), VectorIndexS:$idx))),
7853 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
7854 V128:$Rm, VectorIndexS:$idx)>;
7855 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7856 (vector_extract (v2f32 V64:$Rm), VectorIndexS:$idx))),
7857 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
7858 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
7860 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
7861 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
7862 (vector_extract (v2f64 V128:$Rm), VectorIndexD:$idx))),
7863 (!cast<Instruction>(INST # "v1i64_indexed") FPR64:$Rd, FPR64:$Rn,
7864 V128:$Rm, VectorIndexD:$idx)>;
7867 multiclass SIMDFPIndexedTied<bit U, bits<4> opc, string asm> {
7868 let Predicates = [HasNEON, HasFullFP16] in {
7869 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b00, opc, V64, V64,
7870 V128_lo, VectorIndexH,
7871 asm, ".4h", ".4h", ".4h", ".h", []> {
7873 let Inst{11} = idx{2};
7874 let Inst{21} = idx{1};
7875 let Inst{20} = idx{0};
7878 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b00, opc,
7880 V128_lo, VectorIndexH,
7881 asm, ".8h", ".8h", ".8h", ".h", []> {
7883 let Inst{11} = idx{2};
7884 let Inst{21} = idx{1};
7885 let Inst{20} = idx{0};
7887 } // Predicates = [HasNEON, HasFullFP16]
7889 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc, V64, V64,
7891 asm, ".2s", ".2s", ".2s", ".s", []> {
7893 let Inst{11} = idx{1};
7894 let Inst{21} = idx{0};
7897 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
7900 asm, ".4s", ".4s", ".4s", ".s", []> {
7902 let Inst{11} = idx{1};
7903 let Inst{21} = idx{0};
7906 def v2i64_indexed : BaseSIMDIndexedTied<1, U, 0, 0b11, opc,
7909 asm, ".2d", ".2d", ".2d", ".d", []> {
7911 let Inst{11} = idx{0};
7915 let Predicates = [HasNEON, HasFullFP16] in {
7916 def v1i16_indexed : BaseSIMDIndexedTied<1, U, 1, 0b00, opc,
7917 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7918 asm, ".h", "", "", ".h", []> {
7920 let Inst{11} = idx{2};
7921 let Inst{21} = idx{1};
7922 let Inst{20} = idx{0};
7924 } // Predicates = [HasNEON, HasFullFP16]
7926 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
7927 FPR32Op, FPR32Op, V128, VectorIndexS,
7928 asm, ".s", "", "", ".s", []> {
7930 let Inst{11} = idx{1};
7931 let Inst{21} = idx{0};
7934 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b11, opc,
7935 FPR64Op, FPR64Op, V128, VectorIndexD,
7936 asm, ".d", "", "", ".d", []> {
7938 let Inst{11} = idx{0};
7943 multiclass SIMDIndexedHS<bit U, bits<4> opc, string asm,
7944 SDPatternOperator OpNode> {
7945 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc, V64, V64,
7946 V128_lo, VectorIndexH,
7947 asm, ".4h", ".4h", ".4h", ".h",
7948 [(set (v4i16 V64:$Rd),
7949 (OpNode (v4i16 V64:$Rn),
7950 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7952 let Inst{11} = idx{2};
7953 let Inst{21} = idx{1};
7954 let Inst{20} = idx{0};
7957 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7959 V128_lo, VectorIndexH,
7960 asm, ".8h", ".8h", ".8h", ".h",
7961 [(set (v8i16 V128:$Rd),
7962 (OpNode (v8i16 V128:$Rn),
7963 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7965 let Inst{11} = idx{2};
7966 let Inst{21} = idx{1};
7967 let Inst{20} = idx{0};
7970 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7973 asm, ".2s", ".2s", ".2s", ".s",
7974 [(set (v2i32 V64:$Rd),
7975 (OpNode (v2i32 V64:$Rn),
7976 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7978 let Inst{11} = idx{1};
7979 let Inst{21} = idx{0};
7982 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7985 asm, ".4s", ".4s", ".4s", ".s",
7986 [(set (v4i32 V128:$Rd),
7987 (OpNode (v4i32 V128:$Rn),
7988 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7990 let Inst{11} = idx{1};
7991 let Inst{21} = idx{0};
7994 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
7995 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7996 asm, ".h", "", "", ".h", []> {
7998 let Inst{11} = idx{2};
7999 let Inst{21} = idx{1};
8000 let Inst{20} = idx{0};
8003 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
8004 FPR32Op, FPR32Op, V128, VectorIndexS,
8005 asm, ".s", "", "", ".s",
8006 [(set (i32 FPR32Op:$Rd),
8007 (OpNode FPR32Op:$Rn,
8008 (i32 (vector_extract (v4i32 V128:$Rm),
8009 VectorIndexS:$idx))))]> {
8011 let Inst{11} = idx{1};
8012 let Inst{21} = idx{0};
8016 multiclass SIMDVectorIndexedHS<bit U, bits<4> opc, string asm,
8017 SDPatternOperator OpNode> {
8018 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
8020 V128_lo, VectorIndexH,
8021 asm, ".4h", ".4h", ".4h", ".h",
8022 [(set (v4i16 V64:$Rd),
8023 (OpNode (v4i16 V64:$Rn),
8024 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8026 let Inst{11} = idx{2};
8027 let Inst{21} = idx{1};
8028 let Inst{20} = idx{0};
8031 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
8033 V128_lo, VectorIndexH,
8034 asm, ".8h", ".8h", ".8h", ".h",
8035 [(set (v8i16 V128:$Rd),
8036 (OpNode (v8i16 V128:$Rn),
8037 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8039 let Inst{11} = idx{2};
8040 let Inst{21} = idx{1};
8041 let Inst{20} = idx{0};
8044 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
8047 asm, ".2s", ".2s", ".2s", ".s",
8048 [(set (v2i32 V64:$Rd),
8049 (OpNode (v2i32 V64:$Rn),
8050 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8052 let Inst{11} = idx{1};
8053 let Inst{21} = idx{0};
8056 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
8059 asm, ".4s", ".4s", ".4s", ".s",
8060 [(set (v4i32 V128:$Rd),
8061 (OpNode (v4i32 V128:$Rn),
8062 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8064 let Inst{11} = idx{1};
8065 let Inst{21} = idx{0};
8069 multiclass SIMDVectorIndexedHSTied<bit U, bits<4> opc, string asm,
8070 SDPatternOperator OpNode> {
8071 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc, V64, V64,
8072 V128_lo, VectorIndexH,
8073 asm, ".4h", ".4h", ".4h", ".h",
8074 [(set (v4i16 V64:$dst),
8075 (OpNode (v4i16 V64:$Rd),(v4i16 V64:$Rn),
8076 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8078 let Inst{11} = idx{2};
8079 let Inst{21} = idx{1};
8080 let Inst{20} = idx{0};
8083 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
8085 V128_lo, VectorIndexH,
8086 asm, ".8h", ".8h", ".8h", ".h",
8087 [(set (v8i16 V128:$dst),
8088 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
8089 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8091 let Inst{11} = idx{2};
8092 let Inst{21} = idx{1};
8093 let Inst{20} = idx{0};
8096 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
8099 asm, ".2s", ".2s", ".2s", ".s",
8100 [(set (v2i32 V64:$dst),
8101 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
8102 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8104 let Inst{11} = idx{1};
8105 let Inst{21} = idx{0};
8108 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
8111 asm, ".4s", ".4s", ".4s", ".s",
8112 [(set (v4i32 V128:$dst),
8113 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8114 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8116 let Inst{11} = idx{1};
8117 let Inst{21} = idx{0};
8121 multiclass SIMDIndexedLongSD<bit U, bits<4> opc, string asm,
8122 SDPatternOperator OpNode> {
8123 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
8125 V128_lo, VectorIndexH,
8126 asm, ".4s", ".4s", ".4h", ".h",
8127 [(set (v4i32 V128:$Rd),
8128 (OpNode (v4i16 V64:$Rn),
8129 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8131 let Inst{11} = idx{2};
8132 let Inst{21} = idx{1};
8133 let Inst{20} = idx{0};
8136 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
8138 V128_lo, VectorIndexH,
8139 asm#"2", ".4s", ".4s", ".8h", ".h",
8140 [(set (v4i32 V128:$Rd),
8141 (OpNode (extract_high_v8i16 V128:$Rn),
8142 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8143 VectorIndexH:$idx))))]> {
8146 let Inst{11} = idx{2};
8147 let Inst{21} = idx{1};
8148 let Inst{20} = idx{0};
8151 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
8154 asm, ".2d", ".2d", ".2s", ".s",
8155 [(set (v2i64 V128:$Rd),
8156 (OpNode (v2i32 V64:$Rn),
8157 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8159 let Inst{11} = idx{1};
8160 let Inst{21} = idx{0};
8163 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
8166 asm#"2", ".2d", ".2d", ".4s", ".s",
8167 [(set (v2i64 V128:$Rd),
8168 (OpNode (extract_high_v4i32 V128:$Rn),
8169 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
8170 VectorIndexS:$idx))))]> {
8172 let Inst{11} = idx{1};
8173 let Inst{21} = idx{0};
8176 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
8177 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
8178 asm, ".h", "", "", ".h", []> {
8180 let Inst{11} = idx{2};
8181 let Inst{21} = idx{1};
8182 let Inst{20} = idx{0};
8185 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
8186 FPR64Op, FPR32Op, V128, VectorIndexS,
8187 asm, ".s", "", "", ".s", []> {
8189 let Inst{11} = idx{1};
8190 let Inst{21} = idx{0};
8194 multiclass SIMDIndexedLongSQDMLXSDTied<bit U, bits<4> opc, string asm,
8195 SDPatternOperator Accum> {
8196 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
8198 V128_lo, VectorIndexH,
8199 asm, ".4s", ".4s", ".4h", ".h",
8200 [(set (v4i32 V128:$dst),
8201 (Accum (v4i32 V128:$Rd),
8202 (v4i32 (int_aarch64_neon_sqdmull
8204 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8205 VectorIndexH:$idx))))))]> {
8207 let Inst{11} = idx{2};
8208 let Inst{21} = idx{1};
8209 let Inst{20} = idx{0};
8212 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but an
8213 // intermediate EXTRACT_SUBREG would be untyped.
8214 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
8215 (i32 (vector_extract (v4i32
8216 (int_aarch64_neon_sqdmull (v4i16 V64:$Rn),
8217 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8218 VectorIndexH:$idx)))),
8221 (!cast<Instruction>(NAME # v4i16_indexed)
8222 (SUBREG_TO_REG (i32 0), FPR32Op:$Rd, ssub), V64:$Rn,
8223 V128_lo:$Rm, VectorIndexH:$idx),
8226 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
8228 V128_lo, VectorIndexH,
8229 asm#"2", ".4s", ".4s", ".8h", ".h",
8230 [(set (v4i32 V128:$dst),
8231 (Accum (v4i32 V128:$Rd),
8232 (v4i32 (int_aarch64_neon_sqdmull
8233 (extract_high_v8i16 V128:$Rn),
8235 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8236 VectorIndexH:$idx))))))]> {
8238 let Inst{11} = idx{2};
8239 let Inst{21} = idx{1};
8240 let Inst{20} = idx{0};
8243 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
8246 asm, ".2d", ".2d", ".2s", ".s",
8247 [(set (v2i64 V128:$dst),
8248 (Accum (v2i64 V128:$Rd),
8249 (v2i64 (int_aarch64_neon_sqdmull
8251 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm),
8252 VectorIndexS:$idx))))))]> {
8254 let Inst{11} = idx{1};
8255 let Inst{21} = idx{0};
8258 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
8261 asm#"2", ".2d", ".2d", ".4s", ".s",
8262 [(set (v2i64 V128:$dst),
8263 (Accum (v2i64 V128:$Rd),
8264 (v2i64 (int_aarch64_neon_sqdmull
8265 (extract_high_v4i32 V128:$Rn),
8267 (AArch64duplane32 (v4i32 V128:$Rm),
8268 VectorIndexS:$idx))))))]> {
8270 let Inst{11} = idx{1};
8271 let Inst{21} = idx{0};
8274 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
8275 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
8276 asm, ".h", "", "", ".h", []> {
8278 let Inst{11} = idx{2};
8279 let Inst{21} = idx{1};
8280 let Inst{20} = idx{0};
8284 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
8285 FPR64Op, FPR32Op, V128, VectorIndexS,
8286 asm, ".s", "", "", ".s",
8287 [(set (i64 FPR64Op:$dst),
8288 (Accum (i64 FPR64Op:$Rd),
8289 (i64 (int_aarch64_neon_sqdmulls_scalar
8291 (i32 (vector_extract (v4i32 V128:$Rm),
8292 VectorIndexS:$idx))))))]> {
8295 let Inst{11} = idx{1};
8296 let Inst{21} = idx{0};
8300 multiclass SIMDVectorIndexedLongSD<bit U, bits<4> opc, string asm,
8301 SDPatternOperator OpNode> {
8302 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
8303 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
8305 V128_lo, VectorIndexH,
8306 asm, ".4s", ".4s", ".4h", ".h",
8307 [(set (v4i32 V128:$Rd),
8308 (OpNode (v4i16 V64:$Rn),
8309 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8311 let Inst{11} = idx{2};
8312 let Inst{21} = idx{1};
8313 let Inst{20} = idx{0};
8316 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
8318 V128_lo, VectorIndexH,
8319 asm#"2", ".4s", ".4s", ".8h", ".h",
8320 [(set (v4i32 V128:$Rd),
8321 (OpNode (extract_high_v8i16 V128:$Rn),
8322 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8323 VectorIndexH:$idx))))]> {
8326 let Inst{11} = idx{2};
8327 let Inst{21} = idx{1};
8328 let Inst{20} = idx{0};
8331 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
8334 asm, ".2d", ".2d", ".2s", ".s",
8335 [(set (v2i64 V128:$Rd),
8336 (OpNode (v2i32 V64:$Rn),
8337 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8339 let Inst{11} = idx{1};
8340 let Inst{21} = idx{0};
8343 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
8346 asm#"2", ".2d", ".2d", ".4s", ".s",
8347 [(set (v2i64 V128:$Rd),
8348 (OpNode (extract_high_v4i32 V128:$Rn),
8349 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
8350 VectorIndexS:$idx))))]> {
8352 let Inst{11} = idx{1};
8353 let Inst{21} = idx{0};
8358 multiclass SIMDVectorIndexedLongSDTied<bit U, bits<4> opc, string asm,
8359 SDPatternOperator OpNode> {
8360 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
8361 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
8363 V128_lo, VectorIndexH,
8364 asm, ".4s", ".4s", ".4h", ".h",
8365 [(set (v4i32 V128:$dst),
8366 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn),
8367 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8369 let Inst{11} = idx{2};
8370 let Inst{21} = idx{1};
8371 let Inst{20} = idx{0};
8374 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
8376 V128_lo, VectorIndexH,
8377 asm#"2", ".4s", ".4s", ".8h", ".h",
8378 [(set (v4i32 V128:$dst),
8379 (OpNode (v4i32 V128:$Rd),
8380 (extract_high_v8i16 V128:$Rn),
8381 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8382 VectorIndexH:$idx))))]> {
8384 let Inst{11} = idx{2};
8385 let Inst{21} = idx{1};
8386 let Inst{20} = idx{0};
8389 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
8392 asm, ".2d", ".2d", ".2s", ".s",
8393 [(set (v2i64 V128:$dst),
8394 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn),
8395 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8397 let Inst{11} = idx{1};
8398 let Inst{21} = idx{0};
8401 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
8404 asm#"2", ".2d", ".2d", ".4s", ".s",
8405 [(set (v2i64 V128:$dst),
8406 (OpNode (v2i64 V128:$Rd),
8407 (extract_high_v4i32 V128:$Rn),
8408 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
8409 VectorIndexS:$idx))))]> {
8411 let Inst{11} = idx{1};
8412 let Inst{21} = idx{0};
8417 //----------------------------------------------------------------------------
8418 // AdvSIMD scalar shift by immediate
8419 //----------------------------------------------------------------------------
8421 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8422 class BaseSIMDScalarShift<bit U, bits<5> opc, bits<7> fixed_imm,
8423 RegisterClass regtype1, RegisterClass regtype2,
8424 Operand immtype, string asm, list<dag> pattern>
8425 : I<(outs regtype1:$Rd), (ins regtype2:$Rn, immtype:$imm),
8426 asm, "\t$Rd, $Rn, $imm", "", pattern>,
8431 let Inst{31-30} = 0b01;
8433 let Inst{28-23} = 0b111110;
8434 let Inst{22-16} = fixed_imm;
8435 let Inst{15-11} = opc;
8441 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8442 class BaseSIMDScalarShiftTied<bit U, bits<5> opc, bits<7> fixed_imm,
8443 RegisterClass regtype1, RegisterClass regtype2,
8444 Operand immtype, string asm, list<dag> pattern>
8445 : I<(outs regtype1:$dst), (ins regtype1:$Rd, regtype2:$Rn, immtype:$imm),
8446 asm, "\t$Rd, $Rn, $imm", "$Rd = $dst", pattern>,
8451 let Inst{31-30} = 0b01;
8453 let Inst{28-23} = 0b111110;
8454 let Inst{22-16} = fixed_imm;
8455 let Inst{15-11} = opc;
8462 multiclass SIMDFPScalarRShift<bit U, bits<5> opc, string asm> {
8463 let Predicates = [HasNEON, HasFullFP16] in {
8464 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8465 FPR16, FPR16, vecshiftR16, asm, []> {
8466 let Inst{19-16} = imm{3-0};
8468 } // Predicates = [HasNEON, HasFullFP16]
8469 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8470 FPR32, FPR32, vecshiftR32, asm, []> {
8471 let Inst{20-16} = imm{4-0};
8473 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8474 FPR64, FPR64, vecshiftR64, asm, []> {
8475 let Inst{21-16} = imm{5-0};
8479 multiclass SIMDScalarRShiftD<bit U, bits<5> opc, string asm,
8480 SDPatternOperator OpNode> {
8481 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8482 FPR64, FPR64, vecshiftR64, asm,
8483 [(set (i64 FPR64:$Rd),
8484 (OpNode (i64 FPR64:$Rn), (i32 vecshiftR64:$imm)))]> {
8485 let Inst{21-16} = imm{5-0};
8488 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftR64:$imm))),
8489 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftR64:$imm)>;
8492 multiclass SIMDScalarRShiftDTied<bit U, bits<5> opc, string asm,
8493 SDPatternOperator OpNode = null_frag> {
8494 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
8495 FPR64, FPR64, vecshiftR64, asm,
8496 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn),
8497 (i32 vecshiftR64:$imm)))]> {
8498 let Inst{21-16} = imm{5-0};
8501 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
8502 (i32 vecshiftR64:$imm))),
8503 (!cast<Instruction>(NAME # "d") FPR64:$Rd, FPR64:$Rn,
8507 multiclass SIMDScalarLShiftD<bit U, bits<5> opc, string asm,
8508 SDPatternOperator OpNode> {
8509 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8510 FPR64, FPR64, vecshiftL64, asm,
8511 [(set (v1i64 FPR64:$Rd),
8512 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
8513 let Inst{21-16} = imm{5-0};
8517 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8518 multiclass SIMDScalarLShiftDTied<bit U, bits<5> opc, string asm> {
8519 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
8520 FPR64, FPR64, vecshiftL64, asm, []> {
8521 let Inst{21-16} = imm{5-0};
8525 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8526 multiclass SIMDScalarRShiftBHS<bit U, bits<5> opc, string asm,
8527 SDPatternOperator OpNode = null_frag> {
8528 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8529 FPR8, FPR16, vecshiftR8, asm, []> {
8530 let Inst{18-16} = imm{2-0};
8533 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8534 FPR16, FPR32, vecshiftR16, asm, []> {
8535 let Inst{19-16} = imm{3-0};
8538 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8539 FPR32, FPR64, vecshiftR32, asm,
8540 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn), vecshiftR32:$imm))]> {
8541 let Inst{20-16} = imm{4-0};
8545 multiclass SIMDScalarLShiftBHSD<bit U, bits<5> opc, string asm,
8546 SDPatternOperator OpNode> {
8547 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8548 FPR8, FPR8, vecshiftL8, asm, []> {
8549 let Inst{18-16} = imm{2-0};
8552 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8553 FPR16, FPR16, vecshiftL16, asm, []> {
8554 let Inst{19-16} = imm{3-0};
8557 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8558 FPR32, FPR32, vecshiftL32, asm,
8559 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn), (i32 vecshiftL32:$imm)))]> {
8560 let Inst{20-16} = imm{4-0};
8563 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8564 FPR64, FPR64, vecshiftL64, asm,
8565 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
8566 let Inst{21-16} = imm{5-0};
8569 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm))),
8570 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftL64:$imm)>;
8573 multiclass SIMDScalarRShiftBHSD<bit U, bits<5> opc, string asm> {
8574 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8575 FPR8, FPR8, vecshiftR8, asm, []> {
8576 let Inst{18-16} = imm{2-0};
8579 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8580 FPR16, FPR16, vecshiftR16, asm, []> {
8581 let Inst{19-16} = imm{3-0};
8584 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8585 FPR32, FPR32, vecshiftR32, asm, []> {
8586 let Inst{20-16} = imm{4-0};
8589 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8590 FPR64, FPR64, vecshiftR64, asm, []> {
8591 let Inst{21-16} = imm{5-0};
8595 //----------------------------------------------------------------------------
8596 // AdvSIMD vector x indexed element
8597 //----------------------------------------------------------------------------
8599 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8600 class BaseSIMDVectorShift<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
8601 RegisterOperand dst_reg, RegisterOperand src_reg,
8603 string asm, string dst_kind, string src_kind,
8605 : I<(outs dst_reg:$Rd), (ins src_reg:$Rn, immtype:$imm),
8606 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
8607 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "", pattern>,
8614 let Inst{28-23} = 0b011110;
8615 let Inst{22-16} = fixed_imm;
8616 let Inst{15-11} = opc;
8622 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8623 class BaseSIMDVectorShiftTied<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
8624 RegisterOperand vectype1, RegisterOperand vectype2,
8626 string asm, string dst_kind, string src_kind,
8628 : I<(outs vectype1:$dst), (ins vectype1:$Rd, vectype2:$Rn, immtype:$imm),
8629 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
8630 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "$Rd = $dst", pattern>,
8637 let Inst{28-23} = 0b011110;
8638 let Inst{22-16} = fixed_imm;
8639 let Inst{15-11} = opc;
8645 multiclass SIMDVectorRShiftSD<bit U, bits<5> opc, string asm,
8647 let Predicates = [HasNEON, HasFullFP16] in {
8648 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8649 V64, V64, vecshiftR16,
8651 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (i32 imm:$imm)))]> {
8653 let Inst{19-16} = imm;
8656 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8657 V128, V128, vecshiftR16,
8659 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (i32 imm:$imm)))]> {
8661 let Inst{19-16} = imm;
8663 } // Predicates = [HasNEON, HasFullFP16]
8664 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8665 V64, V64, vecshiftR32,
8667 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (i32 imm:$imm)))]> {
8669 let Inst{20-16} = imm;
8672 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8673 V128, V128, vecshiftR32,
8675 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (i32 imm:$imm)))]> {
8677 let Inst{20-16} = imm;
8680 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8681 V128, V128, vecshiftR64,
8683 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (i32 imm:$imm)))]> {
8685 let Inst{21-16} = imm;
8689 multiclass SIMDVectorRShiftToFP<bit U, bits<5> opc, string asm,
8691 let Predicates = [HasNEON, HasFullFP16] in {
8692 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8693 V64, V64, vecshiftR16,
8695 [(set (v4f16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (i32 imm:$imm)))]> {
8697 let Inst{19-16} = imm;
8700 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8701 V128, V128, vecshiftR16,
8703 [(set (v8f16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (i32 imm:$imm)))]> {
8705 let Inst{19-16} = imm;
8707 } // Predicates = [HasNEON, HasFullFP16]
8709 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8710 V64, V64, vecshiftR32,
8712 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (i32 imm:$imm)))]> {
8714 let Inst{20-16} = imm;
8717 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8718 V128, V128, vecshiftR32,
8720 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (i32 imm:$imm)))]> {
8722 let Inst{20-16} = imm;
8725 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8726 V128, V128, vecshiftR64,
8728 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (i32 imm:$imm)))]> {
8730 let Inst{21-16} = imm;
8734 multiclass SIMDVectorRShiftNarrowBHS<bit U, bits<5> opc, string asm,
8735 SDPatternOperator OpNode> {
8736 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8737 V64, V128, vecshiftR16Narrow,
8739 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))]> {
8741 let Inst{18-16} = imm;
8744 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8745 V128, V128, vecshiftR16Narrow,
8746 asm#"2", ".16b", ".8h", []> {
8748 let Inst{18-16} = imm;
8749 let hasSideEffects = 0;
8752 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8753 V64, V128, vecshiftR32Narrow,
8755 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))]> {
8757 let Inst{19-16} = imm;
8760 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8761 V128, V128, vecshiftR32Narrow,
8762 asm#"2", ".8h", ".4s", []> {
8764 let Inst{19-16} = imm;
8765 let hasSideEffects = 0;
8768 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8769 V64, V128, vecshiftR64Narrow,
8771 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))]> {
8773 let Inst{20-16} = imm;
8776 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8777 V128, V128, vecshiftR64Narrow,
8778 asm#"2", ".4s", ".2d", []> {
8780 let Inst{20-16} = imm;
8781 let hasSideEffects = 0;
8784 // TableGen doesn't like patters w/ INSERT_SUBREG on the instructions
8785 // themselves, so put them here instead.
8787 // Patterns involving what's effectively an insert high and a normal
8788 // intrinsic, represented by CONCAT_VECTORS.
8789 def : Pat<(concat_vectors (v8i8 V64:$Rd),(OpNode (v8i16 V128:$Rn),
8790 vecshiftR16Narrow:$imm)),
8791 (!cast<Instruction>(NAME # "v16i8_shift")
8792 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8793 V128:$Rn, vecshiftR16Narrow:$imm)>;
8794 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn),
8795 vecshiftR32Narrow:$imm)),
8796 (!cast<Instruction>(NAME # "v8i16_shift")
8797 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8798 V128:$Rn, vecshiftR32Narrow:$imm)>;
8799 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn),
8800 vecshiftR64Narrow:$imm)),
8801 (!cast<Instruction>(NAME # "v4i32_shift")
8802 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8803 V128:$Rn, vecshiftR64Narrow:$imm)>;
8806 multiclass SIMDVectorLShiftBHSD<bit U, bits<5> opc, string asm,
8807 SDPatternOperator OpNode> {
8808 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8809 V64, V64, vecshiftL8,
8811 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
8812 (i32 vecshiftL8:$imm)))]> {
8814 let Inst{18-16} = imm;
8817 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8818 V128, V128, vecshiftL8,
8819 asm, ".16b", ".16b",
8820 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
8821 (i32 vecshiftL8:$imm)))]> {
8823 let Inst{18-16} = imm;
8826 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8827 V64, V64, vecshiftL16,
8829 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
8830 (i32 vecshiftL16:$imm)))]> {
8832 let Inst{19-16} = imm;
8835 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8836 V128, V128, vecshiftL16,
8838 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
8839 (i32 vecshiftL16:$imm)))]> {
8841 let Inst{19-16} = imm;
8844 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8845 V64, V64, vecshiftL32,
8847 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
8848 (i32 vecshiftL32:$imm)))]> {
8850 let Inst{20-16} = imm;
8853 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8854 V128, V128, vecshiftL32,
8856 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
8857 (i32 vecshiftL32:$imm)))]> {
8859 let Inst{20-16} = imm;
8862 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8863 V128, V128, vecshiftL64,
8865 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
8866 (i32 vecshiftL64:$imm)))]> {
8868 let Inst{21-16} = imm;
8872 multiclass SIMDVectorRShiftBHSD<bit U, bits<5> opc, string asm,
8873 SDPatternOperator OpNode> {
8874 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8875 V64, V64, vecshiftR8,
8877 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
8878 (i32 vecshiftR8:$imm)))]> {
8880 let Inst{18-16} = imm;
8883 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8884 V128, V128, vecshiftR8,
8885 asm, ".16b", ".16b",
8886 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
8887 (i32 vecshiftR8:$imm)))]> {
8889 let Inst{18-16} = imm;
8892 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8893 V64, V64, vecshiftR16,
8895 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
8896 (i32 vecshiftR16:$imm)))]> {
8898 let Inst{19-16} = imm;
8901 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8902 V128, V128, vecshiftR16,
8904 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
8905 (i32 vecshiftR16:$imm)))]> {
8907 let Inst{19-16} = imm;
8910 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8911 V64, V64, vecshiftR32,
8913 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
8914 (i32 vecshiftR32:$imm)))]> {
8916 let Inst{20-16} = imm;
8919 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8920 V128, V128, vecshiftR32,
8922 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
8923 (i32 vecshiftR32:$imm)))]> {
8925 let Inst{20-16} = imm;
8928 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8929 V128, V128, vecshiftR64,
8931 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
8932 (i32 vecshiftR64:$imm)))]> {
8934 let Inst{21-16} = imm;
8938 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8939 multiclass SIMDVectorRShiftBHSDTied<bit U, bits<5> opc, string asm,
8940 SDPatternOperator OpNode = null_frag> {
8941 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
8942 V64, V64, vecshiftR8, asm, ".8b", ".8b",
8943 [(set (v8i8 V64:$dst),
8944 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
8945 (i32 vecshiftR8:$imm)))]> {
8947 let Inst{18-16} = imm;
8950 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8951 V128, V128, vecshiftR8, asm, ".16b", ".16b",
8952 [(set (v16i8 V128:$dst),
8953 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
8954 (i32 vecshiftR8:$imm)))]> {
8956 let Inst{18-16} = imm;
8959 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
8960 V64, V64, vecshiftR16, asm, ".4h", ".4h",
8961 [(set (v4i16 V64:$dst),
8962 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
8963 (i32 vecshiftR16:$imm)))]> {
8965 let Inst{19-16} = imm;
8968 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8969 V128, V128, vecshiftR16, asm, ".8h", ".8h",
8970 [(set (v8i16 V128:$dst),
8971 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
8972 (i32 vecshiftR16:$imm)))]> {
8974 let Inst{19-16} = imm;
8977 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
8978 V64, V64, vecshiftR32, asm, ".2s", ".2s",
8979 [(set (v2i32 V64:$dst),
8980 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
8981 (i32 vecshiftR32:$imm)))]> {
8983 let Inst{20-16} = imm;
8986 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8987 V128, V128, vecshiftR32, asm, ".4s", ".4s",
8988 [(set (v4i32 V128:$dst),
8989 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8990 (i32 vecshiftR32:$imm)))]> {
8992 let Inst{20-16} = imm;
8995 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
8996 V128, V128, vecshiftR64,
8997 asm, ".2d", ".2d", [(set (v2i64 V128:$dst),
8998 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
8999 (i32 vecshiftR64:$imm)))]> {
9001 let Inst{21-16} = imm;
9005 multiclass SIMDVectorLShiftBHSDTied<bit U, bits<5> opc, string asm,
9006 SDPatternOperator OpNode = null_frag> {
9007 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
9008 V64, V64, vecshiftL8,
9010 [(set (v8i8 V64:$dst),
9011 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
9012 (i32 vecshiftL8:$imm)))]> {
9014 let Inst{18-16} = imm;
9017 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
9018 V128, V128, vecshiftL8,
9019 asm, ".16b", ".16b",
9020 [(set (v16i8 V128:$dst),
9021 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
9022 (i32 vecshiftL8:$imm)))]> {
9024 let Inst{18-16} = imm;
9027 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
9028 V64, V64, vecshiftL16,
9030 [(set (v4i16 V64:$dst),
9031 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
9032 (i32 vecshiftL16:$imm)))]> {
9034 let Inst{19-16} = imm;
9037 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
9038 V128, V128, vecshiftL16,
9040 [(set (v8i16 V128:$dst),
9041 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
9042 (i32 vecshiftL16:$imm)))]> {
9044 let Inst{19-16} = imm;
9047 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
9048 V64, V64, vecshiftL32,
9050 [(set (v2i32 V64:$dst),
9051 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
9052 (i32 vecshiftL32:$imm)))]> {
9054 let Inst{20-16} = imm;
9057 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
9058 V128, V128, vecshiftL32,
9060 [(set (v4i32 V128:$dst),
9061 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
9062 (i32 vecshiftL32:$imm)))]> {
9064 let Inst{20-16} = imm;
9067 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
9068 V128, V128, vecshiftL64,
9070 [(set (v2i64 V128:$dst),
9071 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
9072 (i32 vecshiftL64:$imm)))]> {
9074 let Inst{21-16} = imm;
9078 multiclass SIMDVectorLShiftLongBHSD<bit U, bits<5> opc, string asm,
9079 SDPatternOperator OpNode> {
9080 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
9081 V128, V64, vecshiftL8, asm, ".8h", ".8b",
9082 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), vecshiftL8:$imm))]> {
9084 let Inst{18-16} = imm;
9087 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
9088 V128, V128, vecshiftL8,
9089 asm#"2", ".8h", ".16b",
9090 [(set (v8i16 V128:$Rd),
9091 (OpNode (extract_high_v16i8 V128:$Rn), vecshiftL8:$imm))]> {
9093 let Inst{18-16} = imm;
9096 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
9097 V128, V64, vecshiftL16, asm, ".4s", ".4h",
9098 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), vecshiftL16:$imm))]> {
9100 let Inst{19-16} = imm;
9103 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
9104 V128, V128, vecshiftL16,
9105 asm#"2", ".4s", ".8h",
9106 [(set (v4i32 V128:$Rd),
9107 (OpNode (extract_high_v8i16 V128:$Rn), vecshiftL16:$imm))]> {
9110 let Inst{19-16} = imm;
9113 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
9114 V128, V64, vecshiftL32, asm, ".2d", ".2s",
9115 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), vecshiftL32:$imm))]> {
9117 let Inst{20-16} = imm;
9120 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
9121 V128, V128, vecshiftL32,
9122 asm#"2", ".2d", ".4s",
9123 [(set (v2i64 V128:$Rd),
9124 (OpNode (extract_high_v4i32 V128:$Rn), vecshiftL32:$imm))]> {
9126 let Inst{20-16} = imm;
9132 // Vector load/store
9134 // SIMD ldX/stX no-index memory references don't allow the optional
9135 // ", #0" constant and handle post-indexing explicitly, so we use
9136 // a more specialized parse method for them. Otherwise, it's the same as
9137 // the general GPR64sp handling.
9139 class BaseSIMDLdSt<bit Q, bit L, bits<4> opcode, bits<2> size,
9140 string asm, dag oops, dag iops, list<dag> pattern>
9141 : I<oops, iops, asm, "\t$Vt, [$Rn]", "", pattern> {
9146 let Inst{29-23} = 0b0011000;
9148 let Inst{21-16} = 0b000000;
9149 let Inst{15-12} = opcode;
9150 let Inst{11-10} = size;
9155 class BaseSIMDLdStPost<bit Q, bit L, bits<4> opcode, bits<2> size,
9156 string asm, dag oops, dag iops>
9157 : I<oops, iops, asm, "\t$Vt, [$Rn], $Xm", "$Rn = $wback", []> {
9163 let Inst{29-23} = 0b0011001;
9166 let Inst{20-16} = Xm;
9167 let Inst{15-12} = opcode;
9168 let Inst{11-10} = size;
9173 // The immediate form of AdvSIMD post-indexed addressing is encoded with
9174 // register post-index addressing from the zero register.
9175 multiclass SIMDLdStAliases<string BaseName, string asm, string layout, string Count,
9176 int Offset, int Size> {
9177 // E.g. "ld1 { v0.8b, v1.8b }, [x1], #16"
9178 // "ld1\t$Vt, [$Rn], #16"
9179 // may get mapped to
9180 // (LD1Twov8b_POST VecListTwo8b:$Vt, GPR64sp:$Rn, XZR)
9181 def : InstAlias<asm # "\t$Vt, [$Rn], #" # Offset,
9182 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
9184 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
9187 // E.g. "ld1.8b { v0, v1 }, [x1], #16"
9188 // "ld1.8b\t$Vt, [$Rn], #16"
9189 // may get mapped to
9190 // (LD1Twov8b_POST VecListTwo64:$Vt, GPR64sp:$Rn, XZR)
9191 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], #" # Offset,
9192 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
9194 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9197 // E.g. "ld1.8b { v0, v1 }, [x1]"
9198 // "ld1\t$Vt, [$Rn]"
9199 // may get mapped to
9200 // (LD1Twov8b VecListTwo64:$Vt, GPR64sp:$Rn)
9201 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn]",
9202 (!cast<Instruction>(BaseName # Count # "v" # layout)
9203 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9206 // E.g. "ld1.8b { v0, v1 }, [x1], x2"
9207 // "ld1\t$Vt, [$Rn], $Xm"
9208 // may get mapped to
9209 // (LD1Twov8b_POST VecListTwo64:$Vt, GPR64sp:$Rn, GPR64pi8:$Xm)
9210 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], $Xm",
9211 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
9213 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9214 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9217 multiclass BaseSIMDLdN<string BaseName, string Count, string asm, string veclist,
9218 int Offset128, int Offset64, bits<4> opcode> {
9219 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
9220 def v16b: BaseSIMDLdSt<1, 1, opcode, 0b00, asm,
9221 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
9222 (ins GPR64sp:$Rn), []>;
9223 def v8h : BaseSIMDLdSt<1, 1, opcode, 0b01, asm,
9224 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
9225 (ins GPR64sp:$Rn), []>;
9226 def v4s : BaseSIMDLdSt<1, 1, opcode, 0b10, asm,
9227 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
9228 (ins GPR64sp:$Rn), []>;
9229 def v2d : BaseSIMDLdSt<1, 1, opcode, 0b11, asm,
9230 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
9231 (ins GPR64sp:$Rn), []>;
9232 def v8b : BaseSIMDLdSt<0, 1, opcode, 0b00, asm,
9233 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
9234 (ins GPR64sp:$Rn), []>;
9235 def v4h : BaseSIMDLdSt<0, 1, opcode, 0b01, asm,
9236 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
9237 (ins GPR64sp:$Rn), []>;
9238 def v2s : BaseSIMDLdSt<0, 1, opcode, 0b10, asm,
9239 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
9240 (ins GPR64sp:$Rn), []>;
9243 def v16b_POST: BaseSIMDLdStPost<1, 1, opcode, 0b00, asm,
9244 (outs GPR64sp:$wback,
9245 !cast<RegisterOperand>(veclist # "16b"):$Vt),
9247 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9248 def v8h_POST : BaseSIMDLdStPost<1, 1, opcode, 0b01, asm,
9249 (outs GPR64sp:$wback,
9250 !cast<RegisterOperand>(veclist # "8h"):$Vt),
9252 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9253 def v4s_POST : BaseSIMDLdStPost<1, 1, opcode, 0b10, asm,
9254 (outs GPR64sp:$wback,
9255 !cast<RegisterOperand>(veclist # "4s"):$Vt),
9257 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9258 def v2d_POST : BaseSIMDLdStPost<1, 1, opcode, 0b11, asm,
9259 (outs GPR64sp:$wback,
9260 !cast<RegisterOperand>(veclist # "2d"):$Vt),
9262 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9263 def v8b_POST : BaseSIMDLdStPost<0, 1, opcode, 0b00, asm,
9264 (outs GPR64sp:$wback,
9265 !cast<RegisterOperand>(veclist # "8b"):$Vt),
9267 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9268 def v4h_POST : BaseSIMDLdStPost<0, 1, opcode, 0b01, asm,
9269 (outs GPR64sp:$wback,
9270 !cast<RegisterOperand>(veclist # "4h"):$Vt),
9272 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9273 def v2s_POST : BaseSIMDLdStPost<0, 1, opcode, 0b10, asm,
9274 (outs GPR64sp:$wback,
9275 !cast<RegisterOperand>(veclist # "2s"):$Vt),
9277 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9280 defm : SIMDLdStAliases<BaseName, asm, "16b", Count, Offset128, 128>;
9281 defm : SIMDLdStAliases<BaseName, asm, "8h", Count, Offset128, 128>;
9282 defm : SIMDLdStAliases<BaseName, asm, "4s", Count, Offset128, 128>;
9283 defm : SIMDLdStAliases<BaseName, asm, "2d", Count, Offset128, 128>;
9284 defm : SIMDLdStAliases<BaseName, asm, "8b", Count, Offset64, 64>;
9285 defm : SIMDLdStAliases<BaseName, asm, "4h", Count, Offset64, 64>;
9286 defm : SIMDLdStAliases<BaseName, asm, "2s", Count, Offset64, 64>;
9289 // Only ld1/st1 has a v1d version.
9290 multiclass BaseSIMDStN<string BaseName, string Count, string asm, string veclist,
9291 int Offset128, int Offset64, bits<4> opcode> {
9292 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in {
9293 def v16b : BaseSIMDLdSt<1, 0, opcode, 0b00, asm, (outs),
9294 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
9296 def v8h : BaseSIMDLdSt<1, 0, opcode, 0b01, asm, (outs),
9297 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
9299 def v4s : BaseSIMDLdSt<1, 0, opcode, 0b10, asm, (outs),
9300 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
9302 def v2d : BaseSIMDLdSt<1, 0, opcode, 0b11, asm, (outs),
9303 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
9305 def v8b : BaseSIMDLdSt<0, 0, opcode, 0b00, asm, (outs),
9306 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
9308 def v4h : BaseSIMDLdSt<0, 0, opcode, 0b01, asm, (outs),
9309 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
9311 def v2s : BaseSIMDLdSt<0, 0, opcode, 0b10, asm, (outs),
9312 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
9315 def v16b_POST : BaseSIMDLdStPost<1, 0, opcode, 0b00, asm,
9316 (outs GPR64sp:$wback),
9317 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
9319 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9320 def v8h_POST : BaseSIMDLdStPost<1, 0, opcode, 0b01, asm,
9321 (outs GPR64sp:$wback),
9322 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
9324 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9325 def v4s_POST : BaseSIMDLdStPost<1, 0, opcode, 0b10, asm,
9326 (outs GPR64sp:$wback),
9327 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
9329 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9330 def v2d_POST : BaseSIMDLdStPost<1, 0, opcode, 0b11, asm,
9331 (outs GPR64sp:$wback),
9332 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
9334 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9335 def v8b_POST : BaseSIMDLdStPost<0, 0, opcode, 0b00, asm,
9336 (outs GPR64sp:$wback),
9337 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
9339 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9340 def v4h_POST : BaseSIMDLdStPost<0, 0, opcode, 0b01, asm,
9341 (outs GPR64sp:$wback),
9342 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
9344 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9345 def v2s_POST : BaseSIMDLdStPost<0, 0, opcode, 0b10, asm,
9346 (outs GPR64sp:$wback),
9347 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
9349 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9352 defm : SIMDLdStAliases<BaseName, asm, "16b", Count, Offset128, 128>;
9353 defm : SIMDLdStAliases<BaseName, asm, "8h", Count, Offset128, 128>;
9354 defm : SIMDLdStAliases<BaseName, asm, "4s", Count, Offset128, 128>;
9355 defm : SIMDLdStAliases<BaseName, asm, "2d", Count, Offset128, 128>;
9356 defm : SIMDLdStAliases<BaseName, asm, "8b", Count, Offset64, 64>;
9357 defm : SIMDLdStAliases<BaseName, asm, "4h", Count, Offset64, 64>;
9358 defm : SIMDLdStAliases<BaseName, asm, "2s", Count, Offset64, 64>;
9361 multiclass BaseSIMDLd1<string BaseName, string Count, string asm, string veclist,
9362 int Offset128, int Offset64, bits<4> opcode>
9363 : BaseSIMDLdN<BaseName, Count, asm, veclist, Offset128, Offset64, opcode> {
9365 // LD1 instructions have extra "1d" variants.
9366 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
9367 def v1d : BaseSIMDLdSt<0, 1, opcode, 0b11, asm,
9368 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
9369 (ins GPR64sp:$Rn), []>;
9371 def v1d_POST : BaseSIMDLdStPost<0, 1, opcode, 0b11, asm,
9372 (outs GPR64sp:$wback,
9373 !cast<RegisterOperand>(veclist # "1d"):$Vt),
9375 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9378 defm : SIMDLdStAliases<BaseName, asm, "1d", Count, Offset64, 64>;
9381 multiclass BaseSIMDSt1<string BaseName, string Count, string asm, string veclist,
9382 int Offset128, int Offset64, bits<4> opcode>
9383 : BaseSIMDStN<BaseName, Count, asm, veclist, Offset128, Offset64, opcode> {
9385 // ST1 instructions have extra "1d" variants.
9386 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in {
9387 def v1d : BaseSIMDLdSt<0, 0, opcode, 0b11, asm, (outs),
9388 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
9391 def v1d_POST : BaseSIMDLdStPost<0, 0, opcode, 0b11, asm,
9392 (outs GPR64sp:$wback),
9393 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
9395 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9398 defm : SIMDLdStAliases<BaseName, asm, "1d", Count, Offset64, 64>;
9401 multiclass SIMDLd1Multiple<string asm> {
9402 defm One : BaseSIMDLd1<NAME, "One", asm, "VecListOne", 16, 8, 0b0111>;
9403 defm Two : BaseSIMDLd1<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1010>;
9404 defm Three : BaseSIMDLd1<NAME, "Three", asm, "VecListThree", 48, 24, 0b0110>;
9405 defm Four : BaseSIMDLd1<NAME, "Four", asm, "VecListFour", 64, 32, 0b0010>;
9408 multiclass SIMDSt1Multiple<string asm> {
9409 defm One : BaseSIMDSt1<NAME, "One", asm, "VecListOne", 16, 8, 0b0111>;
9410 defm Two : BaseSIMDSt1<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1010>;
9411 defm Three : BaseSIMDSt1<NAME, "Three", asm, "VecListThree", 48, 24, 0b0110>;
9412 defm Four : BaseSIMDSt1<NAME, "Four", asm, "VecListFour", 64, 32, 0b0010>;
9415 multiclass SIMDLd2Multiple<string asm> {
9416 defm Two : BaseSIMDLdN<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1000>;
9419 multiclass SIMDSt2Multiple<string asm> {
9420 defm Two : BaseSIMDStN<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1000>;
9423 multiclass SIMDLd3Multiple<string asm> {
9424 defm Three : BaseSIMDLdN<NAME, "Three", asm, "VecListThree", 48, 24, 0b0100>;
9427 multiclass SIMDSt3Multiple<string asm> {
9428 defm Three : BaseSIMDStN<NAME, "Three", asm, "VecListThree", 48, 24, 0b0100>;
9431 multiclass SIMDLd4Multiple<string asm> {
9432 defm Four : BaseSIMDLdN<NAME, "Four", asm, "VecListFour", 64, 32, 0b0000>;
9435 multiclass SIMDSt4Multiple<string asm> {
9436 defm Four : BaseSIMDStN<NAME, "Four", asm, "VecListFour", 64, 32, 0b0000>;
9440 // AdvSIMD Load/store single-element
9443 class BaseSIMDLdStSingle<bit L, bit R, bits<3> opcode,
9444 string asm, string operands, string cst,
9445 dag oops, dag iops, list<dag> pattern>
9446 : I<oops, iops, asm, operands, cst, pattern> {
9450 let Inst{29-24} = 0b001101;
9453 let Inst{15-13} = opcode;
9458 class BaseSIMDLdStSingleTied<bit L, bit R, bits<3> opcode,
9459 string asm, string operands, string cst,
9460 dag oops, dag iops, list<dag> pattern>
9461 : I<oops, iops, asm, operands, "$Vt = $dst," # cst, pattern> {
9465 let Inst{29-24} = 0b001101;
9468 let Inst{15-13} = opcode;
9474 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9475 class BaseSIMDLdR<bit Q, bit R, bits<3> opcode, bit S, bits<2> size, string asm,
9476 DAGOperand listtype>
9477 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, [$Rn]", "",
9478 (outs listtype:$Vt), (ins GPR64sp:$Rn),
9482 let Inst{20-16} = 0b00000;
9484 let Inst{11-10} = size;
9486 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9487 class BaseSIMDLdRPost<bit Q, bit R, bits<3> opcode, bit S, bits<2> size,
9488 string asm, DAGOperand listtype, DAGOperand GPR64pi>
9489 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, [$Rn], $Xm",
9491 (outs GPR64sp:$wback, listtype:$Vt),
9492 (ins GPR64sp:$Rn, GPR64pi:$Xm), []> {
9496 let Inst{20-16} = Xm;
9498 let Inst{11-10} = size;
9501 multiclass SIMDLdrAliases<string BaseName, string asm, string layout, string Count,
9502 int Offset, int Size> {
9503 // E.g. "ld1r { v0.8b }, [x1], #1"
9504 // "ld1r.8b\t$Vt, [$Rn], #1"
9505 // may get mapped to
9506 // (LD1Rv8b_POST VecListOne8b:$Vt, GPR64sp:$Rn, XZR)
9507 def : InstAlias<asm # "\t$Vt, [$Rn], #" # Offset,
9508 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9510 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
9513 // E.g. "ld1r.8b { v0 }, [x1], #1"
9514 // "ld1r.8b\t$Vt, [$Rn], #1"
9515 // may get mapped to
9516 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, XZR)
9517 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], #" # Offset,
9518 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9520 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9523 // E.g. "ld1r.8b { v0 }, [x1]"
9524 // "ld1r.8b\t$Vt, [$Rn]"
9525 // may get mapped to
9526 // (LD1Rv8b VecListOne64:$Vt, GPR64sp:$Rn)
9527 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn]",
9528 (!cast<Instruction>(BaseName # "v" # layout)
9529 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9532 // E.g. "ld1r.8b { v0 }, [x1], x2"
9533 // "ld1r.8b\t$Vt, [$Rn], $Xm"
9534 // may get mapped to
9535 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, GPR64pi1:$Xm)
9536 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], $Xm",
9537 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9539 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9540 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9543 multiclass SIMDLdR<bit R, bits<3> opcode, bit S, string asm, string Count,
9544 int Offset1, int Offset2, int Offset4, int Offset8> {
9545 def v8b : BaseSIMDLdR<0, R, opcode, S, 0b00, asm,
9546 !cast<DAGOperand>("VecList" # Count # "8b")>;
9547 def v16b: BaseSIMDLdR<1, R, opcode, S, 0b00, asm,
9548 !cast<DAGOperand>("VecList" # Count #"16b")>;
9549 def v4h : BaseSIMDLdR<0, R, opcode, S, 0b01, asm,
9550 !cast<DAGOperand>("VecList" # Count #"4h")>;
9551 def v8h : BaseSIMDLdR<1, R, opcode, S, 0b01, asm,
9552 !cast<DAGOperand>("VecList" # Count #"8h")>;
9553 def v2s : BaseSIMDLdR<0, R, opcode, S, 0b10, asm,
9554 !cast<DAGOperand>("VecList" # Count #"2s")>;
9555 def v4s : BaseSIMDLdR<1, R, opcode, S, 0b10, asm,
9556 !cast<DAGOperand>("VecList" # Count #"4s")>;
9557 def v1d : BaseSIMDLdR<0, R, opcode, S, 0b11, asm,
9558 !cast<DAGOperand>("VecList" # Count #"1d")>;
9559 def v2d : BaseSIMDLdR<1, R, opcode, S, 0b11, asm,
9560 !cast<DAGOperand>("VecList" # Count #"2d")>;
9562 def v8b_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b00, asm,
9563 !cast<DAGOperand>("VecList" # Count # "8b"),
9564 !cast<DAGOperand>("GPR64pi" # Offset1)>;
9565 def v16b_POST: BaseSIMDLdRPost<1, R, opcode, S, 0b00, asm,
9566 !cast<DAGOperand>("VecList" # Count # "16b"),
9567 !cast<DAGOperand>("GPR64pi" # Offset1)>;
9568 def v4h_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b01, asm,
9569 !cast<DAGOperand>("VecList" # Count # "4h"),
9570 !cast<DAGOperand>("GPR64pi" # Offset2)>;
9571 def v8h_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b01, asm,
9572 !cast<DAGOperand>("VecList" # Count # "8h"),
9573 !cast<DAGOperand>("GPR64pi" # Offset2)>;
9574 def v2s_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b10, asm,
9575 !cast<DAGOperand>("VecList" # Count # "2s"),
9576 !cast<DAGOperand>("GPR64pi" # Offset4)>;
9577 def v4s_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b10, asm,
9578 !cast<DAGOperand>("VecList" # Count # "4s"),
9579 !cast<DAGOperand>("GPR64pi" # Offset4)>;
9580 def v1d_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b11, asm,
9581 !cast<DAGOperand>("VecList" # Count # "1d"),
9582 !cast<DAGOperand>("GPR64pi" # Offset8)>;
9583 def v2d_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b11, asm,
9584 !cast<DAGOperand>("VecList" # Count # "2d"),
9585 !cast<DAGOperand>("GPR64pi" # Offset8)>;
9587 defm : SIMDLdrAliases<NAME, asm, "8b", Count, Offset1, 64>;
9588 defm : SIMDLdrAliases<NAME, asm, "16b", Count, Offset1, 128>;
9589 defm : SIMDLdrAliases<NAME, asm, "4h", Count, Offset2, 64>;
9590 defm : SIMDLdrAliases<NAME, asm, "8h", Count, Offset2, 128>;
9591 defm : SIMDLdrAliases<NAME, asm, "2s", Count, Offset4, 64>;
9592 defm : SIMDLdrAliases<NAME, asm, "4s", Count, Offset4, 128>;
9593 defm : SIMDLdrAliases<NAME, asm, "1d", Count, Offset8, 64>;
9594 defm : SIMDLdrAliases<NAME, asm, "2d", Count, Offset8, 128>;
9597 class SIMDLdStSingleB<bit L, bit R, bits<3> opcode, string asm,
9598 dag oops, dag iops, list<dag> pattern>
9599 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9601 // idx encoded in Q:S:size fields.
9603 let Inst{30} = idx{3};
9605 let Inst{20-16} = 0b00000;
9606 let Inst{12} = idx{2};
9607 let Inst{11-10} = idx{1-0};
9609 class SIMDLdStSingleBTied<bit L, bit R, bits<3> opcode, string asm,
9610 dag oops, dag iops, list<dag> pattern>
9611 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9612 oops, iops, pattern> {
9613 // idx encoded in Q:S:size fields.
9615 let Inst{30} = idx{3};
9617 let Inst{20-16} = 0b00000;
9618 let Inst{12} = idx{2};
9619 let Inst{11-10} = idx{1-0};
9621 class SIMDLdStSingleBPost<bit L, bit R, bits<3> opcode, string asm,
9623 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9624 "$Rn = $wback", oops, iops, []> {
9625 // idx encoded in Q:S:size fields.
9628 let Inst{30} = idx{3};
9630 let Inst{20-16} = Xm;
9631 let Inst{12} = idx{2};
9632 let Inst{11-10} = idx{1-0};
9634 class SIMDLdStSingleBTiedPost<bit L, bit R, bits<3> opcode, string asm,
9636 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9637 "$Rn = $wback", oops, iops, []> {
9638 // idx encoded in Q:S:size fields.
9641 let Inst{30} = idx{3};
9643 let Inst{20-16} = Xm;
9644 let Inst{12} = idx{2};
9645 let Inst{11-10} = idx{1-0};
9648 class SIMDLdStSingleH<bit L, bit R, bits<3> opcode, bit size, string asm,
9649 dag oops, dag iops, list<dag> pattern>
9650 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9652 // idx encoded in Q:S:size<1> fields.
9654 let Inst{30} = idx{2};
9656 let Inst{20-16} = 0b00000;
9657 let Inst{12} = idx{1};
9658 let Inst{11} = idx{0};
9659 let Inst{10} = size;
9661 class SIMDLdStSingleHTied<bit L, bit R, bits<3> opcode, bit size, string asm,
9662 dag oops, dag iops, list<dag> pattern>
9663 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9664 oops, iops, pattern> {
9665 // idx encoded in Q:S:size<1> fields.
9667 let Inst{30} = idx{2};
9669 let Inst{20-16} = 0b00000;
9670 let Inst{12} = idx{1};
9671 let Inst{11} = idx{0};
9672 let Inst{10} = size;
9675 class SIMDLdStSingleHPost<bit L, bit R, bits<3> opcode, bit size, string asm,
9677 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9678 "$Rn = $wback", oops, iops, []> {
9679 // idx encoded in Q:S:size<1> fields.
9682 let Inst{30} = idx{2};
9684 let Inst{20-16} = Xm;
9685 let Inst{12} = idx{1};
9686 let Inst{11} = idx{0};
9687 let Inst{10} = size;
9689 class SIMDLdStSingleHTiedPost<bit L, bit R, bits<3> opcode, bit size, string asm,
9691 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9692 "$Rn = $wback", oops, iops, []> {
9693 // idx encoded in Q:S:size<1> fields.
9696 let Inst{30} = idx{2};
9698 let Inst{20-16} = Xm;
9699 let Inst{12} = idx{1};
9700 let Inst{11} = idx{0};
9701 let Inst{10} = size;
9703 class SIMDLdStSingleS<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9704 dag oops, dag iops, list<dag> pattern>
9705 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9707 // idx encoded in Q:S fields.
9709 let Inst{30} = idx{1};
9711 let Inst{20-16} = 0b00000;
9712 let Inst{12} = idx{0};
9713 let Inst{11-10} = size;
9715 class SIMDLdStSingleSTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9716 dag oops, dag iops, list<dag> pattern>
9717 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9718 oops, iops, pattern> {
9719 // idx encoded in Q:S fields.
9721 let Inst{30} = idx{1};
9723 let Inst{20-16} = 0b00000;
9724 let Inst{12} = idx{0};
9725 let Inst{11-10} = size;
9727 class SIMDLdStSingleSPost<bit L, bit R, bits<3> opcode, bits<2> size,
9728 string asm, dag oops, dag iops>
9729 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9730 "$Rn = $wback", oops, iops, []> {
9731 // idx encoded in Q:S fields.
9734 let Inst{30} = idx{1};
9736 let Inst{20-16} = Xm;
9737 let Inst{12} = idx{0};
9738 let Inst{11-10} = size;
9740 class SIMDLdStSingleSTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
9741 string asm, dag oops, dag iops>
9742 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9743 "$Rn = $wback", oops, iops, []> {
9744 // idx encoded in Q:S fields.
9747 let Inst{30} = idx{1};
9749 let Inst{20-16} = Xm;
9750 let Inst{12} = idx{0};
9751 let Inst{11-10} = size;
9753 class SIMDLdStSingleD<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9754 dag oops, dag iops, list<dag> pattern>
9755 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9757 // idx encoded in Q field.
9761 let Inst{20-16} = 0b00000;
9763 let Inst{11-10} = size;
9765 class SIMDLdStSingleDTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9766 dag oops, dag iops, list<dag> pattern>
9767 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9768 oops, iops, pattern> {
9769 // idx encoded in Q field.
9773 let Inst{20-16} = 0b00000;
9775 let Inst{11-10} = size;
9777 class SIMDLdStSingleDPost<bit L, bit R, bits<3> opcode, bits<2> size,
9778 string asm, dag oops, dag iops>
9779 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9780 "$Rn = $wback", oops, iops, []> {
9781 // idx encoded in Q field.
9786 let Inst{20-16} = Xm;
9788 let Inst{11-10} = size;
9790 class SIMDLdStSingleDTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
9791 string asm, dag oops, dag iops>
9792 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9793 "$Rn = $wback", oops, iops, []> {
9794 // idx encoded in Q field.
9799 let Inst{20-16} = Xm;
9801 let Inst{11-10} = size;
9804 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9805 multiclass SIMDLdSingleBTied<bit R, bits<3> opcode, string asm,
9806 RegisterOperand listtype,
9807 RegisterOperand GPR64pi> {
9808 def i8 : SIMDLdStSingleBTied<1, R, opcode, asm,
9809 (outs listtype:$dst),
9810 (ins listtype:$Vt, VectorIndexB:$idx,
9813 def i8_POST : SIMDLdStSingleBTiedPost<1, R, opcode, asm,
9814 (outs GPR64sp:$wback, listtype:$dst),
9815 (ins listtype:$Vt, VectorIndexB:$idx,
9816 GPR64sp:$Rn, GPR64pi:$Xm)>;
9818 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9819 multiclass SIMDLdSingleHTied<bit R, bits<3> opcode, bit size, string asm,
9820 RegisterOperand listtype,
9821 RegisterOperand GPR64pi> {
9822 def i16 : SIMDLdStSingleHTied<1, R, opcode, size, asm,
9823 (outs listtype:$dst),
9824 (ins listtype:$Vt, VectorIndexH:$idx,
9827 def i16_POST : SIMDLdStSingleHTiedPost<1, R, opcode, size, asm,
9828 (outs GPR64sp:$wback, listtype:$dst),
9829 (ins listtype:$Vt, VectorIndexH:$idx,
9830 GPR64sp:$Rn, GPR64pi:$Xm)>;
9832 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9833 multiclass SIMDLdSingleSTied<bit R, bits<3> opcode, bits<2> size,string asm,
9834 RegisterOperand listtype,
9835 RegisterOperand GPR64pi> {
9836 def i32 : SIMDLdStSingleSTied<1, R, opcode, size, asm,
9837 (outs listtype:$dst),
9838 (ins listtype:$Vt, VectorIndexS:$idx,
9841 def i32_POST : SIMDLdStSingleSTiedPost<1, R, opcode, size, asm,
9842 (outs GPR64sp:$wback, listtype:$dst),
9843 (ins listtype:$Vt, VectorIndexS:$idx,
9844 GPR64sp:$Rn, GPR64pi:$Xm)>;
9846 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9847 multiclass SIMDLdSingleDTied<bit R, bits<3> opcode, bits<2> size, string asm,
9848 RegisterOperand listtype, RegisterOperand GPR64pi> {
9849 def i64 : SIMDLdStSingleDTied<1, R, opcode, size, asm,
9850 (outs listtype:$dst),
9851 (ins listtype:$Vt, VectorIndexD:$idx,
9854 def i64_POST : SIMDLdStSingleDTiedPost<1, R, opcode, size, asm,
9855 (outs GPR64sp:$wback, listtype:$dst),
9856 (ins listtype:$Vt, VectorIndexD:$idx,
9857 GPR64sp:$Rn, GPR64pi:$Xm)>;
9859 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9860 multiclass SIMDStSingleB<bit R, bits<3> opcode, string asm,
9861 RegisterOperand listtype, RegisterOperand GPR64pi> {
9862 def i8 : SIMDLdStSingleB<0, R, opcode, asm,
9863 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
9866 def i8_POST : SIMDLdStSingleBPost<0, R, opcode, asm,
9867 (outs GPR64sp:$wback),
9868 (ins listtype:$Vt, VectorIndexB:$idx,
9869 GPR64sp:$Rn, GPR64pi:$Xm)>;
9871 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9872 multiclass SIMDStSingleH<bit R, bits<3> opcode, bit size, string asm,
9873 RegisterOperand listtype, RegisterOperand GPR64pi> {
9874 def i16 : SIMDLdStSingleH<0, R, opcode, size, asm,
9875 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
9878 def i16_POST : SIMDLdStSingleHPost<0, R, opcode, size, asm,
9879 (outs GPR64sp:$wback),
9880 (ins listtype:$Vt, VectorIndexH:$idx,
9881 GPR64sp:$Rn, GPR64pi:$Xm)>;
9883 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9884 multiclass SIMDStSingleS<bit R, bits<3> opcode, bits<2> size,string asm,
9885 RegisterOperand listtype, RegisterOperand GPR64pi> {
9886 def i32 : SIMDLdStSingleS<0, R, opcode, size, asm,
9887 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
9890 def i32_POST : SIMDLdStSingleSPost<0, R, opcode, size, asm,
9891 (outs GPR64sp:$wback),
9892 (ins listtype:$Vt, VectorIndexS:$idx,
9893 GPR64sp:$Rn, GPR64pi:$Xm)>;
9895 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9896 multiclass SIMDStSingleD<bit R, bits<3> opcode, bits<2> size, string asm,
9897 RegisterOperand listtype, RegisterOperand GPR64pi> {
9898 def i64 : SIMDLdStSingleD<0, R, opcode, size, asm,
9899 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
9902 def i64_POST : SIMDLdStSingleDPost<0, R, opcode, size, asm,
9903 (outs GPR64sp:$wback),
9904 (ins listtype:$Vt, VectorIndexD:$idx,
9905 GPR64sp:$Rn, GPR64pi:$Xm)>;
9908 multiclass SIMDLdStSingleAliases<string asm, string layout, string Type,
9909 string Count, int Offset, Operand idxtype> {
9910 // E.g. "ld1 { v0.8b }[0], [x1], #1"
9911 // "ld1\t$Vt, [$Rn], #1"
9912 // may get mapped to
9913 // (LD1Rv8b_POST VecListOne8b:$Vt, GPR64sp:$Rn, XZR)
9914 def : InstAlias<asm # "\t$Vt$idx, [$Rn], #" # Offset,
9915 (!cast<Instruction>(NAME # Type # "_POST")
9917 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
9918 idxtype:$idx, XZR), 1>;
9920 // E.g. "ld1.8b { v0 }[0], [x1], #1"
9921 // "ld1.8b\t$Vt, [$Rn], #1"
9922 // may get mapped to
9923 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, XZR)
9924 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn], #" # Offset,
9925 (!cast<Instruction>(NAME # Type # "_POST")
9927 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9928 idxtype:$idx, XZR), 0>;
9930 // E.g. "ld1.8b { v0 }[0], [x1]"
9931 // "ld1.8b\t$Vt, [$Rn]"
9932 // may get mapped to
9933 // (LD1Rv8b VecListOne64:$Vt, GPR64sp:$Rn)
9934 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn]",
9935 (!cast<Instruction>(NAME # Type)
9936 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9937 idxtype:$idx, GPR64sp:$Rn), 0>;
9939 // E.g. "ld1.8b { v0 }[0], [x1], x2"
9940 // "ld1.8b\t$Vt, [$Rn], $Xm"
9941 // may get mapped to
9942 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, GPR64pi1:$Xm)
9943 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn], $Xm",
9944 (!cast<Instruction>(NAME # Type # "_POST")
9946 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9948 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9951 multiclass SIMDLdSt1SingleAliases<string asm> {
9952 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "One", 1, VectorIndexB>;
9953 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "One", 2, VectorIndexH>;
9954 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "One", 4, VectorIndexS>;
9955 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "One", 8, VectorIndexD>;
9958 multiclass SIMDLdSt2SingleAliases<string asm> {
9959 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Two", 2, VectorIndexB>;
9960 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Two", 4, VectorIndexH>;
9961 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Two", 8, VectorIndexS>;
9962 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Two", 16, VectorIndexD>;
9965 multiclass SIMDLdSt3SingleAliases<string asm> {
9966 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Three", 3, VectorIndexB>;
9967 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Three", 6, VectorIndexH>;
9968 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Three", 12, VectorIndexS>;
9969 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Three", 24, VectorIndexD>;
9972 multiclass SIMDLdSt4SingleAliases<string asm> {
9973 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Four", 4, VectorIndexB>;
9974 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Four", 8, VectorIndexH>;
9975 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Four", 16, VectorIndexS>;
9976 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Four", 32, VectorIndexD>;
9978 } // end of 'let Predicates = [HasNEON]'
9980 //----------------------------------------------------------------------------
9981 // AdvSIMD v8.1 Rounding Double Multiply Add/Subtract
9982 //----------------------------------------------------------------------------
9984 let Predicates = [HasNEON, HasRDM] in {
9986 class BaseSIMDThreeSameVectorTiedR0<bit Q, bit U, bits<2> size, bits<5> opcode,
9987 RegisterOperand regtype, string asm,
9988 string kind, list<dag> pattern>
9989 : BaseSIMDThreeSameVectorTied<Q, U, {size,0}, opcode, regtype, asm, kind,
9992 multiclass SIMDThreeSameVectorSQRDMLxHTiedHS<bit U, bits<5> opc, string asm,
9993 SDPatternOperator Accum> {
9994 def v4i16 : BaseSIMDThreeSameVectorTiedR0<0, U, 0b01, opc, V64, asm, ".4h",
9995 [(set (v4i16 V64:$dst),
9996 (Accum (v4i16 V64:$Rd),
9997 (v4i16 (int_aarch64_neon_sqrdmulh (v4i16 V64:$Rn),
9998 (v4i16 V64:$Rm)))))]>;
9999 def v8i16 : BaseSIMDThreeSameVectorTiedR0<1, U, 0b01, opc, V128, asm, ".8h",
10000 [(set (v8i16 V128:$dst),
10001 (Accum (v8i16 V128:$Rd),
10002 (v8i16 (int_aarch64_neon_sqrdmulh (v8i16 V128:$Rn),
10003 (v8i16 V128:$Rm)))))]>;
10004 def v2i32 : BaseSIMDThreeSameVectorTiedR0<0, U, 0b10, opc, V64, asm, ".2s",
10005 [(set (v2i32 V64:$dst),
10006 (Accum (v2i32 V64:$Rd),
10007 (v2i32 (int_aarch64_neon_sqrdmulh (v2i32 V64:$Rn),
10008 (v2i32 V64:$Rm)))))]>;
10009 def v4i32 : BaseSIMDThreeSameVectorTiedR0<1, U, 0b10, opc, V128, asm, ".4s",
10010 [(set (v4i32 V128:$dst),
10011 (Accum (v4i32 V128:$Rd),
10012 (v4i32 (int_aarch64_neon_sqrdmulh (v4i32 V128:$Rn),
10013 (v4i32 V128:$Rm)))))]>;
10016 multiclass SIMDIndexedSQRDMLxHSDTied<bit U, bits<4> opc, string asm,
10017 SDPatternOperator Accum> {
10018 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
10019 V64, V64, V128_lo, VectorIndexH,
10020 asm, ".4h", ".4h", ".4h", ".h",
10021 [(set (v4i16 V64:$dst),
10022 (Accum (v4i16 V64:$Rd),
10023 (v4i16 (int_aarch64_neon_sqrdmulh
10025 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
10026 VectorIndexH:$idx))))))]> {
10028 let Inst{11} = idx{2};
10029 let Inst{21} = idx{1};
10030 let Inst{20} = idx{0};
10033 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
10034 V128, V128, V128_lo, VectorIndexH,
10035 asm, ".8h", ".8h", ".8h", ".h",
10036 [(set (v8i16 V128:$dst),
10037 (Accum (v8i16 V128:$Rd),
10038 (v8i16 (int_aarch64_neon_sqrdmulh
10040 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
10041 VectorIndexH:$idx))))))]> {
10043 let Inst{11} = idx{2};
10044 let Inst{21} = idx{1};
10045 let Inst{20} = idx{0};
10048 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
10049 V64, V64, V128, VectorIndexS,
10050 asm, ".2s", ".2s", ".2s", ".s",
10051 [(set (v2i32 V64:$dst),
10052 (Accum (v2i32 V64:$Rd),
10053 (v2i32 (int_aarch64_neon_sqrdmulh
10055 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm),
10056 VectorIndexS:$idx))))))]> {
10058 let Inst{11} = idx{1};
10059 let Inst{21} = idx{0};
10062 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but
10063 // an intermediate EXTRACT_SUBREG would be untyped.
10064 // FIXME: direct EXTRACT_SUBREG from v2i32 to i32 is illegal, that's why we
10065 // got it lowered here as (i32 vector_extract (v4i32 insert_subvector(..)))
10066 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
10067 (i32 (vector_extract
10068 (v4i32 (insert_subvector
10070 (v2i32 (int_aarch64_neon_sqrdmulh
10072 (v2i32 (AArch64duplane32
10074 VectorIndexS:$idx)))),
10078 (v2i32 (!cast<Instruction>(NAME # v2i32_indexed)
10079 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
10084 VectorIndexS:$idx)),
10087 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
10088 V128, V128, V128, VectorIndexS,
10089 asm, ".4s", ".4s", ".4s", ".s",
10090 [(set (v4i32 V128:$dst),
10091 (Accum (v4i32 V128:$Rd),
10092 (v4i32 (int_aarch64_neon_sqrdmulh
10094 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
10095 VectorIndexS:$idx))))))]> {
10097 let Inst{11} = idx{1};
10098 let Inst{21} = idx{0};
10101 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but
10102 // an intermediate EXTRACT_SUBREG would be untyped.
10103 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
10104 (i32 (vector_extract
10105 (v4i32 (int_aarch64_neon_sqrdmulh
10107 (v4i32 (AArch64duplane32
10109 VectorIndexS:$idx)))),
10112 (v4i32 (!cast<Instruction>(NAME # v4i32_indexed)
10113 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
10118 VectorIndexS:$idx)),
10121 def i16_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
10122 FPR16Op, FPR16Op, V128_lo,
10123 VectorIndexH, asm, ".h", "", "", ".h",
10126 let Inst{11} = idx{2};
10127 let Inst{21} = idx{1};
10128 let Inst{20} = idx{0};
10131 def i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
10132 FPR32Op, FPR32Op, V128, VectorIndexS,
10133 asm, ".s", "", "", ".s",
10134 [(set (i32 FPR32Op:$dst),
10135 (Accum (i32 FPR32Op:$Rd),
10136 (i32 (int_aarch64_neon_sqrdmulh
10138 (i32 (vector_extract (v4i32 V128:$Rm),
10139 VectorIndexS:$idx))))))]> {
10141 let Inst{11} = idx{1};
10142 let Inst{21} = idx{0};
10145 } // let Predicates = [HasNeon, HasRDM]
10147 //----------------------------------------------------------------------------
10148 // ARMv8.3 Complex ADD/MLA instructions
10149 //----------------------------------------------------------------------------
10151 class ComplexRotationOperand<int Angle, int Remainder, string Type>
10152 : AsmOperandClass {
10153 let PredicateMethod = "isComplexRotation<" # Angle # ", " # Remainder # ">";
10154 let DiagnosticType = "InvalidComplexRotation" # Type;
10155 let Name = "ComplexRotation" # Type;
10157 def complexrotateop : Operand<i32>, ImmLeaf<i32, [{ return Imm >= 0 && Imm <= 270; }],
10158 SDNodeXForm<imm, [{
10159 return CurDAG->getTargetConstant((N->getSExtValue() / 90), SDLoc(N), MVT::i32);
10161 let ParserMatchClass = ComplexRotationOperand<90, 0, "Even">;
10162 let PrintMethod = "printComplexRotationOp<90, 0>";
10164 def complexrotateopodd : Operand<i32>, ImmLeaf<i32, [{ return Imm >= 0 && Imm <= 270; }],
10165 SDNodeXForm<imm, [{
10166 return CurDAG->getTargetConstant(((N->getSExtValue() - 90) / 180), SDLoc(N), MVT::i32);
10168 let ParserMatchClass = ComplexRotationOperand<180, 90, "Odd">;
10169 let PrintMethod = "printComplexRotationOp<180, 90>";
10171 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10172 class BaseSIMDThreeSameVectorComplex<bit Q, bit U, bits<2> size, bits<3> opcode,
10173 RegisterOperand regtype, Operand rottype,
10174 string asm, string kind, list<dag> pattern>
10175 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, rottype:$rot), asm,
10176 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $rot"
10177 "|" # kind # "\t$Rd, $Rn, $Rm, $rot}", "", pattern>,
10186 let Inst{28-24} = 0b01110;
10187 let Inst{23-22} = size;
10189 let Inst{20-16} = Rm;
10190 let Inst{15-13} = opcode;
10191 // Non-tied version (FCADD) only has one rotation bit
10192 let Inst{12} = rot;
10195 let Inst{9-5} = Rn;
10196 let Inst{4-0} = Rd;
10199 //8.3 CompNum - Floating-point complex number support
10200 multiclass SIMDThreeSameVectorComplexHSD<bit U, bits<3> opcode, Operand rottype,
10201 string asm, SDPatternOperator OpNode>{
10202 let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
10203 def v4f16 : BaseSIMDThreeSameVectorComplex<0, U, 0b01, opcode, V64, rottype,
10205 [(set (v4f16 V64:$dst), (OpNode (v4f16 V64:$Rd),
10208 (rottype i32:$rot)))]>;
10210 def v8f16 : BaseSIMDThreeSameVectorComplex<1, U, 0b01, opcode, V128, rottype,
10212 [(set (v8f16 V128:$dst), (OpNode (v8f16 V128:$Rd),
10215 (rottype i32:$rot)))]>;
10218 let Predicates = [HasComplxNum, HasNEON] in {
10219 def v2f32 : BaseSIMDThreeSameVectorComplex<0, U, 0b10, opcode, V64, rottype,
10221 [(set (v2f32 V64:$dst), (OpNode (v2f32 V64:$Rd),
10224 (rottype i32:$rot)))]>;
10226 def v4f32 : BaseSIMDThreeSameVectorComplex<1, U, 0b10, opcode, V128, rottype,
10228 [(set (v4f32 V128:$dst), (OpNode (v4f32 V128:$Rd),
10231 (rottype i32:$rot)))]>;
10233 def v2f64 : BaseSIMDThreeSameVectorComplex<1, U, 0b11, opcode, V128, rottype,
10235 [(set (v2f64 V128:$dst), (OpNode (v2f64 V128:$Rd),
10238 (rottype i32:$rot)))]>;
10242 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10243 class BaseSIMDThreeSameVectorTiedComplex<bit Q, bit U, bits<2> size,
10245 RegisterOperand regtype,
10246 Operand rottype, string asm,
10247 string kind, list<dag> pattern>
10248 : I<(outs regtype:$dst),
10249 (ins regtype:$Rd, regtype:$Rn, regtype:$Rm, rottype:$rot), asm,
10250 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $rot"
10251 "|" # kind # "\t$Rd, $Rn, $Rm, $rot}", "$Rd = $dst", pattern>,
10260 let Inst{28-24} = 0b01110;
10261 let Inst{23-22} = size;
10263 let Inst{20-16} = Rm;
10264 let Inst{15-13} = opcode;
10265 let Inst{12-11} = rot;
10267 let Inst{9-5} = Rn;
10268 let Inst{4-0} = Rd;
10271 multiclass SIMDThreeSameVectorTiedComplexHSD<bit U, bits<3> opcode,
10272 Operand rottype, string asm,
10273 SDPatternOperator OpNode> {
10274 let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
10275 def v4f16 : BaseSIMDThreeSameVectorTiedComplex<0, U, 0b01, opcode, V64,
10276 rottype, asm, ".4h",
10277 [(set (v4f16 V64:$dst), (OpNode (v4f16 V64:$Rd),
10280 (rottype i32:$rot)))]>;
10282 def v8f16 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b01, opcode, V128,
10283 rottype, asm, ".8h",
10284 [(set (v8f16 V128:$dst), (OpNode (v8f16 V128:$Rd),
10287 (rottype i32:$rot)))]>;
10290 let Predicates = [HasComplxNum, HasNEON] in {
10291 def v2f32 : BaseSIMDThreeSameVectorTiedComplex<0, U, 0b10, opcode, V64,
10292 rottype, asm, ".2s",
10293 [(set (v2f32 V64:$dst), (OpNode (v2f32 V64:$Rd),
10296 (rottype i32:$rot)))]>;
10298 def v4f32 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b10, opcode, V128,
10299 rottype, asm, ".4s",
10300 [(set (v4f32 V128:$dst), (OpNode (v4f32 V128:$Rd),
10303 (rottype i32:$rot)))]>;
10305 def v2f64 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b11, opcode, V128,
10306 rottype, asm, ".2d",
10307 [(set (v2f64 V128:$dst), (OpNode (v2f64 V128:$Rd),
10310 (rottype i32:$rot)))]>;
10314 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10315 class BaseSIMDIndexedTiedComplex<bit Q, bit U, bit Scalar, bits<2> size,
10316 bit opc1, bit opc2, RegisterOperand dst_reg,
10317 RegisterOperand lhs_reg,
10318 RegisterOperand rhs_reg, Operand vec_idx,
10319 Operand rottype, string asm, string apple_kind,
10320 string dst_kind, string lhs_kind,
10321 string rhs_kind, list<dag> pattern>
10322 : I<(outs dst_reg:$dst),
10323 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx, rottype:$rot),
10325 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind #
10326 "$idx, $rot" # "|" # apple_kind #
10327 "\t$Rd, $Rn, $Rm$idx, $rot}", "$Rd = $dst", pattern>,
10337 let Inst{28} = Scalar;
10338 let Inst{27-24} = 0b1111;
10339 let Inst{23-22} = size;
10340 // Bit 21 must be set by the derived class.
10341 let Inst{20-16} = Rm;
10342 let Inst{15} = opc1;
10343 let Inst{14-13} = rot;
10344 let Inst{12} = opc2;
10345 // Bit 11 must be set by the derived class.
10347 let Inst{9-5} = Rn;
10348 let Inst{4-0} = Rd;
10351 // The complex instructions index by pairs of elements, so the VectorIndexes
10352 // don't match the lane types, and the index bits are different to the other
10354 multiclass SIMDIndexedTiedComplexHSD<bit U, bit opc1, bit opc2, Operand rottype,
10355 string asm, SDPatternOperator OpNode> {
10356 let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
10357 def v4f16_indexed : BaseSIMDIndexedTiedComplex<0, 1, 0, 0b01, opc1, opc2, V64,
10358 V64, V128, VectorIndexD, rottype, asm, ".4h", ".4h",
10362 let Inst{21} = idx{0};
10365 def v8f16_indexed : BaseSIMDIndexedTiedComplex<1, 1, 0, 0b01, opc1, opc2,
10366 V128, V128, V128, VectorIndexS, rottype, asm, ".8h",
10367 ".8h", ".8h", ".h", []> {
10369 let Inst{11} = idx{1};
10370 let Inst{21} = idx{0};
10372 } // Predicates = HasComplxNum, HasNEON, HasFullFP16]
10374 let Predicates = [HasComplxNum, HasNEON] in {
10375 def v4f32_indexed : BaseSIMDIndexedTiedComplex<1, 1, 0, 0b10, opc1, opc2,
10376 V128, V128, V128, VectorIndexD, rottype, asm, ".4s",
10377 ".4s", ".4s", ".s", []> {
10379 let Inst{11} = idx{0};
10382 } // Predicates = [HasComplxNum, HasNEON]
10385 //----------------------------------------------------------------------------
10386 // Crypto extensions
10387 //----------------------------------------------------------------------------
10389 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10390 class AESBase<bits<4> opc, string asm, dag outs, dag ins, string cstr,
10392 : I<outs, ins, asm, "{\t$Rd.16b, $Rn.16b|.16b\t$Rd, $Rn}", cstr, pat>,
10396 let Inst{31-16} = 0b0100111000101000;
10397 let Inst{15-12} = opc;
10398 let Inst{11-10} = 0b10;
10399 let Inst{9-5} = Rn;
10400 let Inst{4-0} = Rd;
10403 class AESInst<bits<4> opc, string asm, Intrinsic OpNode>
10404 : AESBase<opc, asm, (outs V128:$Rd), (ins V128:$Rn), "",
10405 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
10407 class AESTiedInst<bits<4> opc, string asm, Intrinsic OpNode>
10408 : AESBase<opc, asm, (outs V128:$dst), (ins V128:$Rd, V128:$Rn),
10410 [(set (v16i8 V128:$dst),
10411 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
10413 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10414 class SHA3OpTiedInst<bits<3> opc, string asm, string dst_lhs_kind,
10415 dag oops, dag iops, list<dag> pat>
10416 : I<oops, iops, asm,
10417 "{\t$Rd" # dst_lhs_kind # ", $Rn" # dst_lhs_kind # ", $Rm.4s" #
10418 "|.4s\t$Rd, $Rn, $Rm}", "$Rd = $dst", pat>,
10423 let Inst{31-21} = 0b01011110000;
10424 let Inst{20-16} = Rm;
10426 let Inst{14-12} = opc;
10427 let Inst{11-10} = 0b00;
10428 let Inst{9-5} = Rn;
10429 let Inst{4-0} = Rd;
10432 class SHATiedInstQSV<bits<3> opc, string asm, Intrinsic OpNode>
10433 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
10434 (ins FPR128:$Rd, FPR32:$Rn, V128:$Rm),
10435 [(set (v4i32 FPR128:$dst),
10436 (OpNode (v4i32 FPR128:$Rd), (i32 FPR32:$Rn),
10437 (v4i32 V128:$Rm)))]>;
10439 class SHATiedInstVVV<bits<3> opc, string asm, Intrinsic OpNode>
10440 : SHA3OpTiedInst<opc, asm, ".4s", (outs V128:$dst),
10441 (ins V128:$Rd, V128:$Rn, V128:$Rm),
10442 [(set (v4i32 V128:$dst),
10443 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
10444 (v4i32 V128:$Rm)))]>;
10446 class SHATiedInstQQV<bits<3> opc, string asm, Intrinsic OpNode>
10447 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
10448 (ins FPR128:$Rd, FPR128:$Rn, V128:$Rm),
10449 [(set (v4i32 FPR128:$dst),
10450 (OpNode (v4i32 FPR128:$Rd), (v4i32 FPR128:$Rn),
10451 (v4i32 V128:$Rm)))]>;
10453 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10454 class SHA2OpInst<bits<4> opc, string asm, string kind,
10455 string cstr, dag oops, dag iops,
10457 : I<oops, iops, asm, "{\t$Rd" # kind # ", $Rn" # kind #
10458 "|" # kind # "\t$Rd, $Rn}", cstr, pat>,
10462 let Inst{31-16} = 0b0101111000101000;
10463 let Inst{15-12} = opc;
10464 let Inst{11-10} = 0b10;
10465 let Inst{9-5} = Rn;
10466 let Inst{4-0} = Rd;
10469 class SHATiedInstVV<bits<4> opc, string asm, Intrinsic OpNode>
10470 : SHA2OpInst<opc, asm, ".4s", "$Rd = $dst", (outs V128:$dst),
10471 (ins V128:$Rd, V128:$Rn),
10472 [(set (v4i32 V128:$dst),
10473 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
10475 class SHAInstSS<bits<4> opc, string asm, Intrinsic OpNode>
10476 : SHA2OpInst<opc, asm, "", "", (outs FPR32:$Rd), (ins FPR32:$Rn),
10477 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
10479 // Armv8.2-A Crypto extensions
10480 class BaseCryptoV82<dag oops, dag iops, string asm, string asmops, string cst,
10482 : I <oops, iops, asm, asmops, cst, pattern>, Sched<[WriteV]> {
10485 let Inst{31-25} = 0b1100111;
10486 let Inst{9-5} = Vn;
10487 let Inst{4-0} = Vd;
10490 class CryptoRRTied<bits<1>op0, bits<2>op1, string asm, string asmops>
10491 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm, asmops,
10493 let Inst{31-25} = 0b1100111;
10494 let Inst{24-21} = 0b0110;
10495 let Inst{20-15} = 0b000001;
10496 let Inst{14} = op0;
10497 let Inst{13-12} = 0b00;
10498 let Inst{11-10} = op1;
10500 class CryptoRRTied_2D<bits<1>op0, bits<2>op1, string asm>
10501 : CryptoRRTied<op0, op1, asm, "{\t$Vd.2d, $Vn.2d|.2d\t$Vd, $Vn}">;
10502 class CryptoRRTied_4S<bits<1>op0, bits<2>op1, string asm>
10503 : CryptoRRTied<op0, op1, asm, "{\t$Vd.4s, $Vn.4s|.4s\t$Vd, $Vn}">;
10505 class CryptoRRR<bits<1> op0, bits<2>op1, dag oops, dag iops, string asm,
10506 string asmops, string cst>
10507 : BaseCryptoV82<oops, iops, asm , asmops, cst, []> {
10509 let Inst{24-21} = 0b0011;
10510 let Inst{20-16} = Vm;
10511 let Inst{15} = 0b1;
10512 let Inst{14} = op0;
10513 let Inst{13-12} = 0b00;
10514 let Inst{11-10} = op1;
10516 class CryptoRRR_2D<bits<1> op0, bits<2>op1, string asm>
10517 : CryptoRRR<op0, op1, (outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm,
10518 "{\t$Vd.2d, $Vn.2d, $Vm.2d|.2d\t$Vd, $Vn, $Vm}", "">;
10519 class CryptoRRRTied_2D<bits<1> op0, bits<2>op1, string asm>
10520 : CryptoRRR<op0, op1, (outs V128:$Vdst), (ins V128:$Vd, V128:$Vn, V128:$Vm), asm,
10521 "{\t$Vd.2d, $Vn.2d, $Vm.2d|.2d\t$Vd, $Vn, $Vm}", "$Vd = $Vdst">;
10522 class CryptoRRR_4S<bits<1> op0, bits<2>op1, string asm>
10523 : CryptoRRR<op0, op1, (outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm,
10524 "{\t$Vd.4s, $Vn.4s, $Vm.4s|.4s\t$Vd, $Vn, $Vm}", "">;
10525 class CryptoRRRTied_4S<bits<1> op0, bits<2>op1, string asm>
10526 : CryptoRRR<op0, op1, (outs V128:$Vdst), (ins V128:$Vd, V128:$Vn, V128:$Vm), asm,
10527 "{\t$Vd.4s, $Vn.4s, $Vm.4s|.4s\t$Vd, $Vn, $Vm}", "$Vd = $Vdst">;
10528 class CryptoRRRTied<bits<1> op0, bits<2>op1, string asm>
10529 : CryptoRRR<op0, op1, (outs FPR128:$Vdst), (ins FPR128:$Vd, FPR128:$Vn, V128:$Vm),
10530 asm, "{\t$Vd, $Vn, $Vm.2d|.2d\t$Vd, $Vn, $Vm}", "$Vd = $Vdst">;
10532 class CryptoRRRR<bits<2>op0, string asm, string asmops>
10533 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm, V128:$Va), asm,
10537 let Inst{24-23} = 0b00;
10538 let Inst{22-21} = op0;
10539 let Inst{20-16} = Vm;
10540 let Inst{15} = 0b0;
10541 let Inst{14-10} = Va;
10543 class CryptoRRRR_16B<bits<2>op0, string asm>
10544 : CryptoRRRR<op0, asm, "{\t$Vd.16b, $Vn.16b, $Vm.16b, $Va.16b" #
10545 "|.16b\t$Vd, $Vn, $Vm, $Va}"> {
10547 class CryptoRRRR_4S<bits<2>op0, string asm>
10548 : CryptoRRRR<op0, asm, "{\t$Vd.4s, $Vn.4s, $Vm.4s, $Va.4s" #
10549 "|.4s\t$Vd, $Vn, $Vm, $Va}"> {
10552 class CryptoRRRi6<string asm>
10553 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm, uimm6:$imm), asm,
10554 "{\t$Vd.2d, $Vn.2d, $Vm.2d, $imm" #
10555 "|.2d\t$Vd, $Vn, $Vm, $imm}", "", []> {
10558 let Inst{24-21} = 0b0100;
10559 let Inst{20-16} = Vm;
10560 let Inst{15-10} = imm;
10561 let Inst{9-5} = Vn;
10562 let Inst{4-0} = Vd;
10565 class CryptoRRRi2Tied<bits<1>op0, bits<2>op1, string asm>
10566 : BaseCryptoV82<(outs V128:$Vdst),
10567 (ins V128:$Vd, V128:$Vn, V128:$Vm, VectorIndexS:$imm),
10568 asm, "{\t$Vd.4s, $Vn.4s, $Vm.s$imm" #
10569 "|.4s\t$Vd, $Vn, $Vm$imm}", "$Vd = $Vdst", []> {
10572 let Inst{24-21} = 0b0010;
10573 let Inst{20-16} = Vm;
10574 let Inst{15} = 0b1;
10575 let Inst{14} = op0;
10576 let Inst{13-12} = imm;
10577 let Inst{11-10} = op1;
10580 //----------------------------------------------------------------------------
10581 // v8.1 atomic instructions extension:
10585 // * LDOPregister<OP>, and aliases STOPregister<OP>
10587 // Instruction encodings:
10589 // 31 30|29 24|23|22|21|20 16|15|14 10|9 5|4 0
10590 // CAS SZ |001000|1 |A |1 |Rs |R |11111 |Rn |Rt
10591 // CASP 0|SZ|001000|0 |A |1 |Rs |R |11111 |Rn |Rt
10592 // SWP SZ |111000|A |R |1 |Rs |1 |OPC|00|Rn |Rt
10593 // LD SZ |111000|A |R |1 |Rs |0 |OPC|00|Rn |Rt
10594 // ST SZ |111000|A |R |1 |Rs |0 |OPC|00|Rn |11111
10596 // Instruction syntax:
10598 // CAS{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10599 // CAS{<order>} <Xs>, <Xt>, [<Xn|SP>]
10600 // CASP{<order>} <Ws>, <W(s+1)>, <Wt>, <W(t+1)>, [<Xn|SP>]
10601 // CASP{<order>} <Xs>, <X(s+1)>, <Xt>, <X(t+1)>, [<Xn|SP>]
10602 // SWP{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10603 // SWP{<order>} <Xs>, <Xt>, [<Xn|SP>]
10604 // LD<OP>{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10605 // LD<OP>{<order>} <Xs>, <Xt>, [<Xn|SP>]
10606 // ST<OP>{<order>}[<size>] <Ws>, [<Xn|SP>]
10607 // ST<OP>{<order>} <Xs>, [<Xn|SP>]
10609 let Predicates = [HasLSE], mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
10610 class BaseCASEncoding<dag oops, dag iops, string asm, string operands,
10611 string cstr, list<dag> pattern>
10612 : I<oops, iops, asm, operands, cstr, pattern> {
10620 let Inst{31-30} = Sz;
10621 let Inst{29-24} = 0b001000;
10623 let Inst{22} = Acq;
10624 let Inst{21} = 0b1;
10625 let Inst{20-16} = Rs;
10626 let Inst{15} = Rel;
10627 let Inst{14-10} = 0b11111;
10628 let Inst{9-5} = Rn;
10629 let Inst{4-0} = Rt;
10630 let Predicates = [HasLSE];
10633 class BaseCAS<string order, string size, RegisterClass RC>
10634 : BaseCASEncoding<(outs RC:$out),(ins RC:$Rs, RC:$Rt, GPR64sp:$Rn),
10635 "cas" # order # size, "\t$Rs, $Rt, [$Rn]",
10637 Sched<[WriteAtomic]> {
10641 multiclass CompareAndSwap<bits<1> Acq, bits<1> Rel, string order> {
10642 let Sz = 0b00, Acq = Acq, Rel = Rel in def B : BaseCAS<order, "b", GPR32>;
10643 let Sz = 0b01, Acq = Acq, Rel = Rel in def H : BaseCAS<order, "h", GPR32>;
10644 let Sz = 0b10, Acq = Acq, Rel = Rel in def W : BaseCAS<order, "", GPR32>;
10645 let Sz = 0b11, Acq = Acq, Rel = Rel in def X : BaseCAS<order, "", GPR64>;
10648 class BaseCASP<string order, string size, RegisterOperand RC>
10649 : BaseCASEncoding<(outs RC:$out),(ins RC:$Rs, RC:$Rt, GPR64sp:$Rn),
10650 "casp" # order # size, "\t$Rs, $Rt, [$Rn]",
10652 Sched<[WriteAtomic]> {
10656 multiclass CompareAndSwapPair<bits<1> Acq, bits<1> Rel, string order> {
10657 let Sz = 0b00, Acq = Acq, Rel = Rel in
10658 def W : BaseCASP<order, "", WSeqPairClassOperand>;
10659 let Sz = 0b01, Acq = Acq, Rel = Rel in
10660 def X : BaseCASP<order, "", XSeqPairClassOperand>;
10663 let Predicates = [HasLSE] in
10664 class BaseSWP<string order, string size, RegisterClass RC>
10665 : I<(outs RC:$Rt),(ins RC:$Rs, GPR64sp:$Rn), "swp" # order # size,
10666 "\t$Rs, $Rt, [$Rn]","",[]>,
10667 Sched<[WriteAtomic]> {
10672 bits<3> opc = 0b000;
10675 let Inst{31-30} = Sz;
10676 let Inst{29-24} = 0b111000;
10677 let Inst{23} = Acq;
10678 let Inst{22} = Rel;
10679 let Inst{21} = 0b1;
10680 let Inst{20-16} = Rs;
10681 let Inst{15} = 0b1;
10682 let Inst{14-12} = opc;
10683 let Inst{11-10} = 0b00;
10684 let Inst{9-5} = Rn;
10685 let Inst{4-0} = Rt;
10686 let Predicates = [HasLSE];
10689 multiclass Swap<bits<1> Acq, bits<1> Rel, string order> {
10690 let Sz = 0b00, Acq = Acq, Rel = Rel in def B : BaseSWP<order, "b", GPR32>;
10691 let Sz = 0b01, Acq = Acq, Rel = Rel in def H : BaseSWP<order, "h", GPR32>;
10692 let Sz = 0b10, Acq = Acq, Rel = Rel in def W : BaseSWP<order, "", GPR32>;
10693 let Sz = 0b11, Acq = Acq, Rel = Rel in def X : BaseSWP<order, "", GPR64>;
10696 let Predicates = [HasLSE], mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
10697 class BaseLDOPregister<string op, string order, string size, RegisterClass RC>
10698 : I<(outs RC:$Rt),(ins RC:$Rs, GPR64sp:$Rn), "ld" # op # order # size,
10699 "\t$Rs, $Rt, [$Rn]","",[]>,
10700 Sched<[WriteAtomic]> {
10708 let Inst{31-30} = Sz;
10709 let Inst{29-24} = 0b111000;
10710 let Inst{23} = Acq;
10711 let Inst{22} = Rel;
10712 let Inst{21} = 0b1;
10713 let Inst{20-16} = Rs;
10714 let Inst{15} = 0b0;
10715 let Inst{14-12} = opc;
10716 let Inst{11-10} = 0b00;
10717 let Inst{9-5} = Rn;
10718 let Inst{4-0} = Rt;
10719 let Predicates = [HasLSE];
10722 multiclass LDOPregister<bits<3> opc, string op, bits<1> Acq, bits<1> Rel,
10724 let Sz = 0b00, Acq = Acq, Rel = Rel, opc = opc in
10725 def B : BaseLDOPregister<op, order, "b", GPR32>;
10726 let Sz = 0b01, Acq = Acq, Rel = Rel, opc = opc in
10727 def H : BaseLDOPregister<op, order, "h", GPR32>;
10728 let Sz = 0b10, Acq = Acq, Rel = Rel, opc = opc in
10729 def W : BaseLDOPregister<op, order, "", GPR32>;
10730 let Sz = 0b11, Acq = Acq, Rel = Rel, opc = opc in
10731 def X : BaseLDOPregister<op, order, "", GPR64>;
10734 // Differing SrcRHS and DstRHS allow you to cover CLR & SUB by giving a more
10735 // complex DAG for DstRHS.
10736 let Predicates = [HasLSE] in
10737 multiclass LDOPregister_patterns_ord_dag<string inst, string suffix, string op,
10738 string size, dag SrcRHS, dag DstRHS> {
10739 def : Pat<(!cast<PatFrag>(op#"_"#size#"_monotonic") GPR64sp:$Rn, SrcRHS),
10740 (!cast<Instruction>(inst # suffix) DstRHS, GPR64sp:$Rn)>;
10741 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acquire") GPR64sp:$Rn, SrcRHS),
10742 (!cast<Instruction>(inst # "A" # suffix) DstRHS, GPR64sp:$Rn)>;
10743 def : Pat<(!cast<PatFrag>(op#"_"#size#"_release") GPR64sp:$Rn, SrcRHS),
10744 (!cast<Instruction>(inst # "L" # suffix) DstRHS, GPR64sp:$Rn)>;
10745 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acq_rel") GPR64sp:$Rn, SrcRHS),
10746 (!cast<Instruction>(inst # "AL" # suffix) DstRHS, GPR64sp:$Rn)>;
10747 def : Pat<(!cast<PatFrag>(op#"_"#size#"_seq_cst") GPR64sp:$Rn, SrcRHS),
10748 (!cast<Instruction>(inst # "AL" # suffix) DstRHS, GPR64sp:$Rn)>;
10751 multiclass LDOPregister_patterns_ord<string inst, string suffix, string op,
10752 string size, dag RHS> {
10753 defm : LDOPregister_patterns_ord_dag<inst, suffix, op, size, RHS, RHS>;
10756 multiclass LDOPregister_patterns_ord_mod<string inst, string suffix, string op,
10757 string size, dag LHS, dag RHS> {
10758 defm : LDOPregister_patterns_ord_dag<inst, suffix, op, size, LHS, RHS>;
10761 multiclass LDOPregister_patterns<string inst, string op> {
10762 defm : LDOPregister_patterns_ord<inst, "X", op, "64", (i64 GPR64:$Rm)>;
10763 defm : LDOPregister_patterns_ord<inst, "W", op, "32", (i32 GPR32:$Rm)>;
10764 defm : LDOPregister_patterns_ord<inst, "H", op, "16", (i32 GPR32:$Rm)>;
10765 defm : LDOPregister_patterns_ord<inst, "B", op, "8", (i32 GPR32:$Rm)>;
10768 multiclass LDOPregister_patterns_mod<string inst, string op, string mod> {
10769 defm : LDOPregister_patterns_ord_mod<inst, "X", op, "64",
10771 (i64 (!cast<Instruction>(mod#Xrr) XZR, GPR64:$Rm))>;
10772 defm : LDOPregister_patterns_ord_mod<inst, "W", op, "32",
10774 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10775 defm : LDOPregister_patterns_ord_mod<inst, "H", op, "16",
10777 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10778 defm : LDOPregister_patterns_ord_mod<inst, "B", op, "8",
10780 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10783 let Predicates = [HasLSE] in
10784 multiclass CASregister_patterns_ord_dag<string inst, string suffix, string op,
10785 string size, dag OLD, dag NEW> {
10786 def : Pat<(!cast<PatFrag>(op#"_"#size#"_monotonic") GPR64sp:$Rn, OLD, NEW),
10787 (!cast<Instruction>(inst # suffix) OLD, NEW, GPR64sp:$Rn)>;
10788 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acquire") GPR64sp:$Rn, OLD, NEW),
10789 (!cast<Instruction>(inst # "A" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10790 def : Pat<(!cast<PatFrag>(op#"_"#size#"_release") GPR64sp:$Rn, OLD, NEW),
10791 (!cast<Instruction>(inst # "L" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10792 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acq_rel") GPR64sp:$Rn, OLD, NEW),
10793 (!cast<Instruction>(inst # "AL" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10794 def : Pat<(!cast<PatFrag>(op#"_"#size#"_seq_cst") GPR64sp:$Rn, OLD, NEW),
10795 (!cast<Instruction>(inst # "AL" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10798 multiclass CASregister_patterns_ord<string inst, string suffix, string op,
10799 string size, dag OLD, dag NEW> {
10800 defm : CASregister_patterns_ord_dag<inst, suffix, op, size, OLD, NEW>;
10803 multiclass CASregister_patterns<string inst, string op> {
10804 defm : CASregister_patterns_ord<inst, "X", op, "64",
10805 (i64 GPR64:$Rold), (i64 GPR64:$Rnew)>;
10806 defm : CASregister_patterns_ord<inst, "W", op, "32",
10807 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10808 defm : CASregister_patterns_ord<inst, "H", op, "16",
10809 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10810 defm : CASregister_patterns_ord<inst, "B", op, "8",
10811 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10814 let Predicates = [HasLSE] in
10815 class BaseSTOPregister<string asm, RegisterClass OP, Register Reg,
10816 Instruction inst> :
10817 InstAlias<asm # "\t$Rs, [$Rn]", (inst Reg, OP:$Rs, GPR64sp:$Rn)>;
10819 multiclass STOPregister<string asm, string instr> {
10820 def : BaseSTOPregister<asm # "lb", GPR32, WZR,
10821 !cast<Instruction>(instr # "LB")>;
10822 def : BaseSTOPregister<asm # "lh", GPR32, WZR,
10823 !cast<Instruction>(instr # "LH")>;
10824 def : BaseSTOPregister<asm # "l", GPR32, WZR,
10825 !cast<Instruction>(instr # "LW")>;
10826 def : BaseSTOPregister<asm # "l", GPR64, XZR,
10827 !cast<Instruction>(instr # "LX")>;
10828 def : BaseSTOPregister<asm # "b", GPR32, WZR,
10829 !cast<Instruction>(instr # "B")>;
10830 def : BaseSTOPregister<asm # "h", GPR32, WZR,
10831 !cast<Instruction>(instr # "H")>;
10832 def : BaseSTOPregister<asm, GPR32, WZR,
10833 !cast<Instruction>(instr # "W")>;
10834 def : BaseSTOPregister<asm, GPR64, XZR,
10835 !cast<Instruction>(instr # "X")>;
10838 //----------------------------------------------------------------------------
10839 // Allow the size specifier tokens to be upper case, not just lower.
10840 def : TokenAlias<".4B", ".4b">; // Add dot product
10841 def : TokenAlias<".8B", ".8b">;
10842 def : TokenAlias<".4H", ".4h">;
10843 def : TokenAlias<".2S", ".2s">;
10844 def : TokenAlias<".1D", ".1d">;
10845 def : TokenAlias<".16B", ".16b">;
10846 def : TokenAlias<".8H", ".8h">;
10847 def : TokenAlias<".4S", ".4s">;
10848 def : TokenAlias<".2D", ".2d">;
10849 def : TokenAlias<".1Q", ".1q">;
10850 def : TokenAlias<".2H", ".2h">;
10851 def : TokenAlias<".B", ".b">;
10852 def : TokenAlias<".H", ".h">;
10853 def : TokenAlias<".S", ".s">;
10854 def : TokenAlias<".D", ".d">;
10855 def : TokenAlias<".Q", ".q">;