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<i64>, ImmLeaf<i64, [{ 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 // uimm5sN predicate - True if the immediate is a multiple of N in the range
363 def UImm5s2Operand : UImmScaledMemoryIndexed<5, 2>;
364 def UImm5s4Operand : UImmScaledMemoryIndexed<5, 4>;
365 def UImm5s8Operand : UImmScaledMemoryIndexed<5, 8>;
367 def uimm5s2 : Operand<i64>, ImmLeaf<i64,
368 [{ return Imm >= 0 && Imm < (32*2) && ((Imm % 2) == 0); }]> {
369 let ParserMatchClass = UImm5s2Operand;
370 let PrintMethod = "printImmScale<2>";
372 def uimm5s4 : Operand<i64>, ImmLeaf<i64,
373 [{ return Imm >= 0 && Imm < (32*4) && ((Imm % 4) == 0); }]> {
374 let ParserMatchClass = UImm5s4Operand;
375 let PrintMethod = "printImmScale<4>";
377 def uimm5s8 : Operand<i64>, ImmLeaf<i64,
378 [{ return Imm >= 0 && Imm < (32*8) && ((Imm % 8) == 0); }]> {
379 let ParserMatchClass = UImm5s8Operand;
380 let PrintMethod = "printImmScale<8>";
383 // uimm6sN predicate - True if the immediate is a multiple of N in the range
385 def UImm6s1Operand : UImmScaledMemoryIndexed<6, 1>;
386 def UImm6s2Operand : UImmScaledMemoryIndexed<6, 2>;
387 def UImm6s4Operand : UImmScaledMemoryIndexed<6, 4>;
388 def UImm6s8Operand : UImmScaledMemoryIndexed<6, 8>;
389 def UImm6s16Operand : UImmScaledMemoryIndexed<6, 16>;
391 def uimm6s1 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= 0 && Imm < 64; }]> {
392 let ParserMatchClass = UImm6s1Operand;
394 def uimm6s2 : Operand<i64>, ImmLeaf<i64,
395 [{ return Imm >= 0 && Imm < (64*2) && ((Imm % 2) == 0); }]> {
396 let PrintMethod = "printImmScale<2>";
397 let ParserMatchClass = UImm6s2Operand;
399 def uimm6s4 : Operand<i64>, ImmLeaf<i64,
400 [{ return Imm >= 0 && Imm < (64*4) && ((Imm % 4) == 0); }]> {
401 let PrintMethod = "printImmScale<4>";
402 let ParserMatchClass = UImm6s4Operand;
404 def uimm6s8 : Operand<i64>, ImmLeaf<i64,
405 [{ return Imm >= 0 && Imm < (64*8) && ((Imm % 8) == 0); }]> {
406 let PrintMethod = "printImmScale<8>";
407 let ParserMatchClass = UImm6s8Operand;
409 def uimm6s16 : Operand<i64>, ImmLeaf<i64,
410 [{ return Imm >= 0 && Imm < (64*16) && ((Imm % 16) == 0); }]> {
411 let PrintMethod = "printImmScale<16>";
412 let ParserMatchClass = UImm6s16Operand;
415 // simm6sN predicate - True if the immediate is a multiple of N in the range
416 // [-32 * N, 31 * N].
417 def SImm6s1Operand : SImmScaledMemoryIndexed<6, 1>;
418 def simm6s1 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -32 && Imm < 32; }]> {
419 let ParserMatchClass = SImm6s1Operand;
420 let DecoderMethod = "DecodeSImm<6>";
423 // simm4sN predicate - True if the immediate is a multiple of N in the range
425 def SImm4s1Operand : SImmScaledMemoryIndexed<4, 1>;
426 def SImm4s2Operand : SImmScaledMemoryIndexed<4, 2>;
427 def SImm4s3Operand : SImmScaledMemoryIndexed<4, 3>;
428 def SImm4s4Operand : SImmScaledMemoryIndexed<4, 4>;
429 def SImm4s16Operand : SImmScaledMemoryIndexed<4, 16>;
431 def simm4s1 : Operand<i64>, ImmLeaf<i64,
432 [{ return Imm >=-8 && Imm <= 7; }]> {
433 let ParserMatchClass = SImm4s1Operand;
434 let DecoderMethod = "DecodeSImm<4>";
437 def simm4s2 : Operand<i64>, ImmLeaf<i64,
438 [{ return Imm >=-16 && Imm <= 14 && (Imm % 2) == 0x0; }]> {
439 let PrintMethod = "printImmScale<2>";
440 let ParserMatchClass = SImm4s2Operand;
441 let DecoderMethod = "DecodeSImm<4>";
444 def simm4s3 : Operand<i64>, ImmLeaf<i64,
445 [{ return Imm >=-24 && Imm <= 21 && (Imm % 3) == 0x0; }]> {
446 let PrintMethod = "printImmScale<3>";
447 let ParserMatchClass = SImm4s3Operand;
448 let DecoderMethod = "DecodeSImm<4>";
451 def simm4s4 : Operand<i64>, ImmLeaf<i64,
452 [{ return Imm >=-32 && Imm <= 28 && (Imm % 4) == 0x0; }]> {
453 let PrintMethod = "printImmScale<4>";
454 let ParserMatchClass = SImm4s4Operand;
455 let DecoderMethod = "DecodeSImm<4>";
457 def simm4s16 : Operand<i64>, ImmLeaf<i64,
458 [{ return Imm >=-128 && Imm <= 112 && (Imm % 16) == 0x0; }]> {
459 let PrintMethod = "printImmScale<16>";
460 let ParserMatchClass = SImm4s16Operand;
461 let DecoderMethod = "DecodeSImm<4>";
464 def Imm1_8Operand : AsmImmRange<1, 8>;
465 def Imm1_16Operand : AsmImmRange<1, 16>;
466 def Imm1_32Operand : AsmImmRange<1, 32>;
467 def Imm1_64Operand : AsmImmRange<1, 64>;
469 class BranchTarget<int N> : AsmOperandClass {
470 let Name = "BranchTarget" # N;
471 let DiagnosticType = "InvalidLabel";
472 let PredicateMethod = "isBranchTarget<" # N # ">";
475 class PCRelLabel<int N> : BranchTarget<N> {
476 let Name = "PCRelLabel" # N;
479 def BranchTarget14Operand : BranchTarget<14>;
480 def BranchTarget26Operand : BranchTarget<26>;
481 def PCRelLabel19Operand : PCRelLabel<19>;
483 def MovZSymbolG3AsmOperand : AsmOperandClass {
484 let Name = "MovZSymbolG3";
485 let RenderMethod = "addImmOperands";
488 def movz_symbol_g3 : Operand<i32> {
489 let ParserMatchClass = MovZSymbolG3AsmOperand;
492 def MovZSymbolG2AsmOperand : AsmOperandClass {
493 let Name = "MovZSymbolG2";
494 let RenderMethod = "addImmOperands";
497 def movz_symbol_g2 : Operand<i32> {
498 let ParserMatchClass = MovZSymbolG2AsmOperand;
501 def MovZSymbolG1AsmOperand : AsmOperandClass {
502 let Name = "MovZSymbolG1";
503 let RenderMethod = "addImmOperands";
506 def movz_symbol_g1 : Operand<i32> {
507 let ParserMatchClass = MovZSymbolG1AsmOperand;
510 def MovZSymbolG0AsmOperand : AsmOperandClass {
511 let Name = "MovZSymbolG0";
512 let RenderMethod = "addImmOperands";
515 def movz_symbol_g0 : Operand<i32> {
516 let ParserMatchClass = MovZSymbolG0AsmOperand;
519 def MovKSymbolG3AsmOperand : AsmOperandClass {
520 let Name = "MovKSymbolG3";
521 let RenderMethod = "addImmOperands";
524 def movk_symbol_g3 : Operand<i32> {
525 let ParserMatchClass = MovKSymbolG3AsmOperand;
528 def MovKSymbolG2AsmOperand : AsmOperandClass {
529 let Name = "MovKSymbolG2";
530 let RenderMethod = "addImmOperands";
533 def movk_symbol_g2 : Operand<i32> {
534 let ParserMatchClass = MovKSymbolG2AsmOperand;
537 def MovKSymbolG1AsmOperand : AsmOperandClass {
538 let Name = "MovKSymbolG1";
539 let RenderMethod = "addImmOperands";
542 def movk_symbol_g1 : Operand<i32> {
543 let ParserMatchClass = MovKSymbolG1AsmOperand;
546 def MovKSymbolG0AsmOperand : AsmOperandClass {
547 let Name = "MovKSymbolG0";
548 let RenderMethod = "addImmOperands";
551 def movk_symbol_g0 : Operand<i32> {
552 let ParserMatchClass = MovKSymbolG0AsmOperand;
555 class fixedpoint_i32<ValueType FloatVT>
557 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<32>", [fpimm, ld]> {
558 let EncoderMethod = "getFixedPointScaleOpValue";
559 let DecoderMethod = "DecodeFixedPointScaleImm32";
560 let ParserMatchClass = Imm1_32Operand;
563 class fixedpoint_i64<ValueType FloatVT>
565 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<64>", [fpimm, ld]> {
566 let EncoderMethod = "getFixedPointScaleOpValue";
567 let DecoderMethod = "DecodeFixedPointScaleImm64";
568 let ParserMatchClass = Imm1_64Operand;
571 def fixedpoint_f16_i32 : fixedpoint_i32<f16>;
572 def fixedpoint_f32_i32 : fixedpoint_i32<f32>;
573 def fixedpoint_f64_i32 : fixedpoint_i32<f64>;
575 def fixedpoint_f16_i64 : fixedpoint_i64<f16>;
576 def fixedpoint_f32_i64 : fixedpoint_i64<f32>;
577 def fixedpoint_f64_i64 : fixedpoint_i64<f64>;
579 def vecshiftR8 : Operand<i32>, ImmLeaf<i32, [{
580 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
582 let EncoderMethod = "getVecShiftR8OpValue";
583 let DecoderMethod = "DecodeVecShiftR8Imm";
584 let ParserMatchClass = Imm1_8Operand;
586 def vecshiftR16 : Operand<i32>, ImmLeaf<i32, [{
587 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
589 let EncoderMethod = "getVecShiftR16OpValue";
590 let DecoderMethod = "DecodeVecShiftR16Imm";
591 let ParserMatchClass = Imm1_16Operand;
593 def vecshiftR16Narrow : Operand<i32>, ImmLeaf<i32, [{
594 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
596 let EncoderMethod = "getVecShiftR16OpValue";
597 let DecoderMethod = "DecodeVecShiftR16ImmNarrow";
598 let ParserMatchClass = Imm1_8Operand;
600 def vecshiftR32 : Operand<i32>, ImmLeaf<i32, [{
601 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
603 let EncoderMethod = "getVecShiftR32OpValue";
604 let DecoderMethod = "DecodeVecShiftR32Imm";
605 let ParserMatchClass = Imm1_32Operand;
607 def vecshiftR32Narrow : Operand<i32>, ImmLeaf<i32, [{
608 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
610 let EncoderMethod = "getVecShiftR32OpValue";
611 let DecoderMethod = "DecodeVecShiftR32ImmNarrow";
612 let ParserMatchClass = Imm1_16Operand;
614 def vecshiftR64 : Operand<i32>, ImmLeaf<i32, [{
615 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 65);
617 let EncoderMethod = "getVecShiftR64OpValue";
618 let DecoderMethod = "DecodeVecShiftR64Imm";
619 let ParserMatchClass = Imm1_64Operand;
621 def vecshiftR64Narrow : Operand<i32>, ImmLeaf<i32, [{
622 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
624 let EncoderMethod = "getVecShiftR64OpValue";
625 let DecoderMethod = "DecodeVecShiftR64ImmNarrow";
626 let ParserMatchClass = Imm1_32Operand;
629 def Imm0_1Operand : AsmImmRange<0, 1>;
630 def Imm0_7Operand : AsmImmRange<0, 7>;
631 def Imm0_15Operand : AsmImmRange<0, 15>;
632 def Imm0_31Operand : AsmImmRange<0, 31>;
633 def Imm0_63Operand : AsmImmRange<0, 63>;
635 def vecshiftL8 : Operand<i32>, ImmLeaf<i32, [{
636 return (((uint32_t)Imm) < 8);
638 let EncoderMethod = "getVecShiftL8OpValue";
639 let DecoderMethod = "DecodeVecShiftL8Imm";
640 let ParserMatchClass = Imm0_7Operand;
642 def vecshiftL16 : Operand<i32>, ImmLeaf<i32, [{
643 return (((uint32_t)Imm) < 16);
645 let EncoderMethod = "getVecShiftL16OpValue";
646 let DecoderMethod = "DecodeVecShiftL16Imm";
647 let ParserMatchClass = Imm0_15Operand;
649 def vecshiftL32 : Operand<i32>, ImmLeaf<i32, [{
650 return (((uint32_t)Imm) < 32);
652 let EncoderMethod = "getVecShiftL32OpValue";
653 let DecoderMethod = "DecodeVecShiftL32Imm";
654 let ParserMatchClass = Imm0_31Operand;
656 def vecshiftL64 : Operand<i32>, ImmLeaf<i32, [{
657 return (((uint32_t)Imm) < 64);
659 let EncoderMethod = "getVecShiftL64OpValue";
660 let DecoderMethod = "DecodeVecShiftL64Imm";
661 let ParserMatchClass = Imm0_63Operand;
665 // Crazy immediate formats used by 32-bit and 64-bit logical immediate
666 // instructions for splatting repeating bit patterns across the immediate.
667 def logical_imm32_XFORM : SDNodeXForm<imm, [{
668 uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 32);
669 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
671 def logical_imm64_XFORM : SDNodeXForm<imm, [{
672 uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 64);
673 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
676 let DiagnosticType = "LogicalSecondSource" in {
677 def LogicalImm32Operand : AsmOperandClass {
678 let Name = "LogicalImm32";
679 let PredicateMethod = "isLogicalImm<int32_t>";
680 let RenderMethod = "addLogicalImmOperands<int32_t>";
682 def LogicalImm64Operand : AsmOperandClass {
683 let Name = "LogicalImm64";
684 let PredicateMethod = "isLogicalImm<int64_t>";
685 let RenderMethod = "addLogicalImmOperands<int64_t>";
687 def LogicalImm32NotOperand : AsmOperandClass {
688 let Name = "LogicalImm32Not";
689 let PredicateMethod = "isLogicalImm<int32_t>";
690 let RenderMethod = "addLogicalImmNotOperands<int32_t>";
692 def LogicalImm64NotOperand : AsmOperandClass {
693 let Name = "LogicalImm64Not";
694 let PredicateMethod = "isLogicalImm<int64_t>";
695 let RenderMethod = "addLogicalImmNotOperands<int64_t>";
698 def logical_imm32 : Operand<i32>, IntImmLeaf<i32, [{
699 return AArch64_AM::isLogicalImmediate(Imm.getZExtValue(), 32);
700 }], logical_imm32_XFORM> {
701 let PrintMethod = "printLogicalImm<int32_t>";
702 let ParserMatchClass = LogicalImm32Operand;
704 def logical_imm64 : Operand<i64>, IntImmLeaf<i64, [{
705 return AArch64_AM::isLogicalImmediate(Imm.getZExtValue(), 64);
706 }], logical_imm64_XFORM> {
707 let PrintMethod = "printLogicalImm<int64_t>";
708 let ParserMatchClass = LogicalImm64Operand;
710 def logical_imm32_not : Operand<i32> {
711 let ParserMatchClass = LogicalImm32NotOperand;
713 def logical_imm64_not : Operand<i64> {
714 let ParserMatchClass = LogicalImm64NotOperand;
717 // imm0_65535 predicate - True if the immediate is in the range [0,65535].
718 def imm0_65535 : Operand<i32>, ImmLeaf<i32, [{
719 return ((uint32_t)Imm) < 65536;
721 let ParserMatchClass = AsmImmRange<0, 65535>;
722 let PrintMethod = "printImmHex";
725 // imm0_255 predicate - True if the immediate is in the range [0,255].
726 def Imm0_255Operand : AsmImmRange<0,255>;
728 def imm0_255 : Operand<i32>, ImmLeaf<i32, [{
729 return ((uint32_t)Imm) < 256;
731 let ParserMatchClass = Imm0_255Operand;
732 let PrintMethod = "printImm";
735 // imm0_127 predicate - True if the immediate is in the range [0,127]
736 def Imm0_127Operand : AsmImmRange<0, 127>;
737 def imm0_127 : Operand<i32>, ImmLeaf<i32, [{
738 return ((uint32_t)Imm) < 128;
740 let ParserMatchClass = Imm0_127Operand;
741 let PrintMethod = "printImm";
744 // NOTE: These imm0_N operands have to be of type i64 because i64 is the size
745 // for all shift-amounts.
747 // imm0_63 predicate - True if the immediate is in the range [0,63]
748 def imm0_63 : Operand<i64>, ImmLeaf<i64, [{
749 return ((uint64_t)Imm) < 64;
751 let ParserMatchClass = Imm0_63Operand;
754 // imm0_31 predicate - True if the immediate is in the range [0,31]
755 def imm0_31 : Operand<i64>, ImmLeaf<i64, [{
756 return ((uint64_t)Imm) < 32;
758 let ParserMatchClass = Imm0_31Operand;
761 // True if the 32-bit immediate is in the range [0,31]
762 def imm32_0_31 : Operand<i32>, ImmLeaf<i32, [{
763 return ((uint64_t)Imm) < 32;
765 let ParserMatchClass = Imm0_31Operand;
768 // imm0_1 predicate - True if the immediate is in the range [0,1]
769 def imm0_1 : Operand<i64>, ImmLeaf<i64, [{
770 return ((uint64_t)Imm) < 2;
772 let ParserMatchClass = Imm0_1Operand;
775 // imm0_15 predicate - True if the immediate is in the range [0,15]
776 def imm0_15 : Operand<i64>, ImmLeaf<i64, [{
777 return ((uint64_t)Imm) < 16;
779 let ParserMatchClass = Imm0_15Operand;
782 // imm0_7 predicate - True if the immediate is in the range [0,7]
783 def imm0_7 : Operand<i64>, ImmLeaf<i64, [{
784 return ((uint64_t)Imm) < 8;
786 let ParserMatchClass = Imm0_7Operand;
789 // imm32_0_15 predicate - True if the 32-bit immediate is in the range [0,15]
790 def imm32_0_15 : Operand<i32>, ImmLeaf<i32, [{
791 return ((uint32_t)Imm) < 16;
793 let ParserMatchClass = Imm0_15Operand;
796 // An arithmetic shifter operand:
797 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr
799 class arith_shift<ValueType Ty, int width> : Operand<Ty> {
800 let PrintMethod = "printShifter";
801 let ParserMatchClass = !cast<AsmOperandClass>(
802 "ArithmeticShifterOperand" # width);
805 def arith_shift32 : arith_shift<i32, 32>;
806 def arith_shift64 : arith_shift<i64, 64>;
808 class arith_shifted_reg<ValueType Ty, RegisterClass regclass, int width>
810 ComplexPattern<Ty, 2, "SelectArithShiftedRegister", []> {
811 let PrintMethod = "printShiftedRegister";
812 let MIOperandInfo = (ops regclass, !cast<Operand>("arith_shift" # width));
815 def arith_shifted_reg32 : arith_shifted_reg<i32, GPR32, 32>;
816 def arith_shifted_reg64 : arith_shifted_reg<i64, GPR64, 64>;
818 // An arithmetic shifter operand:
819 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr, 11 = ror
821 class logical_shift<int width> : Operand<i32> {
822 let PrintMethod = "printShifter";
823 let ParserMatchClass = !cast<AsmOperandClass>(
824 "LogicalShifterOperand" # width);
827 def logical_shift32 : logical_shift<32>;
828 def logical_shift64 : logical_shift<64>;
830 class logical_shifted_reg<ValueType Ty, RegisterClass regclass, Operand shiftop>
832 ComplexPattern<Ty, 2, "SelectLogicalShiftedRegister", []> {
833 let PrintMethod = "printShiftedRegister";
834 let MIOperandInfo = (ops regclass, shiftop);
837 def logical_shifted_reg32 : logical_shifted_reg<i32, GPR32, logical_shift32>;
838 def logical_shifted_reg64 : logical_shifted_reg<i64, GPR64, logical_shift64>;
840 // A logical vector shifter operand:
841 // {7-6} - shift type: 00 = lsl
842 // {5-0} - imm6: #0, #8, #16, or #24
843 def logical_vec_shift : Operand<i32> {
844 let PrintMethod = "printShifter";
845 let EncoderMethod = "getVecShifterOpValue";
846 let ParserMatchClass = LogicalVecShifterOperand;
849 // A logical vector half-word shifter operand:
850 // {7-6} - shift type: 00 = lsl
851 // {5-0} - imm6: #0 or #8
852 def logical_vec_hw_shift : Operand<i32> {
853 let PrintMethod = "printShifter";
854 let EncoderMethod = "getVecShifterOpValue";
855 let ParserMatchClass = LogicalVecHalfWordShifterOperand;
858 // A vector move shifter operand:
859 // {0} - imm1: #8 or #16
860 def move_vec_shift : Operand<i32> {
861 let PrintMethod = "printShifter";
862 let EncoderMethod = "getMoveVecShifterOpValue";
863 let ParserMatchClass = MoveVecShifterOperand;
866 let DiagnosticType = "AddSubSecondSource" in {
867 def AddSubImmOperand : AsmOperandClass {
868 let Name = "AddSubImm";
869 let ParserMethod = "tryParseImmWithOptionalShift";
870 let RenderMethod = "addImmWithOptionalShiftOperands<12>";
872 def AddSubImmNegOperand : AsmOperandClass {
873 let Name = "AddSubImmNeg";
874 let ParserMethod = "tryParseImmWithOptionalShift";
875 let RenderMethod = "addImmNegWithOptionalShiftOperands<12>";
878 // An ADD/SUB immediate shifter operand:
880 // {7-6} - shift type: 00 = lsl
881 // {5-0} - imm6: #0 or #12
882 class addsub_shifted_imm<ValueType Ty>
883 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectArithImmed", [imm]> {
884 let PrintMethod = "printAddSubImm";
885 let EncoderMethod = "getAddSubImmOpValue";
886 let ParserMatchClass = AddSubImmOperand;
887 let MIOperandInfo = (ops i32imm, i32imm);
890 class addsub_shifted_imm_neg<ValueType Ty>
892 let EncoderMethod = "getAddSubImmOpValue";
893 let ParserMatchClass = AddSubImmNegOperand;
894 let MIOperandInfo = (ops i32imm, i32imm);
897 def addsub_shifted_imm32 : addsub_shifted_imm<i32>;
898 def addsub_shifted_imm64 : addsub_shifted_imm<i64>;
899 def addsub_shifted_imm32_neg : addsub_shifted_imm_neg<i32>;
900 def addsub_shifted_imm64_neg : addsub_shifted_imm_neg<i64>;
902 def gi_addsub_shifted_imm32 :
903 GIComplexOperandMatcher<s32, "selectArithImmed">,
904 GIComplexPatternEquiv<addsub_shifted_imm32>;
906 def gi_addsub_shifted_imm64 :
907 GIComplexOperandMatcher<s64, "selectArithImmed">,
908 GIComplexPatternEquiv<addsub_shifted_imm64>;
910 class neg_addsub_shifted_imm<ValueType Ty>
911 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectNegArithImmed", [imm]> {
912 let PrintMethod = "printAddSubImm";
913 let EncoderMethod = "getAddSubImmOpValue";
914 let ParserMatchClass = AddSubImmOperand;
915 let MIOperandInfo = (ops i32imm, i32imm);
918 def neg_addsub_shifted_imm32 : neg_addsub_shifted_imm<i32>;
919 def neg_addsub_shifted_imm64 : neg_addsub_shifted_imm<i64>;
921 // An extend operand:
922 // {5-3} - extend type
924 def arith_extend : Operand<i32> {
925 let PrintMethod = "printArithExtend";
926 let ParserMatchClass = ExtendOperand;
928 def arith_extend64 : Operand<i32> {
929 let PrintMethod = "printArithExtend";
930 let ParserMatchClass = ExtendOperand64;
933 // 'extend' that's a lsl of a 64-bit register.
934 def arith_extendlsl64 : Operand<i32> {
935 let PrintMethod = "printArithExtend";
936 let ParserMatchClass = ExtendOperandLSL64;
939 class arith_extended_reg32<ValueType Ty> : Operand<Ty>,
940 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
941 let PrintMethod = "printExtendedRegister";
942 let MIOperandInfo = (ops GPR32, arith_extend);
945 class arith_extended_reg32to64<ValueType Ty> : Operand<Ty>,
946 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
947 let PrintMethod = "printExtendedRegister";
948 let MIOperandInfo = (ops GPR32, arith_extend64);
951 // Floating-point immediate.
952 def fpimm16 : Operand<f16>,
954 return AArch64_AM::getFP16Imm(Imm) != -1;
955 }], SDNodeXForm<fpimm, [{
956 APFloat InVal = N->getValueAPF();
957 uint32_t enc = AArch64_AM::getFP16Imm(InVal);
958 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
960 let ParserMatchClass = FPImmOperand;
961 let PrintMethod = "printFPImmOperand";
963 def fpimm32 : Operand<f32>,
965 return AArch64_AM::getFP32Imm(Imm) != -1;
966 }], SDNodeXForm<fpimm, [{
967 APFloat InVal = N->getValueAPF();
968 uint32_t enc = AArch64_AM::getFP32Imm(InVal);
969 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
971 let ParserMatchClass = FPImmOperand;
972 let PrintMethod = "printFPImmOperand";
974 def fpimm64 : Operand<f64>,
976 return AArch64_AM::getFP64Imm(Imm) != -1;
977 }], SDNodeXForm<fpimm, [{
978 APFloat InVal = N->getValueAPF();
979 uint32_t enc = AArch64_AM::getFP64Imm(InVal);
980 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
982 let ParserMatchClass = FPImmOperand;
983 let PrintMethod = "printFPImmOperand";
986 def fpimm8 : Operand<i32> {
987 let ParserMatchClass = FPImmOperand;
988 let PrintMethod = "printFPImmOperand";
991 def fpimm0 : FPImmLeaf<fAny, [{
992 return Imm.isExactlyValue(+0.0);
995 // Vector lane operands
996 class AsmVectorIndex<int Min, int Max, string NamePrefix=""> : AsmOperandClass {
997 let Name = NamePrefix # "IndexRange" # Min # "_" # Max;
998 let DiagnosticType = "Invalid" # Name;
999 let PredicateMethod = "isVectorIndex<" # Min # ", " # Max # ">";
1000 let RenderMethod = "addVectorIndexOperands";
1003 class AsmVectorIndexOpnd<AsmOperandClass mc, code pred>
1004 : Operand<i64>, ImmLeaf<i64, pred> {
1005 let ParserMatchClass = mc;
1006 let PrintMethod = "printVectorIndex";
1009 def VectorIndex1Operand : AsmVectorIndex<1, 1>;
1010 def VectorIndexBOperand : AsmVectorIndex<0, 15>;
1011 def VectorIndexHOperand : AsmVectorIndex<0, 7>;
1012 def VectorIndexSOperand : AsmVectorIndex<0, 3>;
1013 def VectorIndexDOperand : AsmVectorIndex<0, 1>;
1015 def VectorIndex1 : AsmVectorIndexOpnd<VectorIndex1Operand, [{ return ((uint64_t)Imm) == 1; }]>;
1016 def VectorIndexB : AsmVectorIndexOpnd<VectorIndexBOperand, [{ return ((uint64_t)Imm) < 16; }]>;
1017 def VectorIndexH : AsmVectorIndexOpnd<VectorIndexHOperand, [{ return ((uint64_t)Imm) < 8; }]>;
1018 def VectorIndexS : AsmVectorIndexOpnd<VectorIndexSOperand, [{ return ((uint64_t)Imm) < 4; }]>;
1019 def VectorIndexD : AsmVectorIndexOpnd<VectorIndexDOperand, [{ return ((uint64_t)Imm) < 2; }]>;
1021 def SVEVectorIndexExtDupBOperand : AsmVectorIndex<0, 63, "SVE">;
1022 def SVEVectorIndexExtDupHOperand : AsmVectorIndex<0, 31, "SVE">;
1023 def SVEVectorIndexExtDupSOperand : AsmVectorIndex<0, 15, "SVE">;
1024 def SVEVectorIndexExtDupDOperand : AsmVectorIndex<0, 7, "SVE">;
1025 def SVEVectorIndexExtDupQOperand : AsmVectorIndex<0, 3, "SVE">;
1027 def sve_elm_idx_extdup_b
1028 : AsmVectorIndexOpnd<SVEVectorIndexExtDupBOperand, [{ return ((uint64_t)Imm) < 64; }]>;
1029 def sve_elm_idx_extdup_h
1030 : AsmVectorIndexOpnd<SVEVectorIndexExtDupHOperand, [{ return ((uint64_t)Imm) < 32; }]>;
1031 def sve_elm_idx_extdup_s
1032 : AsmVectorIndexOpnd<SVEVectorIndexExtDupSOperand, [{ return ((uint64_t)Imm) < 16; }]>;
1033 def sve_elm_idx_extdup_d
1034 : AsmVectorIndexOpnd<SVEVectorIndexExtDupDOperand, [{ return ((uint64_t)Imm) < 8; }]>;
1035 def sve_elm_idx_extdup_q
1036 : AsmVectorIndexOpnd<SVEVectorIndexExtDupQOperand, [{ return ((uint64_t)Imm) < 4; }]>;
1038 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
1039 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
1040 // are encoded as the eight bit value 'abcdefgh'.
1041 def simdimmtype10 : Operand<i32>,
1043 return AArch64_AM::isAdvSIMDModImmType10(
1044 Imm.bitcastToAPInt().getZExtValue());
1045 }], SDNodeXForm<fpimm, [{
1046 APFloat InVal = N->getValueAPF();
1047 uint32_t enc = AArch64_AM::encodeAdvSIMDModImmType10(N->getValueAPF()
1050 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
1052 let ParserMatchClass = SIMDImmType10Operand;
1053 let PrintMethod = "printSIMDType10Operand";
1058 // System management
1061 // Base encoding for system instruction operands.
1062 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
1063 class BaseSystemI<bit L, dag oops, dag iops, string asm, string operands,
1064 list<dag> pattern = []>
1065 : I<oops, iops, asm, operands, "", pattern> {
1066 let Inst{31-22} = 0b1101010100;
1070 // System instructions which do not have an Rt register.
1071 class SimpleSystemI<bit L, dag iops, string asm, string operands,
1072 list<dag> pattern = []>
1073 : BaseSystemI<L, (outs), iops, asm, operands, pattern> {
1074 let Inst{4-0} = 0b11111;
1077 // System instructions which have an Rt register.
1078 class RtSystemI<bit L, dag oops, dag iops, string asm, string operands>
1079 : BaseSystemI<L, oops, iops, asm, operands>,
1085 // Hint instructions that take both a CRm and a 3-bit immediate.
1086 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
1087 // model patterns with sufficiently fine granularity
1088 let mayStore = 1, mayLoad = 1, hasSideEffects = 1 in
1089 class HintI<string mnemonic>
1090 : SimpleSystemI<0, (ins imm0_127:$imm), mnemonic#"\t$imm", "",
1091 [(int_aarch64_hint imm0_127:$imm)]>,
1092 Sched<[WriteHint]> {
1094 let Inst{20-12} = 0b000110010;
1095 let Inst{11-5} = imm;
1098 // System instructions taking a single literal operand which encodes into
1099 // CRm. op2 differentiates the opcodes.
1100 def BarrierAsmOperand : AsmOperandClass {
1101 let Name = "Barrier";
1102 let ParserMethod = "tryParseBarrierOperand";
1104 def barrier_op : Operand<i32> {
1105 let PrintMethod = "printBarrierOption";
1106 let ParserMatchClass = BarrierAsmOperand;
1108 class CRmSystemI<Operand crmtype, bits<3> opc, string asm,
1109 list<dag> pattern = []>
1110 : SimpleSystemI<0, (ins crmtype:$CRm), asm, "\t$CRm", pattern>,
1111 Sched<[WriteBarrier]> {
1113 let Inst{20-12} = 0b000110011;
1114 let Inst{11-8} = CRm;
1115 let Inst{7-5} = opc;
1118 class SystemNoOperands<bits<3> op2, string asm, list<dag> pattern = []>
1119 : SimpleSystemI<0, (ins), asm, "", pattern>,
1123 let Inst{31-12} = 0b11010101000000110010;
1124 let Inst{11-8} = CRm;
1125 let Inst{7-5} = op2;
1126 let Inst{4-0} = 0b11111;
1129 // MRS/MSR system instructions. These have different operand classes because
1130 // a different subset of registers can be accessed through each instruction.
1131 def MRSSystemRegisterOperand : AsmOperandClass {
1132 let Name = "MRSSystemRegister";
1133 let ParserMethod = "tryParseSysReg";
1134 let DiagnosticType = "MRS";
1136 // concatenation of op0, op1, CRn, CRm, op2. 16-bit immediate.
1137 def mrs_sysreg_op : Operand<i32> {
1138 let ParserMatchClass = MRSSystemRegisterOperand;
1139 let DecoderMethod = "DecodeMRSSystemRegister";
1140 let PrintMethod = "printMRSSystemRegister";
1143 def MSRSystemRegisterOperand : AsmOperandClass {
1144 let Name = "MSRSystemRegister";
1145 let ParserMethod = "tryParseSysReg";
1146 let DiagnosticType = "MSR";
1148 def msr_sysreg_op : Operand<i32> {
1149 let ParserMatchClass = MSRSystemRegisterOperand;
1150 let DecoderMethod = "DecodeMSRSystemRegister";
1151 let PrintMethod = "printMSRSystemRegister";
1154 def PSBHintOperand : AsmOperandClass {
1155 let Name = "PSBHint";
1156 let ParserMethod = "tryParsePSBHint";
1158 def psbhint_op : Operand<i32> {
1159 let ParserMatchClass = PSBHintOperand;
1160 let PrintMethod = "printPSBHintOp";
1161 let MCOperandPredicate = [{
1162 // Check, if operand is valid, to fix exhaustive aliasing in disassembly.
1163 // "psb" is an alias to "hint" only for certain values of CRm:Op2 fields.
1166 return AArch64PSBHint::lookupPSBByEncoding(MCOp.getImm()) != nullptr;
1170 def BTIHintOperand : AsmOperandClass {
1171 let Name = "BTIHint";
1172 let ParserMethod = "tryParseBTIHint";
1174 def btihint_op : Operand<i32> {
1175 let ParserMatchClass = BTIHintOperand;
1176 let PrintMethod = "printBTIHintOp";
1177 let MCOperandPredicate = [{
1178 // "bti" is an alias to "hint" only for certain values of CRm:Op2 fields.
1181 return AArch64BTIHint::lookupBTIByEncoding((MCOp.getImm() ^ 32) >> 1) != nullptr;
1185 class MRSI : RtSystemI<1, (outs GPR64:$Rt), (ins mrs_sysreg_op:$systemreg),
1186 "mrs", "\t$Rt, $systemreg"> {
1188 let Inst{20-5} = systemreg;
1191 // FIXME: Some of these def NZCV, others don't. Best way to model that?
1192 // Explicitly modeling each of the system register as a register class
1193 // would do it, but feels like overkill at this point.
1194 class MSRI : RtSystemI<0, (outs), (ins msr_sysreg_op:$systemreg, GPR64:$Rt),
1195 "msr", "\t$systemreg, $Rt"> {
1197 let Inst{20-5} = systemreg;
1200 def SystemPStateFieldWithImm0_15Operand : AsmOperandClass {
1201 let Name = "SystemPStateFieldWithImm0_15";
1202 let ParserMethod = "tryParseSysReg";
1204 def pstatefield4_op : Operand<i32> {
1205 let ParserMatchClass = SystemPStateFieldWithImm0_15Operand;
1206 let PrintMethod = "printSystemPStateField";
1209 // Instructions to modify PSTATE, no input reg
1210 let Defs = [NZCV] in
1211 class PstateWriteSimple<dag iops, string asm, string operands>
1212 : SimpleSystemI<0, iops, asm, operands> {
1214 let Inst{20-19} = 0b00;
1215 let Inst{15-12} = 0b0100;
1218 class MSRpstateImm0_15
1219 : PstateWriteSimple<(ins pstatefield4_op:$pstatefield, imm0_15:$imm), "msr",
1220 "\t$pstatefield, $imm">,
1223 bits<6> pstatefield;
1225 let Inst{18-16} = pstatefield{5-3};
1226 let Inst{11-8} = imm;
1227 let Inst{7-5} = pstatefield{2-0};
1229 let DecoderMethod = "DecodeSystemPStateInstruction";
1230 // MSRpstateI aliases with MSRI. When the MSRpstateI decoder method returns
1231 // Fail the decoder should attempt to decode the instruction as MSRI.
1232 let hasCompleteDecoder = 0;
1235 def SystemPStateFieldWithImm0_1Operand : AsmOperandClass {
1236 let Name = "SystemPStateFieldWithImm0_1";
1237 let ParserMethod = "tryParseSysReg";
1239 def pstatefield1_op : Operand<i32> {
1240 let ParserMatchClass = SystemPStateFieldWithImm0_1Operand;
1241 let PrintMethod = "printSystemPStateField";
1244 class MSRpstateImm0_1
1245 : PstateWriteSimple<(ins pstatefield1_op:$pstatefield, imm0_1:$imm), "msr",
1246 "\t$pstatefield, $imm">,
1249 bits<6> pstatefield;
1251 let Inst{18-16} = pstatefield{5-3};
1252 let Inst{11-9} = 0b000;
1254 let Inst{7-5} = pstatefield{2-0};
1256 let DecoderMethod = "DecodeSystemPStateInstruction";
1257 // MSRpstateI aliases with MSRI. When the MSRpstateI decoder method returns
1258 // Fail the decoder should attempt to decode the instruction as MSRI.
1259 let hasCompleteDecoder = 0;
1262 // SYS and SYSL generic system instructions.
1263 def SysCRAsmOperand : AsmOperandClass {
1265 let ParserMethod = "tryParseSysCROperand";
1268 def sys_cr_op : Operand<i32> {
1269 let PrintMethod = "printSysCROperand";
1270 let ParserMatchClass = SysCRAsmOperand;
1273 class SystemXtI<bit L, string asm>
1274 : RtSystemI<L, (outs),
1275 (ins imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, GPR64:$Rt),
1276 asm, "\t$op1, $Cn, $Cm, $op2, $Rt"> {
1281 let Inst{20-19} = 0b01;
1282 let Inst{18-16} = op1;
1283 let Inst{15-12} = Cn;
1284 let Inst{11-8} = Cm;
1285 let Inst{7-5} = op2;
1288 class SystemLXtI<bit L, string asm>
1289 : RtSystemI<L, (outs),
1290 (ins GPR64:$Rt, imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2),
1291 asm, "\t$Rt, $op1, $Cn, $Cm, $op2"> {
1296 let Inst{20-19} = 0b01;
1297 let Inst{18-16} = op1;
1298 let Inst{15-12} = Cn;
1299 let Inst{11-8} = Cm;
1300 let Inst{7-5} = op2;
1304 // Branch (register) instructions:
1312 // otherwise UNDEFINED
1313 class BaseBranchReg<bits<4> opc, dag oops, dag iops, string asm,
1314 string operands, list<dag> pattern>
1315 : I<oops, iops, asm, operands, "", pattern>, Sched<[WriteBrReg]> {
1316 let Inst{31-25} = 0b1101011;
1317 let Inst{24-21} = opc;
1318 let Inst{20-16} = 0b11111;
1319 let Inst{15-10} = 0b000000;
1320 let Inst{4-0} = 0b00000;
1323 class BranchReg<bits<4> opc, string asm, list<dag> pattern>
1324 : BaseBranchReg<opc, (outs), (ins GPR64:$Rn), asm, "\t$Rn", pattern> {
1329 let mayLoad = 0, mayStore = 0, hasSideEffects = 1, isReturn = 1 in
1330 class SpecialReturn<bits<4> opc, string asm>
1331 : BaseBranchReg<opc, (outs), (ins), asm, "", []> {
1332 let Inst{9-5} = 0b11111;
1336 class RCPCLoad<bits<2> sz, string asm, RegisterClass RC>
1337 : I<(outs RC:$Rt), (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]", "", []>,
1341 let Inst{31-30} = sz;
1342 let Inst{29-10} = 0b11100010111111110000;
1347 class AuthBase<bits<1> M, dag oops, dag iops, string asm, string operands,
1349 : I<oops, iops, asm, operands, "", pattern>, Sched<[]> {
1350 let Inst{31-25} = 0b1101011;
1351 let Inst{20-11} = 0b1111100001;
1353 let Inst{4-0} = 0b11111;
1356 class AuthBranchTwoOperands<bits<1> op, bits<1> M, string asm>
1357 : AuthBase<M, (outs), (ins GPR64:$Rn, GPR64sp:$Rm), asm, "\t$Rn, $Rm", []> {
1360 let Inst{24-22} = 0b100;
1366 class AuthOneOperand<bits<3> opc, bits<1> M, string asm>
1367 : AuthBase<M, (outs), (ins GPR64:$Rn), asm, "\t$Rn", []> {
1370 let Inst{23-21} = opc;
1374 class AuthReturn<bits<3> op, bits<1> M, string asm>
1375 : AuthBase<M, (outs), (ins), asm, "", []> {
1377 let Inst{23-21} = op;
1378 let Inst{9-0} = 0b1111111111;
1382 class BaseAuthLoad<bit M, bit W, dag oops, dag iops, string asm,
1383 string operands, string cstr, Operand opr>
1384 : I<oops, iops, asm, operands, cstr, []>, Sched<[]> {
1388 let Inst{31-24} = 0b11111000;
1390 let Inst{22} = offset{9};
1392 let Inst{20-12} = offset{8-0};
1399 multiclass AuthLoad<bit M, string asm, Operand opr> {
1400 def indexed : BaseAuthLoad<M, 0, (outs GPR64:$Rt),
1401 (ins GPR64sp:$Rn, opr:$offset),
1402 asm, "\t$Rt, [$Rn, $offset]", "", opr>;
1403 def writeback : BaseAuthLoad<M, 1, (outs GPR64sp:$wback, GPR64:$Rt),
1404 (ins GPR64sp:$Rn, opr:$offset),
1405 asm, "\t$Rt, [$Rn, $offset]!",
1406 "$Rn = $wback,@earlyclobber $wback", opr>;
1408 def : InstAlias<asm # "\t$Rt, [$Rn]",
1409 (!cast<Instruction>(NAME # "indexed") GPR64:$Rt, GPR64sp:$Rn, 0)>;
1413 // Conditional branch instruction.
1417 // 4-bit immediate. Pretty-printed as <cc>
1418 def ccode : Operand<i32> {
1419 let PrintMethod = "printCondCode";
1420 let ParserMatchClass = CondCode;
1422 def inv_ccode : Operand<i32> {
1423 // AL and NV are invalid in the aliases which use inv_ccode
1424 let PrintMethod = "printInverseCondCode";
1425 let ParserMatchClass = CondCode;
1426 let MCOperandPredicate = [{
1427 return MCOp.isImm() &&
1428 MCOp.getImm() != AArch64CC::AL &&
1429 MCOp.getImm() != AArch64CC::NV;
1433 // Conditional branch target. 19-bit immediate. The low two bits of the target
1434 // offset are implied zero and so are not part of the immediate.
1435 def am_brcond : Operand<OtherVT> {
1436 let EncoderMethod = "getCondBranchTargetOpValue";
1437 let DecoderMethod = "DecodePCRelLabel19";
1438 let PrintMethod = "printAlignedLabel";
1439 let ParserMatchClass = PCRelLabel19Operand;
1440 let OperandType = "OPERAND_PCREL";
1443 class BranchCond : I<(outs), (ins ccode:$cond, am_brcond:$target),
1444 "b", ".$cond\t$target", "",
1445 [(AArch64brcond bb:$target, imm:$cond, NZCV)]>,
1448 let isTerminator = 1;
1453 let Inst{31-24} = 0b01010100;
1454 let Inst{23-5} = target;
1456 let Inst{3-0} = cond;
1460 // Compare-and-branch instructions.
1462 class BaseCmpBranch<RegisterClass regtype, bit op, string asm, SDNode node>
1463 : I<(outs), (ins regtype:$Rt, am_brcond:$target),
1464 asm, "\t$Rt, $target", "",
1465 [(node regtype:$Rt, bb:$target)]>,
1468 let isTerminator = 1;
1472 let Inst{30-25} = 0b011010;
1474 let Inst{23-5} = target;
1478 multiclass CmpBranch<bit op, string asm, SDNode node> {
1479 def W : BaseCmpBranch<GPR32, op, asm, node> {
1482 def X : BaseCmpBranch<GPR64, op, asm, node> {
1488 // Test-bit-and-branch instructions.
1490 // Test-and-branch target. 14-bit sign-extended immediate. The low two bits of
1491 // the target offset are implied zero and so are not part of the immediate.
1492 def am_tbrcond : Operand<OtherVT> {
1493 let EncoderMethod = "getTestBranchTargetOpValue";
1494 let PrintMethod = "printAlignedLabel";
1495 let ParserMatchClass = BranchTarget14Operand;
1496 let OperandType = "OPERAND_PCREL";
1499 // AsmOperand classes to emit (or not) special diagnostics
1500 def TBZImm0_31Operand : AsmOperandClass {
1501 let Name = "TBZImm0_31";
1502 let PredicateMethod = "isImmInRange<0,31>";
1503 let RenderMethod = "addImmOperands";
1505 def TBZImm32_63Operand : AsmOperandClass {
1506 let Name = "Imm32_63";
1507 let PredicateMethod = "isImmInRange<32,63>";
1508 let DiagnosticType = "InvalidImm0_63";
1509 let RenderMethod = "addImmOperands";
1512 class tbz_imm0_31<AsmOperandClass matcher> : Operand<i64>, ImmLeaf<i64, [{
1513 return (((uint32_t)Imm) < 32);
1515 let ParserMatchClass = matcher;
1518 def tbz_imm0_31_diag : tbz_imm0_31<Imm0_31Operand>;
1519 def tbz_imm0_31_nodiag : tbz_imm0_31<TBZImm0_31Operand>;
1521 def tbz_imm32_63 : Operand<i64>, ImmLeaf<i64, [{
1522 return (((uint32_t)Imm) > 31) && (((uint32_t)Imm) < 64);
1524 let ParserMatchClass = TBZImm32_63Operand;
1527 class BaseTestBranch<RegisterClass regtype, Operand immtype,
1528 bit op, string asm, SDNode node>
1529 : I<(outs), (ins regtype:$Rt, immtype:$bit_off, am_tbrcond:$target),
1530 asm, "\t$Rt, $bit_off, $target", "",
1531 [(node regtype:$Rt, immtype:$bit_off, bb:$target)]>,
1534 let isTerminator = 1;
1540 let Inst{30-25} = 0b011011;
1542 let Inst{23-19} = bit_off{4-0};
1543 let Inst{18-5} = target;
1546 let DecoderMethod = "DecodeTestAndBranch";
1549 multiclass TestBranch<bit op, string asm, SDNode node> {
1550 def W : BaseTestBranch<GPR32, tbz_imm0_31_diag, op, asm, node> {
1554 def X : BaseTestBranch<GPR64, tbz_imm32_63, op, asm, node> {
1558 // Alias X-reg with 0-31 imm to W-Reg.
1559 def : InstAlias<asm # "\t$Rd, $imm, $target",
1560 (!cast<Instruction>(NAME#"W") GPR32as64:$Rd,
1561 tbz_imm0_31_nodiag:$imm, am_tbrcond:$target), 0>;
1562 def : Pat<(node GPR64:$Rn, tbz_imm0_31_diag:$imm, bb:$target),
1563 (!cast<Instruction>(NAME#"W") (EXTRACT_SUBREG GPR64:$Rn, sub_32),
1564 tbz_imm0_31_diag:$imm, bb:$target)>;
1568 // Unconditional branch (immediate) instructions.
1570 def am_b_target : Operand<OtherVT> {
1571 let EncoderMethod = "getBranchTargetOpValue";
1572 let PrintMethod = "printAlignedLabel";
1573 let ParserMatchClass = BranchTarget26Operand;
1574 let OperandType = "OPERAND_PCREL";
1576 def am_bl_target : Operand<i64> {
1577 let EncoderMethod = "getBranchTargetOpValue";
1578 let PrintMethod = "printAlignedLabel";
1579 let ParserMatchClass = BranchTarget26Operand;
1580 let OperandType = "OPERAND_PCREL";
1583 class BImm<bit op, dag iops, string asm, list<dag> pattern>
1584 : I<(outs), iops, asm, "\t$addr", "", pattern>, Sched<[WriteBr]> {
1587 let Inst{30-26} = 0b00101;
1588 let Inst{25-0} = addr;
1590 let DecoderMethod = "DecodeUnconditionalBranch";
1593 class BranchImm<bit op, string asm, list<dag> pattern>
1594 : BImm<op, (ins am_b_target:$addr), asm, pattern>;
1595 class CallImm<bit op, string asm, list<dag> pattern>
1596 : BImm<op, (ins am_bl_target:$addr), asm, pattern>;
1599 // Basic one-operand data processing instructions.
1602 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1603 class BaseOneOperandData<bits<3> opc, RegisterClass regtype, string asm,
1604 SDPatternOperator node>
1605 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
1606 [(set regtype:$Rd, (node regtype:$Rn))]>,
1607 Sched<[WriteI, ReadI]> {
1611 let Inst{30-13} = 0b101101011000000000;
1612 let Inst{12-10} = opc;
1617 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1618 multiclass OneOperandData<bits<3> opc, string asm,
1619 SDPatternOperator node = null_frag> {
1620 def Wr : BaseOneOperandData<opc, GPR32, asm, node> {
1624 def Xr : BaseOneOperandData<opc, GPR64, asm, node> {
1629 class OneWRegData<bits<3> opc, string asm, SDPatternOperator node>
1630 : BaseOneOperandData<opc, GPR32, asm, node> {
1634 class OneXRegData<bits<3> opc, string asm, SDPatternOperator node>
1635 : BaseOneOperandData<opc, GPR64, asm, node> {
1639 class SignAuthOneData<bits<3> opcode_prefix, bits<2> opcode, string asm>
1640 : I<(outs GPR64:$Rd), (ins GPR64sp:$Rn), asm, "\t$Rd, $Rn", "",
1642 Sched<[WriteI, ReadI]> {
1645 let Inst{31-15} = 0b11011010110000010;
1646 let Inst{14-12} = opcode_prefix;
1647 let Inst{11-10} = opcode;
1652 class SignAuthZero<bits<3> opcode_prefix, bits<2> opcode, string asm>
1653 : I<(outs GPR64:$Rd), (ins), asm, "\t$Rd", "", []>, Sched<[]> {
1655 let Inst{31-15} = 0b11011010110000010;
1656 let Inst{14-12} = opcode_prefix;
1657 let Inst{11-10} = opcode;
1658 let Inst{9-5} = 0b11111;
1662 class SignAuthTwoOperand<bits<4> opc, string asm,
1663 SDPatternOperator OpNode>
1664 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64sp:$Rm),
1665 asm, "\t$Rd, $Rn, $Rm", "",
1666 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64sp:$Rm))]>,
1667 Sched<[WriteI, ReadI, ReadI]> {
1671 let Inst{31-21} = 0b10011010110;
1672 let Inst{20-16} = Rm;
1673 let Inst{15-14} = 0b00;
1674 let Inst{13-10} = opc;
1679 // Base class for the Armv8.4-A 8 and 16-bit flag manipulation instructions
1680 class BaseFlagManipulation<bit sf, bit sz, dag iops, string asm, string ops>
1681 : I<(outs), iops, asm, ops, "", []>,
1682 Sched<[WriteI, ReadI, ReadI]> {
1686 let Inst{30-15} = 0b0111010000000000;
1688 let Inst{13-10} = 0b0010;
1690 let Inst{4-0} = 0b01101;
1693 class FlagRotate<dag iops, string asm, string ops>
1694 : BaseFlagManipulation<0b1, 0b0, iops, asm, ops> {
1697 let Inst{20-15} = imm;
1698 let Inst{13-10} = 0b0001;
1700 let Inst{3-0} = mask;
1704 // Basic two-operand data processing instructions.
1706 class BaseBaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1708 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1709 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
1710 Sched<[WriteI, ReadI, ReadI]> {
1715 let Inst{30} = isSub;
1716 let Inst{28-21} = 0b11010000;
1717 let Inst{20-16} = Rm;
1718 let Inst{15-10} = 0;
1723 class BaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1725 : BaseBaseAddSubCarry<isSub, regtype, asm,
1726 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, NZCV))]>;
1728 class BaseAddSubCarrySetFlags<bit isSub, RegisterClass regtype, string asm,
1730 : BaseBaseAddSubCarry<isSub, regtype, asm,
1731 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, NZCV)),
1736 multiclass AddSubCarry<bit isSub, string asm, string asm_setflags,
1737 SDNode OpNode, SDNode OpNode_setflags> {
1738 def Wr : BaseAddSubCarry<isSub, GPR32, asm, OpNode> {
1742 def Xr : BaseAddSubCarry<isSub, GPR64, asm, OpNode> {
1748 def SWr : BaseAddSubCarrySetFlags<isSub, GPR32, asm_setflags,
1753 def SXr : BaseAddSubCarrySetFlags<isSub, GPR64, asm_setflags,
1760 class BaseTwoOperand<bits<4> opc, RegisterClass regtype, string asm,
1761 SDPatternOperator OpNode,
1762 RegisterClass in1regtype = regtype,
1763 RegisterClass in2regtype = regtype>
1764 : I<(outs regtype:$Rd), (ins in1regtype:$Rn, in2regtype:$Rm),
1765 asm, "\t$Rd, $Rn, $Rm", "",
1766 [(set regtype:$Rd, (OpNode in1regtype:$Rn, in2regtype:$Rm))]> {
1770 let Inst{30-21} = 0b0011010110;
1771 let Inst{20-16} = Rm;
1772 let Inst{15-14} = 0b00;
1773 let Inst{13-10} = opc;
1778 class BaseDiv<bit isSigned, RegisterClass regtype, string asm,
1779 SDPatternOperator OpNode>
1780 : BaseTwoOperand<{0,0,1,?}, regtype, asm, OpNode> {
1781 let Inst{10} = isSigned;
1784 multiclass Div<bit isSigned, string asm, SDPatternOperator OpNode> {
1785 def Wr : BaseDiv<isSigned, GPR32, asm, OpNode>,
1786 Sched<[WriteID32, ReadID, ReadID]> {
1789 def Xr : BaseDiv<isSigned, GPR64, asm, OpNode>,
1790 Sched<[WriteID64, ReadID, ReadID]> {
1795 class BaseShift<bits<2> shift_type, RegisterClass regtype, string asm,
1796 SDPatternOperator OpNode = null_frag>
1797 : BaseTwoOperand<{1,0,?,?}, regtype, asm, OpNode>,
1798 Sched<[WriteIS, ReadI]> {
1799 let Inst{11-10} = shift_type;
1802 multiclass Shift<bits<2> shift_type, string asm, SDNode OpNode> {
1803 def Wr : BaseShift<shift_type, GPR32, asm> {
1807 def Xr : BaseShift<shift_type, GPR64, asm, OpNode> {
1811 def : Pat<(i32 (OpNode GPR32:$Rn, i64:$Rm)),
1812 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn,
1813 (EXTRACT_SUBREG i64:$Rm, sub_32))>;
1815 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (zext GPR32:$Rm)))),
1816 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1818 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (anyext GPR32:$Rm)))),
1819 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1821 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (sext GPR32:$Rm)))),
1822 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1824 def : Pat<(i64 (OpNode GPR64:$Rn, (i64 (sext GPR32:$Rm)))),
1825 (!cast<Instruction>(NAME # "Xr") GPR64:$Rn,
1826 (SUBREG_TO_REG (i32 0), GPR32:$Rm, sub_32))>;
1828 def : Pat<(i64 (OpNode GPR64:$Rn, (i64 (zext GPR32:$Rm)))),
1829 (!cast<Instruction>(NAME # "Xr") GPR64:$Rn,
1830 (SUBREG_TO_REG (i32 0), GPR32:$Rm, sub_32))>;
1833 class ShiftAlias<string asm, Instruction inst, RegisterClass regtype>
1834 : InstAlias<asm#"\t$dst, $src1, $src2",
1835 (inst regtype:$dst, regtype:$src1, regtype:$src2), 0>;
1837 class BaseMulAccum<bit isSub, bits<3> opc, RegisterClass multype,
1838 RegisterClass addtype, string asm,
1840 : I<(outs addtype:$Rd), (ins multype:$Rn, multype:$Rm, addtype:$Ra),
1841 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pattern> {
1846 let Inst{30-24} = 0b0011011;
1847 let Inst{23-21} = opc;
1848 let Inst{20-16} = Rm;
1849 let Inst{15} = isSub;
1850 let Inst{14-10} = Ra;
1855 multiclass MulAccum<bit isSub, string asm, SDNode AccNode> {
1856 // MADD/MSUB generation is decided by MachineCombiner.cpp
1857 def Wrrr : BaseMulAccum<isSub, 0b000, GPR32, GPR32, asm,
1858 [/*(set GPR32:$Rd, (AccNode GPR32:$Ra, (mul GPR32:$Rn, GPR32:$Rm)))*/]>,
1859 Sched<[WriteIM32, ReadIM, ReadIM, ReadIMA]> {
1863 def Xrrr : BaseMulAccum<isSub, 0b000, GPR64, GPR64, asm,
1864 [/*(set GPR64:$Rd, (AccNode GPR64:$Ra, (mul GPR64:$Rn, GPR64:$Rm)))*/]>,
1865 Sched<[WriteIM64, ReadIM, ReadIM, ReadIMA]> {
1870 class WideMulAccum<bit isSub, bits<3> opc, string asm,
1871 SDNode AccNode, SDNode ExtNode>
1872 : BaseMulAccum<isSub, opc, GPR32, GPR64, asm,
1873 [(set GPR64:$Rd, (AccNode GPR64:$Ra,
1874 (mul (ExtNode GPR32:$Rn), (ExtNode GPR32:$Rm))))]>,
1875 Sched<[WriteIM32, ReadIM, ReadIM, ReadIMA]> {
1879 class MulHi<bits<3> opc, string asm, SDNode OpNode>
1880 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64:$Rm),
1881 asm, "\t$Rd, $Rn, $Rm", "",
1882 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64:$Rm))]>,
1883 Sched<[WriteIM64, ReadIM, ReadIM]> {
1887 let Inst{31-24} = 0b10011011;
1888 let Inst{23-21} = opc;
1889 let Inst{20-16} = Rm;
1894 // The Ra field of SMULH and UMULH is unused: it should be assembled as 31
1895 // (i.e. all bits 1) but is ignored by the processor.
1896 let PostEncoderMethod = "fixMulHigh";
1899 class MulAccumWAlias<string asm, Instruction inst>
1900 : InstAlias<asm#"\t$dst, $src1, $src2",
1901 (inst GPR32:$dst, GPR32:$src1, GPR32:$src2, WZR)>;
1902 class MulAccumXAlias<string asm, Instruction inst>
1903 : InstAlias<asm#"\t$dst, $src1, $src2",
1904 (inst GPR64:$dst, GPR64:$src1, GPR64:$src2, XZR)>;
1905 class WideMulAccumAlias<string asm, Instruction inst>
1906 : InstAlias<asm#"\t$dst, $src1, $src2",
1907 (inst GPR64:$dst, GPR32:$src1, GPR32:$src2, XZR)>;
1909 class BaseCRC32<bit sf, bits<2> sz, bit C, RegisterClass StreamReg,
1910 SDPatternOperator OpNode, string asm>
1911 : I<(outs GPR32:$Rd), (ins GPR32:$Rn, StreamReg:$Rm),
1912 asm, "\t$Rd, $Rn, $Rm", "",
1913 [(set GPR32:$Rd, (OpNode GPR32:$Rn, StreamReg:$Rm))]>,
1914 Sched<[WriteISReg, ReadI, ReadISReg]> {
1920 let Inst{30-21} = 0b0011010110;
1921 let Inst{20-16} = Rm;
1922 let Inst{15-13} = 0b010;
1924 let Inst{11-10} = sz;
1927 let Predicates = [HasCRC];
1931 // Address generation.
1934 class ADRI<bit page, string asm, Operand adr, list<dag> pattern>
1935 : I<(outs GPR64:$Xd), (ins adr:$label), asm, "\t$Xd, $label", "",
1940 let Inst{31} = page;
1941 let Inst{30-29} = label{1-0};
1942 let Inst{28-24} = 0b10000;
1943 let Inst{23-5} = label{20-2};
1946 let DecoderMethod = "DecodeAdrInstruction";
1953 def movimm32_imm : Operand<i32> {
1954 let ParserMatchClass = AsmImmRange<0, 65535>;
1955 let EncoderMethod = "getMoveWideImmOpValue";
1956 let PrintMethod = "printImm";
1958 def movimm32_shift : Operand<i32> {
1959 let PrintMethod = "printShifter";
1960 let ParserMatchClass = MovImm32ShifterOperand;
1962 def movimm64_shift : Operand<i32> {
1963 let PrintMethod = "printShifter";
1964 let ParserMatchClass = MovImm64ShifterOperand;
1967 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1968 class BaseMoveImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1970 : I<(outs regtype:$Rd), (ins movimm32_imm:$imm, shifter:$shift),
1971 asm, "\t$Rd, $imm$shift", "", []>,
1976 let Inst{30-29} = opc;
1977 let Inst{28-23} = 0b100101;
1978 let Inst{22-21} = shift{5-4};
1979 let Inst{20-5} = imm;
1982 let DecoderMethod = "DecodeMoveImmInstruction";
1985 multiclass MoveImmediate<bits<2> opc, string asm> {
1986 def Wi : BaseMoveImmediate<opc, GPR32, movimm32_shift, asm> {
1990 def Xi : BaseMoveImmediate<opc, GPR64, movimm64_shift, asm> {
1995 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1996 class BaseInsertImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1998 : I<(outs regtype:$Rd),
1999 (ins regtype:$src, movimm32_imm:$imm, shifter:$shift),
2000 asm, "\t$Rd, $imm$shift", "$src = $Rd", []>,
2001 Sched<[WriteI, ReadI]> {
2005 let Inst{30-29} = opc;
2006 let Inst{28-23} = 0b100101;
2007 let Inst{22-21} = shift{5-4};
2008 let Inst{20-5} = imm;
2011 let DecoderMethod = "DecodeMoveImmInstruction";
2014 multiclass InsertImmediate<bits<2> opc, string asm> {
2015 def Wi : BaseInsertImmediate<opc, GPR32, movimm32_shift, asm> {
2019 def Xi : BaseInsertImmediate<opc, GPR64, movimm64_shift, asm> {
2028 class BaseAddSubImm<bit isSub, bit setFlags, RegisterClass dstRegtype,
2029 string asm_inst, string asm_ops,
2030 dag inputs, dag pattern>
2031 : I<(outs dstRegtype:$Rd), inputs, asm_inst, asm_ops, "", [pattern]>,
2032 Sched<[WriteI, ReadI]> {
2035 let Inst{30} = isSub;
2036 let Inst{29} = setFlags;
2037 let Inst{28-24} = 0b10001;
2042 class AddSubImmShift<bit isSub, bit setFlags, RegisterClass dstRegtype,
2043 RegisterClass srcRegtype, addsub_shifted_imm immtype,
2044 string asm_inst, SDPatternOperator OpNode>
2045 : BaseAddSubImm<isSub, setFlags, dstRegtype, asm_inst, "\t$Rd, $Rn, $imm",
2046 (ins srcRegtype:$Rn, immtype:$imm),
2047 (set dstRegtype:$Rd, (OpNode srcRegtype:$Rn, immtype:$imm))> {
2049 let Inst{23-22} = imm{13-12}; // '00' => lsl #0, '01' => lsl #12
2050 let Inst{21-10} = imm{11-0};
2051 let DecoderMethod = "DecodeAddSubImmShift";
2054 class BaseAddSubRegPseudo<RegisterClass regtype,
2055 SDPatternOperator OpNode>
2056 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
2057 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
2058 Sched<[WriteI, ReadI, ReadI]>;
2060 class BaseAddSubSReg<bit isSub, bit setFlags, RegisterClass regtype,
2061 arith_shifted_reg shifted_regtype, string asm,
2062 SDPatternOperator OpNode>
2063 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
2064 asm, "\t$Rd, $Rn, $Rm", "",
2065 [(set regtype:$Rd, (OpNode regtype:$Rn, shifted_regtype:$Rm))]>,
2066 Sched<[WriteISReg, ReadI, ReadISReg]> {
2067 // The operands are in order to match the 'addr' MI operands, so we
2068 // don't need an encoder method and by-name matching. Just use the default
2069 // in-order handling. Since we're using by-order, make sure the names
2075 let Inst{30} = isSub;
2076 let Inst{29} = setFlags;
2077 let Inst{28-24} = 0b01011;
2078 let Inst{23-22} = shift{7-6};
2080 let Inst{20-16} = src2;
2081 let Inst{15-10} = shift{5-0};
2082 let Inst{9-5} = src1;
2083 let Inst{4-0} = dst;
2085 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
2088 class BaseAddSubEReg<bit isSub, bit setFlags, RegisterClass dstRegtype,
2089 RegisterClass src1Regtype, Operand src2Regtype,
2090 string asm, SDPatternOperator OpNode>
2091 : I<(outs dstRegtype:$R1),
2092 (ins src1Regtype:$R2, src2Regtype:$R3),
2093 asm, "\t$R1, $R2, $R3", "",
2094 [(set dstRegtype:$R1, (OpNode src1Regtype:$R2, src2Regtype:$R3))]>,
2095 Sched<[WriteIEReg, ReadI, ReadIEReg]> {
2100 let Inst{30} = isSub;
2101 let Inst{29} = setFlags;
2102 let Inst{28-24} = 0b01011;
2103 let Inst{23-21} = 0b001;
2104 let Inst{20-16} = Rm;
2105 let Inst{15-13} = ext{5-3};
2106 let Inst{12-10} = ext{2-0};
2110 let DecoderMethod = "DecodeAddSubERegInstruction";
2113 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2114 class BaseAddSubEReg64<bit isSub, bit setFlags, RegisterClass dstRegtype,
2115 RegisterClass src1Regtype, RegisterClass src2Regtype,
2116 Operand ext_op, string asm>
2117 : I<(outs dstRegtype:$Rd),
2118 (ins src1Regtype:$Rn, src2Regtype:$Rm, ext_op:$ext),
2119 asm, "\t$Rd, $Rn, $Rm$ext", "", []>,
2120 Sched<[WriteIEReg, ReadI, ReadIEReg]> {
2125 let Inst{30} = isSub;
2126 let Inst{29} = setFlags;
2127 let Inst{28-24} = 0b01011;
2128 let Inst{23-21} = 0b001;
2129 let Inst{20-16} = Rm;
2130 let Inst{15} = ext{5};
2131 let Inst{12-10} = ext{2-0};
2135 let DecoderMethod = "DecodeAddSubERegInstruction";
2138 // Aliases for register+register add/subtract.
2139 class AddSubRegAlias<string asm, Instruction inst, RegisterClass dstRegtype,
2140 RegisterClass src1Regtype, RegisterClass src2Regtype,
2142 : InstAlias<asm#"\t$dst, $src1, $src2",
2143 (inst dstRegtype:$dst, src1Regtype:$src1, src2Regtype:$src2,
2146 multiclass AddSub<bit isSub, string mnemonic, string alias,
2147 SDPatternOperator OpNode = null_frag> {
2148 let hasSideEffects = 0, isReMaterializable = 1, isAsCheapAsAMove = 1 in {
2149 // Add/Subtract immediate
2150 // Increase the weight of the immediate variant to try to match it before
2151 // the extended register variant.
2152 // We used to match the register variant before the immediate when the
2153 // register argument could be implicitly zero-extended.
2154 let AddedComplexity = 6 in
2155 def Wri : AddSubImmShift<isSub, 0, GPR32sp, GPR32sp, addsub_shifted_imm32,
2159 let AddedComplexity = 6 in
2160 def Xri : AddSubImmShift<isSub, 0, GPR64sp, GPR64sp, addsub_shifted_imm64,
2165 // Add/Subtract register - Only used for CodeGen
2166 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
2167 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
2169 // Add/Subtract shifted register
2170 def Wrs : BaseAddSubSReg<isSub, 0, GPR32, arith_shifted_reg32, mnemonic,
2174 def Xrs : BaseAddSubSReg<isSub, 0, GPR64, arith_shifted_reg64, mnemonic,
2180 // Add/Subtract extended register
2181 let AddedComplexity = 1, hasSideEffects = 0 in {
2182 def Wrx : BaseAddSubEReg<isSub, 0, GPR32sp, GPR32sp,
2183 arith_extended_reg32<i32>, mnemonic, OpNode> {
2186 def Xrx : BaseAddSubEReg<isSub, 0, GPR64sp, GPR64sp,
2187 arith_extended_reg32to64<i64>, mnemonic, OpNode> {
2192 def Xrx64 : BaseAddSubEReg64<isSub, 0, GPR64sp, GPR64sp, GPR64,
2193 arith_extendlsl64, mnemonic> {
2194 // UXTX and SXTX only.
2195 let Inst{14-13} = 0b11;
2199 // add Rd, Rb, -imm -> sub Rd, Rn, imm
2200 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2201 (!cast<Instruction>(NAME # "Wri") GPR32sp:$Rd, GPR32sp:$Rn,
2202 addsub_shifted_imm32_neg:$imm), 0>;
2203 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2204 (!cast<Instruction>(NAME # "Xri") GPR64sp:$Rd, GPR64sp:$Rn,
2205 addsub_shifted_imm64_neg:$imm), 0>;
2207 // Register/register aliases with no shift when SP is not used.
2208 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
2209 GPR32, GPR32, GPR32, 0>;
2210 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
2211 GPR64, GPR64, GPR64, 0>;
2213 // Register/register aliases with no shift when either the destination or
2214 // first source register is SP.
2215 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2216 GPR32sponly, GPR32sp, GPR32, 16>; // UXTW #0
2217 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2218 GPR32sp, GPR32sponly, GPR32, 16>; // UXTW #0
2219 def : AddSubRegAlias<mnemonic,
2220 !cast<Instruction>(NAME#"Xrx64"),
2221 GPR64sponly, GPR64sp, GPR64, 24>; // UXTX #0
2222 def : AddSubRegAlias<mnemonic,
2223 !cast<Instruction>(NAME#"Xrx64"),
2224 GPR64sp, GPR64sponly, GPR64, 24>; // UXTX #0
2227 multiclass AddSubS<bit isSub, string mnemonic, SDNode OpNode, string cmp,
2228 string alias, string cmpAlias> {
2229 let isCompare = 1, Defs = [NZCV] in {
2230 // Add/Subtract immediate
2231 def Wri : AddSubImmShift<isSub, 1, GPR32, GPR32sp, addsub_shifted_imm32,
2235 def Xri : AddSubImmShift<isSub, 1, GPR64, GPR64sp, addsub_shifted_imm64,
2240 // Add/Subtract register
2241 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
2242 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
2244 // Add/Subtract shifted register
2245 def Wrs : BaseAddSubSReg<isSub, 1, GPR32, arith_shifted_reg32, mnemonic,
2249 def Xrs : BaseAddSubSReg<isSub, 1, GPR64, arith_shifted_reg64, mnemonic,
2254 // Add/Subtract extended register
2255 let AddedComplexity = 1 in {
2256 def Wrx : BaseAddSubEReg<isSub, 1, GPR32, GPR32sp,
2257 arith_extended_reg32<i32>, mnemonic, OpNode> {
2260 def Xrx : BaseAddSubEReg<isSub, 1, GPR64, GPR64sp,
2261 arith_extended_reg32<i64>, mnemonic, OpNode> {
2266 def Xrx64 : BaseAddSubEReg64<isSub, 1, GPR64, GPR64sp, GPR64,
2267 arith_extendlsl64, mnemonic> {
2268 // UXTX and SXTX only.
2269 let Inst{14-13} = 0b11;
2274 // Support negative immediates, e.g. adds Rd, Rn, -imm -> subs Rd, Rn, imm
2275 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2276 (!cast<Instruction>(NAME # "Wri") GPR32:$Rd, GPR32sp:$Rn,
2277 addsub_shifted_imm32_neg:$imm), 0>;
2278 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2279 (!cast<Instruction>(NAME # "Xri") GPR64:$Rd, GPR64sp:$Rn,
2280 addsub_shifted_imm64_neg:$imm), 0>;
2283 def : InstAlias<cmp#"\t$src, $imm", (!cast<Instruction>(NAME#"Wri")
2284 WZR, GPR32sp:$src, addsub_shifted_imm32:$imm), 5>;
2285 def : InstAlias<cmp#"\t$src, $imm", (!cast<Instruction>(NAME#"Xri")
2286 XZR, GPR64sp:$src, addsub_shifted_imm64:$imm), 5>;
2287 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Wrx")
2288 WZR, GPR32sp:$src1, GPR32:$src2, arith_extend:$sh), 4>;
2289 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrx")
2290 XZR, GPR64sp:$src1, GPR32:$src2, arith_extend:$sh), 4>;
2291 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrx64")
2292 XZR, GPR64sp:$src1, GPR64:$src2, arith_extendlsl64:$sh), 4>;
2293 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Wrs")
2294 WZR, GPR32:$src1, GPR32:$src2, arith_shift32:$sh), 4>;
2295 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrs")
2296 XZR, GPR64:$src1, GPR64:$src2, arith_shift64:$sh), 4>;
2298 // Support negative immediates, e.g. cmp Rn, -imm -> cmn Rn, imm
2299 def : InstSubst<cmpAlias#"\t$src, $imm", (!cast<Instruction>(NAME#"Wri")
2300 WZR, GPR32sp:$src, addsub_shifted_imm32_neg:$imm), 0>;
2301 def : InstSubst<cmpAlias#"\t$src, $imm", (!cast<Instruction>(NAME#"Xri")
2302 XZR, GPR64sp:$src, addsub_shifted_imm64_neg:$imm), 0>;
2304 // Compare shorthands
2305 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Wrs")
2306 WZR, GPR32:$src1, GPR32:$src2, 0), 5>;
2307 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Xrs")
2308 XZR, GPR64:$src1, GPR64:$src2, 0), 5>;
2309 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Wrx")
2310 WZR, GPR32sponly:$src1, GPR32:$src2, 16), 5>;
2311 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Xrx64")
2312 XZR, GPR64sponly:$src1, GPR64:$src2, 24), 5>;
2314 // Register/register aliases with no shift when SP is not used.
2315 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
2316 GPR32, GPR32, GPR32, 0>;
2317 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
2318 GPR64, GPR64, GPR64, 0>;
2320 // Register/register aliases with no shift when the first source register
2322 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2323 GPR32, GPR32sponly, GPR32, 16>; // UXTW #0
2324 def : AddSubRegAlias<mnemonic,
2325 !cast<Instruction>(NAME#"Xrx64"),
2326 GPR64, GPR64sponly, GPR64, 24>; // UXTX #0
2329 class AddSubG<bit isSub, string asm_inst, SDPatternOperator OpNode>
2331 isSub, 0, GPR64sp, asm_inst, "\t$Rd, $Rn, $imm6, $imm4",
2332 (ins GPR64sp:$Rn, uimm6s16:$imm6, imm0_15:$imm4),
2333 (set GPR64sp:$Rd, (OpNode GPR64sp:$Rn, imm0_63:$imm6, imm0_15:$imm4))> {
2337 let Inst{23-22} = 0b10;
2338 let Inst{21-16} = imm6;
2339 let Inst{15-14} = 0b00;
2340 let Inst{13-10} = imm4;
2341 let Unpredictable{15-14} = 0b11;
2344 class SUBP<bit setsFlags, string asm_instr, SDPatternOperator OpNode>
2345 : BaseTwoOperand<0b0000, GPR64, asm_instr, OpNode, GPR64sp, GPR64sp> {
2347 let Inst{29} = setsFlags;
2353 def SDTA64EXTR : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>,
2355 def AArch64Extr : SDNode<"AArch64ISD::EXTR", SDTA64EXTR>;
2357 class BaseExtractImm<RegisterClass regtype, Operand imm_type, string asm,
2359 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, imm_type:$imm),
2360 asm, "\t$Rd, $Rn, $Rm, $imm", "", patterns>,
2361 Sched<[WriteExtr, ReadExtrHi]> {
2367 let Inst{30-23} = 0b00100111;
2369 let Inst{20-16} = Rm;
2370 let Inst{15-10} = imm;
2375 multiclass ExtractImm<string asm> {
2376 def Wrri : BaseExtractImm<GPR32, imm0_31, asm,
2378 (AArch64Extr GPR32:$Rn, GPR32:$Rm, imm0_31:$imm))]> {
2381 // imm<5> must be zero.
2384 def Xrri : BaseExtractImm<GPR64, imm0_63, asm,
2386 (AArch64Extr GPR64:$Rn, GPR64:$Rm, imm0_63:$imm))]> {
2397 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2398 class BaseBitfieldImm<bits<2> opc,
2399 RegisterClass regtype, Operand imm_type, string asm>
2400 : I<(outs regtype:$Rd), (ins regtype:$Rn, imm_type:$immr, imm_type:$imms),
2401 asm, "\t$Rd, $Rn, $immr, $imms", "", []>,
2402 Sched<[WriteIS, ReadI]> {
2408 let Inst{30-29} = opc;
2409 let Inst{28-23} = 0b100110;
2410 let Inst{21-16} = immr;
2411 let Inst{15-10} = imms;
2416 multiclass BitfieldImm<bits<2> opc, string asm> {
2417 def Wri : BaseBitfieldImm<opc, GPR32, imm0_31, asm> {
2420 // imms<5> and immr<5> must be zero, else ReservedValue().
2424 def Xri : BaseBitfieldImm<opc, GPR64, imm0_63, asm> {
2430 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2431 class BaseBitfieldImmWith2RegArgs<bits<2> opc,
2432 RegisterClass regtype, Operand imm_type, string asm>
2433 : I<(outs regtype:$Rd), (ins regtype:$src, regtype:$Rn, imm_type:$immr,
2435 asm, "\t$Rd, $Rn, $immr, $imms", "$src = $Rd", []>,
2436 Sched<[WriteIS, ReadI]> {
2442 let Inst{30-29} = opc;
2443 let Inst{28-23} = 0b100110;
2444 let Inst{21-16} = immr;
2445 let Inst{15-10} = imms;
2450 multiclass BitfieldImmWith2RegArgs<bits<2> opc, string asm> {
2451 def Wri : BaseBitfieldImmWith2RegArgs<opc, GPR32, imm0_31, asm> {
2454 // imms<5> and immr<5> must be zero, else ReservedValue().
2458 def Xri : BaseBitfieldImmWith2RegArgs<opc, GPR64, imm0_63, asm> {
2468 // Logical (immediate)
2469 class BaseLogicalImm<bits<2> opc, RegisterClass dregtype,
2470 RegisterClass sregtype, Operand imm_type, string asm,
2472 : I<(outs dregtype:$Rd), (ins sregtype:$Rn, imm_type:$imm),
2473 asm, "\t$Rd, $Rn, $imm", "", pattern>,
2474 Sched<[WriteI, ReadI]> {
2478 let Inst{30-29} = opc;
2479 let Inst{28-23} = 0b100100;
2480 let Inst{22} = imm{12};
2481 let Inst{21-16} = imm{11-6};
2482 let Inst{15-10} = imm{5-0};
2486 let DecoderMethod = "DecodeLogicalImmInstruction";
2489 // Logical (shifted register)
2490 class BaseLogicalSReg<bits<2> opc, bit N, RegisterClass regtype,
2491 logical_shifted_reg shifted_regtype, string asm,
2493 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
2494 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
2495 Sched<[WriteISReg, ReadI, ReadISReg]> {
2496 // The operands are in order to match the 'addr' MI operands, so we
2497 // don't need an encoder method and by-name matching. Just use the default
2498 // in-order handling. Since we're using by-order, make sure the names
2504 let Inst{30-29} = opc;
2505 let Inst{28-24} = 0b01010;
2506 let Inst{23-22} = shift{7-6};
2508 let Inst{20-16} = src2;
2509 let Inst{15-10} = shift{5-0};
2510 let Inst{9-5} = src1;
2511 let Inst{4-0} = dst;
2513 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
2516 // Aliases for register+register logical instructions.
2517 class LogicalRegAlias<string asm, Instruction inst, RegisterClass regtype>
2518 : InstAlias<asm#"\t$dst, $src1, $src2",
2519 (inst regtype:$dst, regtype:$src1, regtype:$src2, 0)>;
2521 multiclass LogicalImm<bits<2> opc, string mnemonic, SDNode OpNode,
2523 let AddedComplexity = 6, isReMaterializable = 1, isAsCheapAsAMove = 1 in
2524 def Wri : BaseLogicalImm<opc, GPR32sp, GPR32, logical_imm32, mnemonic,
2525 [(set GPR32sp:$Rd, (OpNode GPR32:$Rn,
2526 logical_imm32:$imm))]> {
2528 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
2530 let AddedComplexity = 6, isReMaterializable = 1, isAsCheapAsAMove = 1 in
2531 def Xri : BaseLogicalImm<opc, GPR64sp, GPR64, logical_imm64, mnemonic,
2532 [(set GPR64sp:$Rd, (OpNode GPR64:$Rn,
2533 logical_imm64:$imm))]> {
2537 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2538 (!cast<Instruction>(NAME # "Wri") GPR32sp:$Rd, GPR32:$Rn,
2539 logical_imm32_not:$imm), 0>;
2540 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2541 (!cast<Instruction>(NAME # "Xri") GPR64sp:$Rd, GPR64:$Rn,
2542 logical_imm64_not:$imm), 0>;
2545 multiclass LogicalImmS<bits<2> opc, string mnemonic, SDNode OpNode,
2547 let isCompare = 1, Defs = [NZCV] in {
2548 def Wri : BaseLogicalImm<opc, GPR32, GPR32, logical_imm32, mnemonic,
2549 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_imm32:$imm))]> {
2551 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
2553 def Xri : BaseLogicalImm<opc, GPR64, GPR64, logical_imm64, mnemonic,
2554 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_imm64:$imm))]> {
2557 } // end Defs = [NZCV]
2559 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2560 (!cast<Instruction>(NAME # "Wri") GPR32:$Rd, GPR32:$Rn,
2561 logical_imm32_not:$imm), 0>;
2562 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2563 (!cast<Instruction>(NAME # "Xri") GPR64:$Rd, GPR64:$Rn,
2564 logical_imm64_not:$imm), 0>;
2567 class BaseLogicalRegPseudo<RegisterClass regtype, SDPatternOperator OpNode>
2568 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
2569 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
2570 Sched<[WriteI, ReadI, ReadI]>;
2572 // Split from LogicalImm as not all instructions have both.
2573 multiclass LogicalReg<bits<2> opc, bit N, string mnemonic,
2574 SDPatternOperator OpNode> {
2575 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
2576 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
2577 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
2580 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
2581 [(set GPR32:$Rd, (OpNode GPR32:$Rn,
2582 logical_shifted_reg32:$Rm))]> {
2585 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
2586 [(set GPR64:$Rd, (OpNode GPR64:$Rn,
2587 logical_shifted_reg64:$Rm))]> {
2591 def : LogicalRegAlias<mnemonic,
2592 !cast<Instruction>(NAME#"Wrs"), GPR32>;
2593 def : LogicalRegAlias<mnemonic,
2594 !cast<Instruction>(NAME#"Xrs"), GPR64>;
2597 // Split from LogicalReg to allow setting NZCV Defs
2598 multiclass LogicalRegS<bits<2> opc, bit N, string mnemonic,
2599 SDPatternOperator OpNode = null_frag> {
2600 let Defs = [NZCV], mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
2601 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
2602 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
2604 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
2605 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_shifted_reg32:$Rm))]> {
2608 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
2609 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_shifted_reg64:$Rm))]> {
2614 def : LogicalRegAlias<mnemonic,
2615 !cast<Instruction>(NAME#"Wrs"), GPR32>;
2616 def : LogicalRegAlias<mnemonic,
2617 !cast<Instruction>(NAME#"Xrs"), GPR64>;
2621 // Conditionally set flags
2624 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2625 class BaseCondComparisonImm<bit op, RegisterClass regtype, ImmLeaf immtype,
2626 string mnemonic, SDNode OpNode>
2627 : I<(outs), (ins regtype:$Rn, immtype:$imm, imm32_0_15:$nzcv, ccode:$cond),
2628 mnemonic, "\t$Rn, $imm, $nzcv, $cond", "",
2629 [(set NZCV, (OpNode regtype:$Rn, immtype:$imm, (i32 imm:$nzcv),
2630 (i32 imm:$cond), NZCV))]>,
2631 Sched<[WriteI, ReadI]> {
2641 let Inst{29-21} = 0b111010010;
2642 let Inst{20-16} = imm;
2643 let Inst{15-12} = cond;
2644 let Inst{11-10} = 0b10;
2647 let Inst{3-0} = nzcv;
2650 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2651 class BaseCondComparisonReg<bit op, RegisterClass regtype, string mnemonic,
2653 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm32_0_15:$nzcv, ccode:$cond),
2654 mnemonic, "\t$Rn, $Rm, $nzcv, $cond", "",
2655 [(set NZCV, (OpNode regtype:$Rn, regtype:$Rm, (i32 imm:$nzcv),
2656 (i32 imm:$cond), NZCV))]>,
2657 Sched<[WriteI, ReadI, ReadI]> {
2667 let Inst{29-21} = 0b111010010;
2668 let Inst{20-16} = Rm;
2669 let Inst{15-12} = cond;
2670 let Inst{11-10} = 0b00;
2673 let Inst{3-0} = nzcv;
2676 multiclass CondComparison<bit op, string mnemonic, SDNode OpNode> {
2677 // immediate operand variants
2678 def Wi : BaseCondComparisonImm<op, GPR32, imm32_0_31, mnemonic, OpNode> {
2681 def Xi : BaseCondComparisonImm<op, GPR64, imm0_31, mnemonic, OpNode> {
2684 // register operand variants
2685 def Wr : BaseCondComparisonReg<op, GPR32, mnemonic, OpNode> {
2688 def Xr : BaseCondComparisonReg<op, GPR64, mnemonic, OpNode> {
2694 // Conditional select
2697 class BaseCondSelect<bit op, bits<2> op2, RegisterClass regtype, string asm>
2698 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
2699 asm, "\t$Rd, $Rn, $Rm, $cond", "",
2701 (AArch64csel regtype:$Rn, regtype:$Rm, (i32 imm:$cond), NZCV))]>,
2702 Sched<[WriteI, ReadI, ReadI]> {
2711 let Inst{29-21} = 0b011010100;
2712 let Inst{20-16} = Rm;
2713 let Inst{15-12} = cond;
2714 let Inst{11-10} = op2;
2719 multiclass CondSelect<bit op, bits<2> op2, string asm> {
2720 def Wr : BaseCondSelect<op, op2, GPR32, asm> {
2723 def Xr : BaseCondSelect<op, op2, GPR64, asm> {
2728 class BaseCondSelectOp<bit op, bits<2> op2, RegisterClass regtype, string asm,
2730 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
2731 asm, "\t$Rd, $Rn, $Rm, $cond", "",
2733 (AArch64csel regtype:$Rn, (frag regtype:$Rm),
2734 (i32 imm:$cond), NZCV))]>,
2735 Sched<[WriteI, ReadI, ReadI]> {
2744 let Inst{29-21} = 0b011010100;
2745 let Inst{20-16} = Rm;
2746 let Inst{15-12} = cond;
2747 let Inst{11-10} = op2;
2752 def inv_cond_XFORM : SDNodeXForm<imm, [{
2753 AArch64CC::CondCode CC = static_cast<AArch64CC::CondCode>(N->getZExtValue());
2754 return CurDAG->getTargetConstant(AArch64CC::getInvertedCondCode(CC), SDLoc(N),
2758 multiclass CondSelectOp<bit op, bits<2> op2, string asm, PatFrag frag> {
2759 def Wr : BaseCondSelectOp<op, op2, GPR32, asm, frag> {
2762 def Xr : BaseCondSelectOp<op, op2, GPR64, asm, frag> {
2766 def : Pat<(AArch64csel (frag GPR32:$Rm), GPR32:$Rn, (i32 imm:$cond), NZCV),
2767 (!cast<Instruction>(NAME # Wr) GPR32:$Rn, GPR32:$Rm,
2768 (inv_cond_XFORM imm:$cond))>;
2770 def : Pat<(AArch64csel (frag GPR64:$Rm), GPR64:$Rn, (i32 imm:$cond), NZCV),
2771 (!cast<Instruction>(NAME # Xr) GPR64:$Rn, GPR64:$Rm,
2772 (inv_cond_XFORM imm:$cond))>;
2776 // Special Mask Value
2778 def maski8_or_more : Operand<i32>,
2779 ImmLeaf<i32, [{ return (Imm & 0xff) == 0xff; }]> {
2781 def maski16_or_more : Operand<i32>,
2782 ImmLeaf<i32, [{ return (Imm & 0xffff) == 0xffff; }]> {
2790 // (unsigned immediate)
2791 // Indexed for 8-bit registers. offset is in range [0,4095].
2792 def am_indexed8 : ComplexPattern<i64, 2, "SelectAddrModeIndexed8", []>;
2793 def am_indexed16 : ComplexPattern<i64, 2, "SelectAddrModeIndexed16", []>;
2794 def am_indexed32 : ComplexPattern<i64, 2, "SelectAddrModeIndexed32", []>;
2795 def am_indexed64 : ComplexPattern<i64, 2, "SelectAddrModeIndexed64", []>;
2796 def am_indexed128 : ComplexPattern<i64, 2, "SelectAddrModeIndexed128", []>;
2798 def gi_am_indexed8 :
2799 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<8>">,
2800 GIComplexPatternEquiv<am_indexed8>;
2801 def gi_am_indexed16 :
2802 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<16>">,
2803 GIComplexPatternEquiv<am_indexed16>;
2804 def gi_am_indexed32 :
2805 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<32>">,
2806 GIComplexPatternEquiv<am_indexed32>;
2807 def gi_am_indexed64 :
2808 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<64>">,
2809 GIComplexPatternEquiv<am_indexed64>;
2810 def gi_am_indexed128 :
2811 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<128>">,
2812 GIComplexPatternEquiv<am_indexed128>;
2814 class UImm12OffsetOperand<int Scale> : AsmOperandClass {
2815 let Name = "UImm12Offset" # Scale;
2816 let RenderMethod = "addUImm12OffsetOperands<" # Scale # ">";
2817 let PredicateMethod = "isUImm12Offset<" # Scale # ">";
2818 let DiagnosticType = "InvalidMemoryIndexed" # Scale;
2821 def UImm12OffsetScale1Operand : UImm12OffsetOperand<1>;
2822 def UImm12OffsetScale2Operand : UImm12OffsetOperand<2>;
2823 def UImm12OffsetScale4Operand : UImm12OffsetOperand<4>;
2824 def UImm12OffsetScale8Operand : UImm12OffsetOperand<8>;
2825 def UImm12OffsetScale16Operand : UImm12OffsetOperand<16>;
2827 class uimm12_scaled<int Scale> : Operand<i64> {
2828 let ParserMatchClass
2829 = !cast<AsmOperandClass>("UImm12OffsetScale" # Scale # "Operand");
2831 = "getLdStUImm12OpValue<AArch64::fixup_aarch64_ldst_imm12_scale" # Scale # ">";
2832 let PrintMethod = "printUImm12Offset<" # Scale # ">";
2835 def uimm12s1 : uimm12_scaled<1>;
2836 def uimm12s2 : uimm12_scaled<2>;
2837 def uimm12s4 : uimm12_scaled<4>;
2838 def uimm12s8 : uimm12_scaled<8>;
2839 def uimm12s16 : uimm12_scaled<16>;
2841 class BaseLoadStoreUI<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2842 string asm, list<dag> pattern>
2843 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", pattern> {
2849 let Inst{31-30} = sz;
2850 let Inst{29-27} = 0b111;
2852 let Inst{25-24} = 0b01;
2853 let Inst{23-22} = opc;
2854 let Inst{21-10} = offset;
2858 let DecoderMethod = "DecodeUnsignedLdStInstruction";
2861 multiclass LoadUI<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2862 Operand indextype, string asm, list<dag> pattern> {
2863 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2864 def ui : BaseLoadStoreUI<sz, V, opc, (outs regtype:$Rt),
2865 (ins GPR64sp:$Rn, indextype:$offset),
2869 def : InstAlias<asm # "\t$Rt, [$Rn]",
2870 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
2873 multiclass StoreUI<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2874 Operand indextype, string asm, list<dag> pattern> {
2875 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2876 def ui : BaseLoadStoreUI<sz, V, opc, (outs),
2877 (ins regtype:$Rt, GPR64sp:$Rn, indextype:$offset),
2881 def : InstAlias<asm # "\t$Rt, [$Rn]",
2882 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
2885 // Same as StoreUI, but take a RegisterOperand. This is used by GlobalISel to
2886 // substitute zero-registers automatically.
2888 // TODO: Roll out zero-register subtitution to GPR32/GPR64 and fold this back
2890 multiclass StoreUIz<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2891 Operand indextype, string asm, list<dag> pattern> {
2892 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2893 def ui : BaseLoadStoreUI<sz, V, opc, (outs),
2894 (ins regtype:$Rt, GPR64sp:$Rn, indextype:$offset),
2898 def : InstAlias<asm # "\t$Rt, [$Rn]",
2899 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
2902 def PrefetchOperand : AsmOperandClass {
2903 let Name = "Prefetch";
2904 let ParserMethod = "tryParsePrefetch";
2906 def prfop : Operand<i32> {
2907 let PrintMethod = "printPrefetchOp";
2908 let ParserMatchClass = PrefetchOperand;
2911 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2912 class PrefetchUI<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2913 : BaseLoadStoreUI<sz, V, opc,
2914 (outs), (ins prfop:$Rt, GPR64sp:$Rn, uimm12s8:$offset),
2922 // Load literal address: 19-bit immediate. The low two bits of the target
2923 // offset are implied zero and so are not part of the immediate.
2924 def am_ldrlit : Operand<iPTR> {
2925 let EncoderMethod = "getLoadLiteralOpValue";
2926 let DecoderMethod = "DecodePCRelLabel19";
2927 let PrintMethod = "printAlignedLabel";
2928 let ParserMatchClass = PCRelLabel19Operand;
2929 let OperandType = "OPERAND_PCREL";
2932 let mayLoad = 1, mayStore = 0, hasSideEffects = 0, AddedComplexity = 20 in
2933 class LoadLiteral<bits<2> opc, bit V, RegisterOperand regtype, string asm, list<dag> pat>
2934 : I<(outs regtype:$Rt), (ins am_ldrlit:$label),
2935 asm, "\t$Rt, $label", "", pat>,
2939 let Inst{31-30} = opc;
2940 let Inst{29-27} = 0b011;
2942 let Inst{25-24} = 0b00;
2943 let Inst{23-5} = label;
2947 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2948 class PrefetchLiteral<bits<2> opc, bit V, string asm, list<dag> pat>
2949 : I<(outs), (ins prfop:$Rt, am_ldrlit:$label),
2950 asm, "\t$Rt, $label", "", pat>,
2954 let Inst{31-30} = opc;
2955 let Inst{29-27} = 0b011;
2957 let Inst{25-24} = 0b00;
2958 let Inst{23-5} = label;
2963 // Load/store register offset
2966 def ro_Xindexed8 : ComplexPattern<i64, 4, "SelectAddrModeXRO<8>", []>;
2967 def ro_Xindexed16 : ComplexPattern<i64, 4, "SelectAddrModeXRO<16>", []>;
2968 def ro_Xindexed32 : ComplexPattern<i64, 4, "SelectAddrModeXRO<32>", []>;
2969 def ro_Xindexed64 : ComplexPattern<i64, 4, "SelectAddrModeXRO<64>", []>;
2970 def ro_Xindexed128 : ComplexPattern<i64, 4, "SelectAddrModeXRO<128>", []>;
2972 def ro_Windexed8 : ComplexPattern<i64, 4, "SelectAddrModeWRO<8>", []>;
2973 def ro_Windexed16 : ComplexPattern<i64, 4, "SelectAddrModeWRO<16>", []>;
2974 def ro_Windexed32 : ComplexPattern<i64, 4, "SelectAddrModeWRO<32>", []>;
2975 def ro_Windexed64 : ComplexPattern<i64, 4, "SelectAddrModeWRO<64>", []>;
2976 def ro_Windexed128 : ComplexPattern<i64, 4, "SelectAddrModeWRO<128>", []>;
2978 class MemExtendOperand<string Reg, int Width> : AsmOperandClass {
2979 let Name = "Mem" # Reg # "Extend" # Width;
2980 let PredicateMethod = "isMem" # Reg # "Extend<" # Width # ">";
2981 let RenderMethod = "addMemExtendOperands";
2982 let DiagnosticType = "InvalidMemory" # Reg # "Extend" # Width;
2985 def MemWExtend8Operand : MemExtendOperand<"W", 8> {
2986 // The address "[x0, x1, lsl #0]" actually maps to the variant which performs
2987 // the trivial shift.
2988 let RenderMethod = "addMemExtend8Operands";
2990 def MemWExtend16Operand : MemExtendOperand<"W", 16>;
2991 def MemWExtend32Operand : MemExtendOperand<"W", 32>;
2992 def MemWExtend64Operand : MemExtendOperand<"W", 64>;
2993 def MemWExtend128Operand : MemExtendOperand<"W", 128>;
2995 def MemXExtend8Operand : MemExtendOperand<"X", 8> {
2996 // The address "[x0, x1, lsl #0]" actually maps to the variant which performs
2997 // the trivial shift.
2998 let RenderMethod = "addMemExtend8Operands";
3000 def MemXExtend16Operand : MemExtendOperand<"X", 16>;
3001 def MemXExtend32Operand : MemExtendOperand<"X", 32>;
3002 def MemXExtend64Operand : MemExtendOperand<"X", 64>;
3003 def MemXExtend128Operand : MemExtendOperand<"X", 128>;
3005 class ro_extend<AsmOperandClass ParserClass, string Reg, int Width>
3007 let ParserMatchClass = ParserClass;
3008 let PrintMethod = "printMemExtend<'" # Reg # "', " # Width # ">";
3009 let DecoderMethod = "DecodeMemExtend";
3010 let EncoderMethod = "getMemExtendOpValue";
3011 let MIOperandInfo = (ops i32imm:$signed, i32imm:$doshift);
3014 def ro_Wextend8 : ro_extend<MemWExtend8Operand, "w", 8>;
3015 def ro_Wextend16 : ro_extend<MemWExtend16Operand, "w", 16>;
3016 def ro_Wextend32 : ro_extend<MemWExtend32Operand, "w", 32>;
3017 def ro_Wextend64 : ro_extend<MemWExtend64Operand, "w", 64>;
3018 def ro_Wextend128 : ro_extend<MemWExtend128Operand, "w", 128>;
3020 def ro_Xextend8 : ro_extend<MemXExtend8Operand, "x", 8>;
3021 def ro_Xextend16 : ro_extend<MemXExtend16Operand, "x", 16>;
3022 def ro_Xextend32 : ro_extend<MemXExtend32Operand, "x", 32>;
3023 def ro_Xextend64 : ro_extend<MemXExtend64Operand, "x", 64>;
3024 def ro_Xextend128 : ro_extend<MemXExtend128Operand, "x", 128>;
3026 class ROAddrMode<ComplexPattern windex, ComplexPattern xindex,
3027 Operand wextend, Operand xextend> {
3028 // CodeGen-level pattern covering the entire addressing mode.
3029 ComplexPattern Wpat = windex;
3030 ComplexPattern Xpat = xindex;
3032 // Asm-level Operand covering the valid "uxtw #3" style syntax.
3033 Operand Wext = wextend;
3034 Operand Xext = xextend;
3037 def ro8 : ROAddrMode<ro_Windexed8, ro_Xindexed8, ro_Wextend8, ro_Xextend8>;
3038 def ro16 : ROAddrMode<ro_Windexed16, ro_Xindexed16, ro_Wextend16, ro_Xextend16>;
3039 def ro32 : ROAddrMode<ro_Windexed32, ro_Xindexed32, ro_Wextend32, ro_Xextend32>;
3040 def ro64 : ROAddrMode<ro_Windexed64, ro_Xindexed64, ro_Wextend64, ro_Xextend64>;
3041 def ro128 : ROAddrMode<ro_Windexed128, ro_Xindexed128, ro_Wextend128,
3044 class LoadStore8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3045 string asm, dag ins, dag outs, list<dag> pat>
3046 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3051 let Inst{31-30} = sz;
3052 let Inst{29-27} = 0b111;
3054 let Inst{25-24} = 0b00;
3055 let Inst{23-22} = opc;
3057 let Inst{20-16} = Rm;
3058 let Inst{15} = extend{1}; // sign extend Rm?
3060 let Inst{12} = extend{0}; // do shift?
3061 let Inst{11-10} = 0b10;
3066 class ROInstAlias<string asm, RegisterOperand regtype, Instruction INST>
3067 : InstAlias<asm # "\t$Rt, [$Rn, $Rm]",
3068 (INST regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, 0, 0)>;
3070 multiclass Load8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3071 string asm, ValueType Ty, SDPatternOperator loadop> {
3072 let AddedComplexity = 10 in
3073 def roW : LoadStore8RO<sz, V, opc, regtype, asm,
3075 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$extend),
3076 [(set (Ty regtype:$Rt),
3077 (loadop (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm,
3078 ro_Wextend8:$extend)))]>,
3079 Sched<[WriteLDIdx, ReadAdrBase]> {
3083 let AddedComplexity = 10 in
3084 def roX : LoadStore8RO<sz, V, opc, regtype, asm,
3086 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend8:$extend),
3087 [(set (Ty regtype:$Rt),
3088 (loadop (ro_Xindexed8 GPR64sp:$Rn, GPR64:$Rm,
3089 ro_Xextend8:$extend)))]>,
3090 Sched<[WriteLDIdx, ReadAdrBase]> {
3094 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3097 multiclass Store8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3098 string asm, ValueType Ty, SDPatternOperator storeop> {
3099 let AddedComplexity = 10 in
3100 def roW : LoadStore8RO<sz, V, opc, regtype, asm, (outs),
3101 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$extend),
3102 [(storeop (Ty regtype:$Rt),
3103 (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm,
3104 ro_Wextend8:$extend))]>,
3105 Sched<[WriteSTIdx, ReadAdrBase]> {
3109 let AddedComplexity = 10 in
3110 def roX : LoadStore8RO<sz, V, opc, regtype, asm, (outs),
3111 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend8:$extend),
3112 [(storeop (Ty regtype:$Rt),
3113 (ro_Xindexed8 GPR64sp:$Rn, GPR64:$Rm,
3114 ro_Xextend8:$extend))]>,
3115 Sched<[WriteSTIdx, ReadAdrBase]> {
3119 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3122 class LoadStore16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3123 string asm, dag ins, dag outs, list<dag> pat>
3124 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3129 let Inst{31-30} = sz;
3130 let Inst{29-27} = 0b111;
3132 let Inst{25-24} = 0b00;
3133 let Inst{23-22} = opc;
3135 let Inst{20-16} = Rm;
3136 let Inst{15} = extend{1}; // sign extend Rm?
3138 let Inst{12} = extend{0}; // do shift?
3139 let Inst{11-10} = 0b10;
3144 multiclass Load16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3145 string asm, ValueType Ty, SDPatternOperator loadop> {
3146 let AddedComplexity = 10 in
3147 def roW : LoadStore16RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3148 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend16:$extend),
3149 [(set (Ty regtype:$Rt),
3150 (loadop (ro_Windexed16 GPR64sp:$Rn, GPR32:$Rm,
3151 ro_Wextend16:$extend)))]>,
3152 Sched<[WriteLDIdx, ReadAdrBase]> {
3156 let AddedComplexity = 10 in
3157 def roX : LoadStore16RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3158 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend16:$extend),
3159 [(set (Ty regtype:$Rt),
3160 (loadop (ro_Xindexed16 GPR64sp:$Rn, GPR64:$Rm,
3161 ro_Xextend16:$extend)))]>,
3162 Sched<[WriteLDIdx, ReadAdrBase]> {
3166 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3169 multiclass Store16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3170 string asm, ValueType Ty, SDPatternOperator storeop> {
3171 let AddedComplexity = 10 in
3172 def roW : LoadStore16RO<sz, V, opc, regtype, asm, (outs),
3173 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend16:$extend),
3174 [(storeop (Ty regtype:$Rt),
3175 (ro_Windexed16 GPR64sp:$Rn, GPR32:$Rm,
3176 ro_Wextend16:$extend))]>,
3177 Sched<[WriteSTIdx, ReadAdrBase]> {
3181 let AddedComplexity = 10 in
3182 def roX : LoadStore16RO<sz, V, opc, regtype, asm, (outs),
3183 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend16:$extend),
3184 [(storeop (Ty regtype:$Rt),
3185 (ro_Xindexed16 GPR64sp:$Rn, GPR64:$Rm,
3186 ro_Xextend16:$extend))]>,
3187 Sched<[WriteSTIdx, ReadAdrBase]> {
3191 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3194 class LoadStore32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3195 string asm, dag ins, dag outs, list<dag> pat>
3196 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3201 let Inst{31-30} = sz;
3202 let Inst{29-27} = 0b111;
3204 let Inst{25-24} = 0b00;
3205 let Inst{23-22} = opc;
3207 let Inst{20-16} = Rm;
3208 let Inst{15} = extend{1}; // sign extend Rm?
3210 let Inst{12} = extend{0}; // do shift?
3211 let Inst{11-10} = 0b10;
3216 multiclass Load32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3217 string asm, ValueType Ty, SDPatternOperator loadop> {
3218 let AddedComplexity = 10 in
3219 def roW : LoadStore32RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3220 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend32:$extend),
3221 [(set (Ty regtype:$Rt),
3222 (loadop (ro_Windexed32 GPR64sp:$Rn, GPR32:$Rm,
3223 ro_Wextend32:$extend)))]>,
3224 Sched<[WriteLDIdx, ReadAdrBase]> {
3228 let AddedComplexity = 10 in
3229 def roX : LoadStore32RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3230 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend32:$extend),
3231 [(set (Ty regtype:$Rt),
3232 (loadop (ro_Xindexed32 GPR64sp:$Rn, GPR64:$Rm,
3233 ro_Xextend32:$extend)))]>,
3234 Sched<[WriteLDIdx, ReadAdrBase]> {
3238 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3241 multiclass Store32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3242 string asm, ValueType Ty, SDPatternOperator storeop> {
3243 let AddedComplexity = 10 in
3244 def roW : LoadStore32RO<sz, V, opc, regtype, asm, (outs),
3245 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend32:$extend),
3246 [(storeop (Ty regtype:$Rt),
3247 (ro_Windexed32 GPR64sp:$Rn, GPR32:$Rm,
3248 ro_Wextend32:$extend))]>,
3249 Sched<[WriteSTIdx, ReadAdrBase]> {
3253 let AddedComplexity = 10 in
3254 def roX : LoadStore32RO<sz, V, opc, regtype, asm, (outs),
3255 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend32:$extend),
3256 [(storeop (Ty regtype:$Rt),
3257 (ro_Xindexed32 GPR64sp:$Rn, GPR64:$Rm,
3258 ro_Xextend32:$extend))]>,
3259 Sched<[WriteSTIdx, ReadAdrBase]> {
3263 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3266 class LoadStore64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3267 string asm, dag ins, dag outs, list<dag> pat>
3268 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3273 let Inst{31-30} = sz;
3274 let Inst{29-27} = 0b111;
3276 let Inst{25-24} = 0b00;
3277 let Inst{23-22} = opc;
3279 let Inst{20-16} = Rm;
3280 let Inst{15} = extend{1}; // sign extend Rm?
3282 let Inst{12} = extend{0}; // do shift?
3283 let Inst{11-10} = 0b10;
3288 multiclass Load64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3289 string asm, ValueType Ty, SDPatternOperator loadop> {
3290 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3291 def roW : LoadStore64RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3292 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3293 [(set (Ty regtype:$Rt),
3294 (loadop (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3295 ro_Wextend64:$extend)))]>,
3296 Sched<[WriteLDIdx, ReadAdrBase]> {
3300 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3301 def roX : LoadStore64RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3302 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3303 [(set (Ty regtype:$Rt),
3304 (loadop (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3305 ro_Xextend64:$extend)))]>,
3306 Sched<[WriteLDIdx, ReadAdrBase]> {
3310 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3313 multiclass Store64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3314 string asm, ValueType Ty, SDPatternOperator storeop> {
3315 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3316 def roW : LoadStore64RO<sz, V, opc, regtype, asm, (outs),
3317 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3318 [(storeop (Ty regtype:$Rt),
3319 (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3320 ro_Wextend64:$extend))]>,
3321 Sched<[WriteSTIdx, ReadAdrBase]> {
3325 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3326 def roX : LoadStore64RO<sz, V, opc, regtype, asm, (outs),
3327 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3328 [(storeop (Ty regtype:$Rt),
3329 (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3330 ro_Xextend64:$extend))]>,
3331 Sched<[WriteSTIdx, ReadAdrBase]> {
3335 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3338 class LoadStore128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3339 string asm, dag ins, dag outs, list<dag> pat>
3340 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3345 let Inst{31-30} = sz;
3346 let Inst{29-27} = 0b111;
3348 let Inst{25-24} = 0b00;
3349 let Inst{23-22} = opc;
3351 let Inst{20-16} = Rm;
3352 let Inst{15} = extend{1}; // sign extend Rm?
3354 let Inst{12} = extend{0}; // do shift?
3355 let Inst{11-10} = 0b10;
3360 multiclass Load128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3361 string asm, ValueType Ty, SDPatternOperator loadop> {
3362 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3363 def roW : LoadStore128RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3364 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend),
3365 [(set (Ty regtype:$Rt),
3366 (loadop (ro_Windexed128 GPR64sp:$Rn, GPR32:$Rm,
3367 ro_Wextend128:$extend)))]>,
3368 Sched<[WriteLDIdx, ReadAdrBase]> {
3372 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3373 def roX : LoadStore128RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3374 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend128:$extend),
3375 [(set (Ty regtype:$Rt),
3376 (loadop (ro_Xindexed128 GPR64sp:$Rn, GPR64:$Rm,
3377 ro_Xextend128:$extend)))]>,
3378 Sched<[WriteLDIdx, ReadAdrBase]> {
3382 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3385 multiclass Store128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3386 string asm, ValueType Ty, SDPatternOperator storeop> {
3387 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3388 def roW : LoadStore128RO<sz, V, opc, regtype, asm, (outs),
3389 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend),
3391 Sched<[WriteSTIdx, ReadAdrBase]> {
3395 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3396 def roX : LoadStore128RO<sz, V, opc, regtype, asm, (outs),
3397 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend128:$extend),
3399 Sched<[WriteSTIdx, ReadAdrBase]> {
3403 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3406 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3407 class BasePrefetchRO<bits<2> sz, bit V, bits<2> opc, dag outs, dag ins,
3408 string asm, list<dag> pat>
3409 : I<outs, ins, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat>,
3415 let Inst{31-30} = sz;
3416 let Inst{29-27} = 0b111;
3418 let Inst{25-24} = 0b00;
3419 let Inst{23-22} = opc;
3421 let Inst{20-16} = Rm;
3422 let Inst{15} = extend{1}; // sign extend Rm?
3424 let Inst{12} = extend{0}; // do shift?
3425 let Inst{11-10} = 0b10;
3430 multiclass PrefetchRO<bits<2> sz, bit V, bits<2> opc, string asm> {
3431 def roW : BasePrefetchRO<sz, V, opc, (outs),
3432 (ins prfop:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3433 asm, [(AArch64Prefetch imm:$Rt,
3434 (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3435 ro_Wextend64:$extend))]> {
3439 def roX : BasePrefetchRO<sz, V, opc, (outs),
3440 (ins prfop:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3441 asm, [(AArch64Prefetch imm:$Rt,
3442 (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3443 ro_Xextend64:$extend))]> {
3447 def : InstAlias<"prfm $Rt, [$Rn, $Rm]",
3448 (!cast<Instruction>(NAME # "roX") prfop:$Rt,
3449 GPR64sp:$Rn, GPR64:$Rm, 0, 0)>;
3453 // Load/store unscaled immediate
3456 def am_unscaled8 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled8", []>;
3457 def am_unscaled16 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled16", []>;
3458 def am_unscaled32 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled32", []>;
3459 def am_unscaled64 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled64", []>;
3460 def am_unscaled128 :ComplexPattern<i64, 2, "SelectAddrModeUnscaled128", []>;
3462 def gi_am_unscaled8 :
3463 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled8">,
3464 GIComplexPatternEquiv<am_unscaled8>;
3465 def gi_am_unscaled16 :
3466 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled16">,
3467 GIComplexPatternEquiv<am_unscaled16>;
3468 def gi_am_unscaled32 :
3469 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled32">,
3470 GIComplexPatternEquiv<am_unscaled32>;
3471 def gi_am_unscaled64 :
3472 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled64">,
3473 GIComplexPatternEquiv<am_unscaled64>;
3474 def gi_am_unscaled128 :
3475 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled128">,
3476 GIComplexPatternEquiv<am_unscaled128>;
3479 class BaseLoadStoreUnscale<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3480 string asm, list<dag> pattern>
3481 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", pattern> {
3485 let Inst{31-30} = sz;
3486 let Inst{29-27} = 0b111;
3488 let Inst{25-24} = 0b00;
3489 let Inst{23-22} = opc;
3491 let Inst{20-12} = offset;
3492 let Inst{11-10} = 0b00;
3496 let DecoderMethod = "DecodeSignedLdStInstruction";
3499 // Armv8.4 LDAPR & STLR with Immediate Offset instruction
3500 multiclass BaseLoadUnscaleV84<string asm, bits<2> sz, bits<2> opc,
3501 RegisterOperand regtype > {
3502 def i : BaseLoadStoreUnscale<sz, 0, opc, (outs regtype:$Rt),
3503 (ins GPR64sp:$Rn, simm9:$offset), asm, []>,
3508 def : InstAlias<asm # "\t$Rt, [$Rn]",
3509 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3512 multiclass BaseStoreUnscaleV84<string asm, bits<2> sz, bits<2> opc,
3513 RegisterOperand regtype > {
3514 def i : BaseLoadStoreUnscale<sz, 0, opc, (outs),
3515 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3521 def : InstAlias<asm # "\t$Rt, [$Rn]",
3522 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3525 multiclass LoadUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3526 string asm, list<dag> pattern> {
3527 let AddedComplexity = 1 in // try this before LoadUI
3528 def i : BaseLoadStoreUnscale<sz, V, opc, (outs regtype:$Rt),
3529 (ins GPR64sp:$Rn, simm9:$offset), asm, pattern>,
3532 def : InstAlias<asm # "\t$Rt, [$Rn]",
3533 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3536 multiclass StoreUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3537 string asm, list<dag> pattern> {
3538 let AddedComplexity = 1 in // try this before StoreUI
3539 def i : BaseLoadStoreUnscale<sz, V, opc, (outs),
3540 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3544 def : InstAlias<asm # "\t$Rt, [$Rn]",
3545 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3548 multiclass PrefetchUnscaled<bits<2> sz, bit V, bits<2> opc, string asm,
3550 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3551 def i : BaseLoadStoreUnscale<sz, V, opc, (outs),
3552 (ins prfop:$Rt, GPR64sp:$Rn, simm9:$offset),
3556 def : InstAlias<asm # "\t$Rt, [$Rn]",
3557 (!cast<Instruction>(NAME # "i") prfop:$Rt, GPR64sp:$Rn, 0)>;
3561 // Load/store unscaled immediate, unprivileged
3564 class BaseLoadStoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
3565 dag oops, dag iops, string asm>
3566 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", []> {
3570 let Inst{31-30} = sz;
3571 let Inst{29-27} = 0b111;
3573 let Inst{25-24} = 0b00;
3574 let Inst{23-22} = opc;
3576 let Inst{20-12} = offset;
3577 let Inst{11-10} = 0b10;
3581 let DecoderMethod = "DecodeSignedLdStInstruction";
3584 multiclass LoadUnprivileged<bits<2> sz, bit V, bits<2> opc,
3585 RegisterClass regtype, string asm> {
3586 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in
3587 def i : BaseLoadStoreUnprivileged<sz, V, opc, (outs regtype:$Rt),
3588 (ins GPR64sp:$Rn, simm9:$offset), asm>,
3591 def : InstAlias<asm # "\t$Rt, [$Rn]",
3592 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3595 multiclass StoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
3596 RegisterClass regtype, string asm> {
3597 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
3598 def i : BaseLoadStoreUnprivileged<sz, V, opc, (outs),
3599 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3603 def : InstAlias<asm # "\t$Rt, [$Rn]",
3604 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3608 // Load/store pre-indexed
3611 class BaseLoadStorePreIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3612 string asm, string cstr, list<dag> pat>
3613 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]!", cstr, pat> {
3617 let Inst{31-30} = sz;
3618 let Inst{29-27} = 0b111;
3620 let Inst{25-24} = 0;
3621 let Inst{23-22} = opc;
3623 let Inst{20-12} = offset;
3624 let Inst{11-10} = 0b11;
3628 let DecoderMethod = "DecodeSignedLdStInstruction";
3631 let hasSideEffects = 0 in {
3632 let mayStore = 0, mayLoad = 1 in
3633 class LoadPreIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3635 : BaseLoadStorePreIdx<sz, V, opc,
3636 (outs GPR64sp:$wback, regtype:$Rt),
3637 (ins GPR64sp:$Rn, simm9:$offset), asm,
3638 "$Rn = $wback,@earlyclobber $wback", []>,
3639 Sched<[WriteLD, WriteAdr]>;
3641 let mayStore = 1, mayLoad = 0 in
3642 class StorePreIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3643 string asm, SDPatternOperator storeop, ValueType Ty>
3644 : BaseLoadStorePreIdx<sz, V, opc,
3645 (outs GPR64sp:$wback),
3646 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3647 asm, "$Rn = $wback,@earlyclobber $wback",
3648 [(set GPR64sp:$wback,
3649 (storeop (Ty regtype:$Rt), GPR64sp:$Rn, simm9:$offset))]>,
3650 Sched<[WriteAdr, WriteST]>;
3651 } // hasSideEffects = 0
3654 // Load/store post-indexed
3657 class BaseLoadStorePostIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3658 string asm, string cstr, list<dag> pat>
3659 : I<oops, iops, asm, "\t$Rt, [$Rn], $offset", cstr, pat> {
3663 let Inst{31-30} = sz;
3664 let Inst{29-27} = 0b111;
3666 let Inst{25-24} = 0b00;
3667 let Inst{23-22} = opc;
3669 let Inst{20-12} = offset;
3670 let Inst{11-10} = 0b01;
3674 let DecoderMethod = "DecodeSignedLdStInstruction";
3677 let hasSideEffects = 0 in {
3678 let mayStore = 0, mayLoad = 1 in
3679 class LoadPostIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3681 : BaseLoadStorePostIdx<sz, V, opc,
3682 (outs GPR64sp:$wback, regtype:$Rt),
3683 (ins GPR64sp:$Rn, simm9:$offset),
3684 asm, "$Rn = $wback,@earlyclobber $wback", []>,
3685 Sched<[WriteLD, WriteAdr]>;
3687 let mayStore = 1, mayLoad = 0 in
3688 class StorePostIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3689 string asm, SDPatternOperator storeop, ValueType Ty>
3690 : BaseLoadStorePostIdx<sz, V, opc,
3691 (outs GPR64sp:$wback),
3692 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3693 asm, "$Rn = $wback,@earlyclobber $wback",
3694 [(set GPR64sp:$wback,
3695 (storeop (Ty regtype:$Rt), GPR64sp:$Rn, simm9:$offset))]>,
3696 Sched<[WriteAdr, WriteST]>;
3697 } // hasSideEffects = 0
3704 // (indexed, offset)
3706 class BaseLoadStorePairOffset<bits<2> opc, bit V, bit L, dag oops, dag iops,
3708 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]", "", []> {
3713 let Inst{31-30} = opc;
3714 let Inst{29-27} = 0b101;
3716 let Inst{25-23} = 0b010;
3718 let Inst{21-15} = offset;
3719 let Inst{14-10} = Rt2;
3723 let DecoderMethod = "DecodePairLdStInstruction";
3726 multiclass LoadPairOffset<bits<2> opc, bit V, RegisterOperand regtype,
3727 Operand indextype, string asm> {
3728 let hasSideEffects = 0, mayStore = 0, mayLoad = 1 in
3729 def i : BaseLoadStorePairOffset<opc, V, 1,
3730 (outs regtype:$Rt, regtype:$Rt2),
3731 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3732 Sched<[WriteLD, WriteLDHi]>;
3734 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3735 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3740 multiclass StorePairOffset<bits<2> opc, bit V, RegisterOperand regtype,
3741 Operand indextype, string asm> {
3742 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in
3743 def i : BaseLoadStorePairOffset<opc, V, 0, (outs),
3744 (ins regtype:$Rt, regtype:$Rt2,
3745 GPR64sp:$Rn, indextype:$offset),
3749 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3750 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3755 class BaseLoadStorePairPreIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
3757 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]!", "$Rn = $wback,@earlyclobber $wback", []> {
3762 let Inst{31-30} = opc;
3763 let Inst{29-27} = 0b101;
3765 let Inst{25-23} = 0b011;
3767 let Inst{21-15} = offset;
3768 let Inst{14-10} = Rt2;
3772 let DecoderMethod = "DecodePairLdStInstruction";
3775 let hasSideEffects = 0 in {
3776 let mayStore = 0, mayLoad = 1 in
3777 class LoadPairPreIdx<bits<2> opc, bit V, RegisterOperand regtype,
3778 Operand indextype, string asm>
3779 : BaseLoadStorePairPreIdx<opc, V, 1,
3780 (outs GPR64sp:$wback, regtype:$Rt, regtype:$Rt2),
3781 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3782 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
3784 let mayStore = 1, mayLoad = 0 in
3785 class StorePairPreIdx<bits<2> opc, bit V, RegisterOperand regtype,
3786 Operand indextype, string asm>
3787 : BaseLoadStorePairPreIdx<opc, V, 0, (outs GPR64sp:$wback),
3788 (ins regtype:$Rt, regtype:$Rt2,
3789 GPR64sp:$Rn, indextype:$offset),
3791 Sched<[WriteAdr, WriteSTP]>;
3792 } // hasSideEffects = 0
3796 class BaseLoadStorePairPostIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
3798 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn], $offset", "$Rn = $wback,@earlyclobber $wback", []> {
3803 let Inst{31-30} = opc;
3804 let Inst{29-27} = 0b101;
3806 let Inst{25-23} = 0b001;
3808 let Inst{21-15} = offset;
3809 let Inst{14-10} = Rt2;
3813 let DecoderMethod = "DecodePairLdStInstruction";
3816 let hasSideEffects = 0 in {
3817 let mayStore = 0, mayLoad = 1 in
3818 class LoadPairPostIdx<bits<2> opc, bit V, RegisterOperand regtype,
3819 Operand idxtype, string asm>
3820 : BaseLoadStorePairPostIdx<opc, V, 1,
3821 (outs GPR64sp:$wback, regtype:$Rt, regtype:$Rt2),
3822 (ins GPR64sp:$Rn, idxtype:$offset), asm>,
3823 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
3825 let mayStore = 1, mayLoad = 0 in
3826 class StorePairPostIdx<bits<2> opc, bit V, RegisterOperand regtype,
3827 Operand idxtype, string asm>
3828 : BaseLoadStorePairPostIdx<opc, V, 0, (outs GPR64sp:$wback),
3829 (ins regtype:$Rt, regtype:$Rt2,
3830 GPR64sp:$Rn, idxtype:$offset),
3832 Sched<[WriteAdr, WriteSTP]>;
3833 } // hasSideEffects = 0
3837 class BaseLoadStorePairNoAlloc<bits<2> opc, bit V, bit L, dag oops, dag iops,
3839 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]", "", []> {
3844 let Inst{31-30} = opc;
3845 let Inst{29-27} = 0b101;
3847 let Inst{25-23} = 0b000;
3849 let Inst{21-15} = offset;
3850 let Inst{14-10} = Rt2;
3854 let DecoderMethod = "DecodePairLdStInstruction";
3857 multiclass LoadPairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
3858 Operand indextype, string asm> {
3859 let hasSideEffects = 0, mayStore = 0, mayLoad = 1 in
3860 def i : BaseLoadStorePairNoAlloc<opc, V, 1,
3861 (outs regtype:$Rt, regtype:$Rt2),
3862 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3863 Sched<[WriteLD, WriteLDHi]>;
3866 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3867 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3871 multiclass StorePairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
3872 Operand indextype, string asm> {
3873 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in
3874 def i : BaseLoadStorePairNoAlloc<opc, V, 0, (outs),
3875 (ins regtype:$Rt, regtype:$Rt2,
3876 GPR64sp:$Rn, indextype:$offset),
3880 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3881 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3886 // Load/store exclusive
3889 // True exclusive operations write to and/or read from the system's exclusive
3890 // monitors, which as far as a compiler is concerned can be modelled as a
3891 // random shared memory address. Hence LoadExclusive mayStore.
3893 // Since these instructions have the undefined register bits set to 1 in
3894 // their canonical form, we need a post encoder method to set those bits
3895 // to 1 when encoding these instructions. We do this using the
3896 // fixLoadStoreExclusive function. This function has template parameters:
3898 // fixLoadStoreExclusive<int hasRs, int hasRt2>
3900 // hasRs indicates that the instruction uses the Rs field, so we won't set
3901 // it to 1 (and the same for Rt2). We don't need template parameters for
3902 // the other register fields since Rt and Rn are always used.
3904 let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
3905 class BaseLoadStoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3906 dag oops, dag iops, string asm, string operands>
3907 : I<oops, iops, asm, operands, "", []> {
3908 let Inst{31-30} = sz;
3909 let Inst{29-24} = 0b001000;
3915 let DecoderMethod = "DecodeExclusiveLdStInstruction";
3918 // Neither Rs nor Rt2 operands.
3919 class LoadStoreExclusiveSimple<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3920 dag oops, dag iops, string asm, string operands>
3921 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, oops, iops, asm, operands> {
3924 let Inst{20-16} = 0b11111;
3925 let Unpredictable{20-16} = 0b11111;
3926 let Inst{14-10} = 0b11111;
3927 let Unpredictable{14-10} = 0b11111;
3931 let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
3934 // Simple load acquires don't set the exclusive monitor
3935 let mayLoad = 1, mayStore = 0 in
3936 class LoadAcquire<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3937 RegisterClass regtype, string asm>
3938 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
3939 (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]">,
3942 class LoadExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3943 RegisterClass regtype, string asm>
3944 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
3945 (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]">,
3948 class LoadExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3949 RegisterClass regtype, string asm>
3950 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3951 (outs regtype:$Rt, regtype:$Rt2),
3952 (ins GPR64sp0:$Rn), asm,
3953 "\t$Rt, $Rt2, [$Rn]">,
3954 Sched<[WriteLD, WriteLDHi]> {
3958 let Inst{14-10} = Rt2;
3962 let PostEncoderMethod = "fixLoadStoreExclusive<0,1>";
3965 // Simple store release operations do not check the exclusive monitor.
3966 let mayLoad = 0, mayStore = 1 in
3967 class StoreRelease<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3968 RegisterClass regtype, string asm>
3969 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs),
3970 (ins regtype:$Rt, GPR64sp0:$Rn),
3971 asm, "\t$Rt, [$Rn]">,
3974 let mayLoad = 1, mayStore = 1 in
3975 class StoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3976 RegisterClass regtype, string asm>
3977 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, (outs GPR32:$Ws),
3978 (ins regtype:$Rt, GPR64sp0:$Rn),
3979 asm, "\t$Ws, $Rt, [$Rn]">,
3984 let Inst{20-16} = Ws;
3988 let Constraints = "@earlyclobber $Ws";
3989 let PostEncoderMethod = "fixLoadStoreExclusive<1,0>";
3992 class StoreExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3993 RegisterClass regtype, string asm>
3994 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3996 (ins regtype:$Rt, regtype:$Rt2, GPR64sp0:$Rn),
3997 asm, "\t$Ws, $Rt, $Rt2, [$Rn]">,
4003 let Inst{20-16} = Ws;
4004 let Inst{14-10} = Rt2;
4008 let Constraints = "@earlyclobber $Ws";
4011 // Armv8.5-A Memory Tagging Extension
4012 class BaseMemTag<bits<2> opc1, bits<2> opc2, string asm_insn,
4013 string asm_opnds, string cstr, dag oops, dag iops>
4014 : I<oops, iops, asm_insn, asm_opnds, cstr, []>,
4018 let Inst{31-24} = 0b11011001;
4019 let Inst{23-22} = opc1;
4021 // Inst{20-12} defined by subclass
4022 let Inst{11-10} = opc2;
4024 // Inst{4-0} defined by subclass
4027 class MemTagVector<bit Load, string asm_insn, string asm_opnds,
4029 : BaseMemTag<{0b1, Load}, 0b00, asm_insn, asm_opnds,
4033 let Inst{20-12} = 0b000000000;
4039 class MemTagLoad<string asm_insn, string asm_opnds>
4040 : BaseMemTag<0b01, 0b00, asm_insn, asm_opnds, "$Rt = $wback",
4041 (outs GPR64:$wback),
4042 (ins GPR64:$Rt, GPR64sp:$Rn, simm9s16:$offset)> {
4046 let Inst{20-12} = offset;
4052 class BaseMemTagStore<bits<2> opc1, bits<2> opc2, string asm_insn,
4053 string asm_opnds, string cstr, dag oops, dag iops>
4054 : BaseMemTag<opc1, opc2, asm_insn, asm_opnds, cstr, oops, iops> {
4058 let Inst{20-12} = offset;
4064 multiclass MemTagStore<bits<2> opc1, string insn> {
4066 BaseMemTagStore<opc1, 0b10, insn, "\t$Rt, [$Rn, $offset]", "",
4067 (outs), (ins GPR64sp:$Rt, GPR64sp:$Rn, simm9s16:$offset)>;
4069 BaseMemTagStore<opc1, 0b11, insn, "\t$Rt, [$Rn, $offset]!",
4071 (outs GPR64sp:$wback),
4072 (ins GPR64sp:$Rt, GPR64sp:$Rn, simm9s16:$offset)>;
4074 BaseMemTagStore<opc1, 0b01, insn, "\t$Rt, [$Rn], $offset",
4076 (outs GPR64sp:$wback),
4077 (ins GPR64sp:$Rt, GPR64sp:$Rn, simm9s16:$offset)>;
4079 def : InstAlias<insn # "\t$Rt, [$Rn]",
4080 (!cast<Instruction>(NAME # "Offset") GPR64sp:$Rt, GPR64sp:$Rn, 0)>;
4084 // Exception generation
4087 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
4088 class ExceptionGeneration<bits<3> op1, bits<2> ll, string asm>
4089 : I<(outs), (ins imm0_65535:$imm), asm, "\t$imm", "", []>,
4092 let Inst{31-24} = 0b11010100;
4093 let Inst{23-21} = op1;
4094 let Inst{20-5} = imm;
4095 let Inst{4-2} = 0b000;
4100 // UDF : Permanently UNDEFINED instructions. Format: Opc = 0x0000, 16 bit imm.
4102 let hasSideEffects = 1, isTrap = 1, mayLoad = 0, mayStore = 0 in {
4103 class UDFType<bits<16> opc, string asm>
4104 : I<(outs), (ins uimm16:$imm),
4105 asm, "\t$imm", "", []>,
4108 let Inst{31-16} = opc;
4109 let Inst{15-0} = imm;
4112 let Predicates = [HasFPARMv8] in {
4115 // Floating point to integer conversion
4118 class BaseFPToIntegerUnscaled<bits<2> type, bits<2> rmode, bits<3> opcode,
4119 RegisterClass srcType, RegisterClass dstType,
4120 string asm, list<dag> pattern>
4121 : I<(outs dstType:$Rd), (ins srcType:$Rn),
4122 asm, "\t$Rd, $Rn", "", pattern>,
4123 Sched<[WriteFCvt]> {
4126 let Inst{30-29} = 0b00;
4127 let Inst{28-24} = 0b11110;
4128 let Inst{23-22} = type;
4130 let Inst{20-19} = rmode;
4131 let Inst{18-16} = opcode;
4132 let Inst{15-10} = 0;
4137 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4138 class BaseFPToInteger<bits<2> type, bits<2> rmode, bits<3> opcode,
4139 RegisterClass srcType, RegisterClass dstType,
4140 Operand immType, string asm, list<dag> pattern>
4141 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
4142 asm, "\t$Rd, $Rn, $scale", "", pattern>,
4143 Sched<[WriteFCvt]> {
4147 let Inst{30-29} = 0b00;
4148 let Inst{28-24} = 0b11110;
4149 let Inst{23-22} = type;
4151 let Inst{20-19} = rmode;
4152 let Inst{18-16} = opcode;
4153 let Inst{15-10} = scale;
4158 multiclass FPToIntegerUnscaled<bits<2> rmode, bits<3> opcode, string asm,
4159 SDPatternOperator OpN> {
4160 // Unscaled half-precision to 32-bit
4161 def UWHr : BaseFPToIntegerUnscaled<0b11, rmode, opcode, FPR16, GPR32, asm,
4162 [(set GPR32:$Rd, (OpN FPR16:$Rn))]> {
4163 let Inst{31} = 0; // 32-bit GPR flag
4164 let Predicates = [HasFullFP16];
4167 // Unscaled half-precision to 64-bit
4168 def UXHr : BaseFPToIntegerUnscaled<0b11, rmode, opcode, FPR16, GPR64, asm,
4169 [(set GPR64:$Rd, (OpN FPR16:$Rn))]> {
4170 let Inst{31} = 1; // 64-bit GPR flag
4171 let Predicates = [HasFullFP16];
4174 // Unscaled single-precision to 32-bit
4175 def UWSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR32, asm,
4176 [(set GPR32:$Rd, (OpN FPR32:$Rn))]> {
4177 let Inst{31} = 0; // 32-bit GPR flag
4180 // Unscaled single-precision to 64-bit
4181 def UXSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR64, asm,
4182 [(set GPR64:$Rd, (OpN FPR32:$Rn))]> {
4183 let Inst{31} = 1; // 64-bit GPR flag
4186 // Unscaled double-precision to 32-bit
4187 def UWDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR32, asm,
4188 [(set GPR32:$Rd, (OpN (f64 FPR64:$Rn)))]> {
4189 let Inst{31} = 0; // 32-bit GPR flag
4192 // Unscaled double-precision to 64-bit
4193 def UXDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR64, asm,
4194 [(set GPR64:$Rd, (OpN (f64 FPR64:$Rn)))]> {
4195 let Inst{31} = 1; // 64-bit GPR flag
4199 multiclass FPToIntegerScaled<bits<2> rmode, bits<3> opcode, string asm,
4200 SDPatternOperator OpN> {
4201 // Scaled half-precision to 32-bit
4202 def SWHri : BaseFPToInteger<0b11, rmode, opcode, FPR16, GPR32,
4203 fixedpoint_f16_i32, asm,
4204 [(set GPR32:$Rd, (OpN (fmul FPR16:$Rn,
4205 fixedpoint_f16_i32:$scale)))]> {
4206 let Inst{31} = 0; // 32-bit GPR flag
4208 let Predicates = [HasFullFP16];
4211 // Scaled half-precision to 64-bit
4212 def SXHri : BaseFPToInteger<0b11, rmode, opcode, FPR16, GPR64,
4213 fixedpoint_f16_i64, asm,
4214 [(set GPR64:$Rd, (OpN (fmul FPR16:$Rn,
4215 fixedpoint_f16_i64:$scale)))]> {
4216 let Inst{31} = 1; // 64-bit GPR flag
4217 let Predicates = [HasFullFP16];
4220 // Scaled single-precision to 32-bit
4221 def SWSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR32,
4222 fixedpoint_f32_i32, asm,
4223 [(set GPR32:$Rd, (OpN (fmul FPR32:$Rn,
4224 fixedpoint_f32_i32:$scale)))]> {
4225 let Inst{31} = 0; // 32-bit GPR flag
4229 // Scaled single-precision to 64-bit
4230 def SXSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR64,
4231 fixedpoint_f32_i64, asm,
4232 [(set GPR64:$Rd, (OpN (fmul FPR32:$Rn,
4233 fixedpoint_f32_i64:$scale)))]> {
4234 let Inst{31} = 1; // 64-bit GPR flag
4237 // Scaled double-precision to 32-bit
4238 def SWDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR32,
4239 fixedpoint_f64_i32, asm,
4240 [(set GPR32:$Rd, (OpN (fmul FPR64:$Rn,
4241 fixedpoint_f64_i32:$scale)))]> {
4242 let Inst{31} = 0; // 32-bit GPR flag
4246 // Scaled double-precision to 64-bit
4247 def SXDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR64,
4248 fixedpoint_f64_i64, asm,
4249 [(set GPR64:$Rd, (OpN (fmul FPR64:$Rn,
4250 fixedpoint_f64_i64:$scale)))]> {
4251 let Inst{31} = 1; // 64-bit GPR flag
4256 // Integer to floating point conversion
4259 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
4260 class BaseIntegerToFP<bit isUnsigned,
4261 RegisterClass srcType, RegisterClass dstType,
4262 Operand immType, string asm, list<dag> pattern>
4263 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
4264 asm, "\t$Rd, $Rn, $scale", "", pattern>,
4265 Sched<[WriteFCvt]> {
4269 let Inst{30-24} = 0b0011110;
4270 let Inst{21-17} = 0b00001;
4271 let Inst{16} = isUnsigned;
4272 let Inst{15-10} = scale;
4277 class BaseIntegerToFPUnscaled<bit isUnsigned,
4278 RegisterClass srcType, RegisterClass dstType,
4279 ValueType dvt, string asm, SDNode node>
4280 : I<(outs dstType:$Rd), (ins srcType:$Rn),
4281 asm, "\t$Rd, $Rn", "", [(set (dvt dstType:$Rd), (node srcType:$Rn))]>,
4282 Sched<[WriteFCvt]> {
4286 let Inst{30-24} = 0b0011110;
4287 let Inst{21-17} = 0b10001;
4288 let Inst{16} = isUnsigned;
4289 let Inst{15-10} = 0b000000;
4294 multiclass IntegerToFP<bit isUnsigned, string asm, SDNode node> {
4296 def UWHri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR16, f16, asm, node> {
4297 let Inst{31} = 0; // 32-bit GPR flag
4298 let Inst{23-22} = 0b11; // 16-bit FPR flag
4299 let Predicates = [HasFullFP16];
4302 def UWSri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR32, f32, asm, node> {
4303 let Inst{31} = 0; // 32-bit GPR flag
4304 let Inst{23-22} = 0b00; // 32-bit FPR flag
4307 def UWDri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR64, f64, asm, node> {
4308 let Inst{31} = 0; // 32-bit GPR flag
4309 let Inst{23-22} = 0b01; // 64-bit FPR flag
4312 def UXHri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR16, f16, asm, node> {
4313 let Inst{31} = 1; // 64-bit GPR flag
4314 let Inst{23-22} = 0b11; // 16-bit FPR flag
4315 let Predicates = [HasFullFP16];
4318 def UXSri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR32, f32, asm, node> {
4319 let Inst{31} = 1; // 64-bit GPR flag
4320 let Inst{23-22} = 0b00; // 32-bit FPR flag
4323 def UXDri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR64, f64, asm, node> {
4324 let Inst{31} = 1; // 64-bit GPR flag
4325 let Inst{23-22} = 0b01; // 64-bit FPR flag
4329 def SWHri: BaseIntegerToFP<isUnsigned, GPR32, FPR16, fixedpoint_f16_i32, asm,
4331 (fdiv (node GPR32:$Rn),
4332 fixedpoint_f16_i32:$scale))]> {
4333 let Inst{31} = 0; // 32-bit GPR flag
4334 let Inst{23-22} = 0b11; // 16-bit FPR flag
4336 let Predicates = [HasFullFP16];
4339 def SWSri: BaseIntegerToFP<isUnsigned, GPR32, FPR32, fixedpoint_f32_i32, asm,
4341 (fdiv (node GPR32:$Rn),
4342 fixedpoint_f32_i32:$scale))]> {
4343 let Inst{31} = 0; // 32-bit GPR flag
4344 let Inst{23-22} = 0b00; // 32-bit FPR flag
4348 def SWDri: BaseIntegerToFP<isUnsigned, GPR32, FPR64, fixedpoint_f64_i32, asm,
4350 (fdiv (node GPR32:$Rn),
4351 fixedpoint_f64_i32:$scale))]> {
4352 let Inst{31} = 0; // 32-bit GPR flag
4353 let Inst{23-22} = 0b01; // 64-bit FPR flag
4357 def SXHri: BaseIntegerToFP<isUnsigned, GPR64, FPR16, fixedpoint_f16_i64, asm,
4359 (fdiv (node GPR64:$Rn),
4360 fixedpoint_f16_i64:$scale))]> {
4361 let Inst{31} = 1; // 64-bit GPR flag
4362 let Inst{23-22} = 0b11; // 16-bit FPR flag
4363 let Predicates = [HasFullFP16];
4366 def SXSri: BaseIntegerToFP<isUnsigned, GPR64, FPR32, fixedpoint_f32_i64, asm,
4368 (fdiv (node GPR64:$Rn),
4369 fixedpoint_f32_i64:$scale))]> {
4370 let Inst{31} = 1; // 64-bit GPR flag
4371 let Inst{23-22} = 0b00; // 32-bit FPR flag
4374 def SXDri: BaseIntegerToFP<isUnsigned, GPR64, FPR64, fixedpoint_f64_i64, asm,
4376 (fdiv (node GPR64:$Rn),
4377 fixedpoint_f64_i64:$scale))]> {
4378 let Inst{31} = 1; // 64-bit GPR flag
4379 let Inst{23-22} = 0b01; // 64-bit FPR flag
4384 // Unscaled integer <-> floating point conversion (i.e. FMOV)
4387 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4388 class BaseUnscaledConversion<bits<2> rmode, bits<3> opcode,
4389 RegisterClass srcType, RegisterClass dstType,
4391 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "",
4392 // We use COPY_TO_REGCLASS for these bitconvert operations.
4393 // copyPhysReg() expands the resultant COPY instructions after
4394 // regalloc is done. This gives greater freedom for the allocator
4395 // and related passes (coalescing, copy propagation, et. al.) to
4396 // be more effective.
4397 [/*(set (dvt dstType:$Rd), (bitconvert (svt srcType:$Rn)))*/]>,
4398 Sched<[WriteFCopy]> {
4401 let Inst{30-24} = 0b0011110;
4403 let Inst{20-19} = rmode;
4404 let Inst{18-16} = opcode;
4405 let Inst{15-10} = 0b000000;
4410 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4411 class BaseUnscaledConversionToHigh<bits<2> rmode, bits<3> opcode,
4412 RegisterClass srcType, RegisterOperand dstType, string asm,
4414 : I<(outs dstType:$Rd), (ins srcType:$Rn, VectorIndex1:$idx), asm,
4415 "{\t$Rd"#kind#"$idx, $Rn|"#kind#"\t$Rd$idx, $Rn}", "", []>,
4416 Sched<[WriteFCopy]> {
4419 let Inst{30-23} = 0b00111101;
4421 let Inst{20-19} = rmode;
4422 let Inst{18-16} = opcode;
4423 let Inst{15-10} = 0b000000;
4427 let DecoderMethod = "DecodeFMOVLaneInstruction";
4430 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4431 class BaseUnscaledConversionFromHigh<bits<2> rmode, bits<3> opcode,
4432 RegisterOperand srcType, RegisterClass dstType, string asm,
4434 : I<(outs dstType:$Rd), (ins srcType:$Rn, VectorIndex1:$idx), asm,
4435 "{\t$Rd, $Rn"#kind#"$idx|"#kind#"\t$Rd, $Rn$idx}", "", []>,
4436 Sched<[WriteFCopy]> {
4439 let Inst{30-23} = 0b00111101;
4441 let Inst{20-19} = rmode;
4442 let Inst{18-16} = opcode;
4443 let Inst{15-10} = 0b000000;
4447 let DecoderMethod = "DecodeFMOVLaneInstruction";
4451 multiclass UnscaledConversion<string asm> {
4452 def WHr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR16, asm> {
4453 let Inst{31} = 0; // 32-bit GPR flag
4454 let Inst{23-22} = 0b11; // 16-bit FPR flag
4455 let Predicates = [HasFullFP16];
4458 def XHr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR16, asm> {
4459 let Inst{31} = 1; // 64-bit GPR flag
4460 let Inst{23-22} = 0b11; // 16-bit FPR flag
4461 let Predicates = [HasFullFP16];
4464 def WSr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR32, asm> {
4465 let Inst{31} = 0; // 32-bit GPR flag
4466 let Inst{23-22} = 0b00; // 32-bit FPR flag
4469 def XDr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR64, asm> {
4470 let Inst{31} = 1; // 64-bit GPR flag
4471 let Inst{23-22} = 0b01; // 64-bit FPR flag
4474 def HWr : BaseUnscaledConversion<0b00, 0b110, FPR16, GPR32, asm> {
4475 let Inst{31} = 0; // 32-bit GPR flag
4476 let Inst{23-22} = 0b11; // 16-bit FPR flag
4477 let Predicates = [HasFullFP16];
4480 def HXr : BaseUnscaledConversion<0b00, 0b110, FPR16, GPR64, asm> {
4481 let Inst{31} = 1; // 64-bit GPR flag
4482 let Inst{23-22} = 0b11; // 16-bit FPR flag
4483 let Predicates = [HasFullFP16];
4486 def SWr : BaseUnscaledConversion<0b00, 0b110, FPR32, GPR32, asm> {
4487 let Inst{31} = 0; // 32-bit GPR flag
4488 let Inst{23-22} = 0b00; // 32-bit FPR flag
4491 def DXr : BaseUnscaledConversion<0b00, 0b110, FPR64, GPR64, asm> {
4492 let Inst{31} = 1; // 64-bit GPR flag
4493 let Inst{23-22} = 0b01; // 64-bit FPR flag
4496 def XDHighr : BaseUnscaledConversionToHigh<0b01, 0b111, GPR64, V128,
4502 def DXHighr : BaseUnscaledConversionFromHigh<0b01, 0b110, V128, GPR64,
4510 // Floating point conversion
4513 class BaseFPConversion<bits<2> type, bits<2> opcode, RegisterClass dstType,
4514 RegisterClass srcType, string asm, list<dag> pattern>
4515 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "", pattern>,
4516 Sched<[WriteFCvt]> {
4519 let Inst{31-24} = 0b00011110;
4520 let Inst{23-22} = type;
4521 let Inst{21-17} = 0b10001;
4522 let Inst{16-15} = opcode;
4523 let Inst{14-10} = 0b10000;
4528 multiclass FPConversion<string asm> {
4529 // Double-precision to Half-precision
4530 def HDr : BaseFPConversion<0b01, 0b11, FPR16, FPR64, asm,
4531 [(set FPR16:$Rd, (fpround FPR64:$Rn))]>;
4533 // Double-precision to Single-precision
4534 def SDr : BaseFPConversion<0b01, 0b00, FPR32, FPR64, asm,
4535 [(set FPR32:$Rd, (fpround FPR64:$Rn))]>;
4537 // Half-precision to Double-precision
4538 def DHr : BaseFPConversion<0b11, 0b01, FPR64, FPR16, asm,
4539 [(set FPR64:$Rd, (fpextend FPR16:$Rn))]>;
4541 // Half-precision to Single-precision
4542 def SHr : BaseFPConversion<0b11, 0b00, FPR32, FPR16, asm,
4543 [(set FPR32:$Rd, (fpextend FPR16:$Rn))]>;
4545 // Single-precision to Double-precision
4546 def DSr : BaseFPConversion<0b00, 0b01, FPR64, FPR32, asm,
4547 [(set FPR64:$Rd, (fpextend FPR32:$Rn))]>;
4549 // Single-precision to Half-precision
4550 def HSr : BaseFPConversion<0b00, 0b11, FPR16, FPR32, asm,
4551 [(set FPR16:$Rd, (fpround FPR32:$Rn))]>;
4555 // Single operand floating point data processing
4558 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4559 class BaseSingleOperandFPData<bits<6> opcode, RegisterClass regtype,
4560 ValueType vt, string asm, SDPatternOperator node>
4561 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
4562 [(set (vt regtype:$Rd), (node (vt regtype:$Rn)))]>,
4566 let Inst{31-24} = 0b00011110;
4568 let Inst{20-15} = opcode;
4569 let Inst{14-10} = 0b10000;
4574 multiclass SingleOperandFPData<bits<4> opcode, string asm,
4575 SDPatternOperator node = null_frag> {
4577 def Hr : BaseSingleOperandFPData<{0b00,opcode}, FPR16, f16, asm, node> {
4578 let Inst{23-22} = 0b11; // 16-bit size flag
4579 let Predicates = [HasFullFP16];
4582 def Sr : BaseSingleOperandFPData<{0b00,opcode}, FPR32, f32, asm, node> {
4583 let Inst{23-22} = 0b00; // 32-bit size flag
4586 def Dr : BaseSingleOperandFPData<{0b00,opcode}, FPR64, f64, asm, node> {
4587 let Inst{23-22} = 0b01; // 64-bit size flag
4591 multiclass SingleOperandFPNo16<bits<6> opcode, string asm,
4592 SDPatternOperator node = null_frag>{
4594 def Sr : BaseSingleOperandFPData<opcode, FPR32, f32, asm, node> {
4595 let Inst{23-22} = 0b00; // 32-bit registers
4598 def Dr : BaseSingleOperandFPData<opcode, FPR64, f64, asm, node> {
4599 let Inst{23-22} = 0b01; // 64-bit registers
4603 // FRInt[32|64][Z|N] instructions
4604 multiclass FRIntNNT<bits<2> opcode, string asm, SDPatternOperator node = null_frag> :
4605 SingleOperandFPNo16<{0b0100,opcode}, asm, node>;
4608 // Two operand floating point data processing
4611 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4612 class BaseTwoOperandFPData<bits<4> opcode, RegisterClass regtype,
4613 string asm, list<dag> pat>
4614 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
4615 asm, "\t$Rd, $Rn, $Rm", "", pat>,
4620 let Inst{31-24} = 0b00011110;
4622 let Inst{20-16} = Rm;
4623 let Inst{15-12} = opcode;
4624 let Inst{11-10} = 0b10;
4629 multiclass TwoOperandFPData<bits<4> opcode, string asm,
4630 SDPatternOperator node = null_frag> {
4631 def Hrr : BaseTwoOperandFPData<opcode, FPR16, asm,
4632 [(set (f16 FPR16:$Rd),
4633 (node (f16 FPR16:$Rn), (f16 FPR16:$Rm)))]> {
4634 let Inst{23-22} = 0b11; // 16-bit size flag
4635 let Predicates = [HasFullFP16];
4638 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
4639 [(set (f32 FPR32:$Rd),
4640 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]> {
4641 let Inst{23-22} = 0b00; // 32-bit size flag
4644 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
4645 [(set (f64 FPR64:$Rd),
4646 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]> {
4647 let Inst{23-22} = 0b01; // 64-bit size flag
4651 multiclass TwoOperandFPDataNeg<bits<4> opcode, string asm, SDNode node> {
4652 def Hrr : BaseTwoOperandFPData<opcode, FPR16, asm,
4653 [(set FPR16:$Rd, (fneg (node FPR16:$Rn, (f16 FPR16:$Rm))))]> {
4654 let Inst{23-22} = 0b11; // 16-bit size flag
4655 let Predicates = [HasFullFP16];
4658 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
4659 [(set FPR32:$Rd, (fneg (node FPR32:$Rn, (f32 FPR32:$Rm))))]> {
4660 let Inst{23-22} = 0b00; // 32-bit size flag
4663 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
4664 [(set FPR64:$Rd, (fneg (node FPR64:$Rn, (f64 FPR64:$Rm))))]> {
4665 let Inst{23-22} = 0b01; // 64-bit size flag
4671 // Three operand floating point data processing
4674 class BaseThreeOperandFPData<bit isNegated, bit isSub,
4675 RegisterClass regtype, string asm, list<dag> pat>
4676 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, regtype: $Ra),
4677 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pat>,
4678 Sched<[WriteFMul]> {
4683 let Inst{31-24} = 0b00011111;
4684 let Inst{21} = isNegated;
4685 let Inst{20-16} = Rm;
4686 let Inst{15} = isSub;
4687 let Inst{14-10} = Ra;
4692 multiclass ThreeOperandFPData<bit isNegated, bit isSub,string asm,
4693 SDPatternOperator node> {
4694 def Hrrr : BaseThreeOperandFPData<isNegated, isSub, FPR16, asm,
4696 (node (f16 FPR16:$Rn), (f16 FPR16:$Rm), (f16 FPR16:$Ra)))]> {
4697 let Inst{23-22} = 0b11; // 16-bit size flag
4698 let Predicates = [HasFullFP16];
4701 def Srrr : BaseThreeOperandFPData<isNegated, isSub, FPR32, asm,
4703 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm), (f32 FPR32:$Ra)))]> {
4704 let Inst{23-22} = 0b00; // 32-bit size flag
4707 def Drrr : BaseThreeOperandFPData<isNegated, isSub, FPR64, asm,
4709 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm), (f64 FPR64:$Ra)))]> {
4710 let Inst{23-22} = 0b01; // 64-bit size flag
4715 // Floating point data comparisons
4718 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4719 class BaseOneOperandFPComparison<bit signalAllNans,
4720 RegisterClass regtype, string asm,
4722 : I<(outs), (ins regtype:$Rn), asm, "\t$Rn, #0.0", "", pat>,
4723 Sched<[WriteFCmp]> {
4725 let Inst{31-24} = 0b00011110;
4728 let Inst{15-10} = 0b001000;
4730 let Inst{4} = signalAllNans;
4731 let Inst{3-0} = 0b1000;
4733 // Rm should be 0b00000 canonically, but we need to accept any value.
4734 let PostEncoderMethod = "fixOneOperandFPComparison";
4737 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4738 class BaseTwoOperandFPComparison<bit signalAllNans, RegisterClass regtype,
4739 string asm, list<dag> pat>
4740 : I<(outs), (ins regtype:$Rn, regtype:$Rm), asm, "\t$Rn, $Rm", "", pat>,
4741 Sched<[WriteFCmp]> {
4744 let Inst{31-24} = 0b00011110;
4746 let Inst{20-16} = Rm;
4747 let Inst{15-10} = 0b001000;
4749 let Inst{4} = signalAllNans;
4750 let Inst{3-0} = 0b0000;
4753 multiclass FPComparison<bit signalAllNans, string asm,
4754 SDPatternOperator OpNode = null_frag> {
4755 let Defs = [NZCV] in {
4756 def Hrr : BaseTwoOperandFPComparison<signalAllNans, FPR16, asm,
4757 [(OpNode FPR16:$Rn, (f16 FPR16:$Rm)), (implicit NZCV)]> {
4758 let Inst{23-22} = 0b11;
4759 let Predicates = [HasFullFP16];
4762 def Hri : BaseOneOperandFPComparison<signalAllNans, FPR16, asm,
4763 [(OpNode (f16 FPR16:$Rn), fpimm0), (implicit NZCV)]> {
4764 let Inst{23-22} = 0b11;
4765 let Predicates = [HasFullFP16];
4768 def Srr : BaseTwoOperandFPComparison<signalAllNans, FPR32, asm,
4769 [(OpNode FPR32:$Rn, (f32 FPR32:$Rm)), (implicit NZCV)]> {
4770 let Inst{23-22} = 0b00;
4773 def Sri : BaseOneOperandFPComparison<signalAllNans, FPR32, asm,
4774 [(OpNode (f32 FPR32:$Rn), fpimm0), (implicit NZCV)]> {
4775 let Inst{23-22} = 0b00;
4778 def Drr : BaseTwoOperandFPComparison<signalAllNans, FPR64, asm,
4779 [(OpNode FPR64:$Rn, (f64 FPR64:$Rm)), (implicit NZCV)]> {
4780 let Inst{23-22} = 0b01;
4783 def Dri : BaseOneOperandFPComparison<signalAllNans, FPR64, asm,
4784 [(OpNode (f64 FPR64:$Rn), fpimm0), (implicit NZCV)]> {
4785 let Inst{23-22} = 0b01;
4791 // Floating point conditional comparisons
4794 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4795 class BaseFPCondComparison<bit signalAllNans, RegisterClass regtype,
4796 string mnemonic, list<dag> pat>
4797 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm32_0_15:$nzcv, ccode:$cond),
4798 mnemonic, "\t$Rn, $Rm, $nzcv, $cond", "", pat>,
4799 Sched<[WriteFCmp]> {
4808 let Inst{31-24} = 0b00011110;
4810 let Inst{20-16} = Rm;
4811 let Inst{15-12} = cond;
4812 let Inst{11-10} = 0b01;
4814 let Inst{4} = signalAllNans;
4815 let Inst{3-0} = nzcv;
4818 multiclass FPCondComparison<bit signalAllNans, string mnemonic,
4819 SDPatternOperator OpNode = null_frag> {
4820 def Hrr : BaseFPCondComparison<signalAllNans, FPR16, mnemonic,
4821 [(set NZCV, (OpNode (f16 FPR16:$Rn), (f16 FPR16:$Rm), (i32 imm:$nzcv),
4822 (i32 imm:$cond), NZCV))]> {
4823 let Inst{23-22} = 0b11;
4824 let Predicates = [HasFullFP16];
4827 def Srr : BaseFPCondComparison<signalAllNans, FPR32, mnemonic,
4828 [(set NZCV, (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm), (i32 imm:$nzcv),
4829 (i32 imm:$cond), NZCV))]> {
4830 let Inst{23-22} = 0b00;
4833 def Drr : BaseFPCondComparison<signalAllNans, FPR64, mnemonic,
4834 [(set NZCV, (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm), (i32 imm:$nzcv),
4835 (i32 imm:$cond), NZCV))]> {
4836 let Inst{23-22} = 0b01;
4841 // Floating point conditional select
4844 class BaseFPCondSelect<RegisterClass regtype, ValueType vt, string asm>
4845 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
4846 asm, "\t$Rd, $Rn, $Rm, $cond", "",
4848 (AArch64csel (vt regtype:$Rn), regtype:$Rm,
4849 (i32 imm:$cond), NZCV))]>,
4856 let Inst{31-24} = 0b00011110;
4858 let Inst{20-16} = Rm;
4859 let Inst{15-12} = cond;
4860 let Inst{11-10} = 0b11;
4865 multiclass FPCondSelect<string asm> {
4866 let Uses = [NZCV] in {
4867 def Hrrr : BaseFPCondSelect<FPR16, f16, asm> {
4868 let Inst{23-22} = 0b11;
4869 let Predicates = [HasFullFP16];
4872 def Srrr : BaseFPCondSelect<FPR32, f32, asm> {
4873 let Inst{23-22} = 0b00;
4876 def Drrr : BaseFPCondSelect<FPR64, f64, asm> {
4877 let Inst{23-22} = 0b01;
4883 // Floating move immediate
4886 class BaseFPMoveImmediate<RegisterClass regtype, Operand fpimmtype, string asm>
4887 : I<(outs regtype:$Rd), (ins fpimmtype:$imm), asm, "\t$Rd, $imm", "",
4888 [(set regtype:$Rd, fpimmtype:$imm)]>,
4889 Sched<[WriteFImm]> {
4892 let Inst{31-24} = 0b00011110;
4894 let Inst{20-13} = imm;
4895 let Inst{12-5} = 0b10000000;
4899 multiclass FPMoveImmediate<string asm> {
4900 def Hi : BaseFPMoveImmediate<FPR16, fpimm16, asm> {
4901 let Inst{23-22} = 0b11;
4902 let Predicates = [HasFullFP16];
4905 def Si : BaseFPMoveImmediate<FPR32, fpimm32, asm> {
4906 let Inst{23-22} = 0b00;
4909 def Di : BaseFPMoveImmediate<FPR64, fpimm64, asm> {
4910 let Inst{23-22} = 0b01;
4913 } // end of 'let Predicates = [HasFPARMv8]'
4915 //----------------------------------------------------------------------------
4917 //----------------------------------------------------------------------------
4919 let Predicates = [HasNEON] in {
4921 //----------------------------------------------------------------------------
4922 // AdvSIMD three register vector instructions
4923 //----------------------------------------------------------------------------
4925 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4926 class BaseSIMDThreeSameVector<bit Q, bit U, bits<3> size, bits<5> opcode,
4927 RegisterOperand regtype, string asm, string kind,
4929 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
4930 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
4931 "|" # kind # "\t$Rd, $Rn, $Rm|}", "", pattern>,
4939 let Inst{28-24} = 0b01110;
4940 let Inst{23-21} = size;
4941 let Inst{20-16} = Rm;
4942 let Inst{15-11} = opcode;
4948 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4949 class BaseSIMDThreeSameVectorTied<bit Q, bit U, bits<3> size, bits<5> opcode,
4950 RegisterOperand regtype, string asm, string kind,
4952 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn, regtype:$Rm), asm,
4953 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
4954 "|" # kind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
4962 let Inst{28-24} = 0b01110;
4963 let Inst{23-21} = size;
4964 let Inst{20-16} = Rm;
4965 let Inst{15-11} = opcode;
4971 // All operand sizes distinguished in the encoding.
4972 multiclass SIMDThreeSameVector<bit U, bits<5> opc, string asm,
4973 SDPatternOperator OpNode> {
4974 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
4976 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4977 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
4979 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
4980 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
4982 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4983 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
4985 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
4986 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
4988 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4989 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
4991 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
4992 def v2i64 : BaseSIMDThreeSameVector<1, U, 0b111, opc, V128,
4994 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
4997 // As above, but D sized elements unsupported.
4998 multiclass SIMDThreeSameVectorBHS<bit U, bits<5> opc, string asm,
4999 SDPatternOperator OpNode> {
5000 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
5002 [(set V64:$Rd, (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))]>;
5003 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
5005 [(set V128:$Rd, (v16i8 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm))))]>;
5006 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
5008 [(set V64:$Rd, (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))]>;
5009 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
5011 [(set V128:$Rd, (v8i16 (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm))))]>;
5012 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
5014 [(set V64:$Rd, (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))]>;
5015 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
5017 [(set V128:$Rd, (v4i32 (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm))))]>;
5020 multiclass SIMDThreeSameVectorBHSTied<bit U, bits<5> opc, string asm,
5021 SDPatternOperator OpNode> {
5022 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, 0b001, opc, V64,
5024 [(set (v8i8 V64:$dst),
5025 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5026 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, 0b001, opc, V128,
5028 [(set (v16i8 V128:$dst),
5029 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
5030 def v4i16 : BaseSIMDThreeSameVectorTied<0, U, 0b011, opc, V64,
5032 [(set (v4i16 V64:$dst),
5033 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5034 def v8i16 : BaseSIMDThreeSameVectorTied<1, U, 0b011, opc, V128,
5036 [(set (v8i16 V128:$dst),
5037 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
5038 def v2i32 : BaseSIMDThreeSameVectorTied<0, U, 0b101, opc, V64,
5040 [(set (v2i32 V64:$dst),
5041 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5042 def v4i32 : BaseSIMDThreeSameVectorTied<1, U, 0b101, opc, V128,
5044 [(set (v4i32 V128:$dst),
5045 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
5048 // As above, but only B sized elements supported.
5049 multiclass SIMDThreeSameVectorB<bit U, bits<5> opc, string asm,
5050 SDPatternOperator OpNode> {
5051 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
5053 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5054 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
5056 [(set (v16i8 V128:$Rd),
5057 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
5060 // As above, but only floating point elements supported.
5061 multiclass SIMDThreeSameVectorFP<bit U, bit S, bits<3> opc,
5062 string asm, SDPatternOperator OpNode> {
5063 let Predicates = [HasNEON, HasFullFP16] in {
5064 def v4f16 : BaseSIMDThreeSameVector<0, U, {S,0b10}, {0b00,opc}, V64,
5066 [(set (v4f16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
5067 def v8f16 : BaseSIMDThreeSameVector<1, U, {S,0b10}, {0b00,opc}, V128,
5069 [(set (v8f16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
5070 } // Predicates = [HasNEON, HasFullFP16]
5071 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0b01}, {0b11,opc}, V64,
5073 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
5074 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0b01}, {0b11,opc}, V128,
5076 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
5077 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,0b11}, {0b11,opc}, V128,
5079 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
5082 multiclass SIMDThreeSameVectorFPCmp<bit U, bit S, bits<3> opc,
5084 SDPatternOperator OpNode> {
5085 let Predicates = [HasNEON, HasFullFP16] in {
5086 def v4f16 : BaseSIMDThreeSameVector<0, U, {S,0b10}, {0b00,opc}, V64,
5088 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
5089 def v8f16 : BaseSIMDThreeSameVector<1, U, {S,0b10}, {0b00,opc}, V128,
5091 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
5092 } // Predicates = [HasNEON, HasFullFP16]
5093 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0b01}, {0b11,opc}, V64,
5095 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
5096 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0b01}, {0b11,opc}, V128,
5098 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
5099 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,0b11}, {0b11,opc}, V128,
5101 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
5104 multiclass SIMDThreeSameVectorFPTied<bit U, bit S, bits<3> opc,
5105 string asm, SDPatternOperator OpNode> {
5106 let Predicates = [HasNEON, HasFullFP16] in {
5107 def v4f16 : BaseSIMDThreeSameVectorTied<0, U, {S,0b10}, {0b00,opc}, V64,
5109 [(set (v4f16 V64:$dst),
5110 (OpNode (v4f16 V64:$Rd), (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
5111 def v8f16 : BaseSIMDThreeSameVectorTied<1, U, {S,0b10}, {0b00,opc}, V128,
5113 [(set (v8f16 V128:$dst),
5114 (OpNode (v8f16 V128:$Rd), (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
5115 } // Predicates = [HasNEON, HasFullFP16]
5116 def v2f32 : BaseSIMDThreeSameVectorTied<0, U, {S,0b01}, {0b11,opc}, V64,
5118 [(set (v2f32 V64:$dst),
5119 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
5120 def v4f32 : BaseSIMDThreeSameVectorTied<1, U, {S,0b01}, {0b11,opc}, V128,
5122 [(set (v4f32 V128:$dst),
5123 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
5124 def v2f64 : BaseSIMDThreeSameVectorTied<1, U, {S,0b11}, {0b11,opc}, V128,
5126 [(set (v2f64 V128:$dst),
5127 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
5130 // As above, but D and B sized elements unsupported.
5131 multiclass SIMDThreeSameVectorHS<bit U, bits<5> opc, string asm,
5132 SDPatternOperator OpNode> {
5133 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
5135 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5136 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
5138 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
5139 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
5141 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5142 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
5144 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
5147 // Logical three vector ops share opcode bits, and only use B sized elements.
5148 multiclass SIMDLogicalThreeVector<bit U, bits<2> size, string asm,
5149 SDPatternOperator OpNode = null_frag> {
5150 def v8i8 : BaseSIMDThreeSameVector<0, U, {size,1}, 0b00011, V64,
5152 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn, V64:$Rm))]>;
5153 def v16i8 : BaseSIMDThreeSameVector<1, U, {size,1}, 0b00011, V128,
5155 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn, V128:$Rm))]>;
5157 def : Pat<(v4i16 (OpNode V64:$LHS, V64:$RHS)),
5158 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
5159 def : Pat<(v2i32 (OpNode V64:$LHS, V64:$RHS)),
5160 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
5161 def : Pat<(v1i64 (OpNode V64:$LHS, V64:$RHS)),
5162 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
5164 def : Pat<(v8i16 (OpNode V128:$LHS, V128:$RHS)),
5165 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
5166 def : Pat<(v4i32 (OpNode V128:$LHS, V128:$RHS)),
5167 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
5168 def : Pat<(v2i64 (OpNode V128:$LHS, V128:$RHS)),
5169 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
5172 multiclass SIMDLogicalThreeVectorTied<bit U, bits<2> size,
5173 string asm, SDPatternOperator OpNode> {
5174 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, {size,1}, 0b00011, V64,
5176 [(set (v8i8 V64:$dst),
5177 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5178 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, {size,1}, 0b00011, V128,
5180 [(set (v16i8 V128:$dst),
5181 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
5182 (v16i8 V128:$Rm)))]>;
5184 def : Pat<(v4i16 (OpNode (v4i16 V64:$LHS), (v4i16 V64:$MHS),
5186 (!cast<Instruction>(NAME#"v8i8")
5187 V64:$LHS, V64:$MHS, V64:$RHS)>;
5188 def : Pat<(v2i32 (OpNode (v2i32 V64:$LHS), (v2i32 V64:$MHS),
5190 (!cast<Instruction>(NAME#"v8i8")
5191 V64:$LHS, V64:$MHS, V64:$RHS)>;
5192 def : Pat<(v1i64 (OpNode (v1i64 V64:$LHS), (v1i64 V64:$MHS),
5194 (!cast<Instruction>(NAME#"v8i8")
5195 V64:$LHS, V64:$MHS, V64:$RHS)>;
5197 def : Pat<(v8i16 (OpNode (v8i16 V128:$LHS), (v8i16 V128:$MHS),
5198 (v8i16 V128:$RHS))),
5199 (!cast<Instruction>(NAME#"v16i8")
5200 V128:$LHS, V128:$MHS, V128:$RHS)>;
5201 def : Pat<(v4i32 (OpNode (v4i32 V128:$LHS), (v4i32 V128:$MHS),
5202 (v4i32 V128:$RHS))),
5203 (!cast<Instruction>(NAME#"v16i8")
5204 V128:$LHS, V128:$MHS, V128:$RHS)>;
5205 def : Pat<(v2i64 (OpNode (v2i64 V128:$LHS), (v2i64 V128:$MHS),
5206 (v2i64 V128:$RHS))),
5207 (!cast<Instruction>(NAME#"v16i8")
5208 V128:$LHS, V128:$MHS, V128:$RHS)>;
5211 // ARMv8.2-A Dot Product Instructions (Vector): These instructions extract
5212 // bytes from S-sized elements.
5213 class BaseSIMDThreeSameVectorDot<bit Q, bit U, string asm, string kind1,
5214 string kind2, RegisterOperand RegType,
5215 ValueType AccumType, ValueType InputType,
5216 SDPatternOperator OpNode> :
5217 BaseSIMDThreeSameVectorTied<Q, U, 0b100, 0b10010, RegType, asm, kind1,
5218 [(set (AccumType RegType:$dst),
5219 (OpNode (AccumType RegType:$Rd),
5220 (InputType RegType:$Rn),
5221 (InputType RegType:$Rm)))]> {
5222 let AsmString = !strconcat(asm, "{\t$Rd" # kind1 # ", $Rn" # kind2 # ", $Rm" # kind2 # "}");
5225 multiclass SIMDThreeSameVectorDot<bit U, string asm, SDPatternOperator OpNode> {
5226 def v8i8 : BaseSIMDThreeSameVectorDot<0, U, asm, ".2s", ".8b", V64,
5227 v2i32, v8i8, OpNode>;
5228 def v16i8 : BaseSIMDThreeSameVectorDot<1, U, asm, ".4s", ".16b", V128,
5229 v4i32, v16i8, OpNode>;
5232 // ARMv8.2-A Fused Multiply Add-Long Instructions (Vector): These instructions
5233 // select inputs from 4H vectors and accumulate outputs to a 2S vector (or from
5234 // 8H to 4S, when Q=1).
5235 class BaseSIMDThreeSameVectorFML<bit Q, bit U, bit b13, bits<3> size, string asm, string kind1,
5236 string kind2, RegisterOperand RegType,
5237 ValueType AccumType, ValueType InputType,
5238 SDPatternOperator OpNode> :
5239 BaseSIMDThreeSameVectorTied<Q, U, size, 0b11101, RegType, asm, kind1,
5240 [(set (AccumType RegType:$dst),
5241 (OpNode (AccumType RegType:$Rd),
5242 (InputType RegType:$Rn),
5243 (InputType RegType:$Rm)))]> {
5244 let AsmString = !strconcat(asm, "{\t$Rd" # kind1 # ", $Rn" # kind2 # ", $Rm" # kind2 # "}");
5248 multiclass SIMDThreeSameVectorFML<bit U, bit b13, bits<3> size, string asm,
5249 SDPatternOperator OpNode> {
5250 def v4f16 : BaseSIMDThreeSameVectorFML<0, U, b13, size, asm, ".2s", ".2h", V64,
5251 v2f32, v4f16, OpNode>;
5252 def v8f16 : BaseSIMDThreeSameVectorFML<1, U, b13, size, asm, ".4s", ".4h", V128,
5253 v4f32, v8f16, OpNode>;
5257 //----------------------------------------------------------------------------
5258 // AdvSIMD two register vector instructions.
5259 //----------------------------------------------------------------------------
5261 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5262 class BaseSIMDTwoSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5263 bits<2> size2, RegisterOperand regtype, string asm,
5264 string dstkind, string srckind, list<dag> pattern>
5265 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5266 "{\t$Rd" # dstkind # ", $Rn" # srckind #
5267 "|" # dstkind # "\t$Rd, $Rn}", "", pattern>,
5274 let Inst{28-24} = 0b01110;
5275 let Inst{23-22} = size;
5277 let Inst{20-19} = size2;
5278 let Inst{18-17} = 0b00;
5279 let Inst{16-12} = opcode;
5280 let Inst{11-10} = 0b10;
5285 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5286 class BaseSIMDTwoSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5287 bits<2> size2, RegisterOperand regtype,
5288 string asm, string dstkind, string srckind,
5290 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn), asm,
5291 "{\t$Rd" # dstkind # ", $Rn" # srckind #
5292 "|" # dstkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
5299 let Inst{28-24} = 0b01110;
5300 let Inst{23-22} = size;
5302 let Inst{20-19} = size2;
5303 let Inst{18-17} = 0b00;
5304 let Inst{16-12} = opcode;
5305 let Inst{11-10} = 0b10;
5310 // Supports B, H, and S element sizes.
5311 multiclass SIMDTwoVectorBHS<bit U, bits<5> opc, string asm,
5312 SDPatternOperator OpNode> {
5313 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5315 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5316 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5317 asm, ".16b", ".16b",
5318 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5319 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5321 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5322 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5324 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5325 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5327 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5328 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5330 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5333 class BaseSIMDVectorLShiftLongBySize<bit Q, bits<2> size,
5334 RegisterOperand regtype, string asm, string dstkind,
5335 string srckind, string amount>
5336 : I<(outs V128:$Rd), (ins regtype:$Rn), asm,
5337 "{\t$Rd" # dstkind # ", $Rn" # srckind # ", #" # amount #
5338 "|" # dstkind # "\t$Rd, $Rn, #" # amount # "}", "", []>,
5344 let Inst{29-24} = 0b101110;
5345 let Inst{23-22} = size;
5346 let Inst{21-10} = 0b100001001110;
5351 multiclass SIMDVectorLShiftLongBySizeBHS {
5352 let hasSideEffects = 0 in {
5353 def v8i8 : BaseSIMDVectorLShiftLongBySize<0, 0b00, V64,
5354 "shll", ".8h", ".8b", "8">;
5355 def v16i8 : BaseSIMDVectorLShiftLongBySize<1, 0b00, V128,
5356 "shll2", ".8h", ".16b", "8">;
5357 def v4i16 : BaseSIMDVectorLShiftLongBySize<0, 0b01, V64,
5358 "shll", ".4s", ".4h", "16">;
5359 def v8i16 : BaseSIMDVectorLShiftLongBySize<1, 0b01, V128,
5360 "shll2", ".4s", ".8h", "16">;
5361 def v2i32 : BaseSIMDVectorLShiftLongBySize<0, 0b10, V64,
5362 "shll", ".2d", ".2s", "32">;
5363 def v4i32 : BaseSIMDVectorLShiftLongBySize<1, 0b10, V128,
5364 "shll2", ".2d", ".4s", "32">;
5368 // Supports all element sizes.
5369 multiclass SIMDLongTwoVector<bit U, bits<5> opc, string asm,
5370 SDPatternOperator OpNode> {
5371 def v8i8_v4i16 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5373 [(set (v4i16 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5374 def v16i8_v8i16 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5376 [(set (v8i16 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5377 def v4i16_v2i32 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5379 [(set (v2i32 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5380 def v8i16_v4i32 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5382 [(set (v4i32 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5383 def v2i32_v1i64 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5385 [(set (v1i64 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5386 def v4i32_v2i64 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5388 [(set (v2i64 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5391 multiclass SIMDLongTwoVectorTied<bit U, bits<5> opc, string asm,
5392 SDPatternOperator OpNode> {
5393 def v8i8_v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, 0b00, V64,
5395 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd),
5397 def v16i8_v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, 0b00, V128,
5399 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd),
5400 (v16i8 V128:$Rn)))]>;
5401 def v4i16_v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, 0b00, V64,
5403 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd),
5404 (v4i16 V64:$Rn)))]>;
5405 def v8i16_v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, 0b00, V128,
5407 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd),
5408 (v8i16 V128:$Rn)))]>;
5409 def v2i32_v1i64 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, 0b00, V64,
5411 [(set (v1i64 V64:$dst), (OpNode (v1i64 V64:$Rd),
5412 (v2i32 V64:$Rn)))]>;
5413 def v4i32_v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, 0b00, V128,
5415 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd),
5416 (v4i32 V128:$Rn)))]>;
5419 // Supports all element sizes, except 1xD.
5420 multiclass SIMDTwoVectorBHSDTied<bit U, bits<5> opc, string asm,
5421 SDPatternOperator OpNode> {
5422 def v8i8 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, 0b00, V64,
5424 [(set (v8i8 V64:$dst), (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn)))]>;
5425 def v16i8 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, 0b00, V128,
5426 asm, ".16b", ".16b",
5427 [(set (v16i8 V128:$dst), (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
5428 def v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, 0b00, V64,
5430 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn)))]>;
5431 def v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, 0b00, V128,
5433 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn)))]>;
5434 def v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, 0b00, V64,
5436 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn)))]>;
5437 def v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, 0b00, V128,
5439 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
5440 def v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b11, opc, 0b00, V128,
5442 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn)))]>;
5445 multiclass SIMDTwoVectorBHSD<bit U, bits<5> opc, string asm,
5446 SDPatternOperator OpNode = null_frag> {
5447 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5449 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5450 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5451 asm, ".16b", ".16b",
5452 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5453 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5455 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5456 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5458 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5459 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5461 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5462 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5464 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5465 def v2i64 : BaseSIMDTwoSameVector<1, U, 0b11, opc, 0b00, V128,
5467 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5471 // Supports only B element sizes.
5472 multiclass SIMDTwoVectorB<bit U, bits<2> size, bits<5> opc, string asm,
5473 SDPatternOperator OpNode> {
5474 def v8i8 : BaseSIMDTwoSameVector<0, U, size, opc, 0b00, V64,
5476 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5477 def v16i8 : BaseSIMDTwoSameVector<1, U, size, opc, 0b00, V128,
5478 asm, ".16b", ".16b",
5479 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5483 // Supports only B and H element sizes.
5484 multiclass SIMDTwoVectorBH<bit U, bits<5> opc, string asm,
5485 SDPatternOperator OpNode> {
5486 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5488 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn))]>;
5489 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5490 asm, ".16b", ".16b",
5491 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn))]>;
5492 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5494 [(set (v4i16 V64:$Rd), (OpNode V64:$Rn))]>;
5495 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5497 [(set (v8i16 V128:$Rd), (OpNode V128:$Rn))]>;
5500 // Supports H, S and D element sizes, uses high bit of the size field
5501 // as an extra opcode bit.
5502 multiclass SIMDTwoVectorFP<bit U, bit S, bits<5> opc, string asm,
5503 SDPatternOperator OpNode> {
5504 let Predicates = [HasNEON, HasFullFP16] in {
5505 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5507 [(set (v4f16 V64:$Rd), (OpNode (v4f16 V64:$Rn)))]>;
5508 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5510 [(set (v8f16 V128:$Rd), (OpNode (v8f16 V128:$Rn)))]>;
5511 } // Predicates = [HasNEON, HasFullFP16]
5512 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5514 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5515 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5517 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5518 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5520 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5523 // Supports only S and D element sizes
5524 multiclass SIMDTwoVectorSD<bit U, bits<5> opc, string asm,
5525 SDPatternOperator OpNode = null_frag> {
5527 def v2f32 : BaseSIMDTwoSameVector<0, U, 00, opc, 0b00, V64,
5529 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5530 def v4f32 : BaseSIMDTwoSameVector<1, U, 00, opc, 0b00, V128,
5532 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5533 def v2f64 : BaseSIMDTwoSameVector<1, U, 01, opc, 0b00, V128,
5535 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5538 multiclass FRIntNNTVector<bit U, bit op, string asm,
5539 SDPatternOperator OpNode = null_frag> :
5540 SIMDTwoVectorSD<U, {0b1111,op}, asm, OpNode>;
5542 // Supports only S element size.
5543 multiclass SIMDTwoVectorS<bit U, bit S, bits<5> opc, string asm,
5544 SDPatternOperator OpNode> {
5545 def v2i32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5547 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5548 def v4i32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5550 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5554 multiclass SIMDTwoVectorFPToInt<bit U, bit S, bits<5> opc, string asm,
5555 SDPatternOperator OpNode> {
5556 let Predicates = [HasNEON, HasFullFP16] in {
5557 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5559 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn)))]>;
5560 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5562 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn)))]>;
5563 } // Predicates = [HasNEON, HasFullFP16]
5564 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5566 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5567 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5569 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5570 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5572 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5575 multiclass SIMDTwoVectorIntToFP<bit U, bit S, bits<5> opc, string asm,
5576 SDPatternOperator OpNode> {
5577 let Predicates = [HasNEON, HasFullFP16] in {
5578 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5580 [(set (v4f16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5581 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5583 [(set (v8f16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5584 } // Predicates = [HasNEON, HasFullFP16]
5585 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5587 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5588 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5590 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5591 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5593 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5597 class BaseSIMDMixedTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5598 RegisterOperand inreg, RegisterOperand outreg,
5599 string asm, string outkind, string inkind,
5601 : I<(outs outreg:$Rd), (ins inreg:$Rn), asm,
5602 "{\t$Rd" # outkind # ", $Rn" # inkind #
5603 "|" # outkind # "\t$Rd, $Rn}", "", pattern>,
5610 let Inst{28-24} = 0b01110;
5611 let Inst{23-22} = size;
5612 let Inst{21-17} = 0b10000;
5613 let Inst{16-12} = opcode;
5614 let Inst{11-10} = 0b10;
5619 class BaseSIMDMixedTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5620 RegisterOperand inreg, RegisterOperand outreg,
5621 string asm, string outkind, string inkind,
5623 : I<(outs outreg:$dst), (ins outreg:$Rd, inreg:$Rn), asm,
5624 "{\t$Rd" # outkind # ", $Rn" # inkind #
5625 "|" # outkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
5632 let Inst{28-24} = 0b01110;
5633 let Inst{23-22} = size;
5634 let Inst{21-17} = 0b10000;
5635 let Inst{16-12} = opcode;
5636 let Inst{11-10} = 0b10;
5641 multiclass SIMDMixedTwoVector<bit U, bits<5> opc, string asm,
5642 SDPatternOperator OpNode> {
5643 def v8i8 : BaseSIMDMixedTwoVector<0, U, 0b00, opc, V128, V64,
5645 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5646 def v16i8 : BaseSIMDMixedTwoVectorTied<1, U, 0b00, opc, V128, V128,
5647 asm#"2", ".16b", ".8h", []>;
5648 def v4i16 : BaseSIMDMixedTwoVector<0, U, 0b01, opc, V128, V64,
5650 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5651 def v8i16 : BaseSIMDMixedTwoVectorTied<1, U, 0b01, opc, V128, V128,
5652 asm#"2", ".8h", ".4s", []>;
5653 def v2i32 : BaseSIMDMixedTwoVector<0, U, 0b10, opc, V128, V64,
5655 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5656 def v4i32 : BaseSIMDMixedTwoVectorTied<1, U, 0b10, opc, V128, V128,
5657 asm#"2", ".4s", ".2d", []>;
5659 def : Pat<(concat_vectors (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn))),
5660 (!cast<Instruction>(NAME # "v16i8")
5661 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5662 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn))),
5663 (!cast<Instruction>(NAME # "v8i16")
5664 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5665 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn))),
5666 (!cast<Instruction>(NAME # "v4i32")
5667 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5670 class BaseSIMDCmpTwoVector<bit Q, bit U, bits<2> size, bits<2> size2,
5671 bits<5> opcode, RegisterOperand regtype, string asm,
5672 string kind, string zero, ValueType dty,
5673 ValueType sty, SDNode OpNode>
5674 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5675 "{\t$Rd" # kind # ", $Rn" # kind # ", #" # zero #
5676 "|" # kind # "\t$Rd, $Rn, #" # zero # "}", "",
5677 [(set (dty regtype:$Rd), (OpNode (sty regtype:$Rn)))]>,
5684 let Inst{28-24} = 0b01110;
5685 let Inst{23-22} = size;
5687 let Inst{20-19} = size2;
5688 let Inst{18-17} = 0b00;
5689 let Inst{16-12} = opcode;
5690 let Inst{11-10} = 0b10;
5695 // Comparisons support all element sizes, except 1xD.
5696 multiclass SIMDCmpTwoVector<bit U, bits<5> opc, string asm,
5698 def v8i8rz : BaseSIMDCmpTwoVector<0, U, 0b00, 0b00, opc, V64,
5700 v8i8, v8i8, OpNode>;
5701 def v16i8rz : BaseSIMDCmpTwoVector<1, U, 0b00, 0b00, opc, V128,
5703 v16i8, v16i8, OpNode>;
5704 def v4i16rz : BaseSIMDCmpTwoVector<0, U, 0b01, 0b00, opc, V64,
5706 v4i16, v4i16, OpNode>;
5707 def v8i16rz : BaseSIMDCmpTwoVector<1, U, 0b01, 0b00, opc, V128,
5709 v8i16, v8i16, OpNode>;
5710 def v2i32rz : BaseSIMDCmpTwoVector<0, U, 0b10, 0b00, opc, V64,
5712 v2i32, v2i32, OpNode>;
5713 def v4i32rz : BaseSIMDCmpTwoVector<1, U, 0b10, 0b00, opc, V128,
5715 v4i32, v4i32, OpNode>;
5716 def v2i64rz : BaseSIMDCmpTwoVector<1, U, 0b11, 0b00, opc, V128,
5718 v2i64, v2i64, OpNode>;
5721 // FP Comparisons support only S and D element sizes (and H for v8.2a).
5722 multiclass SIMDFPCmpTwoVector<bit U, bit S, bits<5> opc,
5723 string asm, SDNode OpNode> {
5725 let Predicates = [HasNEON, HasFullFP16] in {
5726 def v4i16rz : BaseSIMDCmpTwoVector<0, U, {S,1}, 0b11, opc, V64,
5728 v4i16, v4f16, OpNode>;
5729 def v8i16rz : BaseSIMDCmpTwoVector<1, U, {S,1}, 0b11, opc, V128,
5731 v8i16, v8f16, OpNode>;
5732 } // Predicates = [HasNEON, HasFullFP16]
5733 def v2i32rz : BaseSIMDCmpTwoVector<0, U, {S,0}, 0b00, opc, V64,
5735 v2i32, v2f32, OpNode>;
5736 def v4i32rz : BaseSIMDCmpTwoVector<1, U, {S,0}, 0b00, opc, V128,
5738 v4i32, v4f32, OpNode>;
5739 def v2i64rz : BaseSIMDCmpTwoVector<1, U, {S,1}, 0b00, opc, V128,
5741 v2i64, v2f64, OpNode>;
5743 let Predicates = [HasNEON, HasFullFP16] in {
5744 def : InstAlias<asm # "\t$Vd.4h, $Vn.4h, #0",
5745 (!cast<Instruction>(NAME # v4i16rz) V64:$Vd, V64:$Vn), 0>;
5746 def : InstAlias<asm # "\t$Vd.8h, $Vn.8h, #0",
5747 (!cast<Instruction>(NAME # v8i16rz) V128:$Vd, V128:$Vn), 0>;
5749 def : InstAlias<asm # "\t$Vd.2s, $Vn.2s, #0",
5750 (!cast<Instruction>(NAME # v2i32rz) V64:$Vd, V64:$Vn), 0>;
5751 def : InstAlias<asm # "\t$Vd.4s, $Vn.4s, #0",
5752 (!cast<Instruction>(NAME # v4i32rz) V128:$Vd, V128:$Vn), 0>;
5753 def : InstAlias<asm # "\t$Vd.2d, $Vn.2d, #0",
5754 (!cast<Instruction>(NAME # v2i64rz) V128:$Vd, V128:$Vn), 0>;
5755 let Predicates = [HasNEON, HasFullFP16] in {
5756 def : InstAlias<asm # ".4h\t$Vd, $Vn, #0",
5757 (!cast<Instruction>(NAME # v4i16rz) V64:$Vd, V64:$Vn), 0>;
5758 def : InstAlias<asm # ".8h\t$Vd, $Vn, #0",
5759 (!cast<Instruction>(NAME # v8i16rz) V128:$Vd, V128:$Vn), 0>;
5761 def : InstAlias<asm # ".2s\t$Vd, $Vn, #0",
5762 (!cast<Instruction>(NAME # v2i32rz) V64:$Vd, V64:$Vn), 0>;
5763 def : InstAlias<asm # ".4s\t$Vd, $Vn, #0",
5764 (!cast<Instruction>(NAME # v4i32rz) V128:$Vd, V128:$Vn), 0>;
5765 def : InstAlias<asm # ".2d\t$Vd, $Vn, #0",
5766 (!cast<Instruction>(NAME # v2i64rz) V128:$Vd, V128:$Vn), 0>;
5769 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5770 class BaseSIMDFPCvtTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5771 RegisterOperand outtype, RegisterOperand intype,
5772 string asm, string VdTy, string VnTy,
5774 : I<(outs outtype:$Rd), (ins intype:$Rn), asm,
5775 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "", pattern>,
5782 let Inst{28-24} = 0b01110;
5783 let Inst{23-22} = size;
5784 let Inst{21-17} = 0b10000;
5785 let Inst{16-12} = opcode;
5786 let Inst{11-10} = 0b10;
5791 class BaseSIMDFPCvtTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5792 RegisterOperand outtype, RegisterOperand intype,
5793 string asm, string VdTy, string VnTy,
5795 : I<(outs outtype:$dst), (ins outtype:$Rd, intype:$Rn), asm,
5796 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "$Rd = $dst", pattern>,
5803 let Inst{28-24} = 0b01110;
5804 let Inst{23-22} = size;
5805 let Inst{21-17} = 0b10000;
5806 let Inst{16-12} = opcode;
5807 let Inst{11-10} = 0b10;
5812 multiclass SIMDFPWidenTwoVector<bit U, bit S, bits<5> opc, string asm> {
5813 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V128, V64,
5814 asm, ".4s", ".4h", []>;
5815 def v8i16 : BaseSIMDFPCvtTwoVector<1, U, {S,0}, opc, V128, V128,
5816 asm#"2", ".4s", ".8h", []>;
5817 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V128, V64,
5818 asm, ".2d", ".2s", []>;
5819 def v4i32 : BaseSIMDFPCvtTwoVector<1, U, {S,1}, opc, V128, V128,
5820 asm#"2", ".2d", ".4s", []>;
5823 multiclass SIMDFPNarrowTwoVector<bit U, bit S, bits<5> opc, string asm> {
5824 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V64, V128,
5825 asm, ".4h", ".4s", []>;
5826 def v8i16 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,0}, opc, V128, V128,
5827 asm#"2", ".8h", ".4s", []>;
5828 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
5829 asm, ".2s", ".2d", []>;
5830 def v4i32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
5831 asm#"2", ".4s", ".2d", []>;
5834 multiclass SIMDFPInexactCvtTwoVector<bit U, bit S, bits<5> opc, string asm,
5836 def v2f32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
5838 [(set (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5839 def v4f32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
5840 asm#"2", ".4s", ".2d", []>;
5842 def : Pat<(concat_vectors (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn))),
5843 (!cast<Instruction>(NAME # "v4f32")
5844 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5847 //----------------------------------------------------------------------------
5848 // AdvSIMD three register different-size vector instructions.
5849 //----------------------------------------------------------------------------
5851 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5852 class BaseSIMDDifferentThreeVector<bit U, bits<3> size, bits<4> opcode,
5853 RegisterOperand outtype, RegisterOperand intype1,
5854 RegisterOperand intype2, string asm,
5855 string outkind, string inkind1, string inkind2,
5857 : I<(outs outtype:$Rd), (ins intype1:$Rn, intype2:$Rm), asm,
5858 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
5859 "|" # outkind # "\t$Rd, $Rn, $Rm}", "", pattern>,
5865 let Inst{30} = size{0};
5867 let Inst{28-24} = 0b01110;
5868 let Inst{23-22} = size{2-1};
5870 let Inst{20-16} = Rm;
5871 let Inst{15-12} = opcode;
5872 let Inst{11-10} = 0b00;
5877 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5878 class BaseSIMDDifferentThreeVectorTied<bit U, bits<3> size, bits<4> opcode,
5879 RegisterOperand outtype, RegisterOperand intype1,
5880 RegisterOperand intype2, string asm,
5881 string outkind, string inkind1, string inkind2,
5883 : I<(outs outtype:$dst), (ins outtype:$Rd, intype1:$Rn, intype2:$Rm), asm,
5884 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
5885 "|" # outkind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
5891 let Inst{30} = size{0};
5893 let Inst{28-24} = 0b01110;
5894 let Inst{23-22} = size{2-1};
5896 let Inst{20-16} = Rm;
5897 let Inst{15-12} = opcode;
5898 let Inst{11-10} = 0b00;
5903 // FIXME: TableGen doesn't know how to deal with expanded types that also
5904 // change the element count (in this case, placing the results in
5905 // the high elements of the result register rather than the low
5906 // elements). Until that's fixed, we can't code-gen those.
5907 multiclass SIMDNarrowThreeVectorBHS<bit U, bits<4> opc, string asm,
5909 def v8i16_v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
5911 asm, ".8b", ".8h", ".8h",
5912 [(set (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
5913 def v8i16_v16i8 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
5915 asm#"2", ".16b", ".8h", ".8h",
5917 def v4i32_v4i16 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
5919 asm, ".4h", ".4s", ".4s",
5920 [(set (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
5921 def v4i32_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
5923 asm#"2", ".8h", ".4s", ".4s",
5925 def v2i64_v2i32 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
5927 asm, ".2s", ".2d", ".2d",
5928 [(set (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
5929 def v2i64_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
5931 asm#"2", ".4s", ".2d", ".2d",
5935 // Patterns for the '2' variants involve INSERT_SUBREG, which you can't put in
5936 // a version attached to an instruction.
5937 def : Pat<(concat_vectors (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn),
5939 (!cast<Instruction>(NAME # "v8i16_v16i8")
5940 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5941 V128:$Rn, V128:$Rm)>;
5942 def : Pat<(concat_vectors (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn),
5944 (!cast<Instruction>(NAME # "v4i32_v8i16")
5945 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5946 V128:$Rn, V128:$Rm)>;
5947 def : Pat<(concat_vectors (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn),
5949 (!cast<Instruction>(NAME # "v2i64_v4i32")
5950 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5951 V128:$Rn, V128:$Rm)>;
5954 multiclass SIMDDifferentThreeVectorBD<bit U, bits<4> opc, string asm,
5956 def v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
5958 asm, ".8h", ".8b", ".8b",
5959 [(set (v8i16 V128:$Rd), (IntOp (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5960 def v16i8 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
5962 asm#"2", ".8h", ".16b", ".16b", []>;
5963 let Predicates = [HasAES] in {
5964 def v1i64 : BaseSIMDDifferentThreeVector<U, 0b110, opc,
5966 asm, ".1q", ".1d", ".1d", []>;
5967 def v2i64 : BaseSIMDDifferentThreeVector<U, 0b111, opc,
5969 asm#"2", ".1q", ".2d", ".2d", []>;
5972 def : Pat<(v8i16 (IntOp (v8i8 (extract_high_v16i8 V128:$Rn)),
5973 (v8i8 (extract_high_v16i8 V128:$Rm)))),
5974 (!cast<Instruction>(NAME#"v16i8") V128:$Rn, V128:$Rm)>;
5977 multiclass SIMDLongThreeVectorHS<bit U, bits<4> opc, string asm,
5978 SDPatternOperator OpNode> {
5979 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
5981 asm, ".4s", ".4h", ".4h",
5982 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5983 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
5985 asm#"2", ".4s", ".8h", ".8h",
5986 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
5987 (extract_high_v8i16 V128:$Rm)))]>;
5988 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
5990 asm, ".2d", ".2s", ".2s",
5991 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5992 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
5994 asm#"2", ".2d", ".4s", ".4s",
5995 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
5996 (extract_high_v4i32 V128:$Rm)))]>;
5999 multiclass SIMDLongThreeVectorBHSabdl<bit U, bits<4> opc, string asm,
6000 SDPatternOperator OpNode = null_frag> {
6001 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6003 asm, ".8h", ".8b", ".8b",
6004 [(set (v8i16 V128:$Rd),
6005 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))))]>;
6006 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
6008 asm#"2", ".8h", ".16b", ".16b",
6009 [(set (v8i16 V128:$Rd),
6010 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
6011 (extract_high_v16i8 V128:$Rm)))))]>;
6012 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6014 asm, ".4s", ".4h", ".4h",
6015 [(set (v4i32 V128:$Rd),
6016 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))))]>;
6017 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
6019 asm#"2", ".4s", ".8h", ".8h",
6020 [(set (v4i32 V128:$Rd),
6021 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
6022 (extract_high_v8i16 V128:$Rm)))))]>;
6023 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6025 asm, ".2d", ".2s", ".2s",
6026 [(set (v2i64 V128:$Rd),
6027 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))))]>;
6028 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
6030 asm#"2", ".2d", ".4s", ".4s",
6031 [(set (v2i64 V128:$Rd),
6032 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
6033 (extract_high_v4i32 V128:$Rm)))))]>;
6036 multiclass SIMDLongThreeVectorTiedBHSabal<bit U, bits<4> opc,
6038 SDPatternOperator OpNode> {
6039 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
6041 asm, ".8h", ".8b", ".8b",
6042 [(set (v8i16 V128:$dst),
6043 (add (v8i16 V128:$Rd),
6044 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))))]>;
6045 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
6047 asm#"2", ".8h", ".16b", ".16b",
6048 [(set (v8i16 V128:$dst),
6049 (add (v8i16 V128:$Rd),
6050 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
6051 (extract_high_v16i8 V128:$Rm))))))]>;
6052 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
6054 asm, ".4s", ".4h", ".4h",
6055 [(set (v4i32 V128:$dst),
6056 (add (v4i32 V128:$Rd),
6057 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))))]>;
6058 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
6060 asm#"2", ".4s", ".8h", ".8h",
6061 [(set (v4i32 V128:$dst),
6062 (add (v4i32 V128:$Rd),
6063 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
6064 (extract_high_v8i16 V128:$Rm))))))]>;
6065 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
6067 asm, ".2d", ".2s", ".2s",
6068 [(set (v2i64 V128:$dst),
6069 (add (v2i64 V128:$Rd),
6070 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))))]>;
6071 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
6073 asm#"2", ".2d", ".4s", ".4s",
6074 [(set (v2i64 V128:$dst),
6075 (add (v2i64 V128:$Rd),
6076 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
6077 (extract_high_v4i32 V128:$Rm))))))]>;
6080 multiclass SIMDLongThreeVectorBHS<bit U, bits<4> opc, string asm,
6081 SDPatternOperator OpNode = null_frag> {
6082 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6084 asm, ".8h", ".8b", ".8b",
6085 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
6086 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
6088 asm#"2", ".8h", ".16b", ".16b",
6089 [(set (v8i16 V128:$Rd), (OpNode (extract_high_v16i8 V128:$Rn),
6090 (extract_high_v16i8 V128:$Rm)))]>;
6091 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6093 asm, ".4s", ".4h", ".4h",
6094 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
6095 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
6097 asm#"2", ".4s", ".8h", ".8h",
6098 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
6099 (extract_high_v8i16 V128:$Rm)))]>;
6100 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6102 asm, ".2d", ".2s", ".2s",
6103 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
6104 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
6106 asm#"2", ".2d", ".4s", ".4s",
6107 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
6108 (extract_high_v4i32 V128:$Rm)))]>;
6111 multiclass SIMDLongThreeVectorTiedBHS<bit U, bits<4> opc,
6113 SDPatternOperator OpNode> {
6114 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
6116 asm, ".8h", ".8b", ".8b",
6117 [(set (v8i16 V128:$dst),
6118 (OpNode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
6119 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
6121 asm#"2", ".8h", ".16b", ".16b",
6122 [(set (v8i16 V128:$dst),
6123 (OpNode (v8i16 V128:$Rd),
6124 (extract_high_v16i8 V128:$Rn),
6125 (extract_high_v16i8 V128:$Rm)))]>;
6126 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
6128 asm, ".4s", ".4h", ".4h",
6129 [(set (v4i32 V128:$dst),
6130 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
6131 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
6133 asm#"2", ".4s", ".8h", ".8h",
6134 [(set (v4i32 V128:$dst),
6135 (OpNode (v4i32 V128:$Rd),
6136 (extract_high_v8i16 V128:$Rn),
6137 (extract_high_v8i16 V128:$Rm)))]>;
6138 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
6140 asm, ".2d", ".2s", ".2s",
6141 [(set (v2i64 V128:$dst),
6142 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
6143 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
6145 asm#"2", ".2d", ".4s", ".4s",
6146 [(set (v2i64 V128:$dst),
6147 (OpNode (v2i64 V128:$Rd),
6148 (extract_high_v4i32 V128:$Rn),
6149 (extract_high_v4i32 V128:$Rm)))]>;
6152 multiclass SIMDLongThreeVectorSQDMLXTiedHS<bit U, bits<4> opc, string asm,
6153 SDPatternOperator Accum> {
6154 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
6156 asm, ".4s", ".4h", ".4h",
6157 [(set (v4i32 V128:$dst),
6158 (Accum (v4i32 V128:$Rd),
6159 (v4i32 (int_aarch64_neon_sqdmull (v4i16 V64:$Rn),
6160 (v4i16 V64:$Rm)))))]>;
6161 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
6163 asm#"2", ".4s", ".8h", ".8h",
6164 [(set (v4i32 V128:$dst),
6165 (Accum (v4i32 V128:$Rd),
6166 (v4i32 (int_aarch64_neon_sqdmull (extract_high_v8i16 V128:$Rn),
6167 (extract_high_v8i16 V128:$Rm)))))]>;
6168 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
6170 asm, ".2d", ".2s", ".2s",
6171 [(set (v2i64 V128:$dst),
6172 (Accum (v2i64 V128:$Rd),
6173 (v2i64 (int_aarch64_neon_sqdmull (v2i32 V64:$Rn),
6174 (v2i32 V64:$Rm)))))]>;
6175 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
6177 asm#"2", ".2d", ".4s", ".4s",
6178 [(set (v2i64 V128:$dst),
6179 (Accum (v2i64 V128:$Rd),
6180 (v2i64 (int_aarch64_neon_sqdmull (extract_high_v4i32 V128:$Rn),
6181 (extract_high_v4i32 V128:$Rm)))))]>;
6184 multiclass SIMDWideThreeVectorBHS<bit U, bits<4> opc, string asm,
6185 SDPatternOperator OpNode> {
6186 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
6188 asm, ".8h", ".8h", ".8b",
6189 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i8 V64:$Rm)))]>;
6190 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
6192 asm#"2", ".8h", ".8h", ".16b",
6193 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
6194 (extract_high_v16i8 V128:$Rm)))]>;
6195 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
6197 asm, ".4s", ".4s", ".4h",
6198 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i16 V64:$Rm)))]>;
6199 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
6201 asm#"2", ".4s", ".4s", ".8h",
6202 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
6203 (extract_high_v8i16 V128:$Rm)))]>;
6204 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
6206 asm, ".2d", ".2d", ".2s",
6207 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i32 V64:$Rm)))]>;
6208 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
6210 asm#"2", ".2d", ".2d", ".4s",
6211 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
6212 (extract_high_v4i32 V128:$Rm)))]>;
6215 //----------------------------------------------------------------------------
6216 // AdvSIMD bitwise extract from vector
6217 //----------------------------------------------------------------------------
6219 class BaseSIMDBitwiseExtract<bit size, RegisterOperand regtype, ValueType vty,
6220 string asm, string kind>
6221 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, i32imm:$imm), asm,
6222 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $imm" #
6223 "|" # kind # "\t$Rd, $Rn, $Rm, $imm}", "",
6224 [(set (vty regtype:$Rd),
6225 (AArch64ext regtype:$Rn, regtype:$Rm, (i32 imm:$imm)))]>,
6232 let Inst{30} = size;
6233 let Inst{29-21} = 0b101110000;
6234 let Inst{20-16} = Rm;
6236 let Inst{14-11} = imm;
6243 multiclass SIMDBitwiseExtract<string asm> {
6244 def v8i8 : BaseSIMDBitwiseExtract<0, V64, v8i8, asm, ".8b"> {
6247 def v16i8 : BaseSIMDBitwiseExtract<1, V128, v16i8, asm, ".16b">;
6250 //----------------------------------------------------------------------------
6251 // AdvSIMD zip vector
6252 //----------------------------------------------------------------------------
6254 class BaseSIMDZipVector<bits<3> size, bits<3> opc, RegisterOperand regtype,
6255 string asm, string kind, SDNode OpNode, ValueType valty>
6256 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
6257 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
6258 "|" # kind # "\t$Rd, $Rn, $Rm}", "",
6259 [(set (valty regtype:$Rd), (OpNode regtype:$Rn, regtype:$Rm))]>,
6265 let Inst{30} = size{0};
6266 let Inst{29-24} = 0b001110;
6267 let Inst{23-22} = size{2-1};
6269 let Inst{20-16} = Rm;
6271 let Inst{14-12} = opc;
6272 let Inst{11-10} = 0b10;
6277 multiclass SIMDZipVector<bits<3>opc, string asm,
6279 def v8i8 : BaseSIMDZipVector<0b000, opc, V64,
6280 asm, ".8b", OpNode, v8i8>;
6281 def v16i8 : BaseSIMDZipVector<0b001, opc, V128,
6282 asm, ".16b", OpNode, v16i8>;
6283 def v4i16 : BaseSIMDZipVector<0b010, opc, V64,
6284 asm, ".4h", OpNode, v4i16>;
6285 def v8i16 : BaseSIMDZipVector<0b011, opc, V128,
6286 asm, ".8h", OpNode, v8i16>;
6287 def v2i32 : BaseSIMDZipVector<0b100, opc, V64,
6288 asm, ".2s", OpNode, v2i32>;
6289 def v4i32 : BaseSIMDZipVector<0b101, opc, V128,
6290 asm, ".4s", OpNode, v4i32>;
6291 def v2i64 : BaseSIMDZipVector<0b111, opc, V128,
6292 asm, ".2d", OpNode, v2i64>;
6294 def : Pat<(v4f16 (OpNode V64:$Rn, V64:$Rm)),
6295 (!cast<Instruction>(NAME#"v4i16") V64:$Rn, V64:$Rm)>;
6296 def : Pat<(v8f16 (OpNode V128:$Rn, V128:$Rm)),
6297 (!cast<Instruction>(NAME#"v8i16") V128:$Rn, V128:$Rm)>;
6298 def : Pat<(v2f32 (OpNode V64:$Rn, V64:$Rm)),
6299 (!cast<Instruction>(NAME#"v2i32") V64:$Rn, V64:$Rm)>;
6300 def : Pat<(v4f32 (OpNode V128:$Rn, V128:$Rm)),
6301 (!cast<Instruction>(NAME#"v4i32") V128:$Rn, V128:$Rm)>;
6302 def : Pat<(v2f64 (OpNode V128:$Rn, V128:$Rm)),
6303 (!cast<Instruction>(NAME#"v2i64") V128:$Rn, V128:$Rm)>;
6306 //----------------------------------------------------------------------------
6307 // AdvSIMD three register scalar instructions
6308 //----------------------------------------------------------------------------
6310 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6311 class BaseSIMDThreeScalar<bit U, bits<3> size, bits<5> opcode,
6312 RegisterClass regtype, string asm,
6314 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
6315 "\t$Rd, $Rn, $Rm", "", pattern>,
6320 let Inst{31-30} = 0b01;
6322 let Inst{28-24} = 0b11110;
6323 let Inst{23-21} = size;
6324 let Inst{20-16} = Rm;
6325 let Inst{15-11} = opcode;
6331 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6332 class BaseSIMDThreeScalarTied<bit U, bits<2> size, bit R, bits<5> opcode,
6333 dag oops, dag iops, string asm,
6335 : I<oops, iops, asm, "\t$Rd, $Rn, $Rm", "$Rd = $dst", pattern>,
6340 let Inst{31-30} = 0b01;
6342 let Inst{28-24} = 0b11110;
6343 let Inst{23-22} = size;
6345 let Inst{20-16} = Rm;
6346 let Inst{15-11} = opcode;
6352 multiclass SIMDThreeScalarD<bit U, bits<5> opc, string asm,
6353 SDPatternOperator OpNode> {
6354 def v1i64 : BaseSIMDThreeScalar<U, 0b111, opc, FPR64, asm,
6355 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
6358 multiclass SIMDThreeScalarBHSD<bit U, bits<5> opc, string asm,
6359 SDPatternOperator OpNode> {
6360 def v1i64 : BaseSIMDThreeScalar<U, 0b111, opc, FPR64, asm,
6361 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
6362 def v1i32 : BaseSIMDThreeScalar<U, 0b101, opc, FPR32, asm, []>;
6363 def v1i16 : BaseSIMDThreeScalar<U, 0b011, opc, FPR16, asm, []>;
6364 def v1i8 : BaseSIMDThreeScalar<U, 0b001, opc, FPR8 , asm, []>;
6366 def : Pat<(i64 (OpNode (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6367 (!cast<Instruction>(NAME#"v1i64") FPR64:$Rn, FPR64:$Rm)>;
6368 def : Pat<(i32 (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm))),
6369 (!cast<Instruction>(NAME#"v1i32") FPR32:$Rn, FPR32:$Rm)>;
6372 multiclass SIMDThreeScalarHS<bit U, bits<5> opc, string asm,
6373 SDPatternOperator OpNode> {
6374 def v1i32 : BaseSIMDThreeScalar<U, 0b101, opc, FPR32, asm,
6375 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
6376 def v1i16 : BaseSIMDThreeScalar<U, 0b011, opc, FPR16, asm, []>;
6379 multiclass SIMDThreeScalarHSTied<bit U, bit R, bits<5> opc, string asm,
6380 SDPatternOperator OpNode = null_frag> {
6381 def v1i32: BaseSIMDThreeScalarTied<U, 0b10, R, opc, (outs FPR32:$dst),
6382 (ins FPR32:$Rd, FPR32:$Rn, FPR32:$Rm),
6384 def v1i16: BaseSIMDThreeScalarTied<U, 0b01, R, opc, (outs FPR16:$dst),
6385 (ins FPR16:$Rd, FPR16:$Rn, FPR16:$Rm),
6389 multiclass SIMDFPThreeScalar<bit U, bit S, bits<3> opc, string asm,
6390 SDPatternOperator OpNode = null_frag> {
6391 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6392 def #NAME#64 : BaseSIMDThreeScalar<U, {S,0b11}, {0b11,opc}, FPR64, asm,
6393 [(set (f64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
6394 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0b01}, {0b11,opc}, FPR32, asm,
6395 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
6396 let Predicates = [HasNEON, HasFullFP16] in {
6397 def #NAME#16 : BaseSIMDThreeScalar<U, {S,0b10}, {0b00,opc}, FPR16, asm,
6398 [(set FPR16:$Rd, (OpNode FPR16:$Rn, FPR16:$Rm))]>;
6399 } // Predicates = [HasNEON, HasFullFP16]
6402 def : Pat<(v1f64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
6403 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
6406 multiclass SIMDThreeScalarFPCmp<bit U, bit S, bits<3> opc, string asm,
6407 SDPatternOperator OpNode = null_frag> {
6408 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6409 def #NAME#64 : BaseSIMDThreeScalar<U, {S,0b11}, {0b11,opc}, FPR64, asm,
6410 [(set (i64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
6411 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0b01}, {0b11,opc}, FPR32, asm,
6412 [(set (i32 FPR32:$Rd), (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]>;
6413 let Predicates = [HasNEON, HasFullFP16] in {
6414 def #NAME#16 : BaseSIMDThreeScalar<U, {S,0b10}, {0b00,opc}, FPR16, asm,
6416 } // Predicates = [HasNEON, HasFullFP16]
6419 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
6420 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
6423 class BaseSIMDThreeScalarMixed<bit U, bits<2> size, bits<5> opcode,
6424 dag oops, dag iops, string asm, string cstr, list<dag> pat>
6425 : I<oops, iops, asm,
6426 "\t$Rd, $Rn, $Rm", cstr, pat>,
6431 let Inst{31-30} = 0b01;
6433 let Inst{28-24} = 0b11110;
6434 let Inst{23-22} = size;
6436 let Inst{20-16} = Rm;
6437 let Inst{15-11} = opcode;
6443 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6444 multiclass SIMDThreeScalarMixedHS<bit U, bits<5> opc, string asm,
6445 SDPatternOperator OpNode = null_frag> {
6446 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
6448 (ins FPR16:$Rn, FPR16:$Rm), asm, "", []>;
6449 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
6451 (ins FPR32:$Rn, FPR32:$Rm), asm, "",
6452 [(set (i64 FPR64:$Rd), (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
6455 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6456 multiclass SIMDThreeScalarMixedTiedHS<bit U, bits<5> opc, string asm,
6457 SDPatternOperator OpNode = null_frag> {
6458 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
6460 (ins FPR32:$Rd, FPR16:$Rn, FPR16:$Rm),
6461 asm, "$Rd = $dst", []>;
6462 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
6464 (ins FPR64:$Rd, FPR32:$Rn, FPR32:$Rm),
6466 [(set (i64 FPR64:$dst),
6467 (OpNode (i64 FPR64:$Rd), (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
6470 //----------------------------------------------------------------------------
6471 // AdvSIMD two register scalar instructions
6472 //----------------------------------------------------------------------------
6474 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6475 class BaseSIMDTwoScalar<bit U, bits<2> size, bits<2> size2, bits<5> opcode,
6476 RegisterClass regtype, RegisterClass regtype2,
6477 string asm, list<dag> pat>
6478 : I<(outs regtype:$Rd), (ins regtype2:$Rn), asm,
6479 "\t$Rd, $Rn", "", pat>,
6483 let Inst{31-30} = 0b01;
6485 let Inst{28-24} = 0b11110;
6486 let Inst{23-22} = size;
6488 let Inst{20-19} = size2;
6489 let Inst{18-17} = 0b00;
6490 let Inst{16-12} = opcode;
6491 let Inst{11-10} = 0b10;
6496 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6497 class BaseSIMDTwoScalarTied<bit U, bits<2> size, bits<5> opcode,
6498 RegisterClass regtype, RegisterClass regtype2,
6499 string asm, list<dag> pat>
6500 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype2:$Rn), asm,
6501 "\t$Rd, $Rn", "$Rd = $dst", pat>,
6505 let Inst{31-30} = 0b01;
6507 let Inst{28-24} = 0b11110;
6508 let Inst{23-22} = size;
6509 let Inst{21-17} = 0b10000;
6510 let Inst{16-12} = opcode;
6511 let Inst{11-10} = 0b10;
6517 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6518 class BaseSIMDCmpTwoScalar<bit U, bits<2> size, bits<2> size2, bits<5> opcode,
6519 RegisterClass regtype, string asm, string zero>
6520 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
6521 "\t$Rd, $Rn, #" # zero, "", []>,
6525 let Inst{31-30} = 0b01;
6527 let Inst{28-24} = 0b11110;
6528 let Inst{23-22} = size;
6530 let Inst{20-19} = size2;
6531 let Inst{18-17} = 0b00;
6532 let Inst{16-12} = opcode;
6533 let Inst{11-10} = 0b10;
6538 class SIMDInexactCvtTwoScalar<bits<5> opcode, string asm>
6539 : I<(outs FPR32:$Rd), (ins FPR64:$Rn), asm, "\t$Rd, $Rn", "",
6540 [(set (f32 FPR32:$Rd), (int_aarch64_sisd_fcvtxn (f64 FPR64:$Rn)))]>,
6544 let Inst{31-17} = 0b011111100110000;
6545 let Inst{16-12} = opcode;
6546 let Inst{11-10} = 0b10;
6551 multiclass SIMDCmpTwoScalarD<bit U, bits<5> opc, string asm,
6552 SDPatternOperator OpNode> {
6553 def v1i64rz : BaseSIMDCmpTwoScalar<U, 0b11, 0b00, opc, FPR64, asm, "0">;
6555 def : Pat<(v1i64 (OpNode FPR64:$Rn)),
6556 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
6559 multiclass SIMDFPCmpTwoScalar<bit U, bit S, bits<5> opc, string asm,
6560 SDPatternOperator OpNode> {
6561 def v1i64rz : BaseSIMDCmpTwoScalar<U, {S,1}, 0b00, opc, FPR64, asm, "0.0">;
6562 def v1i32rz : BaseSIMDCmpTwoScalar<U, {S,0}, 0b00, opc, FPR32, asm, "0.0">;
6563 let Predicates = [HasNEON, HasFullFP16] in {
6564 def v1i16rz : BaseSIMDCmpTwoScalar<U, {S,1}, 0b11, opc, FPR16, asm, "0.0">;
6567 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6568 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rd, FPR64:$Rn), 0>;
6569 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6570 (!cast<Instruction>(NAME # v1i32rz) FPR32:$Rd, FPR32:$Rn), 0>;
6571 let Predicates = [HasNEON, HasFullFP16] in {
6572 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6573 (!cast<Instruction>(NAME # v1i16rz) FPR16:$Rd, FPR16:$Rn), 0>;
6576 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn))),
6577 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
6580 multiclass SIMDTwoScalarD<bit U, bits<5> opc, string asm,
6581 SDPatternOperator OpNode = null_frag> {
6582 def v1i64 : BaseSIMDTwoScalar<U, 0b11, 0b00, opc, FPR64, FPR64, asm,
6583 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn)))]>;
6585 def : Pat<(i64 (OpNode (i64 FPR64:$Rn))),
6586 (!cast<Instruction>(NAME # "v1i64") FPR64:$Rn)>;
6589 multiclass SIMDFPTwoScalar<bit U, bit S, bits<5> opc, string asm> {
6590 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, 0b00, opc, FPR64, FPR64, asm,[]>;
6591 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, 0b00, opc, FPR32, FPR32, asm,[]>;
6592 let Predicates = [HasNEON, HasFullFP16] in {
6593 def v1f16 : BaseSIMDTwoScalar<U, {S,1}, 0b11, opc, FPR16, FPR16, asm,[]>;
6597 multiclass SIMDFPTwoScalarCVT<bit U, bit S, bits<5> opc, string asm,
6598 SDPatternOperator OpNode> {
6599 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, 0b00, opc, FPR64, FPR64, asm,
6600 [(set FPR64:$Rd, (OpNode (f64 FPR64:$Rn)))]>;
6601 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, 0b00, opc, FPR32, FPR32, asm,
6602 [(set FPR32:$Rd, (OpNode (f32 FPR32:$Rn)))]>;
6603 let Predicates = [HasNEON, HasFullFP16] in {
6604 def v1i16 : BaseSIMDTwoScalar<U, {S,1}, 0b11, opc, FPR16, FPR16, asm,
6605 [(set FPR16:$Rd, (OpNode (f16 FPR16:$Rn)))]>;
6609 multiclass SIMDTwoScalarBHSD<bit U, bits<5> opc, string asm,
6610 SDPatternOperator OpNode = null_frag> {
6611 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6612 def v1i64 : BaseSIMDTwoScalar<U, 0b11, 0b00, opc, FPR64, FPR64, asm,
6613 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
6614 def v1i32 : BaseSIMDTwoScalar<U, 0b10, 0b00, opc, FPR32, FPR32, asm,
6615 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
6616 def v1i16 : BaseSIMDTwoScalar<U, 0b01, 0b00, opc, FPR16, FPR16, asm, []>;
6617 def v1i8 : BaseSIMDTwoScalar<U, 0b00, 0b00, opc, FPR8 , FPR8 , asm, []>;
6620 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn))),
6621 (!cast<Instruction>(NAME # v1i64) FPR64:$Rn)>;
6624 multiclass SIMDTwoScalarBHSDTied<bit U, bits<5> opc, string asm,
6626 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6627 def v1i64 : BaseSIMDTwoScalarTied<U, 0b11, opc, FPR64, FPR64, asm,
6628 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn)))]>;
6629 def v1i32 : BaseSIMDTwoScalarTied<U, 0b10, opc, FPR32, FPR32, asm,
6630 [(set (i32 FPR32:$dst), (OpNode (i32 FPR32:$Rd), (i32 FPR32:$Rn)))]>;
6631 def v1i16 : BaseSIMDTwoScalarTied<U, 0b01, opc, FPR16, FPR16, asm, []>;
6632 def v1i8 : BaseSIMDTwoScalarTied<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
6635 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn))),
6636 (!cast<Instruction>(NAME # v1i64) FPR64:$Rd, FPR64:$Rn)>;
6641 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6642 multiclass SIMDTwoScalarMixedBHS<bit U, bits<5> opc, string asm,
6643 SDPatternOperator OpNode = null_frag> {
6644 def v1i32 : BaseSIMDTwoScalar<U, 0b10, 0b00, opc, FPR32, FPR64, asm,
6645 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
6646 def v1i16 : BaseSIMDTwoScalar<U, 0b01, 0b00, opc, FPR16, FPR32, asm, []>;
6647 def v1i8 : BaseSIMDTwoScalar<U, 0b00, 0b00, opc, FPR8 , FPR16, asm, []>;
6650 //----------------------------------------------------------------------------
6651 // AdvSIMD scalar pairwise instructions
6652 //----------------------------------------------------------------------------
6654 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6655 class BaseSIMDPairwiseScalar<bit U, bits<2> size, bits<5> opcode,
6656 RegisterOperand regtype, RegisterOperand vectype,
6657 string asm, string kind>
6658 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
6659 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", []>,
6663 let Inst{31-30} = 0b01;
6665 let Inst{28-24} = 0b11110;
6666 let Inst{23-22} = size;
6667 let Inst{21-17} = 0b11000;
6668 let Inst{16-12} = opcode;
6669 let Inst{11-10} = 0b10;
6674 multiclass SIMDPairwiseScalarD<bit U, bits<5> opc, string asm> {
6675 def v2i64p : BaseSIMDPairwiseScalar<U, 0b11, opc, FPR64Op, V128,
6679 multiclass SIMDFPPairwiseScalar<bit S, bits<5> opc, string asm> {
6680 let Predicates = [HasNEON, HasFullFP16] in {
6681 def v2i16p : BaseSIMDPairwiseScalar<0, {S,0}, opc, FPR16Op, V64,
6684 def v2i32p : BaseSIMDPairwiseScalar<1, {S,0}, opc, FPR32Op, V64,
6686 def v2i64p : BaseSIMDPairwiseScalar<1, {S,1}, opc, FPR64Op, V128,
6690 //----------------------------------------------------------------------------
6691 // AdvSIMD across lanes instructions
6692 //----------------------------------------------------------------------------
6694 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6695 class BaseSIMDAcrossLanes<bit Q, bit U, bits<2> size, bits<5> opcode,
6696 RegisterClass regtype, RegisterOperand vectype,
6697 string asm, string kind, list<dag> pattern>
6698 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
6699 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", pattern>,
6706 let Inst{28-24} = 0b01110;
6707 let Inst{23-22} = size;
6708 let Inst{21-17} = 0b11000;
6709 let Inst{16-12} = opcode;
6710 let Inst{11-10} = 0b10;
6715 multiclass SIMDAcrossLanesBHS<bit U, bits<5> opcode,
6717 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR8, V64,
6719 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR8, V128,
6721 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR16, V64,
6723 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR16, V128,
6725 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR32, V128,
6729 multiclass SIMDAcrossLanesHSD<bit U, bits<5> opcode, string asm> {
6730 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR16, V64,
6732 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR16, V128,
6734 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR32, V64,
6736 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR32, V128,
6738 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR64, V128,
6742 multiclass SIMDFPAcrossLanes<bits<5> opcode, bit sz1, string asm,
6744 let Predicates = [HasNEON, HasFullFP16] in {
6745 def v4i16v : BaseSIMDAcrossLanes<0, 0, {sz1, 0}, opcode, FPR16, V64,
6747 [(set FPR16:$Rd, (intOp (v4f16 V64:$Rn)))]>;
6748 def v8i16v : BaseSIMDAcrossLanes<1, 0, {sz1, 0}, opcode, FPR16, V128,
6750 [(set FPR16:$Rd, (intOp (v8f16 V128:$Rn)))]>;
6751 } // Predicates = [HasNEON, HasFullFP16]
6752 def v4i32v : BaseSIMDAcrossLanes<1, 1, {sz1, 0}, opcode, FPR32, V128,
6754 [(set FPR32:$Rd, (intOp (v4f32 V128:$Rn)))]>;
6757 //----------------------------------------------------------------------------
6758 // AdvSIMD INS/DUP instructions
6759 //----------------------------------------------------------------------------
6761 // FIXME: There has got to be a better way to factor these. ugh.
6763 class BaseSIMDInsDup<bit Q, bit op, dag outs, dag ins, string asm,
6764 string operands, string constraints, list<dag> pattern>
6765 : I<outs, ins, asm, operands, constraints, pattern>,
6772 let Inst{28-21} = 0b01110000;
6779 class SIMDDupFromMain<bit Q, bits<5> imm5, string size, ValueType vectype,
6780 RegisterOperand vecreg, RegisterClass regtype>
6781 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins regtype:$Rn), "dup",
6782 "{\t$Rd" # size # ", $Rn" #
6783 "|" # size # "\t$Rd, $Rn}", "",
6784 [(set (vectype vecreg:$Rd), (AArch64dup regtype:$Rn))]> {
6785 let Inst{20-16} = imm5;
6786 let Inst{14-11} = 0b0001;
6789 class SIMDDupFromElement<bit Q, string dstkind, string srckind,
6790 ValueType vectype, ValueType insreg,
6791 RegisterOperand vecreg, Operand idxtype,
6792 ValueType elttype, SDNode OpNode>
6793 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins V128:$Rn, idxtype:$idx), "dup",
6794 "{\t$Rd" # dstkind # ", $Rn" # srckind # "$idx" #
6795 "|" # dstkind # "\t$Rd, $Rn$idx}", "",
6796 [(set (vectype vecreg:$Rd),
6797 (OpNode (insreg V128:$Rn), idxtype:$idx))]> {
6798 let Inst{14-11} = 0b0000;
6801 class SIMDDup64FromElement
6802 : SIMDDupFromElement<1, ".2d", ".d", v2i64, v2i64, V128,
6803 VectorIndexD, i64, AArch64duplane64> {
6806 let Inst{19-16} = 0b1000;
6809 class SIMDDup32FromElement<bit Q, string size, ValueType vectype,
6810 RegisterOperand vecreg>
6811 : SIMDDupFromElement<Q, size, ".s", vectype, v4i32, vecreg,
6812 VectorIndexS, i64, AArch64duplane32> {
6814 let Inst{20-19} = idx;
6815 let Inst{18-16} = 0b100;
6818 class SIMDDup16FromElement<bit Q, string size, ValueType vectype,
6819 RegisterOperand vecreg>
6820 : SIMDDupFromElement<Q, size, ".h", vectype, v8i16, vecreg,
6821 VectorIndexH, i64, AArch64duplane16> {
6823 let Inst{20-18} = idx;
6824 let Inst{17-16} = 0b10;
6827 class SIMDDup8FromElement<bit Q, string size, ValueType vectype,
6828 RegisterOperand vecreg>
6829 : SIMDDupFromElement<Q, size, ".b", vectype, v16i8, vecreg,
6830 VectorIndexB, i64, AArch64duplane8> {
6832 let Inst{20-17} = idx;
6836 class BaseSIMDMov<bit Q, string size, bits<4> imm4, RegisterClass regtype,
6837 Operand idxtype, string asm, list<dag> pattern>
6838 : BaseSIMDInsDup<Q, 0, (outs regtype:$Rd), (ins V128:$Rn, idxtype:$idx), asm,
6839 "{\t$Rd, $Rn" # size # "$idx" #
6840 "|" # size # "\t$Rd, $Rn$idx}", "", pattern> {
6841 let Inst{14-11} = imm4;
6844 class SIMDSMov<bit Q, string size, RegisterClass regtype,
6846 : BaseSIMDMov<Q, size, 0b0101, regtype, idxtype, "smov", []>;
6847 class SIMDUMov<bit Q, string size, ValueType vectype, RegisterClass regtype,
6849 : BaseSIMDMov<Q, size, 0b0111, regtype, idxtype, "umov",
6850 [(set regtype:$Rd, (vector_extract (vectype V128:$Rn), idxtype:$idx))]>;
6852 class SIMDMovAlias<string asm, string size, Instruction inst,
6853 RegisterClass regtype, Operand idxtype>
6854 : InstAlias<asm#"{\t$dst, $src"#size#"$idx" #
6855 "|" # size # "\t$dst, $src$idx}",
6856 (inst regtype:$dst, V128:$src, idxtype:$idx)>;
6859 def vi8to32 : SIMDSMov<0, ".b", GPR32, VectorIndexB> {
6861 let Inst{20-17} = idx;
6864 def vi8to64 : SIMDSMov<1, ".b", GPR64, VectorIndexB> {
6866 let Inst{20-17} = idx;
6869 def vi16to32 : SIMDSMov<0, ".h", GPR32, VectorIndexH> {
6871 let Inst{20-18} = idx;
6872 let Inst{17-16} = 0b10;
6874 def vi16to64 : SIMDSMov<1, ".h", GPR64, VectorIndexH> {
6876 let Inst{20-18} = idx;
6877 let Inst{17-16} = 0b10;
6879 def vi32to64 : SIMDSMov<1, ".s", GPR64, VectorIndexS> {
6881 let Inst{20-19} = idx;
6882 let Inst{18-16} = 0b100;
6887 def vi8 : SIMDUMov<0, ".b", v16i8, GPR32, VectorIndexB> {
6889 let Inst{20-17} = idx;
6892 def vi16 : SIMDUMov<0, ".h", v8i16, GPR32, VectorIndexH> {
6894 let Inst{20-18} = idx;
6895 let Inst{17-16} = 0b10;
6897 def vi32 : SIMDUMov<0, ".s", v4i32, GPR32, VectorIndexS> {
6899 let Inst{20-19} = idx;
6900 let Inst{18-16} = 0b100;
6902 def vi64 : SIMDUMov<1, ".d", v2i64, GPR64, VectorIndexD> {
6905 let Inst{19-16} = 0b1000;
6907 def : SIMDMovAlias<"mov", ".s",
6908 !cast<Instruction>(NAME#"vi32"),
6909 GPR32, VectorIndexS>;
6910 def : SIMDMovAlias<"mov", ".d",
6911 !cast<Instruction>(NAME#"vi64"),
6912 GPR64, VectorIndexD>;
6915 class SIMDInsFromMain<string size, ValueType vectype,
6916 RegisterClass regtype, Operand idxtype>
6917 : BaseSIMDInsDup<1, 0, (outs V128:$dst),
6918 (ins V128:$Rd, idxtype:$idx, regtype:$Rn), "ins",
6919 "{\t$Rd" # size # "$idx, $Rn" #
6920 "|" # size # "\t$Rd$idx, $Rn}",
6923 (vector_insert (vectype V128:$Rd), regtype:$Rn, idxtype:$idx))]> {
6924 let Inst{14-11} = 0b0011;
6927 class SIMDInsFromElement<string size, ValueType vectype,
6928 ValueType elttype, Operand idxtype>
6929 : BaseSIMDInsDup<1, 1, (outs V128:$dst),
6930 (ins V128:$Rd, idxtype:$idx, V128:$Rn, idxtype:$idx2), "ins",
6931 "{\t$Rd" # size # "$idx, $Rn" # size # "$idx2" #
6932 "|" # size # "\t$Rd$idx, $Rn$idx2}",
6937 (elttype (vector_extract (vectype V128:$Rn), idxtype:$idx2)),
6940 class SIMDInsMainMovAlias<string size, Instruction inst,
6941 RegisterClass regtype, Operand idxtype>
6942 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" #
6943 "|" # size #"\t$dst$idx, $src}",
6944 (inst V128:$dst, idxtype:$idx, regtype:$src)>;
6945 class SIMDInsElementMovAlias<string size, Instruction inst,
6947 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" # size # "$idx2" #
6948 # "|" # size #"\t$dst$idx, $src$idx2}",
6949 (inst V128:$dst, idxtype:$idx, V128:$src, idxtype:$idx2)>;
6952 multiclass SIMDIns {
6953 def vi8gpr : SIMDInsFromMain<".b", v16i8, GPR32, VectorIndexB> {
6955 let Inst{20-17} = idx;
6958 def vi16gpr : SIMDInsFromMain<".h", v8i16, GPR32, VectorIndexH> {
6960 let Inst{20-18} = idx;
6961 let Inst{17-16} = 0b10;
6963 def vi32gpr : SIMDInsFromMain<".s", v4i32, GPR32, VectorIndexS> {
6965 let Inst{20-19} = idx;
6966 let Inst{18-16} = 0b100;
6968 def vi64gpr : SIMDInsFromMain<".d", v2i64, GPR64, VectorIndexD> {
6971 let Inst{19-16} = 0b1000;
6974 def vi8lane : SIMDInsFromElement<".b", v16i8, i32, VectorIndexB> {
6977 let Inst{20-17} = idx;
6979 let Inst{14-11} = idx2;
6981 def vi16lane : SIMDInsFromElement<".h", v8i16, i32, VectorIndexH> {
6984 let Inst{20-18} = idx;
6985 let Inst{17-16} = 0b10;
6986 let Inst{14-12} = idx2;
6989 def vi32lane : SIMDInsFromElement<".s", v4i32, i32, VectorIndexS> {
6992 let Inst{20-19} = idx;
6993 let Inst{18-16} = 0b100;
6994 let Inst{14-13} = idx2;
6995 let Inst{12-11} = {?,?};
6997 def vi64lane : SIMDInsFromElement<".d", v2i64, i64, VectorIndexD> {
7001 let Inst{19-16} = 0b1000;
7002 let Inst{14} = idx2;
7003 let Inst{13-11} = {?,?,?};
7006 // For all forms of the INS instruction, the "mov" mnemonic is the
7007 // preferred alias. Why they didn't just call the instruction "mov" in
7008 // the first place is a very good question indeed...
7009 def : SIMDInsMainMovAlias<".b", !cast<Instruction>(NAME#"vi8gpr"),
7010 GPR32, VectorIndexB>;
7011 def : SIMDInsMainMovAlias<".h", !cast<Instruction>(NAME#"vi16gpr"),
7012 GPR32, VectorIndexH>;
7013 def : SIMDInsMainMovAlias<".s", !cast<Instruction>(NAME#"vi32gpr"),
7014 GPR32, VectorIndexS>;
7015 def : SIMDInsMainMovAlias<".d", !cast<Instruction>(NAME#"vi64gpr"),
7016 GPR64, VectorIndexD>;
7018 def : SIMDInsElementMovAlias<".b", !cast<Instruction>(NAME#"vi8lane"),
7020 def : SIMDInsElementMovAlias<".h", !cast<Instruction>(NAME#"vi16lane"),
7022 def : SIMDInsElementMovAlias<".s", !cast<Instruction>(NAME#"vi32lane"),
7024 def : SIMDInsElementMovAlias<".d", !cast<Instruction>(NAME#"vi64lane"),
7028 //----------------------------------------------------------------------------
7030 //----------------------------------------------------------------------------
7032 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
7033 class BaseSIMDTableLookup<bit Q, bits<2> len, bit op, RegisterOperand vectype,
7034 RegisterOperand listtype, string asm, string kind>
7035 : I<(outs vectype:$Vd), (ins listtype:$Vn, vectype:$Vm), asm,
7036 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "", []>,
7043 let Inst{29-21} = 0b001110000;
7044 let Inst{20-16} = Vm;
7046 let Inst{14-13} = len;
7048 let Inst{11-10} = 0b00;
7053 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
7054 class BaseSIMDTableLookupTied<bit Q, bits<2> len, bit op, RegisterOperand vectype,
7055 RegisterOperand listtype, string asm, string kind>
7056 : I<(outs vectype:$dst), (ins vectype:$Vd, listtype:$Vn, vectype:$Vm), asm,
7057 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "$Vd = $dst", []>,
7064 let Inst{29-21} = 0b001110000;
7065 let Inst{20-16} = Vm;
7067 let Inst{14-13} = len;
7069 let Inst{11-10} = 0b00;
7074 class SIMDTableLookupAlias<string asm, Instruction inst,
7075 RegisterOperand vectype, RegisterOperand listtype>
7076 : InstAlias<!strconcat(asm, "\t$dst, $lst, $index"),
7077 (inst vectype:$dst, listtype:$lst, vectype:$index), 0>;
7079 multiclass SIMDTableLookup<bit op, string asm> {
7080 def v8i8One : BaseSIMDTableLookup<0, 0b00, op, V64, VecListOne16b,
7082 def v8i8Two : BaseSIMDTableLookup<0, 0b01, op, V64, VecListTwo16b,
7084 def v8i8Three : BaseSIMDTableLookup<0, 0b10, op, V64, VecListThree16b,
7086 def v8i8Four : BaseSIMDTableLookup<0, 0b11, op, V64, VecListFour16b,
7088 def v16i8One : BaseSIMDTableLookup<1, 0b00, op, V128, VecListOne16b,
7090 def v16i8Two : BaseSIMDTableLookup<1, 0b01, op, V128, VecListTwo16b,
7092 def v16i8Three: BaseSIMDTableLookup<1, 0b10, op, V128, VecListThree16b,
7094 def v16i8Four : BaseSIMDTableLookup<1, 0b11, op, V128, VecListFour16b,
7097 def : SIMDTableLookupAlias<asm # ".8b",
7098 !cast<Instruction>(NAME#"v8i8One"),
7099 V64, VecListOne128>;
7100 def : SIMDTableLookupAlias<asm # ".8b",
7101 !cast<Instruction>(NAME#"v8i8Two"),
7102 V64, VecListTwo128>;
7103 def : SIMDTableLookupAlias<asm # ".8b",
7104 !cast<Instruction>(NAME#"v8i8Three"),
7105 V64, VecListThree128>;
7106 def : SIMDTableLookupAlias<asm # ".8b",
7107 !cast<Instruction>(NAME#"v8i8Four"),
7108 V64, VecListFour128>;
7109 def : SIMDTableLookupAlias<asm # ".16b",
7110 !cast<Instruction>(NAME#"v16i8One"),
7111 V128, VecListOne128>;
7112 def : SIMDTableLookupAlias<asm # ".16b",
7113 !cast<Instruction>(NAME#"v16i8Two"),
7114 V128, VecListTwo128>;
7115 def : SIMDTableLookupAlias<asm # ".16b",
7116 !cast<Instruction>(NAME#"v16i8Three"),
7117 V128, VecListThree128>;
7118 def : SIMDTableLookupAlias<asm # ".16b",
7119 !cast<Instruction>(NAME#"v16i8Four"),
7120 V128, VecListFour128>;
7123 multiclass SIMDTableLookupTied<bit op, string asm> {
7124 def v8i8One : BaseSIMDTableLookupTied<0, 0b00, op, V64, VecListOne16b,
7126 def v8i8Two : BaseSIMDTableLookupTied<0, 0b01, op, V64, VecListTwo16b,
7128 def v8i8Three : BaseSIMDTableLookupTied<0, 0b10, op, V64, VecListThree16b,
7130 def v8i8Four : BaseSIMDTableLookupTied<0, 0b11, op, V64, VecListFour16b,
7132 def v16i8One : BaseSIMDTableLookupTied<1, 0b00, op, V128, VecListOne16b,
7134 def v16i8Two : BaseSIMDTableLookupTied<1, 0b01, op, V128, VecListTwo16b,
7136 def v16i8Three: BaseSIMDTableLookupTied<1, 0b10, op, V128, VecListThree16b,
7138 def v16i8Four : BaseSIMDTableLookupTied<1, 0b11, op, V128, VecListFour16b,
7141 def : SIMDTableLookupAlias<asm # ".8b",
7142 !cast<Instruction>(NAME#"v8i8One"),
7143 V64, VecListOne128>;
7144 def : SIMDTableLookupAlias<asm # ".8b",
7145 !cast<Instruction>(NAME#"v8i8Two"),
7146 V64, VecListTwo128>;
7147 def : SIMDTableLookupAlias<asm # ".8b",
7148 !cast<Instruction>(NAME#"v8i8Three"),
7149 V64, VecListThree128>;
7150 def : SIMDTableLookupAlias<asm # ".8b",
7151 !cast<Instruction>(NAME#"v8i8Four"),
7152 V64, VecListFour128>;
7153 def : SIMDTableLookupAlias<asm # ".16b",
7154 !cast<Instruction>(NAME#"v16i8One"),
7155 V128, VecListOne128>;
7156 def : SIMDTableLookupAlias<asm # ".16b",
7157 !cast<Instruction>(NAME#"v16i8Two"),
7158 V128, VecListTwo128>;
7159 def : SIMDTableLookupAlias<asm # ".16b",
7160 !cast<Instruction>(NAME#"v16i8Three"),
7161 V128, VecListThree128>;
7162 def : SIMDTableLookupAlias<asm # ".16b",
7163 !cast<Instruction>(NAME#"v16i8Four"),
7164 V128, VecListFour128>;
7168 //----------------------------------------------------------------------------
7169 // AdvSIMD scalar CPY
7170 //----------------------------------------------------------------------------
7171 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7172 class BaseSIMDScalarCPY<RegisterClass regtype, RegisterOperand vectype,
7173 string kind, Operand idxtype>
7174 : I<(outs regtype:$dst), (ins vectype:$src, idxtype:$idx), "mov",
7175 "{\t$dst, $src" # kind # "$idx" #
7176 "|\t$dst, $src$idx}", "", []>,
7180 let Inst{31-21} = 0b01011110000;
7181 let Inst{15-10} = 0b000001;
7182 let Inst{9-5} = src;
7183 let Inst{4-0} = dst;
7186 class SIMDScalarCPYAlias<string asm, string size, Instruction inst,
7187 RegisterClass regtype, RegisterOperand vectype, Operand idxtype>
7188 : InstAlias<asm # "{\t$dst, $src" # size # "$index" #
7189 # "|\t$dst, $src$index}",
7190 (inst regtype:$dst, vectype:$src, idxtype:$index), 0>;
7193 multiclass SIMDScalarCPY<string asm> {
7194 def i8 : BaseSIMDScalarCPY<FPR8, V128, ".b", VectorIndexB> {
7196 let Inst{20-17} = idx;
7199 def i16 : BaseSIMDScalarCPY<FPR16, V128, ".h", VectorIndexH> {
7201 let Inst{20-18} = idx;
7202 let Inst{17-16} = 0b10;
7204 def i32 : BaseSIMDScalarCPY<FPR32, V128, ".s", VectorIndexS> {
7206 let Inst{20-19} = idx;
7207 let Inst{18-16} = 0b100;
7209 def i64 : BaseSIMDScalarCPY<FPR64, V128, ".d", VectorIndexD> {
7212 let Inst{19-16} = 0b1000;
7215 def : Pat<(v1i64 (scalar_to_vector (i64 (vector_extract (v2i64 V128:$src),
7216 VectorIndexD:$idx)))),
7217 (!cast<Instruction>(NAME # i64) V128:$src, VectorIndexD:$idx)>;
7219 // 'DUP' mnemonic aliases.
7220 def : SIMDScalarCPYAlias<"dup", ".b",
7221 !cast<Instruction>(NAME#"i8"),
7222 FPR8, V128, VectorIndexB>;
7223 def : SIMDScalarCPYAlias<"dup", ".h",
7224 !cast<Instruction>(NAME#"i16"),
7225 FPR16, V128, VectorIndexH>;
7226 def : SIMDScalarCPYAlias<"dup", ".s",
7227 !cast<Instruction>(NAME#"i32"),
7228 FPR32, V128, VectorIndexS>;
7229 def : SIMDScalarCPYAlias<"dup", ".d",
7230 !cast<Instruction>(NAME#"i64"),
7231 FPR64, V128, VectorIndexD>;
7234 //----------------------------------------------------------------------------
7235 // AdvSIMD modified immediate instructions
7236 //----------------------------------------------------------------------------
7238 class BaseSIMDModifiedImm<bit Q, bit op, bit op2, dag oops, dag iops,
7239 string asm, string op_string,
7240 string cstr, list<dag> pattern>
7241 : I<oops, iops, asm, op_string, cstr, pattern>,
7248 let Inst{28-19} = 0b0111100000;
7249 let Inst{18-16} = imm8{7-5};
7252 let Inst{9-5} = imm8{4-0};
7256 class BaseSIMDModifiedImmVector<bit Q, bit op, bit op2, RegisterOperand vectype,
7257 Operand immtype, dag opt_shift_iop,
7258 string opt_shift, string asm, string kind,
7260 : BaseSIMDModifiedImm<Q, op, op2, (outs vectype:$Rd),
7261 !con((ins immtype:$imm8), opt_shift_iop), asm,
7262 "{\t$Rd" # kind # ", $imm8" # opt_shift #
7263 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
7265 let DecoderMethod = "DecodeModImmInstruction";
7268 class BaseSIMDModifiedImmVectorTied<bit Q, bit op, RegisterOperand vectype,
7269 Operand immtype, dag opt_shift_iop,
7270 string opt_shift, string asm, string kind,
7272 : BaseSIMDModifiedImm<Q, op, 0, (outs vectype:$dst),
7273 !con((ins vectype:$Rd, immtype:$imm8), opt_shift_iop),
7274 asm, "{\t$Rd" # kind # ", $imm8" # opt_shift #
7275 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
7276 "$Rd = $dst", pattern> {
7277 let DecoderMethod = "DecodeModImmTiedInstruction";
7280 class BaseSIMDModifiedImmVectorShift<bit Q, bit op, bits<2> b15_b12,
7281 RegisterOperand vectype, string asm,
7282 string kind, list<dag> pattern>
7283 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7284 (ins logical_vec_shift:$shift),
7285 "$shift", asm, kind, pattern> {
7287 let Inst{15} = b15_b12{1};
7288 let Inst{14-13} = shift;
7289 let Inst{12} = b15_b12{0};
7292 class BaseSIMDModifiedImmVectorShiftTied<bit Q, bit op, bits<2> b15_b12,
7293 RegisterOperand vectype, string asm,
7294 string kind, list<dag> pattern>
7295 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
7296 (ins logical_vec_shift:$shift),
7297 "$shift", asm, kind, pattern> {
7299 let Inst{15} = b15_b12{1};
7300 let Inst{14-13} = shift;
7301 let Inst{12} = b15_b12{0};
7305 class BaseSIMDModifiedImmVectorShiftHalf<bit Q, bit op, bits<2> b15_b12,
7306 RegisterOperand vectype, string asm,
7307 string kind, list<dag> pattern>
7308 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7309 (ins logical_vec_hw_shift:$shift),
7310 "$shift", asm, kind, pattern> {
7312 let Inst{15} = b15_b12{1};
7314 let Inst{13} = shift{0};
7315 let Inst{12} = b15_b12{0};
7318 class BaseSIMDModifiedImmVectorShiftHalfTied<bit Q, bit op, bits<2> b15_b12,
7319 RegisterOperand vectype, string asm,
7320 string kind, list<dag> pattern>
7321 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
7322 (ins logical_vec_hw_shift:$shift),
7323 "$shift", asm, kind, pattern> {
7325 let Inst{15} = b15_b12{1};
7327 let Inst{13} = shift{0};
7328 let Inst{12} = b15_b12{0};
7331 multiclass SIMDModifiedImmVectorShift<bit op, bits<2> hw_cmode, bits<2> w_cmode,
7333 def v4i16 : BaseSIMDModifiedImmVectorShiftHalf<0, op, hw_cmode, V64,
7335 def v8i16 : BaseSIMDModifiedImmVectorShiftHalf<1, op, hw_cmode, V128,
7338 def v2i32 : BaseSIMDModifiedImmVectorShift<0, op, w_cmode, V64,
7340 def v4i32 : BaseSIMDModifiedImmVectorShift<1, op, w_cmode, V128,
7344 multiclass SIMDModifiedImmVectorShiftTied<bit op, bits<2> hw_cmode,
7345 bits<2> w_cmode, string asm,
7347 def v4i16 : BaseSIMDModifiedImmVectorShiftHalfTied<0, op, hw_cmode, V64,
7349 [(set (v4i16 V64:$dst), (OpNode V64:$Rd,
7351 (i32 imm:$shift)))]>;
7352 def v8i16 : BaseSIMDModifiedImmVectorShiftHalfTied<1, op, hw_cmode, V128,
7354 [(set (v8i16 V128:$dst), (OpNode V128:$Rd,
7356 (i32 imm:$shift)))]>;
7358 def v2i32 : BaseSIMDModifiedImmVectorShiftTied<0, op, w_cmode, V64,
7360 [(set (v2i32 V64:$dst), (OpNode V64:$Rd,
7362 (i32 imm:$shift)))]>;
7363 def v4i32 : BaseSIMDModifiedImmVectorShiftTied<1, op, w_cmode, V128,
7365 [(set (v4i32 V128:$dst), (OpNode V128:$Rd,
7367 (i32 imm:$shift)))]>;
7370 class SIMDModifiedImmMoveMSL<bit Q, bit op, bits<4> cmode,
7371 RegisterOperand vectype, string asm,
7372 string kind, list<dag> pattern>
7373 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7374 (ins move_vec_shift:$shift),
7375 "$shift", asm, kind, pattern> {
7377 let Inst{15-13} = cmode{3-1};
7378 let Inst{12} = shift;
7381 class SIMDModifiedImmVectorNoShift<bit Q, bit op, bit op2, bits<4> cmode,
7382 RegisterOperand vectype,
7383 Operand imm_type, string asm,
7384 string kind, list<dag> pattern>
7385 : BaseSIMDModifiedImmVector<Q, op, op2, vectype, imm_type, (ins), "",
7386 asm, kind, pattern> {
7387 let Inst{15-12} = cmode;
7390 class SIMDModifiedImmScalarNoShift<bit Q, bit op, bits<4> cmode, string asm,
7392 : BaseSIMDModifiedImm<Q, op, 0, (outs FPR64:$Rd), (ins simdimmtype10:$imm8), asm,
7393 "\t$Rd, $imm8", "", pattern> {
7394 let Inst{15-12} = cmode;
7395 let DecoderMethod = "DecodeModImmInstruction";
7398 //----------------------------------------------------------------------------
7399 // AdvSIMD indexed element
7400 //----------------------------------------------------------------------------
7402 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7403 class BaseSIMDIndexed<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
7404 RegisterOperand dst_reg, RegisterOperand lhs_reg,
7405 RegisterOperand rhs_reg, Operand vec_idx, string asm,
7406 string apple_kind, string dst_kind, string lhs_kind,
7407 string rhs_kind, list<dag> pattern>
7408 : I<(outs dst_reg:$Rd), (ins lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx),
7410 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
7411 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "", pattern>,
7420 let Inst{28} = Scalar;
7421 let Inst{27-24} = 0b1111;
7422 let Inst{23-22} = size;
7423 // Bit 21 must be set by the derived class.
7424 let Inst{20-16} = Rm;
7425 let Inst{15-12} = opc;
7426 // Bit 11 must be set by the derived class.
7432 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7433 class BaseSIMDIndexedTied<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
7434 RegisterOperand dst_reg, RegisterOperand lhs_reg,
7435 RegisterOperand rhs_reg, Operand vec_idx, string asm,
7436 string apple_kind, string dst_kind, string lhs_kind,
7437 string rhs_kind, list<dag> pattern>
7438 : I<(outs dst_reg:$dst),
7439 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx), asm,
7440 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
7441 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "$Rd = $dst", pattern>,
7450 let Inst{28} = Scalar;
7451 let Inst{27-24} = 0b1111;
7452 let Inst{23-22} = size;
7453 // Bit 21 must be set by the derived class.
7454 let Inst{20-16} = Rm;
7455 let Inst{15-12} = opc;
7456 // Bit 11 must be set by the derived class.
7462 // ARMv8.2-A Dot Product Instructions (Indexed)
7463 class BaseSIMDThreeSameVectorDotIndex<bit Q, bit U, string asm, string dst_kind,
7464 string lhs_kind, string rhs_kind,
7465 RegisterOperand RegType,
7466 ValueType AccumType, ValueType InputType,
7467 SDPatternOperator OpNode> :
7468 BaseSIMDIndexedTied<Q, U, 0b0, 0b10, 0b1110, RegType, RegType, V128,
7469 VectorIndexS, asm, "", dst_kind, lhs_kind, rhs_kind,
7470 [(set (AccumType RegType:$dst),
7471 (AccumType (OpNode (AccumType RegType:$Rd),
7472 (InputType RegType:$Rn),
7473 (InputType (bitconvert (AccumType
7474 (AArch64duplane32 (v4i32 V128:$Rm),
7475 VectorIndexS:$idx)))))))]> {
7477 let Inst{21} = idx{0}; // L
7478 let Inst{11} = idx{1}; // H
7481 multiclass SIMDThreeSameVectorDotIndex<bit U, string asm,
7482 SDPatternOperator OpNode> {
7483 def v8i8 : BaseSIMDThreeSameVectorDotIndex<0, U, asm, ".2s", ".8b", ".4b",
7484 V64, v2i32, v8i8, OpNode>;
7485 def v16i8 : BaseSIMDThreeSameVectorDotIndex<1, U, asm, ".4s", ".16b", ".4b",
7486 V128, v4i32, v16i8, OpNode>;
7489 // ARMv8.2-A Fused Multiply Add-Long Instructions (Indexed)
7490 class BaseSIMDThreeSameVectorFMLIndex<bit Q, bit U, bits<4> opc, string asm,
7491 string dst_kind, string lhs_kind,
7492 string rhs_kind, RegisterOperand RegType,
7493 ValueType AccumType, ValueType InputType,
7494 SDPatternOperator OpNode> :
7495 BaseSIMDIndexedTied<Q, U, 0, 0b10, opc, RegType, RegType, V128,
7496 VectorIndexH, asm, "", dst_kind, lhs_kind, rhs_kind,
7497 [(set (AccumType RegType:$dst),
7498 (AccumType (OpNode (AccumType RegType:$Rd),
7499 (InputType RegType:$Rn),
7500 (InputType (AArch64duplane16 (v8f16 V128:$Rm),
7501 VectorIndexH:$idx)))))]> {
7504 let Inst{11} = idx{2}; // H
7505 let Inst{21} = idx{1}; // L
7506 let Inst{20} = idx{0}; // M
7509 multiclass SIMDThreeSameVectorFMLIndex<bit U, bits<4> opc, string asm,
7510 SDPatternOperator OpNode> {
7511 def v4f16 : BaseSIMDThreeSameVectorFMLIndex<0, U, opc, asm, ".2s", ".2h", ".h",
7512 V64, v2f32, v4f16, OpNode>;
7513 def v8f16 : BaseSIMDThreeSameVectorFMLIndex<1, U, opc, asm, ".4s", ".4h", ".h",
7514 V128, v4f32, v8f16, OpNode>;
7517 multiclass SIMDFPIndexed<bit U, bits<4> opc, string asm,
7518 SDPatternOperator OpNode> {
7519 let Predicates = [HasNEON, HasFullFP16] in {
7520 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b00, opc,
7522 V128_lo, VectorIndexH,
7523 asm, ".4h", ".4h", ".4h", ".h",
7524 [(set (v4f16 V64:$Rd),
7525 (OpNode (v4f16 V64:$Rn),
7526 (v4f16 (AArch64duplane16 (v8f16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7528 let Inst{11} = idx{2};
7529 let Inst{21} = idx{1};
7530 let Inst{20} = idx{0};
7533 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b00, opc,
7535 V128_lo, VectorIndexH,
7536 asm, ".8h", ".8h", ".8h", ".h",
7537 [(set (v8f16 V128:$Rd),
7538 (OpNode (v8f16 V128:$Rn),
7539 (v8f16 (AArch64duplane16 (v8f16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7541 let Inst{11} = idx{2};
7542 let Inst{21} = idx{1};
7543 let Inst{20} = idx{0};
7545 } // Predicates = [HasNEON, HasFullFP16]
7547 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7550 asm, ".2s", ".2s", ".2s", ".s",
7551 [(set (v2f32 V64:$Rd),
7552 (OpNode (v2f32 V64:$Rn),
7553 (v2f32 (AArch64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
7555 let Inst{11} = idx{1};
7556 let Inst{21} = idx{0};
7559 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7562 asm, ".4s", ".4s", ".4s", ".s",
7563 [(set (v4f32 V128:$Rd),
7564 (OpNode (v4f32 V128:$Rn),
7565 (v4f32 (AArch64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
7567 let Inst{11} = idx{1};
7568 let Inst{21} = idx{0};
7571 def v2i64_indexed : BaseSIMDIndexed<1, U, 0, 0b11, opc,
7574 asm, ".2d", ".2d", ".2d", ".d",
7575 [(set (v2f64 V128:$Rd),
7576 (OpNode (v2f64 V128:$Rn),
7577 (v2f64 (AArch64duplane64 (v2f64 V128:$Rm), VectorIndexD:$idx))))]> {
7579 let Inst{11} = idx{0};
7583 let Predicates = [HasNEON, HasFullFP16] in {
7584 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b00, opc,
7585 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7586 asm, ".h", "", "", ".h",
7587 [(set (f16 FPR16Op:$Rd),
7588 (OpNode (f16 FPR16Op:$Rn),
7589 (f16 (vector_extract (v8f16 V128_lo:$Rm),
7590 VectorIndexH:$idx))))]> {
7592 let Inst{11} = idx{2};
7593 let Inst{21} = idx{1};
7594 let Inst{20} = idx{0};
7596 } // Predicates = [HasNEON, HasFullFP16]
7598 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
7599 FPR32Op, FPR32Op, V128, VectorIndexS,
7600 asm, ".s", "", "", ".s",
7601 [(set (f32 FPR32Op:$Rd),
7602 (OpNode (f32 FPR32Op:$Rn),
7603 (f32 (vector_extract (v4f32 V128:$Rm),
7604 VectorIndexS:$idx))))]> {
7606 let Inst{11} = idx{1};
7607 let Inst{21} = idx{0};
7610 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b11, opc,
7611 FPR64Op, FPR64Op, V128, VectorIndexD,
7612 asm, ".d", "", "", ".d",
7613 [(set (f64 FPR64Op:$Rd),
7614 (OpNode (f64 FPR64Op:$Rn),
7615 (f64 (vector_extract (v2f64 V128:$Rm),
7616 VectorIndexD:$idx))))]> {
7618 let Inst{11} = idx{0};
7623 multiclass SIMDFPIndexedTiedPatterns<string INST, SDPatternOperator OpNode> {
7624 // 2 variants for the .2s version: DUPLANE from 128-bit and DUP scalar.
7625 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7626 (AArch64duplane32 (v4f32 V128:$Rm),
7627 VectorIndexS:$idx))),
7628 (!cast<Instruction>(INST # v2i32_indexed)
7629 V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7630 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7631 (AArch64dup (f32 FPR32Op:$Rm)))),
7632 (!cast<Instruction>(INST # "v2i32_indexed") V64:$Rd, V64:$Rn,
7633 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7636 // 2 variants for the .4s version: DUPLANE from 128-bit and DUP scalar.
7637 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7638 (AArch64duplane32 (v4f32 V128:$Rm),
7639 VectorIndexS:$idx))),
7640 (!cast<Instruction>(INST # "v4i32_indexed")
7641 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7642 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7643 (AArch64dup (f32 FPR32Op:$Rm)))),
7644 (!cast<Instruction>(INST # "v4i32_indexed") V128:$Rd, V128:$Rn,
7645 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7647 // 2 variants for the .2d version: DUPLANE from 128-bit and DUP scalar.
7648 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7649 (AArch64duplane64 (v2f64 V128:$Rm),
7650 VectorIndexD:$idx))),
7651 (!cast<Instruction>(INST # "v2i64_indexed")
7652 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7653 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7654 (AArch64dup (f64 FPR64Op:$Rm)))),
7655 (!cast<Instruction>(INST # "v2i64_indexed") V128:$Rd, V128:$Rn,
7656 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
7658 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
7659 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7660 (vector_extract (v4f32 V128:$Rm), VectorIndexS:$idx))),
7661 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
7662 V128:$Rm, VectorIndexS:$idx)>;
7663 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7664 (vector_extract (v2f32 V64:$Rm), VectorIndexS:$idx))),
7665 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
7666 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
7668 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
7669 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
7670 (vector_extract (v2f64 V128:$Rm), VectorIndexD:$idx))),
7671 (!cast<Instruction>(INST # "v1i64_indexed") FPR64:$Rd, FPR64:$Rn,
7672 V128:$Rm, VectorIndexD:$idx)>;
7675 multiclass SIMDFPIndexedTied<bit U, bits<4> opc, string asm> {
7676 let Predicates = [HasNEON, HasFullFP16] in {
7677 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b00, opc, V64, V64,
7678 V128_lo, VectorIndexH,
7679 asm, ".4h", ".4h", ".4h", ".h", []> {
7681 let Inst{11} = idx{2};
7682 let Inst{21} = idx{1};
7683 let Inst{20} = idx{0};
7686 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b00, opc,
7688 V128_lo, VectorIndexH,
7689 asm, ".8h", ".8h", ".8h", ".h", []> {
7691 let Inst{11} = idx{2};
7692 let Inst{21} = idx{1};
7693 let Inst{20} = idx{0};
7695 } // Predicates = [HasNEON, HasFullFP16]
7697 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc, V64, V64,
7699 asm, ".2s", ".2s", ".2s", ".s", []> {
7701 let Inst{11} = idx{1};
7702 let Inst{21} = idx{0};
7705 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
7708 asm, ".4s", ".4s", ".4s", ".s", []> {
7710 let Inst{11} = idx{1};
7711 let Inst{21} = idx{0};
7714 def v2i64_indexed : BaseSIMDIndexedTied<1, U, 0, 0b11, opc,
7717 asm, ".2d", ".2d", ".2d", ".d", []> {
7719 let Inst{11} = idx{0};
7723 let Predicates = [HasNEON, HasFullFP16] in {
7724 def v1i16_indexed : BaseSIMDIndexedTied<1, U, 1, 0b00, opc,
7725 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7726 asm, ".h", "", "", ".h", []> {
7728 let Inst{11} = idx{2};
7729 let Inst{21} = idx{1};
7730 let Inst{20} = idx{0};
7732 } // Predicates = [HasNEON, HasFullFP16]
7734 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
7735 FPR32Op, FPR32Op, V128, VectorIndexS,
7736 asm, ".s", "", "", ".s", []> {
7738 let Inst{11} = idx{1};
7739 let Inst{21} = idx{0};
7742 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b11, opc,
7743 FPR64Op, FPR64Op, V128, VectorIndexD,
7744 asm, ".d", "", "", ".d", []> {
7746 let Inst{11} = idx{0};
7751 multiclass SIMDIndexedHS<bit U, bits<4> opc, string asm,
7752 SDPatternOperator OpNode> {
7753 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc, V64, V64,
7754 V128_lo, VectorIndexH,
7755 asm, ".4h", ".4h", ".4h", ".h",
7756 [(set (v4i16 V64:$Rd),
7757 (OpNode (v4i16 V64:$Rn),
7758 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7760 let Inst{11} = idx{2};
7761 let Inst{21} = idx{1};
7762 let Inst{20} = idx{0};
7765 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7767 V128_lo, VectorIndexH,
7768 asm, ".8h", ".8h", ".8h", ".h",
7769 [(set (v8i16 V128:$Rd),
7770 (OpNode (v8i16 V128:$Rn),
7771 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7773 let Inst{11} = idx{2};
7774 let Inst{21} = idx{1};
7775 let Inst{20} = idx{0};
7778 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7781 asm, ".2s", ".2s", ".2s", ".s",
7782 [(set (v2i32 V64:$Rd),
7783 (OpNode (v2i32 V64:$Rn),
7784 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7786 let Inst{11} = idx{1};
7787 let Inst{21} = idx{0};
7790 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7793 asm, ".4s", ".4s", ".4s", ".s",
7794 [(set (v4i32 V128:$Rd),
7795 (OpNode (v4i32 V128:$Rn),
7796 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7798 let Inst{11} = idx{1};
7799 let Inst{21} = idx{0};
7802 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
7803 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7804 asm, ".h", "", "", ".h", []> {
7806 let Inst{11} = idx{2};
7807 let Inst{21} = idx{1};
7808 let Inst{20} = idx{0};
7811 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
7812 FPR32Op, FPR32Op, V128, VectorIndexS,
7813 asm, ".s", "", "", ".s",
7814 [(set (i32 FPR32Op:$Rd),
7815 (OpNode FPR32Op:$Rn,
7816 (i32 (vector_extract (v4i32 V128:$Rm),
7817 VectorIndexS:$idx))))]> {
7819 let Inst{11} = idx{1};
7820 let Inst{21} = idx{0};
7824 multiclass SIMDVectorIndexedHS<bit U, bits<4> opc, string asm,
7825 SDPatternOperator OpNode> {
7826 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
7828 V128_lo, VectorIndexH,
7829 asm, ".4h", ".4h", ".4h", ".h",
7830 [(set (v4i16 V64:$Rd),
7831 (OpNode (v4i16 V64:$Rn),
7832 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7834 let Inst{11} = idx{2};
7835 let Inst{21} = idx{1};
7836 let Inst{20} = idx{0};
7839 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7841 V128_lo, VectorIndexH,
7842 asm, ".8h", ".8h", ".8h", ".h",
7843 [(set (v8i16 V128:$Rd),
7844 (OpNode (v8i16 V128:$Rn),
7845 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7847 let Inst{11} = idx{2};
7848 let Inst{21} = idx{1};
7849 let Inst{20} = idx{0};
7852 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7855 asm, ".2s", ".2s", ".2s", ".s",
7856 [(set (v2i32 V64:$Rd),
7857 (OpNode (v2i32 V64:$Rn),
7858 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7860 let Inst{11} = idx{1};
7861 let Inst{21} = idx{0};
7864 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7867 asm, ".4s", ".4s", ".4s", ".s",
7868 [(set (v4i32 V128:$Rd),
7869 (OpNode (v4i32 V128:$Rn),
7870 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7872 let Inst{11} = idx{1};
7873 let Inst{21} = idx{0};
7877 multiclass SIMDVectorIndexedHSTied<bit U, bits<4> opc, string asm,
7878 SDPatternOperator OpNode> {
7879 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc, V64, V64,
7880 V128_lo, VectorIndexH,
7881 asm, ".4h", ".4h", ".4h", ".h",
7882 [(set (v4i16 V64:$dst),
7883 (OpNode (v4i16 V64:$Rd),(v4i16 V64:$Rn),
7884 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7886 let Inst{11} = idx{2};
7887 let Inst{21} = idx{1};
7888 let Inst{20} = idx{0};
7891 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
7893 V128_lo, VectorIndexH,
7894 asm, ".8h", ".8h", ".8h", ".h",
7895 [(set (v8i16 V128:$dst),
7896 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7897 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7899 let Inst{11} = idx{2};
7900 let Inst{21} = idx{1};
7901 let Inst{20} = idx{0};
7904 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
7907 asm, ".2s", ".2s", ".2s", ".s",
7908 [(set (v2i32 V64:$dst),
7909 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7910 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7912 let Inst{11} = idx{1};
7913 let Inst{21} = idx{0};
7916 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
7919 asm, ".4s", ".4s", ".4s", ".s",
7920 [(set (v4i32 V128:$dst),
7921 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7922 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7924 let Inst{11} = idx{1};
7925 let Inst{21} = idx{0};
7929 multiclass SIMDIndexedLongSD<bit U, bits<4> opc, string asm,
7930 SDPatternOperator OpNode> {
7931 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
7933 V128_lo, VectorIndexH,
7934 asm, ".4s", ".4s", ".4h", ".h",
7935 [(set (v4i32 V128:$Rd),
7936 (OpNode (v4i16 V64:$Rn),
7937 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7939 let Inst{11} = idx{2};
7940 let Inst{21} = idx{1};
7941 let Inst{20} = idx{0};
7944 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7946 V128_lo, VectorIndexH,
7947 asm#"2", ".4s", ".4s", ".8h", ".h",
7948 [(set (v4i32 V128:$Rd),
7949 (OpNode (extract_high_v8i16 V128:$Rn),
7950 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
7951 VectorIndexH:$idx))))]> {
7954 let Inst{11} = idx{2};
7955 let Inst{21} = idx{1};
7956 let Inst{20} = idx{0};
7959 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7962 asm, ".2d", ".2d", ".2s", ".s",
7963 [(set (v2i64 V128:$Rd),
7964 (OpNode (v2i32 V64:$Rn),
7965 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7967 let Inst{11} = idx{1};
7968 let Inst{21} = idx{0};
7971 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7974 asm#"2", ".2d", ".2d", ".4s", ".s",
7975 [(set (v2i64 V128:$Rd),
7976 (OpNode (extract_high_v4i32 V128:$Rn),
7977 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
7978 VectorIndexS:$idx))))]> {
7980 let Inst{11} = idx{1};
7981 let Inst{21} = idx{0};
7984 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
7985 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
7986 asm, ".h", "", "", ".h", []> {
7988 let Inst{11} = idx{2};
7989 let Inst{21} = idx{1};
7990 let Inst{20} = idx{0};
7993 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
7994 FPR64Op, FPR32Op, V128, VectorIndexS,
7995 asm, ".s", "", "", ".s", []> {
7997 let Inst{11} = idx{1};
7998 let Inst{21} = idx{0};
8002 multiclass SIMDIndexedLongSQDMLXSDTied<bit U, bits<4> opc, string asm,
8003 SDPatternOperator Accum> {
8004 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
8006 V128_lo, VectorIndexH,
8007 asm, ".4s", ".4s", ".4h", ".h",
8008 [(set (v4i32 V128:$dst),
8009 (Accum (v4i32 V128:$Rd),
8010 (v4i32 (int_aarch64_neon_sqdmull
8012 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8013 VectorIndexH:$idx))))))]> {
8015 let Inst{11} = idx{2};
8016 let Inst{21} = idx{1};
8017 let Inst{20} = idx{0};
8020 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but an
8021 // intermediate EXTRACT_SUBREG would be untyped.
8022 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
8023 (i32 (vector_extract (v4i32
8024 (int_aarch64_neon_sqdmull (v4i16 V64:$Rn),
8025 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8026 VectorIndexH:$idx)))),
8029 (!cast<Instruction>(NAME # v4i16_indexed)
8030 (SUBREG_TO_REG (i32 0), FPR32Op:$Rd, ssub), V64:$Rn,
8031 V128_lo:$Rm, VectorIndexH:$idx),
8034 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
8036 V128_lo, VectorIndexH,
8037 asm#"2", ".4s", ".4s", ".8h", ".h",
8038 [(set (v4i32 V128:$dst),
8039 (Accum (v4i32 V128:$Rd),
8040 (v4i32 (int_aarch64_neon_sqdmull
8041 (extract_high_v8i16 V128:$Rn),
8043 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8044 VectorIndexH:$idx))))))]> {
8046 let Inst{11} = idx{2};
8047 let Inst{21} = idx{1};
8048 let Inst{20} = idx{0};
8051 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
8054 asm, ".2d", ".2d", ".2s", ".s",
8055 [(set (v2i64 V128:$dst),
8056 (Accum (v2i64 V128:$Rd),
8057 (v2i64 (int_aarch64_neon_sqdmull
8059 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm),
8060 VectorIndexS:$idx))))))]> {
8062 let Inst{11} = idx{1};
8063 let Inst{21} = idx{0};
8066 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
8069 asm#"2", ".2d", ".2d", ".4s", ".s",
8070 [(set (v2i64 V128:$dst),
8071 (Accum (v2i64 V128:$Rd),
8072 (v2i64 (int_aarch64_neon_sqdmull
8073 (extract_high_v4i32 V128:$Rn),
8075 (AArch64duplane32 (v4i32 V128:$Rm),
8076 VectorIndexS:$idx))))))]> {
8078 let Inst{11} = idx{1};
8079 let Inst{21} = idx{0};
8082 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
8083 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
8084 asm, ".h", "", "", ".h", []> {
8086 let Inst{11} = idx{2};
8087 let Inst{21} = idx{1};
8088 let Inst{20} = idx{0};
8092 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
8093 FPR64Op, FPR32Op, V128, VectorIndexS,
8094 asm, ".s", "", "", ".s",
8095 [(set (i64 FPR64Op:$dst),
8096 (Accum (i64 FPR64Op:$Rd),
8097 (i64 (int_aarch64_neon_sqdmulls_scalar
8099 (i32 (vector_extract (v4i32 V128:$Rm),
8100 VectorIndexS:$idx))))))]> {
8103 let Inst{11} = idx{1};
8104 let Inst{21} = idx{0};
8108 multiclass SIMDVectorIndexedLongSD<bit U, bits<4> opc, string asm,
8109 SDPatternOperator OpNode> {
8110 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
8111 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
8113 V128_lo, VectorIndexH,
8114 asm, ".4s", ".4s", ".4h", ".h",
8115 [(set (v4i32 V128:$Rd),
8116 (OpNode (v4i16 V64:$Rn),
8117 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8119 let Inst{11} = idx{2};
8120 let Inst{21} = idx{1};
8121 let Inst{20} = idx{0};
8124 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
8126 V128_lo, VectorIndexH,
8127 asm#"2", ".4s", ".4s", ".8h", ".h",
8128 [(set (v4i32 V128:$Rd),
8129 (OpNode (extract_high_v8i16 V128:$Rn),
8130 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8131 VectorIndexH:$idx))))]> {
8134 let Inst{11} = idx{2};
8135 let Inst{21} = idx{1};
8136 let Inst{20} = idx{0};
8139 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
8142 asm, ".2d", ".2d", ".2s", ".s",
8143 [(set (v2i64 V128:$Rd),
8144 (OpNode (v2i32 V64:$Rn),
8145 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8147 let Inst{11} = idx{1};
8148 let Inst{21} = idx{0};
8151 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
8154 asm#"2", ".2d", ".2d", ".4s", ".s",
8155 [(set (v2i64 V128:$Rd),
8156 (OpNode (extract_high_v4i32 V128:$Rn),
8157 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
8158 VectorIndexS:$idx))))]> {
8160 let Inst{11} = idx{1};
8161 let Inst{21} = idx{0};
8166 multiclass SIMDVectorIndexedLongSDTied<bit U, bits<4> opc, string asm,
8167 SDPatternOperator OpNode> {
8168 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
8169 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
8171 V128_lo, VectorIndexH,
8172 asm, ".4s", ".4s", ".4h", ".h",
8173 [(set (v4i32 V128:$dst),
8174 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn),
8175 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
8177 let Inst{11} = idx{2};
8178 let Inst{21} = idx{1};
8179 let Inst{20} = idx{0};
8182 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
8184 V128_lo, VectorIndexH,
8185 asm#"2", ".4s", ".4s", ".8h", ".h",
8186 [(set (v4i32 V128:$dst),
8187 (OpNode (v4i32 V128:$Rd),
8188 (extract_high_v8i16 V128:$Rn),
8189 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
8190 VectorIndexH:$idx))))]> {
8192 let Inst{11} = idx{2};
8193 let Inst{21} = idx{1};
8194 let Inst{20} = idx{0};
8197 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
8200 asm, ".2d", ".2d", ".2s", ".s",
8201 [(set (v2i64 V128:$dst),
8202 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn),
8203 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
8205 let Inst{11} = idx{1};
8206 let Inst{21} = idx{0};
8209 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
8212 asm#"2", ".2d", ".2d", ".4s", ".s",
8213 [(set (v2i64 V128:$dst),
8214 (OpNode (v2i64 V128:$Rd),
8215 (extract_high_v4i32 V128:$Rn),
8216 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
8217 VectorIndexS:$idx))))]> {
8219 let Inst{11} = idx{1};
8220 let Inst{21} = idx{0};
8225 //----------------------------------------------------------------------------
8226 // AdvSIMD scalar shift by immediate
8227 //----------------------------------------------------------------------------
8229 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8230 class BaseSIMDScalarShift<bit U, bits<5> opc, bits<7> fixed_imm,
8231 RegisterClass regtype1, RegisterClass regtype2,
8232 Operand immtype, string asm, list<dag> pattern>
8233 : I<(outs regtype1:$Rd), (ins regtype2:$Rn, immtype:$imm),
8234 asm, "\t$Rd, $Rn, $imm", "", pattern>,
8239 let Inst{31-30} = 0b01;
8241 let Inst{28-23} = 0b111110;
8242 let Inst{22-16} = fixed_imm;
8243 let Inst{15-11} = opc;
8249 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8250 class BaseSIMDScalarShiftTied<bit U, bits<5> opc, bits<7> fixed_imm,
8251 RegisterClass regtype1, RegisterClass regtype2,
8252 Operand immtype, string asm, list<dag> pattern>
8253 : I<(outs regtype1:$dst), (ins regtype1:$Rd, regtype2:$Rn, immtype:$imm),
8254 asm, "\t$Rd, $Rn, $imm", "$Rd = $dst", pattern>,
8259 let Inst{31-30} = 0b01;
8261 let Inst{28-23} = 0b111110;
8262 let Inst{22-16} = fixed_imm;
8263 let Inst{15-11} = opc;
8270 multiclass SIMDFPScalarRShift<bit U, bits<5> opc, string asm> {
8271 let Predicates = [HasNEON, HasFullFP16] in {
8272 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8273 FPR16, FPR16, vecshiftR16, asm, []> {
8274 let Inst{19-16} = imm{3-0};
8276 } // Predicates = [HasNEON, HasFullFP16]
8277 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8278 FPR32, FPR32, vecshiftR32, asm, []> {
8279 let Inst{20-16} = imm{4-0};
8281 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8282 FPR64, FPR64, vecshiftR64, asm, []> {
8283 let Inst{21-16} = imm{5-0};
8287 multiclass SIMDScalarRShiftD<bit U, bits<5> opc, string asm,
8288 SDPatternOperator OpNode> {
8289 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8290 FPR64, FPR64, vecshiftR64, asm,
8291 [(set (i64 FPR64:$Rd),
8292 (OpNode (i64 FPR64:$Rn), (i32 vecshiftR64:$imm)))]> {
8293 let Inst{21-16} = imm{5-0};
8296 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftR64:$imm))),
8297 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftR64:$imm)>;
8300 multiclass SIMDScalarRShiftDTied<bit U, bits<5> opc, string asm,
8301 SDPatternOperator OpNode = null_frag> {
8302 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
8303 FPR64, FPR64, vecshiftR64, asm,
8304 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn),
8305 (i32 vecshiftR64:$imm)))]> {
8306 let Inst{21-16} = imm{5-0};
8309 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
8310 (i32 vecshiftR64:$imm))),
8311 (!cast<Instruction>(NAME # "d") FPR64:$Rd, FPR64:$Rn,
8315 multiclass SIMDScalarLShiftD<bit U, bits<5> opc, string asm,
8316 SDPatternOperator OpNode> {
8317 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8318 FPR64, FPR64, vecshiftL64, asm,
8319 [(set (v1i64 FPR64:$Rd),
8320 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
8321 let Inst{21-16} = imm{5-0};
8325 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8326 multiclass SIMDScalarLShiftDTied<bit U, bits<5> opc, string asm> {
8327 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
8328 FPR64, FPR64, vecshiftL64, asm, []> {
8329 let Inst{21-16} = imm{5-0};
8333 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8334 multiclass SIMDScalarRShiftBHS<bit U, bits<5> opc, string asm,
8335 SDPatternOperator OpNode = null_frag> {
8336 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8337 FPR8, FPR16, vecshiftR8, asm, []> {
8338 let Inst{18-16} = imm{2-0};
8341 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8342 FPR16, FPR32, vecshiftR16, asm, []> {
8343 let Inst{19-16} = imm{3-0};
8346 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8347 FPR32, FPR64, vecshiftR32, asm,
8348 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn), vecshiftR32:$imm))]> {
8349 let Inst{20-16} = imm{4-0};
8353 multiclass SIMDScalarLShiftBHSD<bit U, bits<5> opc, string asm,
8354 SDPatternOperator OpNode> {
8355 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8356 FPR8, FPR8, vecshiftL8, asm, []> {
8357 let Inst{18-16} = imm{2-0};
8360 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8361 FPR16, FPR16, vecshiftL16, asm, []> {
8362 let Inst{19-16} = imm{3-0};
8365 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8366 FPR32, FPR32, vecshiftL32, asm,
8367 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn), (i32 vecshiftL32:$imm)))]> {
8368 let Inst{20-16} = imm{4-0};
8371 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8372 FPR64, FPR64, vecshiftL64, asm,
8373 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
8374 let Inst{21-16} = imm{5-0};
8377 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm))),
8378 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftL64:$imm)>;
8381 multiclass SIMDScalarRShiftBHSD<bit U, bits<5> opc, string asm> {
8382 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8383 FPR8, FPR8, vecshiftR8, asm, []> {
8384 let Inst{18-16} = imm{2-0};
8387 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8388 FPR16, FPR16, vecshiftR16, asm, []> {
8389 let Inst{19-16} = imm{3-0};
8392 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8393 FPR32, FPR32, vecshiftR32, asm, []> {
8394 let Inst{20-16} = imm{4-0};
8397 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8398 FPR64, FPR64, vecshiftR64, asm, []> {
8399 let Inst{21-16} = imm{5-0};
8403 //----------------------------------------------------------------------------
8404 // AdvSIMD vector x indexed element
8405 //----------------------------------------------------------------------------
8407 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8408 class BaseSIMDVectorShift<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
8409 RegisterOperand dst_reg, RegisterOperand src_reg,
8411 string asm, string dst_kind, string src_kind,
8413 : I<(outs dst_reg:$Rd), (ins src_reg:$Rn, immtype:$imm),
8414 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
8415 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "", pattern>,
8422 let Inst{28-23} = 0b011110;
8423 let Inst{22-16} = fixed_imm;
8424 let Inst{15-11} = opc;
8430 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8431 class BaseSIMDVectorShiftTied<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
8432 RegisterOperand vectype1, RegisterOperand vectype2,
8434 string asm, string dst_kind, string src_kind,
8436 : I<(outs vectype1:$dst), (ins vectype1:$Rd, vectype2:$Rn, immtype:$imm),
8437 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
8438 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "$Rd = $dst", pattern>,
8445 let Inst{28-23} = 0b011110;
8446 let Inst{22-16} = fixed_imm;
8447 let Inst{15-11} = opc;
8453 multiclass SIMDVectorRShiftSD<bit U, bits<5> opc, string asm,
8455 let Predicates = [HasNEON, HasFullFP16] in {
8456 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8457 V64, V64, vecshiftR16,
8459 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (i32 imm:$imm)))]> {
8461 let Inst{19-16} = imm;
8464 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8465 V128, V128, vecshiftR16,
8467 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (i32 imm:$imm)))]> {
8469 let Inst{19-16} = imm;
8471 } // Predicates = [HasNEON, HasFullFP16]
8472 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8473 V64, V64, vecshiftR32,
8475 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (i32 imm:$imm)))]> {
8477 let Inst{20-16} = imm;
8480 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8481 V128, V128, vecshiftR32,
8483 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (i32 imm:$imm)))]> {
8485 let Inst{20-16} = imm;
8488 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8489 V128, V128, vecshiftR64,
8491 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (i32 imm:$imm)))]> {
8493 let Inst{21-16} = imm;
8497 multiclass SIMDVectorRShiftToFP<bit U, bits<5> opc, string asm,
8499 let Predicates = [HasNEON, HasFullFP16] in {
8500 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8501 V64, V64, vecshiftR16,
8503 [(set (v4f16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (i32 imm:$imm)))]> {
8505 let Inst{19-16} = imm;
8508 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8509 V128, V128, vecshiftR16,
8511 [(set (v8f16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (i32 imm:$imm)))]> {
8513 let Inst{19-16} = imm;
8515 } // Predicates = [HasNEON, HasFullFP16]
8517 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8518 V64, V64, vecshiftR32,
8520 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (i32 imm:$imm)))]> {
8522 let Inst{20-16} = imm;
8525 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8526 V128, V128, vecshiftR32,
8528 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (i32 imm:$imm)))]> {
8530 let Inst{20-16} = imm;
8533 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8534 V128, V128, vecshiftR64,
8536 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (i32 imm:$imm)))]> {
8538 let Inst{21-16} = imm;
8542 multiclass SIMDVectorRShiftNarrowBHS<bit U, bits<5> opc, string asm,
8543 SDPatternOperator OpNode> {
8544 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8545 V64, V128, vecshiftR16Narrow,
8547 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))]> {
8549 let Inst{18-16} = imm;
8552 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8553 V128, V128, vecshiftR16Narrow,
8554 asm#"2", ".16b", ".8h", []> {
8556 let Inst{18-16} = imm;
8557 let hasSideEffects = 0;
8560 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8561 V64, V128, vecshiftR32Narrow,
8563 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))]> {
8565 let Inst{19-16} = imm;
8568 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8569 V128, V128, vecshiftR32Narrow,
8570 asm#"2", ".8h", ".4s", []> {
8572 let Inst{19-16} = imm;
8573 let hasSideEffects = 0;
8576 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8577 V64, V128, vecshiftR64Narrow,
8579 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))]> {
8581 let Inst{20-16} = imm;
8584 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8585 V128, V128, vecshiftR64Narrow,
8586 asm#"2", ".4s", ".2d", []> {
8588 let Inst{20-16} = imm;
8589 let hasSideEffects = 0;
8592 // TableGen doesn't like patters w/ INSERT_SUBREG on the instructions
8593 // themselves, so put them here instead.
8595 // Patterns involving what's effectively an insert high and a normal
8596 // intrinsic, represented by CONCAT_VECTORS.
8597 def : Pat<(concat_vectors (v8i8 V64:$Rd),(OpNode (v8i16 V128:$Rn),
8598 vecshiftR16Narrow:$imm)),
8599 (!cast<Instruction>(NAME # "v16i8_shift")
8600 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8601 V128:$Rn, vecshiftR16Narrow:$imm)>;
8602 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn),
8603 vecshiftR32Narrow:$imm)),
8604 (!cast<Instruction>(NAME # "v8i16_shift")
8605 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8606 V128:$Rn, vecshiftR32Narrow:$imm)>;
8607 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn),
8608 vecshiftR64Narrow:$imm)),
8609 (!cast<Instruction>(NAME # "v4i32_shift")
8610 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8611 V128:$Rn, vecshiftR64Narrow:$imm)>;
8614 multiclass SIMDVectorLShiftBHSD<bit U, bits<5> opc, string asm,
8615 SDPatternOperator OpNode> {
8616 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8617 V64, V64, vecshiftL8,
8619 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
8620 (i32 vecshiftL8:$imm)))]> {
8622 let Inst{18-16} = imm;
8625 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8626 V128, V128, vecshiftL8,
8627 asm, ".16b", ".16b",
8628 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
8629 (i32 vecshiftL8:$imm)))]> {
8631 let Inst{18-16} = imm;
8634 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8635 V64, V64, vecshiftL16,
8637 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
8638 (i32 vecshiftL16:$imm)))]> {
8640 let Inst{19-16} = imm;
8643 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8644 V128, V128, vecshiftL16,
8646 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
8647 (i32 vecshiftL16:$imm)))]> {
8649 let Inst{19-16} = imm;
8652 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8653 V64, V64, vecshiftL32,
8655 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
8656 (i32 vecshiftL32:$imm)))]> {
8658 let Inst{20-16} = imm;
8661 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8662 V128, V128, vecshiftL32,
8664 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
8665 (i32 vecshiftL32:$imm)))]> {
8667 let Inst{20-16} = imm;
8670 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8671 V128, V128, vecshiftL64,
8673 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
8674 (i32 vecshiftL64:$imm)))]> {
8676 let Inst{21-16} = imm;
8680 multiclass SIMDVectorRShiftBHSD<bit U, bits<5> opc, string asm,
8681 SDPatternOperator OpNode> {
8682 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8683 V64, V64, vecshiftR8,
8685 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
8686 (i32 vecshiftR8:$imm)))]> {
8688 let Inst{18-16} = imm;
8691 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8692 V128, V128, vecshiftR8,
8693 asm, ".16b", ".16b",
8694 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
8695 (i32 vecshiftR8:$imm)))]> {
8697 let Inst{18-16} = imm;
8700 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8701 V64, V64, vecshiftR16,
8703 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
8704 (i32 vecshiftR16:$imm)))]> {
8706 let Inst{19-16} = imm;
8709 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8710 V128, V128, vecshiftR16,
8712 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
8713 (i32 vecshiftR16:$imm)))]> {
8715 let Inst{19-16} = imm;
8718 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8719 V64, V64, vecshiftR32,
8721 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
8722 (i32 vecshiftR32:$imm)))]> {
8724 let Inst{20-16} = imm;
8727 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8728 V128, V128, vecshiftR32,
8730 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
8731 (i32 vecshiftR32:$imm)))]> {
8733 let Inst{20-16} = imm;
8736 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8737 V128, V128, vecshiftR64,
8739 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
8740 (i32 vecshiftR64:$imm)))]> {
8742 let Inst{21-16} = imm;
8746 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8747 multiclass SIMDVectorRShiftBHSDTied<bit U, bits<5> opc, string asm,
8748 SDPatternOperator OpNode = null_frag> {
8749 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
8750 V64, V64, vecshiftR8, asm, ".8b", ".8b",
8751 [(set (v8i8 V64:$dst),
8752 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
8753 (i32 vecshiftR8:$imm)))]> {
8755 let Inst{18-16} = imm;
8758 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8759 V128, V128, vecshiftR8, asm, ".16b", ".16b",
8760 [(set (v16i8 V128:$dst),
8761 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
8762 (i32 vecshiftR8:$imm)))]> {
8764 let Inst{18-16} = imm;
8767 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
8768 V64, V64, vecshiftR16, asm, ".4h", ".4h",
8769 [(set (v4i16 V64:$dst),
8770 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
8771 (i32 vecshiftR16:$imm)))]> {
8773 let Inst{19-16} = imm;
8776 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8777 V128, V128, vecshiftR16, asm, ".8h", ".8h",
8778 [(set (v8i16 V128:$dst),
8779 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
8780 (i32 vecshiftR16:$imm)))]> {
8782 let Inst{19-16} = imm;
8785 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
8786 V64, V64, vecshiftR32, asm, ".2s", ".2s",
8787 [(set (v2i32 V64:$dst),
8788 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
8789 (i32 vecshiftR32:$imm)))]> {
8791 let Inst{20-16} = imm;
8794 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8795 V128, V128, vecshiftR32, asm, ".4s", ".4s",
8796 [(set (v4i32 V128:$dst),
8797 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8798 (i32 vecshiftR32:$imm)))]> {
8800 let Inst{20-16} = imm;
8803 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
8804 V128, V128, vecshiftR64,
8805 asm, ".2d", ".2d", [(set (v2i64 V128:$dst),
8806 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
8807 (i32 vecshiftR64:$imm)))]> {
8809 let Inst{21-16} = imm;
8813 multiclass SIMDVectorLShiftBHSDTied<bit U, bits<5> opc, string asm,
8814 SDPatternOperator OpNode = null_frag> {
8815 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
8816 V64, V64, vecshiftL8,
8818 [(set (v8i8 V64:$dst),
8819 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
8820 (i32 vecshiftL8:$imm)))]> {
8822 let Inst{18-16} = imm;
8825 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8826 V128, V128, vecshiftL8,
8827 asm, ".16b", ".16b",
8828 [(set (v16i8 V128:$dst),
8829 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
8830 (i32 vecshiftL8:$imm)))]> {
8832 let Inst{18-16} = imm;
8835 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
8836 V64, V64, vecshiftL16,
8838 [(set (v4i16 V64:$dst),
8839 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
8840 (i32 vecshiftL16:$imm)))]> {
8842 let Inst{19-16} = imm;
8845 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8846 V128, V128, vecshiftL16,
8848 [(set (v8i16 V128:$dst),
8849 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
8850 (i32 vecshiftL16:$imm)))]> {
8852 let Inst{19-16} = imm;
8855 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
8856 V64, V64, vecshiftL32,
8858 [(set (v2i32 V64:$dst),
8859 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
8860 (i32 vecshiftL32:$imm)))]> {
8862 let Inst{20-16} = imm;
8865 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8866 V128, V128, vecshiftL32,
8868 [(set (v4i32 V128:$dst),
8869 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8870 (i32 vecshiftL32:$imm)))]> {
8872 let Inst{20-16} = imm;
8875 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
8876 V128, V128, vecshiftL64,
8878 [(set (v2i64 V128:$dst),
8879 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
8880 (i32 vecshiftL64:$imm)))]> {
8882 let Inst{21-16} = imm;
8886 multiclass SIMDVectorLShiftLongBHSD<bit U, bits<5> opc, string asm,
8887 SDPatternOperator OpNode> {
8888 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8889 V128, V64, vecshiftL8, asm, ".8h", ".8b",
8890 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), vecshiftL8:$imm))]> {
8892 let Inst{18-16} = imm;
8895 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8896 V128, V128, vecshiftL8,
8897 asm#"2", ".8h", ".16b",
8898 [(set (v8i16 V128:$Rd),
8899 (OpNode (extract_high_v16i8 V128:$Rn), vecshiftL8:$imm))]> {
8901 let Inst{18-16} = imm;
8904 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8905 V128, V64, vecshiftL16, asm, ".4s", ".4h",
8906 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), vecshiftL16:$imm))]> {
8908 let Inst{19-16} = imm;
8911 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8912 V128, V128, vecshiftL16,
8913 asm#"2", ".4s", ".8h",
8914 [(set (v4i32 V128:$Rd),
8915 (OpNode (extract_high_v8i16 V128:$Rn), vecshiftL16:$imm))]> {
8918 let Inst{19-16} = imm;
8921 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8922 V128, V64, vecshiftL32, asm, ".2d", ".2s",
8923 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), vecshiftL32:$imm))]> {
8925 let Inst{20-16} = imm;
8928 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8929 V128, V128, vecshiftL32,
8930 asm#"2", ".2d", ".4s",
8931 [(set (v2i64 V128:$Rd),
8932 (OpNode (extract_high_v4i32 V128:$Rn), vecshiftL32:$imm))]> {
8934 let Inst{20-16} = imm;
8940 // Vector load/store
8942 // SIMD ldX/stX no-index memory references don't allow the optional
8943 // ", #0" constant and handle post-indexing explicitly, so we use
8944 // a more specialized parse method for them. Otherwise, it's the same as
8945 // the general GPR64sp handling.
8947 class BaseSIMDLdSt<bit Q, bit L, bits<4> opcode, bits<2> size,
8948 string asm, dag oops, dag iops, list<dag> pattern>
8949 : I<oops, iops, asm, "\t$Vt, [$Rn]", "", pattern> {
8954 let Inst{29-23} = 0b0011000;
8956 let Inst{21-16} = 0b000000;
8957 let Inst{15-12} = opcode;
8958 let Inst{11-10} = size;
8963 class BaseSIMDLdStPost<bit Q, bit L, bits<4> opcode, bits<2> size,
8964 string asm, dag oops, dag iops>
8965 : I<oops, iops, asm, "\t$Vt, [$Rn], $Xm", "$Rn = $wback", []> {
8971 let Inst{29-23} = 0b0011001;
8974 let Inst{20-16} = Xm;
8975 let Inst{15-12} = opcode;
8976 let Inst{11-10} = size;
8981 // The immediate form of AdvSIMD post-indexed addressing is encoded with
8982 // register post-index addressing from the zero register.
8983 multiclass SIMDLdStAliases<string BaseName, string asm, string layout, string Count,
8984 int Offset, int Size> {
8985 // E.g. "ld1 { v0.8b, v1.8b }, [x1], #16"
8986 // "ld1\t$Vt, [$Rn], #16"
8987 // may get mapped to
8988 // (LD1Twov8b_POST VecListTwo8b:$Vt, GPR64sp:$Rn, XZR)
8989 def : InstAlias<asm # "\t$Vt, [$Rn], #" # Offset,
8990 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
8992 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
8995 // E.g. "ld1.8b { v0, v1 }, [x1], #16"
8996 // "ld1.8b\t$Vt, [$Rn], #16"
8997 // may get mapped to
8998 // (LD1Twov8b_POST VecListTwo64:$Vt, GPR64sp:$Rn, XZR)
8999 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], #" # Offset,
9000 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
9002 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9005 // E.g. "ld1.8b { v0, v1 }, [x1]"
9006 // "ld1\t$Vt, [$Rn]"
9007 // may get mapped to
9008 // (LD1Twov8b VecListTwo64:$Vt, GPR64sp:$Rn)
9009 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn]",
9010 (!cast<Instruction>(BaseName # Count # "v" # layout)
9011 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9014 // E.g. "ld1.8b { v0, v1 }, [x1], x2"
9015 // "ld1\t$Vt, [$Rn], $Xm"
9016 // may get mapped to
9017 // (LD1Twov8b_POST VecListTwo64:$Vt, GPR64sp:$Rn, GPR64pi8:$Xm)
9018 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], $Xm",
9019 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
9021 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9022 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9025 multiclass BaseSIMDLdN<string BaseName, string Count, string asm, string veclist,
9026 int Offset128, int Offset64, bits<4> opcode> {
9027 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
9028 def v16b: BaseSIMDLdSt<1, 1, opcode, 0b00, asm,
9029 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
9030 (ins GPR64sp:$Rn), []>;
9031 def v8h : BaseSIMDLdSt<1, 1, opcode, 0b01, asm,
9032 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
9033 (ins GPR64sp:$Rn), []>;
9034 def v4s : BaseSIMDLdSt<1, 1, opcode, 0b10, asm,
9035 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
9036 (ins GPR64sp:$Rn), []>;
9037 def v2d : BaseSIMDLdSt<1, 1, opcode, 0b11, asm,
9038 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
9039 (ins GPR64sp:$Rn), []>;
9040 def v8b : BaseSIMDLdSt<0, 1, opcode, 0b00, asm,
9041 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
9042 (ins GPR64sp:$Rn), []>;
9043 def v4h : BaseSIMDLdSt<0, 1, opcode, 0b01, asm,
9044 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
9045 (ins GPR64sp:$Rn), []>;
9046 def v2s : BaseSIMDLdSt<0, 1, opcode, 0b10, asm,
9047 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
9048 (ins GPR64sp:$Rn), []>;
9051 def v16b_POST: BaseSIMDLdStPost<1, 1, opcode, 0b00, asm,
9052 (outs GPR64sp:$wback,
9053 !cast<RegisterOperand>(veclist # "16b"):$Vt),
9055 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9056 def v8h_POST : BaseSIMDLdStPost<1, 1, opcode, 0b01, asm,
9057 (outs GPR64sp:$wback,
9058 !cast<RegisterOperand>(veclist # "8h"):$Vt),
9060 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9061 def v4s_POST : BaseSIMDLdStPost<1, 1, opcode, 0b10, asm,
9062 (outs GPR64sp:$wback,
9063 !cast<RegisterOperand>(veclist # "4s"):$Vt),
9065 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9066 def v2d_POST : BaseSIMDLdStPost<1, 1, opcode, 0b11, asm,
9067 (outs GPR64sp:$wback,
9068 !cast<RegisterOperand>(veclist # "2d"):$Vt),
9070 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9071 def v8b_POST : BaseSIMDLdStPost<0, 1, opcode, 0b00, asm,
9072 (outs GPR64sp:$wback,
9073 !cast<RegisterOperand>(veclist # "8b"):$Vt),
9075 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9076 def v4h_POST : BaseSIMDLdStPost<0, 1, opcode, 0b01, asm,
9077 (outs GPR64sp:$wback,
9078 !cast<RegisterOperand>(veclist # "4h"):$Vt),
9080 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9081 def v2s_POST : BaseSIMDLdStPost<0, 1, opcode, 0b10, asm,
9082 (outs GPR64sp:$wback,
9083 !cast<RegisterOperand>(veclist # "2s"):$Vt),
9085 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9088 defm : SIMDLdStAliases<BaseName, asm, "16b", Count, Offset128, 128>;
9089 defm : SIMDLdStAliases<BaseName, asm, "8h", Count, Offset128, 128>;
9090 defm : SIMDLdStAliases<BaseName, asm, "4s", Count, Offset128, 128>;
9091 defm : SIMDLdStAliases<BaseName, asm, "2d", Count, Offset128, 128>;
9092 defm : SIMDLdStAliases<BaseName, asm, "8b", Count, Offset64, 64>;
9093 defm : SIMDLdStAliases<BaseName, asm, "4h", Count, Offset64, 64>;
9094 defm : SIMDLdStAliases<BaseName, asm, "2s", Count, Offset64, 64>;
9097 // Only ld1/st1 has a v1d version.
9098 multiclass BaseSIMDStN<string BaseName, string Count, string asm, string veclist,
9099 int Offset128, int Offset64, bits<4> opcode> {
9100 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in {
9101 def v16b : BaseSIMDLdSt<1, 0, opcode, 0b00, asm, (outs),
9102 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
9104 def v8h : BaseSIMDLdSt<1, 0, opcode, 0b01, asm, (outs),
9105 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
9107 def v4s : BaseSIMDLdSt<1, 0, opcode, 0b10, asm, (outs),
9108 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
9110 def v2d : BaseSIMDLdSt<1, 0, opcode, 0b11, asm, (outs),
9111 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
9113 def v8b : BaseSIMDLdSt<0, 0, opcode, 0b00, asm, (outs),
9114 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
9116 def v4h : BaseSIMDLdSt<0, 0, opcode, 0b01, asm, (outs),
9117 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
9119 def v2s : BaseSIMDLdSt<0, 0, opcode, 0b10, asm, (outs),
9120 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
9123 def v16b_POST : BaseSIMDLdStPost<1, 0, opcode, 0b00, asm,
9124 (outs GPR64sp:$wback),
9125 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
9127 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9128 def v8h_POST : BaseSIMDLdStPost<1, 0, opcode, 0b01, asm,
9129 (outs GPR64sp:$wback),
9130 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
9132 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9133 def v4s_POST : BaseSIMDLdStPost<1, 0, opcode, 0b10, asm,
9134 (outs GPR64sp:$wback),
9135 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
9137 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9138 def v2d_POST : BaseSIMDLdStPost<1, 0, opcode, 0b11, asm,
9139 (outs GPR64sp:$wback),
9140 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
9142 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
9143 def v8b_POST : BaseSIMDLdStPost<0, 0, opcode, 0b00, asm,
9144 (outs GPR64sp:$wback),
9145 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
9147 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9148 def v4h_POST : BaseSIMDLdStPost<0, 0, opcode, 0b01, asm,
9149 (outs GPR64sp:$wback),
9150 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
9152 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9153 def v2s_POST : BaseSIMDLdStPost<0, 0, opcode, 0b10, asm,
9154 (outs GPR64sp:$wback),
9155 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
9157 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9160 defm : SIMDLdStAliases<BaseName, asm, "16b", Count, Offset128, 128>;
9161 defm : SIMDLdStAliases<BaseName, asm, "8h", Count, Offset128, 128>;
9162 defm : SIMDLdStAliases<BaseName, asm, "4s", Count, Offset128, 128>;
9163 defm : SIMDLdStAliases<BaseName, asm, "2d", Count, Offset128, 128>;
9164 defm : SIMDLdStAliases<BaseName, asm, "8b", Count, Offset64, 64>;
9165 defm : SIMDLdStAliases<BaseName, asm, "4h", Count, Offset64, 64>;
9166 defm : SIMDLdStAliases<BaseName, asm, "2s", Count, Offset64, 64>;
9169 multiclass BaseSIMDLd1<string BaseName, string Count, string asm, string veclist,
9170 int Offset128, int Offset64, bits<4> opcode>
9171 : BaseSIMDLdN<BaseName, Count, asm, veclist, Offset128, Offset64, opcode> {
9173 // LD1 instructions have extra "1d" variants.
9174 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
9175 def v1d : BaseSIMDLdSt<0, 1, opcode, 0b11, asm,
9176 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
9177 (ins GPR64sp:$Rn), []>;
9179 def v1d_POST : BaseSIMDLdStPost<0, 1, opcode, 0b11, asm,
9180 (outs GPR64sp:$wback,
9181 !cast<RegisterOperand>(veclist # "1d"):$Vt),
9183 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9186 defm : SIMDLdStAliases<BaseName, asm, "1d", Count, Offset64, 64>;
9189 multiclass BaseSIMDSt1<string BaseName, string Count, string asm, string veclist,
9190 int Offset128, int Offset64, bits<4> opcode>
9191 : BaseSIMDStN<BaseName, Count, asm, veclist, Offset128, Offset64, opcode> {
9193 // ST1 instructions have extra "1d" variants.
9194 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in {
9195 def v1d : BaseSIMDLdSt<0, 0, opcode, 0b11, asm, (outs),
9196 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
9199 def v1d_POST : BaseSIMDLdStPost<0, 0, opcode, 0b11, asm,
9200 (outs GPR64sp:$wback),
9201 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
9203 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
9206 defm : SIMDLdStAliases<BaseName, asm, "1d", Count, Offset64, 64>;
9209 multiclass SIMDLd1Multiple<string asm> {
9210 defm One : BaseSIMDLd1<NAME, "One", asm, "VecListOne", 16, 8, 0b0111>;
9211 defm Two : BaseSIMDLd1<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1010>;
9212 defm Three : BaseSIMDLd1<NAME, "Three", asm, "VecListThree", 48, 24, 0b0110>;
9213 defm Four : BaseSIMDLd1<NAME, "Four", asm, "VecListFour", 64, 32, 0b0010>;
9216 multiclass SIMDSt1Multiple<string asm> {
9217 defm One : BaseSIMDSt1<NAME, "One", asm, "VecListOne", 16, 8, 0b0111>;
9218 defm Two : BaseSIMDSt1<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1010>;
9219 defm Three : BaseSIMDSt1<NAME, "Three", asm, "VecListThree", 48, 24, 0b0110>;
9220 defm Four : BaseSIMDSt1<NAME, "Four", asm, "VecListFour", 64, 32, 0b0010>;
9223 multiclass SIMDLd2Multiple<string asm> {
9224 defm Two : BaseSIMDLdN<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1000>;
9227 multiclass SIMDSt2Multiple<string asm> {
9228 defm Two : BaseSIMDStN<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1000>;
9231 multiclass SIMDLd3Multiple<string asm> {
9232 defm Three : BaseSIMDLdN<NAME, "Three", asm, "VecListThree", 48, 24, 0b0100>;
9235 multiclass SIMDSt3Multiple<string asm> {
9236 defm Three : BaseSIMDStN<NAME, "Three", asm, "VecListThree", 48, 24, 0b0100>;
9239 multiclass SIMDLd4Multiple<string asm> {
9240 defm Four : BaseSIMDLdN<NAME, "Four", asm, "VecListFour", 64, 32, 0b0000>;
9243 multiclass SIMDSt4Multiple<string asm> {
9244 defm Four : BaseSIMDStN<NAME, "Four", asm, "VecListFour", 64, 32, 0b0000>;
9248 // AdvSIMD Load/store single-element
9251 class BaseSIMDLdStSingle<bit L, bit R, bits<3> opcode,
9252 string asm, string operands, string cst,
9253 dag oops, dag iops, list<dag> pattern>
9254 : I<oops, iops, asm, operands, cst, pattern> {
9258 let Inst{29-24} = 0b001101;
9261 let Inst{15-13} = opcode;
9266 class BaseSIMDLdStSingleTied<bit L, bit R, bits<3> opcode,
9267 string asm, string operands, string cst,
9268 dag oops, dag iops, list<dag> pattern>
9269 : I<oops, iops, asm, operands, "$Vt = $dst," # cst, pattern> {
9273 let Inst{29-24} = 0b001101;
9276 let Inst{15-13} = opcode;
9282 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9283 class BaseSIMDLdR<bit Q, bit R, bits<3> opcode, bit S, bits<2> size, string asm,
9284 DAGOperand listtype>
9285 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, [$Rn]", "",
9286 (outs listtype:$Vt), (ins GPR64sp:$Rn),
9290 let Inst{20-16} = 0b00000;
9292 let Inst{11-10} = size;
9294 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9295 class BaseSIMDLdRPost<bit Q, bit R, bits<3> opcode, bit S, bits<2> size,
9296 string asm, DAGOperand listtype, DAGOperand GPR64pi>
9297 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, [$Rn], $Xm",
9299 (outs GPR64sp:$wback, listtype:$Vt),
9300 (ins GPR64sp:$Rn, GPR64pi:$Xm), []> {
9304 let Inst{20-16} = Xm;
9306 let Inst{11-10} = size;
9309 multiclass SIMDLdrAliases<string BaseName, string asm, string layout, string Count,
9310 int Offset, int Size> {
9311 // E.g. "ld1r { v0.8b }, [x1], #1"
9312 // "ld1r.8b\t$Vt, [$Rn], #1"
9313 // may get mapped to
9314 // (LD1Rv8b_POST VecListOne8b:$Vt, GPR64sp:$Rn, XZR)
9315 def : InstAlias<asm # "\t$Vt, [$Rn], #" # Offset,
9316 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9318 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
9321 // E.g. "ld1r.8b { v0 }, [x1], #1"
9322 // "ld1r.8b\t$Vt, [$Rn], #1"
9323 // may get mapped to
9324 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, XZR)
9325 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], #" # Offset,
9326 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9328 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9331 // E.g. "ld1r.8b { v0 }, [x1]"
9332 // "ld1r.8b\t$Vt, [$Rn]"
9333 // may get mapped to
9334 // (LD1Rv8b VecListOne64:$Vt, GPR64sp:$Rn)
9335 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn]",
9336 (!cast<Instruction>(BaseName # "v" # layout)
9337 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9340 // E.g. "ld1r.8b { v0 }, [x1], x2"
9341 // "ld1r.8b\t$Vt, [$Rn], $Xm"
9342 // may get mapped to
9343 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, GPR64pi1:$Xm)
9344 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], $Xm",
9345 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9347 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9348 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9351 multiclass SIMDLdR<bit R, bits<3> opcode, bit S, string asm, string Count,
9352 int Offset1, int Offset2, int Offset4, int Offset8> {
9353 def v8b : BaseSIMDLdR<0, R, opcode, S, 0b00, asm,
9354 !cast<DAGOperand>("VecList" # Count # "8b")>;
9355 def v16b: BaseSIMDLdR<1, R, opcode, S, 0b00, asm,
9356 !cast<DAGOperand>("VecList" # Count #"16b")>;
9357 def v4h : BaseSIMDLdR<0, R, opcode, S, 0b01, asm,
9358 !cast<DAGOperand>("VecList" # Count #"4h")>;
9359 def v8h : BaseSIMDLdR<1, R, opcode, S, 0b01, asm,
9360 !cast<DAGOperand>("VecList" # Count #"8h")>;
9361 def v2s : BaseSIMDLdR<0, R, opcode, S, 0b10, asm,
9362 !cast<DAGOperand>("VecList" # Count #"2s")>;
9363 def v4s : BaseSIMDLdR<1, R, opcode, S, 0b10, asm,
9364 !cast<DAGOperand>("VecList" # Count #"4s")>;
9365 def v1d : BaseSIMDLdR<0, R, opcode, S, 0b11, asm,
9366 !cast<DAGOperand>("VecList" # Count #"1d")>;
9367 def v2d : BaseSIMDLdR<1, R, opcode, S, 0b11, asm,
9368 !cast<DAGOperand>("VecList" # Count #"2d")>;
9370 def v8b_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b00, asm,
9371 !cast<DAGOperand>("VecList" # Count # "8b"),
9372 !cast<DAGOperand>("GPR64pi" # Offset1)>;
9373 def v16b_POST: BaseSIMDLdRPost<1, R, opcode, S, 0b00, asm,
9374 !cast<DAGOperand>("VecList" # Count # "16b"),
9375 !cast<DAGOperand>("GPR64pi" # Offset1)>;
9376 def v4h_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b01, asm,
9377 !cast<DAGOperand>("VecList" # Count # "4h"),
9378 !cast<DAGOperand>("GPR64pi" # Offset2)>;
9379 def v8h_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b01, asm,
9380 !cast<DAGOperand>("VecList" # Count # "8h"),
9381 !cast<DAGOperand>("GPR64pi" # Offset2)>;
9382 def v2s_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b10, asm,
9383 !cast<DAGOperand>("VecList" # Count # "2s"),
9384 !cast<DAGOperand>("GPR64pi" # Offset4)>;
9385 def v4s_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b10, asm,
9386 !cast<DAGOperand>("VecList" # Count # "4s"),
9387 !cast<DAGOperand>("GPR64pi" # Offset4)>;
9388 def v1d_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b11, asm,
9389 !cast<DAGOperand>("VecList" # Count # "1d"),
9390 !cast<DAGOperand>("GPR64pi" # Offset8)>;
9391 def v2d_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b11, asm,
9392 !cast<DAGOperand>("VecList" # Count # "2d"),
9393 !cast<DAGOperand>("GPR64pi" # Offset8)>;
9395 defm : SIMDLdrAliases<NAME, asm, "8b", Count, Offset1, 64>;
9396 defm : SIMDLdrAliases<NAME, asm, "16b", Count, Offset1, 128>;
9397 defm : SIMDLdrAliases<NAME, asm, "4h", Count, Offset2, 64>;
9398 defm : SIMDLdrAliases<NAME, asm, "8h", Count, Offset2, 128>;
9399 defm : SIMDLdrAliases<NAME, asm, "2s", Count, Offset4, 64>;
9400 defm : SIMDLdrAliases<NAME, asm, "4s", Count, Offset4, 128>;
9401 defm : SIMDLdrAliases<NAME, asm, "1d", Count, Offset8, 64>;
9402 defm : SIMDLdrAliases<NAME, asm, "2d", Count, Offset8, 128>;
9405 class SIMDLdStSingleB<bit L, bit R, bits<3> opcode, string asm,
9406 dag oops, dag iops, list<dag> pattern>
9407 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9409 // idx encoded in Q:S:size fields.
9411 let Inst{30} = idx{3};
9413 let Inst{20-16} = 0b00000;
9414 let Inst{12} = idx{2};
9415 let Inst{11-10} = idx{1-0};
9417 class SIMDLdStSingleBTied<bit L, bit R, bits<3> opcode, string asm,
9418 dag oops, dag iops, list<dag> pattern>
9419 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9420 oops, iops, pattern> {
9421 // idx encoded in Q:S:size fields.
9423 let Inst{30} = idx{3};
9425 let Inst{20-16} = 0b00000;
9426 let Inst{12} = idx{2};
9427 let Inst{11-10} = idx{1-0};
9429 class SIMDLdStSingleBPost<bit L, bit R, bits<3> opcode, string asm,
9431 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9432 "$Rn = $wback", oops, iops, []> {
9433 // idx encoded in Q:S:size fields.
9436 let Inst{30} = idx{3};
9438 let Inst{20-16} = Xm;
9439 let Inst{12} = idx{2};
9440 let Inst{11-10} = idx{1-0};
9442 class SIMDLdStSingleBTiedPost<bit L, bit R, bits<3> opcode, string asm,
9444 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9445 "$Rn = $wback", oops, iops, []> {
9446 // idx encoded in Q:S:size fields.
9449 let Inst{30} = idx{3};
9451 let Inst{20-16} = Xm;
9452 let Inst{12} = idx{2};
9453 let Inst{11-10} = idx{1-0};
9456 class SIMDLdStSingleH<bit L, bit R, bits<3> opcode, bit size, string asm,
9457 dag oops, dag iops, list<dag> pattern>
9458 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9460 // idx encoded in Q:S:size<1> fields.
9462 let Inst{30} = idx{2};
9464 let Inst{20-16} = 0b00000;
9465 let Inst{12} = idx{1};
9466 let Inst{11} = idx{0};
9467 let Inst{10} = size;
9469 class SIMDLdStSingleHTied<bit L, bit R, bits<3> opcode, bit size, string asm,
9470 dag oops, dag iops, list<dag> pattern>
9471 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9472 oops, iops, pattern> {
9473 // idx encoded in Q:S:size<1> fields.
9475 let Inst{30} = idx{2};
9477 let Inst{20-16} = 0b00000;
9478 let Inst{12} = idx{1};
9479 let Inst{11} = idx{0};
9480 let Inst{10} = size;
9483 class SIMDLdStSingleHPost<bit L, bit R, bits<3> opcode, bit size, string asm,
9485 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9486 "$Rn = $wback", oops, iops, []> {
9487 // idx encoded in Q:S:size<1> fields.
9490 let Inst{30} = idx{2};
9492 let Inst{20-16} = Xm;
9493 let Inst{12} = idx{1};
9494 let Inst{11} = idx{0};
9495 let Inst{10} = size;
9497 class SIMDLdStSingleHTiedPost<bit L, bit R, bits<3> opcode, bit size, string asm,
9499 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9500 "$Rn = $wback", oops, iops, []> {
9501 // idx encoded in Q:S:size<1> fields.
9504 let Inst{30} = idx{2};
9506 let Inst{20-16} = Xm;
9507 let Inst{12} = idx{1};
9508 let Inst{11} = idx{0};
9509 let Inst{10} = size;
9511 class SIMDLdStSingleS<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9512 dag oops, dag iops, list<dag> pattern>
9513 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9515 // idx encoded in Q:S fields.
9517 let Inst{30} = idx{1};
9519 let Inst{20-16} = 0b00000;
9520 let Inst{12} = idx{0};
9521 let Inst{11-10} = size;
9523 class SIMDLdStSingleSTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9524 dag oops, dag iops, list<dag> pattern>
9525 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9526 oops, iops, pattern> {
9527 // idx encoded in Q:S fields.
9529 let Inst{30} = idx{1};
9531 let Inst{20-16} = 0b00000;
9532 let Inst{12} = idx{0};
9533 let Inst{11-10} = size;
9535 class SIMDLdStSingleSPost<bit L, bit R, bits<3> opcode, bits<2> size,
9536 string asm, dag oops, dag iops>
9537 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9538 "$Rn = $wback", oops, iops, []> {
9539 // idx encoded in Q:S fields.
9542 let Inst{30} = idx{1};
9544 let Inst{20-16} = Xm;
9545 let Inst{12} = idx{0};
9546 let Inst{11-10} = size;
9548 class SIMDLdStSingleSTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
9549 string asm, dag oops, dag iops>
9550 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9551 "$Rn = $wback", oops, iops, []> {
9552 // idx encoded in Q:S fields.
9555 let Inst{30} = idx{1};
9557 let Inst{20-16} = Xm;
9558 let Inst{12} = idx{0};
9559 let Inst{11-10} = size;
9561 class SIMDLdStSingleD<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9562 dag oops, dag iops, list<dag> pattern>
9563 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9565 // idx encoded in Q field.
9569 let Inst{20-16} = 0b00000;
9571 let Inst{11-10} = size;
9573 class SIMDLdStSingleDTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9574 dag oops, dag iops, list<dag> pattern>
9575 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9576 oops, iops, pattern> {
9577 // idx encoded in Q field.
9581 let Inst{20-16} = 0b00000;
9583 let Inst{11-10} = size;
9585 class SIMDLdStSingleDPost<bit L, bit R, bits<3> opcode, bits<2> size,
9586 string asm, dag oops, dag iops>
9587 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9588 "$Rn = $wback", oops, iops, []> {
9589 // idx encoded in Q field.
9594 let Inst{20-16} = Xm;
9596 let Inst{11-10} = size;
9598 class SIMDLdStSingleDTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
9599 string asm, dag oops, dag iops>
9600 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9601 "$Rn = $wback", oops, iops, []> {
9602 // idx encoded in Q field.
9607 let Inst{20-16} = Xm;
9609 let Inst{11-10} = size;
9612 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9613 multiclass SIMDLdSingleBTied<bit R, bits<3> opcode, string asm,
9614 RegisterOperand listtype,
9615 RegisterOperand GPR64pi> {
9616 def i8 : SIMDLdStSingleBTied<1, R, opcode, asm,
9617 (outs listtype:$dst),
9618 (ins listtype:$Vt, VectorIndexB:$idx,
9621 def i8_POST : SIMDLdStSingleBTiedPost<1, R, opcode, asm,
9622 (outs GPR64sp:$wback, listtype:$dst),
9623 (ins listtype:$Vt, VectorIndexB:$idx,
9624 GPR64sp:$Rn, GPR64pi:$Xm)>;
9626 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9627 multiclass SIMDLdSingleHTied<bit R, bits<3> opcode, bit size, string asm,
9628 RegisterOperand listtype,
9629 RegisterOperand GPR64pi> {
9630 def i16 : SIMDLdStSingleHTied<1, R, opcode, size, asm,
9631 (outs listtype:$dst),
9632 (ins listtype:$Vt, VectorIndexH:$idx,
9635 def i16_POST : SIMDLdStSingleHTiedPost<1, R, opcode, size, asm,
9636 (outs GPR64sp:$wback, listtype:$dst),
9637 (ins listtype:$Vt, VectorIndexH:$idx,
9638 GPR64sp:$Rn, GPR64pi:$Xm)>;
9640 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9641 multiclass SIMDLdSingleSTied<bit R, bits<3> opcode, bits<2> size,string asm,
9642 RegisterOperand listtype,
9643 RegisterOperand GPR64pi> {
9644 def i32 : SIMDLdStSingleSTied<1, R, opcode, size, asm,
9645 (outs listtype:$dst),
9646 (ins listtype:$Vt, VectorIndexS:$idx,
9649 def i32_POST : SIMDLdStSingleSTiedPost<1, R, opcode, size, asm,
9650 (outs GPR64sp:$wback, listtype:$dst),
9651 (ins listtype:$Vt, VectorIndexS:$idx,
9652 GPR64sp:$Rn, GPR64pi:$Xm)>;
9654 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9655 multiclass SIMDLdSingleDTied<bit R, bits<3> opcode, bits<2> size, string asm,
9656 RegisterOperand listtype, RegisterOperand GPR64pi> {
9657 def i64 : SIMDLdStSingleDTied<1, R, opcode, size, asm,
9658 (outs listtype:$dst),
9659 (ins listtype:$Vt, VectorIndexD:$idx,
9662 def i64_POST : SIMDLdStSingleDTiedPost<1, R, opcode, size, asm,
9663 (outs GPR64sp:$wback, listtype:$dst),
9664 (ins listtype:$Vt, VectorIndexD:$idx,
9665 GPR64sp:$Rn, GPR64pi:$Xm)>;
9667 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9668 multiclass SIMDStSingleB<bit R, bits<3> opcode, string asm,
9669 RegisterOperand listtype, RegisterOperand GPR64pi> {
9670 def i8 : SIMDLdStSingleB<0, R, opcode, asm,
9671 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
9674 def i8_POST : SIMDLdStSingleBPost<0, R, opcode, asm,
9675 (outs GPR64sp:$wback),
9676 (ins listtype:$Vt, VectorIndexB:$idx,
9677 GPR64sp:$Rn, GPR64pi:$Xm)>;
9679 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9680 multiclass SIMDStSingleH<bit R, bits<3> opcode, bit size, string asm,
9681 RegisterOperand listtype, RegisterOperand GPR64pi> {
9682 def i16 : SIMDLdStSingleH<0, R, opcode, size, asm,
9683 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
9686 def i16_POST : SIMDLdStSingleHPost<0, R, opcode, size, asm,
9687 (outs GPR64sp:$wback),
9688 (ins listtype:$Vt, VectorIndexH:$idx,
9689 GPR64sp:$Rn, GPR64pi:$Xm)>;
9691 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9692 multiclass SIMDStSingleS<bit R, bits<3> opcode, bits<2> size,string asm,
9693 RegisterOperand listtype, RegisterOperand GPR64pi> {
9694 def i32 : SIMDLdStSingleS<0, R, opcode, size, asm,
9695 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
9698 def i32_POST : SIMDLdStSingleSPost<0, R, opcode, size, asm,
9699 (outs GPR64sp:$wback),
9700 (ins listtype:$Vt, VectorIndexS:$idx,
9701 GPR64sp:$Rn, GPR64pi:$Xm)>;
9703 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9704 multiclass SIMDStSingleD<bit R, bits<3> opcode, bits<2> size, string asm,
9705 RegisterOperand listtype, RegisterOperand GPR64pi> {
9706 def i64 : SIMDLdStSingleD<0, R, opcode, size, asm,
9707 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
9710 def i64_POST : SIMDLdStSingleDPost<0, R, opcode, size, asm,
9711 (outs GPR64sp:$wback),
9712 (ins listtype:$Vt, VectorIndexD:$idx,
9713 GPR64sp:$Rn, GPR64pi:$Xm)>;
9716 multiclass SIMDLdStSingleAliases<string asm, string layout, string Type,
9717 string Count, int Offset, Operand idxtype> {
9718 // E.g. "ld1 { v0.8b }[0], [x1], #1"
9719 // "ld1\t$Vt, [$Rn], #1"
9720 // may get mapped to
9721 // (LD1Rv8b_POST VecListOne8b:$Vt, GPR64sp:$Rn, XZR)
9722 def : InstAlias<asm # "\t$Vt$idx, [$Rn], #" # Offset,
9723 (!cast<Instruction>(NAME # Type # "_POST")
9725 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
9726 idxtype:$idx, XZR), 1>;
9728 // E.g. "ld1.8b { v0 }[0], [x1], #1"
9729 // "ld1.8b\t$Vt, [$Rn], #1"
9730 // may get mapped to
9731 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, XZR)
9732 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn], #" # Offset,
9733 (!cast<Instruction>(NAME # Type # "_POST")
9735 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9736 idxtype:$idx, XZR), 0>;
9738 // E.g. "ld1.8b { v0 }[0], [x1]"
9739 // "ld1.8b\t$Vt, [$Rn]"
9740 // may get mapped to
9741 // (LD1Rv8b VecListOne64:$Vt, GPR64sp:$Rn)
9742 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn]",
9743 (!cast<Instruction>(NAME # Type)
9744 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9745 idxtype:$idx, GPR64sp:$Rn), 0>;
9747 // E.g. "ld1.8b { v0 }[0], [x1], x2"
9748 // "ld1.8b\t$Vt, [$Rn], $Xm"
9749 // may get mapped to
9750 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, GPR64pi1:$Xm)
9751 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn], $Xm",
9752 (!cast<Instruction>(NAME # Type # "_POST")
9754 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9756 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9759 multiclass SIMDLdSt1SingleAliases<string asm> {
9760 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "One", 1, VectorIndexB>;
9761 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "One", 2, VectorIndexH>;
9762 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "One", 4, VectorIndexS>;
9763 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "One", 8, VectorIndexD>;
9766 multiclass SIMDLdSt2SingleAliases<string asm> {
9767 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Two", 2, VectorIndexB>;
9768 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Two", 4, VectorIndexH>;
9769 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Two", 8, VectorIndexS>;
9770 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Two", 16, VectorIndexD>;
9773 multiclass SIMDLdSt3SingleAliases<string asm> {
9774 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Three", 3, VectorIndexB>;
9775 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Three", 6, VectorIndexH>;
9776 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Three", 12, VectorIndexS>;
9777 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Three", 24, VectorIndexD>;
9780 multiclass SIMDLdSt4SingleAliases<string asm> {
9781 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Four", 4, VectorIndexB>;
9782 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Four", 8, VectorIndexH>;
9783 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Four", 16, VectorIndexS>;
9784 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Four", 32, VectorIndexD>;
9786 } // end of 'let Predicates = [HasNEON]'
9788 //----------------------------------------------------------------------------
9789 // AdvSIMD v8.1 Rounding Double Multiply Add/Subtract
9790 //----------------------------------------------------------------------------
9792 let Predicates = [HasNEON, HasRDM] in {
9794 class BaseSIMDThreeSameVectorTiedR0<bit Q, bit U, bits<2> size, bits<5> opcode,
9795 RegisterOperand regtype, string asm,
9796 string kind, list<dag> pattern>
9797 : BaseSIMDThreeSameVectorTied<Q, U, {size,0}, opcode, regtype, asm, kind,
9800 multiclass SIMDThreeSameVectorSQRDMLxHTiedHS<bit U, bits<5> opc, string asm,
9801 SDPatternOperator Accum> {
9802 def v4i16 : BaseSIMDThreeSameVectorTiedR0<0, U, 0b01, opc, V64, asm, ".4h",
9803 [(set (v4i16 V64:$dst),
9804 (Accum (v4i16 V64:$Rd),
9805 (v4i16 (int_aarch64_neon_sqrdmulh (v4i16 V64:$Rn),
9806 (v4i16 V64:$Rm)))))]>;
9807 def v8i16 : BaseSIMDThreeSameVectorTiedR0<1, U, 0b01, opc, V128, asm, ".8h",
9808 [(set (v8i16 V128:$dst),
9809 (Accum (v8i16 V128:$Rd),
9810 (v8i16 (int_aarch64_neon_sqrdmulh (v8i16 V128:$Rn),
9811 (v8i16 V128:$Rm)))))]>;
9812 def v2i32 : BaseSIMDThreeSameVectorTiedR0<0, U, 0b10, opc, V64, asm, ".2s",
9813 [(set (v2i32 V64:$dst),
9814 (Accum (v2i32 V64:$Rd),
9815 (v2i32 (int_aarch64_neon_sqrdmulh (v2i32 V64:$Rn),
9816 (v2i32 V64:$Rm)))))]>;
9817 def v4i32 : BaseSIMDThreeSameVectorTiedR0<1, U, 0b10, opc, V128, asm, ".4s",
9818 [(set (v4i32 V128:$dst),
9819 (Accum (v4i32 V128:$Rd),
9820 (v4i32 (int_aarch64_neon_sqrdmulh (v4i32 V128:$Rn),
9821 (v4i32 V128:$Rm)))))]>;
9824 multiclass SIMDIndexedSQRDMLxHSDTied<bit U, bits<4> opc, string asm,
9825 SDPatternOperator Accum> {
9826 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
9827 V64, V64, V128_lo, VectorIndexH,
9828 asm, ".4h", ".4h", ".4h", ".h",
9829 [(set (v4i16 V64:$dst),
9830 (Accum (v4i16 V64:$Rd),
9831 (v4i16 (int_aarch64_neon_sqrdmulh
9833 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
9834 VectorIndexH:$idx))))))]> {
9836 let Inst{11} = idx{2};
9837 let Inst{21} = idx{1};
9838 let Inst{20} = idx{0};
9841 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
9842 V128, V128, V128_lo, VectorIndexH,
9843 asm, ".8h", ".8h", ".8h", ".h",
9844 [(set (v8i16 V128:$dst),
9845 (Accum (v8i16 V128:$Rd),
9846 (v8i16 (int_aarch64_neon_sqrdmulh
9848 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
9849 VectorIndexH:$idx))))))]> {
9851 let Inst{11} = idx{2};
9852 let Inst{21} = idx{1};
9853 let Inst{20} = idx{0};
9856 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
9857 V64, V64, V128, VectorIndexS,
9858 asm, ".2s", ".2s", ".2s", ".s",
9859 [(set (v2i32 V64:$dst),
9860 (Accum (v2i32 V64:$Rd),
9861 (v2i32 (int_aarch64_neon_sqrdmulh
9863 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm),
9864 VectorIndexS:$idx))))))]> {
9866 let Inst{11} = idx{1};
9867 let Inst{21} = idx{0};
9870 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but
9871 // an intermediate EXTRACT_SUBREG would be untyped.
9872 // FIXME: direct EXTRACT_SUBREG from v2i32 to i32 is illegal, that's why we
9873 // got it lowered here as (i32 vector_extract (v4i32 insert_subvector(..)))
9874 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
9875 (i32 (vector_extract
9876 (v4i32 (insert_subvector
9878 (v2i32 (int_aarch64_neon_sqrdmulh
9880 (v2i32 (AArch64duplane32
9882 VectorIndexS:$idx)))),
9886 (v2i32 (!cast<Instruction>(NAME # v2i32_indexed)
9887 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
9892 VectorIndexS:$idx)),
9895 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
9896 V128, V128, V128, VectorIndexS,
9897 asm, ".4s", ".4s", ".4s", ".s",
9898 [(set (v4i32 V128:$dst),
9899 (Accum (v4i32 V128:$Rd),
9900 (v4i32 (int_aarch64_neon_sqrdmulh
9902 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
9903 VectorIndexS:$idx))))))]> {
9905 let Inst{11} = idx{1};
9906 let Inst{21} = idx{0};
9909 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but
9910 // an intermediate EXTRACT_SUBREG would be untyped.
9911 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
9912 (i32 (vector_extract
9913 (v4i32 (int_aarch64_neon_sqrdmulh
9915 (v4i32 (AArch64duplane32
9917 VectorIndexS:$idx)))),
9920 (v4i32 (!cast<Instruction>(NAME # v4i32_indexed)
9921 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
9926 VectorIndexS:$idx)),
9929 def i16_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
9930 FPR16Op, FPR16Op, V128_lo,
9931 VectorIndexH, asm, ".h", "", "", ".h",
9934 let Inst{11} = idx{2};
9935 let Inst{21} = idx{1};
9936 let Inst{20} = idx{0};
9939 def i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
9940 FPR32Op, FPR32Op, V128, VectorIndexS,
9941 asm, ".s", "", "", ".s",
9942 [(set (i32 FPR32Op:$dst),
9943 (Accum (i32 FPR32Op:$Rd),
9944 (i32 (int_aarch64_neon_sqrdmulh
9946 (i32 (vector_extract (v4i32 V128:$Rm),
9947 VectorIndexS:$idx))))))]> {
9949 let Inst{11} = idx{1};
9950 let Inst{21} = idx{0};
9953 } // let Predicates = [HasNeon, HasRDM]
9955 //----------------------------------------------------------------------------
9956 // ARMv8.3 Complex ADD/MLA instructions
9957 //----------------------------------------------------------------------------
9959 class ComplexRotationOperand<int Angle, int Remainder, string Type>
9961 let PredicateMethod = "isComplexRotation<" # Angle # ", " # Remainder # ">";
9962 let DiagnosticType = "InvalidComplexRotation" # Type;
9963 let Name = "ComplexRotation" # Type;
9965 def complexrotateop : Operand<i32> {
9966 let ParserMatchClass = ComplexRotationOperand<90, 0, "Even">;
9967 let PrintMethod = "printComplexRotationOp<90, 0>";
9969 def complexrotateopodd : Operand<i32> {
9970 let ParserMatchClass = ComplexRotationOperand<180, 90, "Odd">;
9971 let PrintMethod = "printComplexRotationOp<180, 90>";
9974 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
9975 class BaseSIMDThreeSameVectorComplex<bit Q, bit U, bits<2> size, bits<3> opcode,
9976 RegisterOperand regtype, Operand rottype,
9977 string asm, string kind, list<dag> pattern>
9978 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, rottype:$rot), asm,
9979 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $rot"
9980 "|" # kind # "\t$Rd, $Rn, $Rm, $rot}", "", pattern>,
9989 let Inst{28-24} = 0b01110;
9990 let Inst{23-22} = size;
9992 let Inst{20-16} = Rm;
9993 let Inst{15-13} = opcode;
9994 // Non-tied version (FCADD) only has one rotation bit
10002 //8.3 CompNum - Floating-point complex number support
10003 multiclass SIMDThreeSameVectorComplexHSD<bit U, bits<3> opcode, Operand rottype,
10004 string asm, SDPatternOperator OpNode>{
10005 let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
10006 def v4f16 : BaseSIMDThreeSameVectorComplex<0, U, 0b01, opcode, V64, rottype,
10008 [(set (v4f16 V64:$dst), (OpNode (v4f16 V64:$Rd),
10011 (rottype i32:$rot)))]>;
10013 def v8f16 : BaseSIMDThreeSameVectorComplex<1, U, 0b01, opcode, V128, rottype,
10015 [(set (v8f16 V128:$dst), (OpNode (v8f16 V128:$Rd),
10018 (rottype i32:$rot)))]>;
10021 let Predicates = [HasComplxNum, HasNEON] in {
10022 def v2f32 : BaseSIMDThreeSameVectorComplex<0, U, 0b10, opcode, V64, rottype,
10024 [(set (v2f32 V64:$dst), (OpNode (v2f32 V64:$Rd),
10027 (rottype i32:$rot)))]>;
10029 def v4f32 : BaseSIMDThreeSameVectorComplex<1, U, 0b10, opcode, V128, rottype,
10031 [(set (v4f32 V128:$dst), (OpNode (v4f32 V128:$Rd),
10034 (rottype i32:$rot)))]>;
10036 def v2f64 : BaseSIMDThreeSameVectorComplex<1, U, 0b11, opcode, V128, rottype,
10038 [(set (v2f64 V128:$dst), (OpNode (v2f64 V128:$Rd),
10041 (rottype i32:$rot)))]>;
10045 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10046 class BaseSIMDThreeSameVectorTiedComplex<bit Q, bit U, bits<2> size,
10048 RegisterOperand regtype,
10049 Operand rottype, string asm,
10050 string kind, list<dag> pattern>
10051 : I<(outs regtype:$dst),
10052 (ins regtype:$Rd, regtype:$Rn, regtype:$Rm, rottype:$rot), asm,
10053 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $rot"
10054 "|" # kind # "\t$Rd, $Rn, $Rm, $rot}", "$Rd = $dst", pattern>,
10063 let Inst{28-24} = 0b01110;
10064 let Inst{23-22} = size;
10066 let Inst{20-16} = Rm;
10067 let Inst{15-13} = opcode;
10068 let Inst{12-11} = rot;
10070 let Inst{9-5} = Rn;
10071 let Inst{4-0} = Rd;
10074 multiclass SIMDThreeSameVectorTiedComplexHSD<bit U, bits<3> opcode,
10075 Operand rottype, string asm,
10076 SDPatternOperator OpNode> {
10077 let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
10078 def v4f16 : BaseSIMDThreeSameVectorTiedComplex<0, U, 0b01, opcode, V64,
10079 rottype, asm, ".4h",
10080 [(set (v4f16 V64:$dst), (OpNode (v4f16 V64:$Rd),
10083 (rottype i32:$rot)))]>;
10085 def v8f16 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b01, opcode, V128,
10086 rottype, asm, ".8h",
10087 [(set (v8f16 V128:$dst), (OpNode (v8f16 V128:$Rd),
10090 (rottype i32:$rot)))]>;
10093 let Predicates = [HasComplxNum, HasNEON] in {
10094 def v2f32 : BaseSIMDThreeSameVectorTiedComplex<0, U, 0b10, opcode, V64,
10095 rottype, asm, ".2s",
10096 [(set (v2f32 V64:$dst), (OpNode (v2f32 V64:$Rd),
10099 (rottype i32:$rot)))]>;
10101 def v4f32 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b10, opcode, V128,
10102 rottype, asm, ".4s",
10103 [(set (v4f32 V128:$dst), (OpNode (v4f32 V128:$Rd),
10106 (rottype i32:$rot)))]>;
10108 def v2f64 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b11, opcode, V128,
10109 rottype, asm, ".2d",
10110 [(set (v2f64 V128:$dst), (OpNode (v2f64 V128:$Rd),
10113 (rottype i32:$rot)))]>;
10117 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10118 class BaseSIMDIndexedTiedComplex<bit Q, bit U, bit Scalar, bits<2> size,
10119 bit opc1, bit opc2, RegisterOperand dst_reg,
10120 RegisterOperand lhs_reg,
10121 RegisterOperand rhs_reg, Operand vec_idx,
10122 Operand rottype, string asm, string apple_kind,
10123 string dst_kind, string lhs_kind,
10124 string rhs_kind, list<dag> pattern>
10125 : I<(outs dst_reg:$dst),
10126 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx, rottype:$rot),
10128 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind #
10129 "$idx, $rot" # "|" # apple_kind #
10130 "\t$Rd, $Rn, $Rm$idx, $rot}", "$Rd = $dst", pattern>,
10140 let Inst{28} = Scalar;
10141 let Inst{27-24} = 0b1111;
10142 let Inst{23-22} = size;
10143 // Bit 21 must be set by the derived class.
10144 let Inst{20-16} = Rm;
10145 let Inst{15} = opc1;
10146 let Inst{14-13} = rot;
10147 let Inst{12} = opc2;
10148 // Bit 11 must be set by the derived class.
10150 let Inst{9-5} = Rn;
10151 let Inst{4-0} = Rd;
10154 // The complex instructions index by pairs of elements, so the VectorIndexes
10155 // don't match the lane types, and the index bits are different to the other
10157 multiclass SIMDIndexedTiedComplexHSD<bit U, bit opc1, bit opc2, Operand rottype,
10158 string asm, SDPatternOperator OpNode> {
10159 let Predicates = [HasComplxNum, HasNEON, HasFullFP16] in {
10160 def v4f16_indexed : BaseSIMDIndexedTiedComplex<0, 1, 0, 0b01, opc1, opc2, V64,
10161 V64, V128, VectorIndexD, rottype, asm, ".4h", ".4h",
10165 let Inst{21} = idx{0};
10168 def v8f16_indexed : BaseSIMDIndexedTiedComplex<1, 1, 0, 0b01, opc1, opc2,
10169 V128, V128, V128, VectorIndexS, rottype, asm, ".8h",
10170 ".8h", ".8h", ".h", []> {
10172 let Inst{11} = idx{1};
10173 let Inst{21} = idx{0};
10175 } // Predicates = HasComplxNum, HasNEON, HasFullFP16]
10177 let Predicates = [HasComplxNum, HasNEON] in {
10178 def v4f32_indexed : BaseSIMDIndexedTiedComplex<1, 1, 0, 0b10, opc1, opc2,
10179 V128, V128, V128, VectorIndexD, rottype, asm, ".4s",
10180 ".4s", ".4s", ".s", []> {
10182 let Inst{11} = idx{0};
10185 } // Predicates = [HasComplxNum, HasNEON]
10188 //----------------------------------------------------------------------------
10189 // Crypto extensions
10190 //----------------------------------------------------------------------------
10192 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10193 class AESBase<bits<4> opc, string asm, dag outs, dag ins, string cstr,
10195 : I<outs, ins, asm, "{\t$Rd.16b, $Rn.16b|.16b\t$Rd, $Rn}", cstr, pat>,
10199 let Inst{31-16} = 0b0100111000101000;
10200 let Inst{15-12} = opc;
10201 let Inst{11-10} = 0b10;
10202 let Inst{9-5} = Rn;
10203 let Inst{4-0} = Rd;
10206 class AESInst<bits<4> opc, string asm, Intrinsic OpNode>
10207 : AESBase<opc, asm, (outs V128:$Rd), (ins V128:$Rn), "",
10208 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
10210 class AESTiedInst<bits<4> opc, string asm, Intrinsic OpNode>
10211 : AESBase<opc, asm, (outs V128:$dst), (ins V128:$Rd, V128:$Rn),
10213 [(set (v16i8 V128:$dst),
10214 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
10216 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10217 class SHA3OpTiedInst<bits<3> opc, string asm, string dst_lhs_kind,
10218 dag oops, dag iops, list<dag> pat>
10219 : I<oops, iops, asm,
10220 "{\t$Rd" # dst_lhs_kind # ", $Rn" # dst_lhs_kind # ", $Rm.4s" #
10221 "|.4s\t$Rd, $Rn, $Rm}", "$Rd = $dst", pat>,
10226 let Inst{31-21} = 0b01011110000;
10227 let Inst{20-16} = Rm;
10229 let Inst{14-12} = opc;
10230 let Inst{11-10} = 0b00;
10231 let Inst{9-5} = Rn;
10232 let Inst{4-0} = Rd;
10235 class SHATiedInstQSV<bits<3> opc, string asm, Intrinsic OpNode>
10236 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
10237 (ins FPR128:$Rd, FPR32:$Rn, V128:$Rm),
10238 [(set (v4i32 FPR128:$dst),
10239 (OpNode (v4i32 FPR128:$Rd), (i32 FPR32:$Rn),
10240 (v4i32 V128:$Rm)))]>;
10242 class SHATiedInstVVV<bits<3> opc, string asm, Intrinsic OpNode>
10243 : SHA3OpTiedInst<opc, asm, ".4s", (outs V128:$dst),
10244 (ins V128:$Rd, V128:$Rn, V128:$Rm),
10245 [(set (v4i32 V128:$dst),
10246 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
10247 (v4i32 V128:$Rm)))]>;
10249 class SHATiedInstQQV<bits<3> opc, string asm, Intrinsic OpNode>
10250 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
10251 (ins FPR128:$Rd, FPR128:$Rn, V128:$Rm),
10252 [(set (v4i32 FPR128:$dst),
10253 (OpNode (v4i32 FPR128:$Rd), (v4i32 FPR128:$Rn),
10254 (v4i32 V128:$Rm)))]>;
10256 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
10257 class SHA2OpInst<bits<4> opc, string asm, string kind,
10258 string cstr, dag oops, dag iops,
10260 : I<oops, iops, asm, "{\t$Rd" # kind # ", $Rn" # kind #
10261 "|" # kind # "\t$Rd, $Rn}", cstr, pat>,
10265 let Inst{31-16} = 0b0101111000101000;
10266 let Inst{15-12} = opc;
10267 let Inst{11-10} = 0b10;
10268 let Inst{9-5} = Rn;
10269 let Inst{4-0} = Rd;
10272 class SHATiedInstVV<bits<4> opc, string asm, Intrinsic OpNode>
10273 : SHA2OpInst<opc, asm, ".4s", "$Rd = $dst", (outs V128:$dst),
10274 (ins V128:$Rd, V128:$Rn),
10275 [(set (v4i32 V128:$dst),
10276 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
10278 class SHAInstSS<bits<4> opc, string asm, Intrinsic OpNode>
10279 : SHA2OpInst<opc, asm, "", "", (outs FPR32:$Rd), (ins FPR32:$Rn),
10280 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
10282 // Armv8.2-A Crypto extensions
10283 class BaseCryptoV82<dag oops, dag iops, string asm, string asmops, string cst,
10285 : I <oops, iops, asm, asmops, cst, pattern>, Sched<[WriteV]> {
10288 let Inst{31-25} = 0b1100111;
10289 let Inst{9-5} = Vn;
10290 let Inst{4-0} = Vd;
10293 class CryptoRRTied<bits<1>op0, bits<2>op1, string asm, string asmops>
10294 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm, asmops,
10296 let Inst{31-25} = 0b1100111;
10297 let Inst{24-21} = 0b0110;
10298 let Inst{20-15} = 0b000001;
10299 let Inst{14} = op0;
10300 let Inst{13-12} = 0b00;
10301 let Inst{11-10} = op1;
10303 class CryptoRRTied_2D<bits<1>op0, bits<2>op1, string asm>
10304 : CryptoRRTied<op0, op1, asm, "{\t$Vd.2d, $Vn.2d}">;
10305 class CryptoRRTied_4S<bits<1>op0, bits<2>op1, string asm>
10306 : CryptoRRTied<op0, op1, asm, "{\t$Vd.4s, $Vn.4s}">;
10308 class CryptoRRR<bits<1> op0, bits<2>op1, dag oops, dag iops, string asm,
10309 string asmops, string cst>
10310 : BaseCryptoV82<oops, iops, asm , asmops, cst, []> {
10312 let Inst{24-21} = 0b0011;
10313 let Inst{20-16} = Vm;
10314 let Inst{15} = 0b1;
10315 let Inst{14} = op0;
10316 let Inst{13-12} = 0b00;
10317 let Inst{11-10} = op1;
10319 class CryptoRRR_2D<bits<1> op0, bits<2>op1, string asm>
10320 : CryptoRRR<op0, op1, (outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm,
10321 "{\t$Vd.2d, $Vn.2d, $Vm.2d}", "">;
10322 class CryptoRRRTied_2D<bits<1> op0, bits<2>op1, string asm>
10323 : CryptoRRR<op0, op1, (outs V128:$Vdst), (ins V128:$Vd, V128:$Vn, V128:$Vm), asm,
10324 "{\t$Vd.2d, $Vn.2d, $Vm.2d}", "$Vd = $Vdst">;
10325 class CryptoRRR_4S<bits<1> op0, bits<2>op1, string asm>
10326 : CryptoRRR<op0, op1, (outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm,
10327 "{\t$Vd.4s, $Vn.4s, $Vm.4s}", "">;
10328 class CryptoRRRTied_4S<bits<1> op0, bits<2>op1, string asm>
10329 : CryptoRRR<op0, op1, (outs V128:$Vdst), (ins V128:$Vd, V128:$Vn, V128:$Vm), asm,
10330 "{\t$Vd.4s, $Vn.4s, $Vm.4s}", "$Vd = $Vdst">;
10331 class CryptoRRRTied<bits<1> op0, bits<2>op1, string asm>
10332 : CryptoRRR<op0, op1, (outs FPR128:$Vdst), (ins FPR128:$Vd, FPR128:$Vn, V128:$Vm),
10333 asm, "{\t$Vd, $Vn, $Vm.2d}", "$Vd = $Vdst">;
10335 class CryptoRRRR<bits<2>op0, string asm, string asmops>
10336 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm, V128:$Va), asm,
10340 let Inst{24-23} = 0b00;
10341 let Inst{22-21} = op0;
10342 let Inst{20-16} = Vm;
10343 let Inst{15} = 0b0;
10344 let Inst{14-10} = Va;
10346 class CryptoRRRR_16B<bits<2>op0, string asm>
10347 : CryptoRRRR<op0, asm, "{\t$Vd.16b, $Vn.16b, $Vm.16b, $Va.16b}"> {
10349 class CryptoRRRR_4S<bits<2>op0, string asm>
10350 : CryptoRRRR<op0, asm, "{\t$Vd.4s, $Vn.4s, $Vm.4s, $Va.4s}"> {
10353 class CryptoRRRi6<string asm>
10354 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm, uimm6:$imm), asm,
10355 "{\t$Vd.2d, $Vn.2d, $Vm.2d, $imm}", "", []> {
10358 let Inst{24-21} = 0b0100;
10359 let Inst{20-16} = Vm;
10360 let Inst{15-10} = imm;
10361 let Inst{9-5} = Vn;
10362 let Inst{4-0} = Vd;
10365 class CryptoRRRi2Tied<bits<1>op0, bits<2>op1, string asm>
10366 : BaseCryptoV82<(outs V128:$Vdst),
10367 (ins V128:$Vd, V128:$Vn, V128:$Vm, VectorIndexS:$imm),
10368 asm, "{\t$Vd.4s, $Vn.4s, $Vm.s$imm}", "$Vd = $Vdst", []> {
10371 let Inst{24-21} = 0b0010;
10372 let Inst{20-16} = Vm;
10373 let Inst{15} = 0b1;
10374 let Inst{14} = op0;
10375 let Inst{13-12} = imm;
10376 let Inst{11-10} = op1;
10379 //----------------------------------------------------------------------------
10380 // v8.1 atomic instructions extension:
10384 // * LDOPregister<OP>, and aliases STOPregister<OP>
10386 // Instruction encodings:
10388 // 31 30|29 24|23|22|21|20 16|15|14 10|9 5|4 0
10389 // CAS SZ |001000|1 |A |1 |Rs |R |11111 |Rn |Rt
10390 // CASP 0|SZ|001000|0 |A |1 |Rs |R |11111 |Rn |Rt
10391 // SWP SZ |111000|A |R |1 |Rs |1 |OPC|00|Rn |Rt
10392 // LD SZ |111000|A |R |1 |Rs |0 |OPC|00|Rn |Rt
10393 // ST SZ |111000|A |R |1 |Rs |0 |OPC|00|Rn |11111
10395 // Instruction syntax:
10397 // CAS{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10398 // CAS{<order>} <Xs>, <Xt>, [<Xn|SP>]
10399 // CASP{<order>} <Ws>, <W(s+1)>, <Wt>, <W(t+1)>, [<Xn|SP>]
10400 // CASP{<order>} <Xs>, <X(s+1)>, <Xt>, <X(t+1)>, [<Xn|SP>]
10401 // SWP{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10402 // SWP{<order>} <Xs>, <Xt>, [<Xn|SP>]
10403 // LD<OP>{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10404 // LD<OP>{<order>} <Xs>, <Xt>, [<Xn|SP>]
10405 // ST<OP>{<order>}[<size>] <Ws>, [<Xn|SP>]
10406 // ST<OP>{<order>} <Xs>, [<Xn|SP>]
10408 let Predicates = [HasLSE], mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
10409 class BaseCASEncoding<dag oops, dag iops, string asm, string operands,
10410 string cstr, list<dag> pattern>
10411 : I<oops, iops, asm, operands, cstr, pattern> {
10419 let Inst{31-30} = Sz;
10420 let Inst{29-24} = 0b001000;
10422 let Inst{22} = Acq;
10423 let Inst{21} = 0b1;
10424 let Inst{20-16} = Rs;
10425 let Inst{15} = Rel;
10426 let Inst{14-10} = 0b11111;
10427 let Inst{9-5} = Rn;
10428 let Inst{4-0} = Rt;
10429 let Predicates = [HasLSE];
10432 class BaseCAS<string order, string size, RegisterClass RC>
10433 : BaseCASEncoding<(outs RC:$out),(ins RC:$Rs, RC:$Rt, GPR64sp:$Rn),
10434 "cas" # order # size, "\t$Rs, $Rt, [$Rn]",
10436 Sched<[WriteAtomic]> {
10440 multiclass CompareAndSwap<bits<1> Acq, bits<1> Rel, string order> {
10441 let Sz = 0b00, Acq = Acq, Rel = Rel in def B : BaseCAS<order, "b", GPR32>;
10442 let Sz = 0b01, Acq = Acq, Rel = Rel in def H : BaseCAS<order, "h", GPR32>;
10443 let Sz = 0b10, Acq = Acq, Rel = Rel in def W : BaseCAS<order, "", GPR32>;
10444 let Sz = 0b11, Acq = Acq, Rel = Rel in def X : BaseCAS<order, "", GPR64>;
10447 class BaseCASP<string order, string size, RegisterOperand RC>
10448 : BaseCASEncoding<(outs RC:$out),(ins RC:$Rs, RC:$Rt, GPR64sp:$Rn),
10449 "casp" # order # size, "\t$Rs, $Rt, [$Rn]",
10451 Sched<[WriteAtomic]> {
10455 multiclass CompareAndSwapPair<bits<1> Acq, bits<1> Rel, string order> {
10456 let Sz = 0b00, Acq = Acq, Rel = Rel in
10457 def W : BaseCASP<order, "", WSeqPairClassOperand>;
10458 let Sz = 0b01, Acq = Acq, Rel = Rel in
10459 def X : BaseCASP<order, "", XSeqPairClassOperand>;
10462 let Predicates = [HasLSE] in
10463 class BaseSWP<string order, string size, RegisterClass RC>
10464 : I<(outs RC:$Rt),(ins RC:$Rs, GPR64sp:$Rn), "swp" # order # size,
10465 "\t$Rs, $Rt, [$Rn]","",[]>,
10466 Sched<[WriteAtomic]> {
10471 bits<3> opc = 0b000;
10474 let Inst{31-30} = Sz;
10475 let Inst{29-24} = 0b111000;
10476 let Inst{23} = Acq;
10477 let Inst{22} = Rel;
10478 let Inst{21} = 0b1;
10479 let Inst{20-16} = Rs;
10480 let Inst{15} = 0b1;
10481 let Inst{14-12} = opc;
10482 let Inst{11-10} = 0b00;
10483 let Inst{9-5} = Rn;
10484 let Inst{4-0} = Rt;
10485 let Predicates = [HasLSE];
10488 multiclass Swap<bits<1> Acq, bits<1> Rel, string order> {
10489 let Sz = 0b00, Acq = Acq, Rel = Rel in def B : BaseSWP<order, "b", GPR32>;
10490 let Sz = 0b01, Acq = Acq, Rel = Rel in def H : BaseSWP<order, "h", GPR32>;
10491 let Sz = 0b10, Acq = Acq, Rel = Rel in def W : BaseSWP<order, "", GPR32>;
10492 let Sz = 0b11, Acq = Acq, Rel = Rel in def X : BaseSWP<order, "", GPR64>;
10495 let Predicates = [HasLSE], mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
10496 class BaseLDOPregister<string op, string order, string size, RegisterClass RC>
10497 : I<(outs RC:$Rt),(ins RC:$Rs, GPR64sp:$Rn), "ld" # op # order # size,
10498 "\t$Rs, $Rt, [$Rn]","",[]>,
10499 Sched<[WriteAtomic]> {
10507 let Inst{31-30} = Sz;
10508 let Inst{29-24} = 0b111000;
10509 let Inst{23} = Acq;
10510 let Inst{22} = Rel;
10511 let Inst{21} = 0b1;
10512 let Inst{20-16} = Rs;
10513 let Inst{15} = 0b0;
10514 let Inst{14-12} = opc;
10515 let Inst{11-10} = 0b00;
10516 let Inst{9-5} = Rn;
10517 let Inst{4-0} = Rt;
10518 let Predicates = [HasLSE];
10521 multiclass LDOPregister<bits<3> opc, string op, bits<1> Acq, bits<1> Rel,
10523 let Sz = 0b00, Acq = Acq, Rel = Rel, opc = opc in
10524 def B : BaseLDOPregister<op, order, "b", GPR32>;
10525 let Sz = 0b01, Acq = Acq, Rel = Rel, opc = opc in
10526 def H : BaseLDOPregister<op, order, "h", GPR32>;
10527 let Sz = 0b10, Acq = Acq, Rel = Rel, opc = opc in
10528 def W : BaseLDOPregister<op, order, "", GPR32>;
10529 let Sz = 0b11, Acq = Acq, Rel = Rel, opc = opc in
10530 def X : BaseLDOPregister<op, order, "", GPR64>;
10533 // Differing SrcRHS and DstRHS allow you to cover CLR & SUB by giving a more
10534 // complex DAG for DstRHS.
10535 let Predicates = [HasLSE] in
10536 multiclass LDOPregister_patterns_ord_dag<string inst, string suffix, string op,
10537 string size, dag SrcRHS, dag DstRHS> {
10538 def : Pat<(!cast<PatFrag>(op#"_"#size#"_monotonic") GPR64sp:$Rn, SrcRHS),
10539 (!cast<Instruction>(inst # suffix) DstRHS, GPR64sp:$Rn)>;
10540 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acquire") GPR64sp:$Rn, SrcRHS),
10541 (!cast<Instruction>(inst # "A" # suffix) DstRHS, GPR64sp:$Rn)>;
10542 def : Pat<(!cast<PatFrag>(op#"_"#size#"_release") GPR64sp:$Rn, SrcRHS),
10543 (!cast<Instruction>(inst # "L" # suffix) DstRHS, GPR64sp:$Rn)>;
10544 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acq_rel") GPR64sp:$Rn, SrcRHS),
10545 (!cast<Instruction>(inst # "AL" # suffix) DstRHS, GPR64sp:$Rn)>;
10546 def : Pat<(!cast<PatFrag>(op#"_"#size#"_seq_cst") GPR64sp:$Rn, SrcRHS),
10547 (!cast<Instruction>(inst # "AL" # suffix) DstRHS, GPR64sp:$Rn)>;
10550 multiclass LDOPregister_patterns_ord<string inst, string suffix, string op,
10551 string size, dag RHS> {
10552 defm : LDOPregister_patterns_ord_dag<inst, suffix, op, size, RHS, RHS>;
10555 multiclass LDOPregister_patterns_ord_mod<string inst, string suffix, string op,
10556 string size, dag LHS, dag RHS> {
10557 defm : LDOPregister_patterns_ord_dag<inst, suffix, op, size, LHS, RHS>;
10560 multiclass LDOPregister_patterns<string inst, string op> {
10561 defm : LDOPregister_patterns_ord<inst, "X", op, "64", (i64 GPR64:$Rm)>;
10562 defm : LDOPregister_patterns_ord<inst, "W", op, "32", (i32 GPR32:$Rm)>;
10563 defm : LDOPregister_patterns_ord<inst, "H", op, "16", (i32 GPR32:$Rm)>;
10564 defm : LDOPregister_patterns_ord<inst, "B", op, "8", (i32 GPR32:$Rm)>;
10567 multiclass LDOPregister_patterns_mod<string inst, string op, string mod> {
10568 defm : LDOPregister_patterns_ord_mod<inst, "X", op, "64",
10570 (i64 (!cast<Instruction>(mod#Xrr) XZR, GPR64:$Rm))>;
10571 defm : LDOPregister_patterns_ord_mod<inst, "W", op, "32",
10573 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10574 defm : LDOPregister_patterns_ord_mod<inst, "H", op, "16",
10576 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10577 defm : LDOPregister_patterns_ord_mod<inst, "B", op, "8",
10579 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10582 let Predicates = [HasLSE] in
10583 multiclass CASregister_patterns_ord_dag<string inst, string suffix, string op,
10584 string size, dag OLD, dag NEW> {
10585 def : Pat<(!cast<PatFrag>(op#"_"#size#"_monotonic") GPR64sp:$Rn, OLD, NEW),
10586 (!cast<Instruction>(inst # suffix) OLD, NEW, GPR64sp:$Rn)>;
10587 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acquire") GPR64sp:$Rn, OLD, NEW),
10588 (!cast<Instruction>(inst # "A" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10589 def : Pat<(!cast<PatFrag>(op#"_"#size#"_release") GPR64sp:$Rn, OLD, NEW),
10590 (!cast<Instruction>(inst # "L" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10591 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acq_rel") GPR64sp:$Rn, OLD, NEW),
10592 (!cast<Instruction>(inst # "AL" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10593 def : Pat<(!cast<PatFrag>(op#"_"#size#"_seq_cst") GPR64sp:$Rn, OLD, NEW),
10594 (!cast<Instruction>(inst # "AL" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10597 multiclass CASregister_patterns_ord<string inst, string suffix, string op,
10598 string size, dag OLD, dag NEW> {
10599 defm : CASregister_patterns_ord_dag<inst, suffix, op, size, OLD, NEW>;
10602 multiclass CASregister_patterns<string inst, string op> {
10603 defm : CASregister_patterns_ord<inst, "X", op, "64",
10604 (i64 GPR64:$Rold), (i64 GPR64:$Rnew)>;
10605 defm : CASregister_patterns_ord<inst, "W", op, "32",
10606 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10607 defm : CASregister_patterns_ord<inst, "H", op, "16",
10608 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10609 defm : CASregister_patterns_ord<inst, "B", op, "8",
10610 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10613 let Predicates = [HasLSE] in
10614 class BaseSTOPregister<string asm, RegisterClass OP, Register Reg,
10615 Instruction inst> :
10616 InstAlias<asm # "\t$Rs, [$Rn]", (inst Reg, OP:$Rs, GPR64sp:$Rn)>;
10618 multiclass STOPregister<string asm, string instr> {
10619 def : BaseSTOPregister<asm # "lb", GPR32, WZR,
10620 !cast<Instruction>(instr # "LB")>;
10621 def : BaseSTOPregister<asm # "lh", GPR32, WZR,
10622 !cast<Instruction>(instr # "LH")>;
10623 def : BaseSTOPregister<asm # "l", GPR32, WZR,
10624 !cast<Instruction>(instr # "LW")>;
10625 def : BaseSTOPregister<asm # "l", GPR64, XZR,
10626 !cast<Instruction>(instr # "LX")>;
10627 def : BaseSTOPregister<asm # "b", GPR32, WZR,
10628 !cast<Instruction>(instr # "B")>;
10629 def : BaseSTOPregister<asm # "h", GPR32, WZR,
10630 !cast<Instruction>(instr # "H")>;
10631 def : BaseSTOPregister<asm, GPR32, WZR,
10632 !cast<Instruction>(instr # "W")>;
10633 def : BaseSTOPregister<asm, GPR64, XZR,
10634 !cast<Instruction>(instr # "X")>;
10637 //----------------------------------------------------------------------------
10638 // Allow the size specifier tokens to be upper case, not just lower.
10639 def : TokenAlias<".4B", ".4b">; // Add dot product
10640 def : TokenAlias<".8B", ".8b">;
10641 def : TokenAlias<".4H", ".4h">;
10642 def : TokenAlias<".2S", ".2s">;
10643 def : TokenAlias<".1D", ".1d">;
10644 def : TokenAlias<".16B", ".16b">;
10645 def : TokenAlias<".8H", ".8h">;
10646 def : TokenAlias<".4S", ".4s">;
10647 def : TokenAlias<".2D", ".2d">;
10648 def : TokenAlias<".1Q", ".1q">;
10649 def : TokenAlias<".2H", ".2h">;
10650 def : TokenAlias<".B", ".b">;
10651 def : TokenAlias<".H", ".h">;
10652 def : TokenAlias<".S", ".s">;
10653 def : TokenAlias<".D", ".d">;
10654 def : TokenAlias<".Q", ".q">;