1 //===- AArch64InstrFormats.td - AArch64 Instruction Formats --*- tblgen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 //===----------------------------------------------------------------------===//
11 // Describe AArch64 instructions format here
14 // Format specifies the encoding used by the instruction. This is part of the
15 // ad-hoc solution used to emit machine instruction encodings by our machine
17 class Format<bits<2> val> {
21 def PseudoFrm : Format<0>;
22 def NormalFrm : Format<1>; // Do we need any others?
24 // AArch64 Instruction Format
25 class AArch64Inst<Format f, string cstr> : Instruction {
26 field bits<32> Inst; // Instruction encoding.
27 // Mask of bits that cause an encoding to be UNPREDICTABLE.
28 // If a bit is set, then if the corresponding bit in the
29 // target encoding differs from its value in the "Inst" field,
30 // the instruction is UNPREDICTABLE (SoftFail in abstract parlance).
31 field bits<32> Unpredictable = 0;
32 // SoftFail is the generic name for this field, but we alias it so
33 // as to make it more obvious what it means in ARM-land.
34 field bits<32> SoftFail = Unpredictable;
35 let Namespace = "AArch64";
37 bits<2> Form = F.Value;
39 let Constraints = cstr;
42 class InstSubst<string Asm, dag Result, bit EmitPriority = 0>
43 : InstAlias<Asm, Result, EmitPriority>, Requires<[UseNegativeImmediates]>;
45 // Pseudo instructions (don't have encoding information)
46 class Pseudo<dag oops, dag iops, list<dag> pattern, string cstr = "">
47 : AArch64Inst<PseudoFrm, cstr> {
48 dag OutOperandList = oops;
49 dag InOperandList = iops;
50 let Pattern = pattern;
51 let isCodeGenOnly = 1;
54 // Real instructions (have encoding information)
55 class EncodedI<string cstr, list<dag> pattern> : AArch64Inst<NormalFrm, cstr> {
56 let Pattern = pattern;
60 // Normal instructions
61 class I<dag oops, dag iops, string asm, string operands, string cstr,
63 : EncodedI<cstr, pattern> {
64 dag OutOperandList = oops;
65 dag InOperandList = iops;
66 let AsmString = !strconcat(asm, operands);
69 class TriOpFrag<dag res> : PatFrag<(ops node:$LHS, node:$MHS, node:$RHS), res>;
70 class BinOpFrag<dag res> : PatFrag<(ops node:$LHS, node:$RHS), res>;
71 class UnOpFrag<dag res> : PatFrag<(ops node:$LHS), res>;
73 // Helper fragment for an extract of the high portion of a 128-bit vector.
74 def extract_high_v16i8 :
75 UnOpFrag<(extract_subvector (v16i8 node:$LHS), (i64 8))>;
76 def extract_high_v8i16 :
77 UnOpFrag<(extract_subvector (v8i16 node:$LHS), (i64 4))>;
78 def extract_high_v4i32 :
79 UnOpFrag<(extract_subvector (v4i32 node:$LHS), (i64 2))>;
80 def extract_high_v2i64 :
81 UnOpFrag<(extract_subvector (v2i64 node:$LHS), (i64 1))>;
83 //===----------------------------------------------------------------------===//
84 // Asm Operand Classes.
87 // Shifter operand for arithmetic shifted encodings.
88 def ShifterOperand : AsmOperandClass {
92 // Shifter operand for mov immediate encodings.
93 def MovImm32ShifterOperand : AsmOperandClass {
94 let SuperClasses = [ShifterOperand];
95 let Name = "MovImm32Shifter";
96 let RenderMethod = "addShifterOperands";
97 let DiagnosticType = "InvalidMovImm32Shift";
99 def MovImm64ShifterOperand : AsmOperandClass {
100 let SuperClasses = [ShifterOperand];
101 let Name = "MovImm64Shifter";
102 let RenderMethod = "addShifterOperands";
103 let DiagnosticType = "InvalidMovImm64Shift";
106 // Shifter operand for arithmetic register shifted encodings.
107 class ArithmeticShifterOperand<int width> : AsmOperandClass {
108 let SuperClasses = [ShifterOperand];
109 let Name = "ArithmeticShifter" # width;
110 let PredicateMethod = "isArithmeticShifter<" # width # ">";
111 let RenderMethod = "addShifterOperands";
112 let DiagnosticType = "AddSubRegShift" # width;
115 def ArithmeticShifterOperand32 : ArithmeticShifterOperand<32>;
116 def ArithmeticShifterOperand64 : ArithmeticShifterOperand<64>;
118 // Shifter operand for logical register shifted encodings.
119 class LogicalShifterOperand<int width> : AsmOperandClass {
120 let SuperClasses = [ShifterOperand];
121 let Name = "LogicalShifter" # width;
122 let PredicateMethod = "isLogicalShifter<" # width # ">";
123 let RenderMethod = "addShifterOperands";
124 let DiagnosticType = "AddSubRegShift" # width;
127 def LogicalShifterOperand32 : LogicalShifterOperand<32>;
128 def LogicalShifterOperand64 : LogicalShifterOperand<64>;
130 // Shifter operand for logical vector 128/64-bit shifted encodings.
131 def LogicalVecShifterOperand : AsmOperandClass {
132 let SuperClasses = [ShifterOperand];
133 let Name = "LogicalVecShifter";
134 let RenderMethod = "addShifterOperands";
136 def LogicalVecHalfWordShifterOperand : AsmOperandClass {
137 let SuperClasses = [LogicalVecShifterOperand];
138 let Name = "LogicalVecHalfWordShifter";
139 let RenderMethod = "addShifterOperands";
142 // The "MSL" shifter on the vector MOVI instruction.
143 def MoveVecShifterOperand : AsmOperandClass {
144 let SuperClasses = [ShifterOperand];
145 let Name = "MoveVecShifter";
146 let RenderMethod = "addShifterOperands";
149 // Extend operand for arithmetic encodings.
150 def ExtendOperand : AsmOperandClass {
152 let DiagnosticType = "AddSubRegExtendLarge";
154 def ExtendOperand64 : AsmOperandClass {
155 let SuperClasses = [ExtendOperand];
156 let Name = "Extend64";
157 let DiagnosticType = "AddSubRegExtendSmall";
159 // 'extend' that's a lsl of a 64-bit register.
160 def ExtendOperandLSL64 : AsmOperandClass {
161 let SuperClasses = [ExtendOperand];
162 let Name = "ExtendLSL64";
163 let RenderMethod = "addExtend64Operands";
164 let DiagnosticType = "AddSubRegExtendLarge";
167 // 8-bit floating-point immediate encodings.
168 def FPImmOperand : AsmOperandClass {
170 let ParserMethod = "tryParseFPImm<true>";
171 let DiagnosticType = "InvalidFPImm";
174 def CondCode : AsmOperandClass {
175 let Name = "CondCode";
176 let DiagnosticType = "InvalidCondCode";
179 // A 32-bit register pasrsed as 64-bit
180 def GPR32as64Operand : AsmOperandClass {
181 let Name = "GPR32as64";
183 "tryParseGPROperand<false, RegConstraintEqualityTy::EqualsSubReg>";
185 def GPR32as64 : RegisterOperand<GPR32> {
186 let ParserMatchClass = GPR32as64Operand;
189 // A 64-bit register pasrsed as 32-bit
190 def GPR64as32Operand : AsmOperandClass {
191 let Name = "GPR64as32";
193 "tryParseGPROperand<false, RegConstraintEqualityTy::EqualsSuperReg>";
195 def GPR64as32 : RegisterOperand<GPR64, "printGPR64as32"> {
196 let ParserMatchClass = GPR64as32Operand;
199 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
200 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
201 // are encoded as the eight bit value 'abcdefgh'.
202 def SIMDImmType10Operand : AsmOperandClass { let Name = "SIMDImmType10"; }
204 class UImmScaledMemoryIndexed<int Width, int Scale> : AsmOperandClass {
205 let Name = "UImm" # Width # "s" # Scale;
206 let DiagnosticType = "InvalidMemoryIndexed" # Scale # "UImm" # Width;
207 let RenderMethod = "addImmScaledOperands<" # Scale # ">";
208 let PredicateMethod = "isUImmScaled<" # Width # ", " # Scale # ">";
211 class SImmScaledMemoryIndexed<int Width, int Scale> : AsmOperandClass {
212 let Name = "SImm" # Width # "s" # Scale;
213 let DiagnosticType = "InvalidMemoryIndexed" # Scale # "SImm" # Width;
214 let RenderMethod = "addImmScaledOperands<" # Scale # ">";
215 let PredicateMethod = "isSImmScaled<" # Width # ", " # Scale # ">";
218 //===----------------------------------------------------------------------===//
219 // Operand Definitions.
222 // ADR[P] instruction labels.
223 def AdrpOperand : AsmOperandClass {
224 let Name = "AdrpLabel";
225 let ParserMethod = "tryParseAdrpLabel";
226 let DiagnosticType = "InvalidLabel";
228 def adrplabel : Operand<i64> {
229 let EncoderMethod = "getAdrLabelOpValue";
230 let PrintMethod = "printAdrpLabel";
231 let ParserMatchClass = AdrpOperand;
234 def AdrOperand : AsmOperandClass {
235 let Name = "AdrLabel";
236 let ParserMethod = "tryParseAdrLabel";
237 let DiagnosticType = "InvalidLabel";
239 def adrlabel : Operand<i64> {
240 let EncoderMethod = "getAdrLabelOpValue";
241 let ParserMatchClass = AdrOperand;
244 class SImmOperand<int width> : AsmOperandClass {
245 let Name = "SImm" # width;
246 let DiagnosticType = "InvalidMemoryIndexedSImm" # width;
247 let RenderMethod = "addImmOperands";
248 let PredicateMethod = "isSImm<" # width # ">";
251 // Authenticated loads for v8.3 can have scaled 10-bit immediate offsets.
252 def SImm10s8Operand : SImmScaledMemoryIndexed<10, 8>;
253 def simm10Scaled : Operand<i64> {
254 let ParserMatchClass = SImm10s8Operand;
255 let DecoderMethod = "DecodeSImm<10>";
256 let PrintMethod = "printImmScale<8>";
259 // uimm6 predicate - True if the immediate is in the range [0, 63].
260 def UImm6Operand : AsmOperandClass {
262 let DiagnosticType = "InvalidImm0_63";
265 def uimm6 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= 0 && Imm < 64; }]> {
266 let ParserMatchClass = UImm6Operand;
269 def SImm9Operand : SImmOperand<9>;
270 def simm9 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -256 && Imm < 256; }]> {
271 let ParserMatchClass = SImm9Operand;
272 let DecoderMethod = "DecodeSImm<9>";
275 def SImm8Operand : SImmOperand<8>;
276 def simm8 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -128 && Imm < 127; }]> {
277 let ParserMatchClass = SImm8Operand;
278 let DecoderMethod = "DecodeSImm<8>";
281 def SImm6Operand : SImmOperand<6>;
282 def simm6_32b : Operand<i32>, ImmLeaf<i32, [{ return Imm >= -32 && Imm < 32; }]> {
283 let ParserMatchClass = SImm6Operand;
284 let DecoderMethod = "DecodeSImm<6>";
287 def SImm5Operand : SImmOperand<5>;
288 def simm5_64b : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -16 && Imm < 16; }]> {
289 let ParserMatchClass = SImm5Operand;
290 let DecoderMethod = "DecodeSImm<5>";
293 def simm5_32b : Operand<i32>, ImmLeaf<i32, [{ return Imm >= -16 && Imm < 16; }]> {
294 let ParserMatchClass = SImm5Operand;
295 let DecoderMethod = "DecodeSImm<5>";
298 // simm7sN predicate - True if the immediate is a multiple of N in the range
299 // [-64 * N, 63 * N].
301 def SImm7s4Operand : SImmScaledMemoryIndexed<7, 4>;
302 def SImm7s8Operand : SImmScaledMemoryIndexed<7, 8>;
303 def SImm7s16Operand : SImmScaledMemoryIndexed<7, 16>;
305 def simm7s4 : Operand<i32> {
306 let ParserMatchClass = SImm7s4Operand;
307 let PrintMethod = "printImmScale<4>";
310 def simm7s8 : Operand<i32> {
311 let ParserMatchClass = SImm7s8Operand;
312 let PrintMethod = "printImmScale<8>";
315 def simm7s16 : Operand<i32> {
316 let ParserMatchClass = SImm7s16Operand;
317 let PrintMethod = "printImmScale<16>";
320 def am_indexed7s8 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S8", []>;
321 def am_indexed7s16 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S16", []>;
322 def am_indexed7s32 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S32", []>;
323 def am_indexed7s64 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S64", []>;
324 def am_indexed7s128 : ComplexPattern<i64, 2, "SelectAddrModeIndexed7S128", []>;
326 // uimm5sN predicate - True if the immediate is a multiple of N in the range
328 def UImm5s2Operand : UImmScaledMemoryIndexed<5, 2>;
329 def UImm5s4Operand : UImmScaledMemoryIndexed<5, 4>;
330 def UImm5s8Operand : UImmScaledMemoryIndexed<5, 8>;
332 def uimm5s2 : Operand<i64>, ImmLeaf<i64,
333 [{ return Imm >= 0 && Imm < (32*2) && ((Imm % 2) == 0); }]> {
334 let ParserMatchClass = UImm5s2Operand;
335 let PrintMethod = "printImmScale<2>";
337 def uimm5s4 : Operand<i64>, ImmLeaf<i64,
338 [{ return Imm >= 0 && Imm < (32*4) && ((Imm % 4) == 0); }]> {
339 let ParserMatchClass = UImm5s4Operand;
340 let PrintMethod = "printImmScale<4>";
342 def uimm5s8 : Operand<i64>, ImmLeaf<i64,
343 [{ return Imm >= 0 && Imm < (32*8) && ((Imm % 8) == 0); }]> {
344 let ParserMatchClass = UImm5s8Operand;
345 let PrintMethod = "printImmScale<8>";
348 // uimm6sN predicate - True if the immediate is a multiple of N in the range
350 def UImm6s1Operand : UImmScaledMemoryIndexed<6, 1>;
351 def UImm6s2Operand : UImmScaledMemoryIndexed<6, 2>;
352 def UImm6s4Operand : UImmScaledMemoryIndexed<6, 4>;
353 def UImm6s8Operand : UImmScaledMemoryIndexed<6, 8>;
355 def uimm6s1 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= 0 && Imm < 64; }]> {
356 let ParserMatchClass = UImm6s1Operand;
358 def uimm6s2 : Operand<i64>, ImmLeaf<i64,
359 [{ return Imm >= 0 && Imm < (64*2) && ((Imm % 2) == 0); }]> {
360 let PrintMethod = "printImmScale<2>";
361 let ParserMatchClass = UImm6s2Operand;
363 def uimm6s4 : Operand<i64>, ImmLeaf<i64,
364 [{ return Imm >= 0 && Imm < (64*4) && ((Imm % 4) == 0); }]> {
365 let PrintMethod = "printImmScale<4>";
366 let ParserMatchClass = UImm6s4Operand;
368 def uimm6s8 : Operand<i64>, ImmLeaf<i64,
369 [{ return Imm >= 0 && Imm < (64*8) && ((Imm % 8) == 0); }]> {
370 let PrintMethod = "printImmScale<8>";
371 let ParserMatchClass = UImm6s8Operand;
374 // simm6sN predicate - True if the immediate is a multiple of N in the range
375 // [-32 * N, 31 * N].
376 def SImm6s1Operand : SImmScaledMemoryIndexed<6, 1>;
377 def simm6s1 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -32 && Imm < 32; }]> {
378 let ParserMatchClass = SImm6s1Operand;
379 let DecoderMethod = "DecodeSImm<6>";
382 // simm4sN predicate - True if the immediate is a multiple of N in the range
384 def SImm4s1Operand : SImmScaledMemoryIndexed<4, 1>;
385 def SImm4s2Operand : SImmScaledMemoryIndexed<4, 2>;
386 def SImm4s3Operand : SImmScaledMemoryIndexed<4, 3>;
387 def SImm4s4Operand : SImmScaledMemoryIndexed<4, 4>;
388 def SImm4s16Operand : SImmScaledMemoryIndexed<4, 16>;
390 def simm4s1 : Operand<i64>, ImmLeaf<i64,
391 [{ return Imm >=-8 && Imm <= 7; }]> {
392 let ParserMatchClass = SImm4s1Operand;
393 let DecoderMethod = "DecodeSImm<4>";
396 def simm4s2 : Operand<i64>, ImmLeaf<i64,
397 [{ return Imm >=-16 && Imm <= 14 && (Imm % 2) == 0x0; }]> {
398 let PrintMethod = "printImmScale<2>";
399 let ParserMatchClass = SImm4s2Operand;
400 let DecoderMethod = "DecodeSImm<4>";
403 def simm4s3 : Operand<i64>, ImmLeaf<i64,
404 [{ return Imm >=-24 && Imm <= 21 && (Imm % 3) == 0x0; }]> {
405 let PrintMethod = "printImmScale<3>";
406 let ParserMatchClass = SImm4s3Operand;
407 let DecoderMethod = "DecodeSImm<4>";
410 def simm4s4 : Operand<i64>, ImmLeaf<i64,
411 [{ return Imm >=-32 && Imm <= 28 && (Imm % 4) == 0x0; }]> {
412 let PrintMethod = "printImmScale<4>";
413 let ParserMatchClass = SImm4s4Operand;
414 let DecoderMethod = "DecodeSImm<4>";
416 def simm4s16 : Operand<i64>, ImmLeaf<i64,
417 [{ return Imm >=-128 && Imm <= 112 && (Imm % 16) == 0x0; }]> {
418 let PrintMethod = "printImmScale<16>";
419 let ParserMatchClass = SImm4s16Operand;
420 let DecoderMethod = "DecodeSImm<4>";
423 class AsmImmRange<int Low, int High> : AsmOperandClass {
424 let Name = "Imm" # Low # "_" # High;
425 let DiagnosticType = "InvalidImm" # Low # "_" # High;
426 let RenderMethod = "addImmOperands";
427 let PredicateMethod = "isImmInRange<" # Low # "," # High # ">";
430 def Imm1_8Operand : AsmImmRange<1, 8>;
431 def Imm1_16Operand : AsmImmRange<1, 16>;
432 def Imm1_32Operand : AsmImmRange<1, 32>;
433 def Imm1_64Operand : AsmImmRange<1, 64>;
435 class BranchTarget<int N> : AsmOperandClass {
436 let Name = "BranchTarget" # N;
437 let DiagnosticType = "InvalidLabel";
438 let PredicateMethod = "isBranchTarget<" # N # ">";
441 class PCRelLabel<int N> : BranchTarget<N> {
442 let Name = "PCRelLabel" # N;
445 def BranchTarget14Operand : BranchTarget<14>;
446 def BranchTarget26Operand : BranchTarget<26>;
447 def PCRelLabel19Operand : PCRelLabel<19>;
449 def MovZSymbolG3AsmOperand : AsmOperandClass {
450 let Name = "MovZSymbolG3";
451 let RenderMethod = "addImmOperands";
454 def movz_symbol_g3 : Operand<i32> {
455 let ParserMatchClass = MovZSymbolG3AsmOperand;
458 def MovZSymbolG2AsmOperand : AsmOperandClass {
459 let Name = "MovZSymbolG2";
460 let RenderMethod = "addImmOperands";
463 def movz_symbol_g2 : Operand<i32> {
464 let ParserMatchClass = MovZSymbolG2AsmOperand;
467 def MovZSymbolG1AsmOperand : AsmOperandClass {
468 let Name = "MovZSymbolG1";
469 let RenderMethod = "addImmOperands";
472 def movz_symbol_g1 : Operand<i32> {
473 let ParserMatchClass = MovZSymbolG1AsmOperand;
476 def MovZSymbolG0AsmOperand : AsmOperandClass {
477 let Name = "MovZSymbolG0";
478 let RenderMethod = "addImmOperands";
481 def movz_symbol_g0 : Operand<i32> {
482 let ParserMatchClass = MovZSymbolG0AsmOperand;
485 def MovKSymbolG3AsmOperand : AsmOperandClass {
486 let Name = "MovKSymbolG3";
487 let RenderMethod = "addImmOperands";
490 def movk_symbol_g3 : Operand<i32> {
491 let ParserMatchClass = MovKSymbolG3AsmOperand;
494 def MovKSymbolG2AsmOperand : AsmOperandClass {
495 let Name = "MovKSymbolG2";
496 let RenderMethod = "addImmOperands";
499 def movk_symbol_g2 : Operand<i32> {
500 let ParserMatchClass = MovKSymbolG2AsmOperand;
503 def MovKSymbolG1AsmOperand : AsmOperandClass {
504 let Name = "MovKSymbolG1";
505 let RenderMethod = "addImmOperands";
508 def movk_symbol_g1 : Operand<i32> {
509 let ParserMatchClass = MovKSymbolG1AsmOperand;
512 def MovKSymbolG0AsmOperand : AsmOperandClass {
513 let Name = "MovKSymbolG0";
514 let RenderMethod = "addImmOperands";
517 def movk_symbol_g0 : Operand<i32> {
518 let ParserMatchClass = MovKSymbolG0AsmOperand;
521 class fixedpoint_i32<ValueType FloatVT>
523 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<32>", [fpimm, ld]> {
524 let EncoderMethod = "getFixedPointScaleOpValue";
525 let DecoderMethod = "DecodeFixedPointScaleImm32";
526 let ParserMatchClass = Imm1_32Operand;
529 class fixedpoint_i64<ValueType FloatVT>
531 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<64>", [fpimm, ld]> {
532 let EncoderMethod = "getFixedPointScaleOpValue";
533 let DecoderMethod = "DecodeFixedPointScaleImm64";
534 let ParserMatchClass = Imm1_64Operand;
537 def fixedpoint_f16_i32 : fixedpoint_i32<f16>;
538 def fixedpoint_f32_i32 : fixedpoint_i32<f32>;
539 def fixedpoint_f64_i32 : fixedpoint_i32<f64>;
541 def fixedpoint_f16_i64 : fixedpoint_i64<f16>;
542 def fixedpoint_f32_i64 : fixedpoint_i64<f32>;
543 def fixedpoint_f64_i64 : fixedpoint_i64<f64>;
545 def vecshiftR8 : Operand<i32>, ImmLeaf<i32, [{
546 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
548 let EncoderMethod = "getVecShiftR8OpValue";
549 let DecoderMethod = "DecodeVecShiftR8Imm";
550 let ParserMatchClass = Imm1_8Operand;
552 def vecshiftR16 : Operand<i32>, ImmLeaf<i32, [{
553 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
555 let EncoderMethod = "getVecShiftR16OpValue";
556 let DecoderMethod = "DecodeVecShiftR16Imm";
557 let ParserMatchClass = Imm1_16Operand;
559 def vecshiftR16Narrow : Operand<i32>, ImmLeaf<i32, [{
560 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
562 let EncoderMethod = "getVecShiftR16OpValue";
563 let DecoderMethod = "DecodeVecShiftR16ImmNarrow";
564 let ParserMatchClass = Imm1_8Operand;
566 def vecshiftR32 : Operand<i32>, ImmLeaf<i32, [{
567 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
569 let EncoderMethod = "getVecShiftR32OpValue";
570 let DecoderMethod = "DecodeVecShiftR32Imm";
571 let ParserMatchClass = Imm1_32Operand;
573 def vecshiftR32Narrow : Operand<i32>, ImmLeaf<i32, [{
574 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
576 let EncoderMethod = "getVecShiftR32OpValue";
577 let DecoderMethod = "DecodeVecShiftR32ImmNarrow";
578 let ParserMatchClass = Imm1_16Operand;
580 def vecshiftR64 : Operand<i32>, ImmLeaf<i32, [{
581 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 65);
583 let EncoderMethod = "getVecShiftR64OpValue";
584 let DecoderMethod = "DecodeVecShiftR64Imm";
585 let ParserMatchClass = Imm1_64Operand;
587 def vecshiftR64Narrow : Operand<i32>, ImmLeaf<i32, [{
588 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
590 let EncoderMethod = "getVecShiftR64OpValue";
591 let DecoderMethod = "DecodeVecShiftR64ImmNarrow";
592 let ParserMatchClass = Imm1_32Operand;
595 def Imm0_1Operand : AsmImmRange<0, 1>;
596 def Imm0_7Operand : AsmImmRange<0, 7>;
597 def Imm0_15Operand : AsmImmRange<0, 15>;
598 def Imm0_31Operand : AsmImmRange<0, 31>;
599 def Imm0_63Operand : AsmImmRange<0, 63>;
601 def vecshiftL8 : Operand<i32>, ImmLeaf<i32, [{
602 return (((uint32_t)Imm) < 8);
604 let EncoderMethod = "getVecShiftL8OpValue";
605 let DecoderMethod = "DecodeVecShiftL8Imm";
606 let ParserMatchClass = Imm0_7Operand;
608 def vecshiftL16 : Operand<i32>, ImmLeaf<i32, [{
609 return (((uint32_t)Imm) < 16);
611 let EncoderMethod = "getVecShiftL16OpValue";
612 let DecoderMethod = "DecodeVecShiftL16Imm";
613 let ParserMatchClass = Imm0_15Operand;
615 def vecshiftL32 : Operand<i32>, ImmLeaf<i32, [{
616 return (((uint32_t)Imm) < 32);
618 let EncoderMethod = "getVecShiftL32OpValue";
619 let DecoderMethod = "DecodeVecShiftL32Imm";
620 let ParserMatchClass = Imm0_31Operand;
622 def vecshiftL64 : Operand<i32>, ImmLeaf<i32, [{
623 return (((uint32_t)Imm) < 64);
625 let EncoderMethod = "getVecShiftL64OpValue";
626 let DecoderMethod = "DecodeVecShiftL64Imm";
627 let ParserMatchClass = Imm0_63Operand;
631 // Crazy immediate formats used by 32-bit and 64-bit logical immediate
632 // instructions for splatting repeating bit patterns across the immediate.
633 def logical_imm32_XFORM : SDNodeXForm<imm, [{
634 uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 32);
635 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
637 def logical_imm64_XFORM : SDNodeXForm<imm, [{
638 uint64_t enc = AArch64_AM::encodeLogicalImmediate(N->getZExtValue(), 64);
639 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
642 let DiagnosticType = "LogicalSecondSource" in {
643 def LogicalImm32Operand : AsmOperandClass {
644 let Name = "LogicalImm32";
645 let PredicateMethod = "isLogicalImm<int32_t>";
646 let RenderMethod = "addLogicalImmOperands<int32_t>";
648 def LogicalImm64Operand : AsmOperandClass {
649 let Name = "LogicalImm64";
650 let PredicateMethod = "isLogicalImm<int64_t>";
651 let RenderMethod = "addLogicalImmOperands<int64_t>";
653 def LogicalImm32NotOperand : AsmOperandClass {
654 let Name = "LogicalImm32Not";
655 let PredicateMethod = "isLogicalImm<int32_t>";
656 let RenderMethod = "addLogicalImmNotOperands<int32_t>";
658 def LogicalImm64NotOperand : AsmOperandClass {
659 let Name = "LogicalImm64Not";
660 let PredicateMethod = "isLogicalImm<int64_t>";
661 let RenderMethod = "addLogicalImmNotOperands<int64_t>";
664 def logical_imm32 : Operand<i32>, IntImmLeaf<i32, [{
665 return AArch64_AM::isLogicalImmediate(Imm.getZExtValue(), 32);
666 }], logical_imm32_XFORM> {
667 let PrintMethod = "printLogicalImm<int32_t>";
668 let ParserMatchClass = LogicalImm32Operand;
670 def logical_imm64 : Operand<i64>, IntImmLeaf<i64, [{
671 return AArch64_AM::isLogicalImmediate(Imm.getZExtValue(), 64);
672 }], logical_imm64_XFORM> {
673 let PrintMethod = "printLogicalImm<int64_t>";
674 let ParserMatchClass = LogicalImm64Operand;
676 def logical_imm32_not : Operand<i32> {
677 let ParserMatchClass = LogicalImm32NotOperand;
679 def logical_imm64_not : Operand<i64> {
680 let ParserMatchClass = LogicalImm64NotOperand;
683 // imm0_65535 predicate - True if the immediate is in the range [0,65535].
684 def Imm0_65535Operand : AsmImmRange<0, 65535>;
685 def imm0_65535 : Operand<i32>, ImmLeaf<i32, [{
686 return ((uint32_t)Imm) < 65536;
688 let ParserMatchClass = Imm0_65535Operand;
689 let PrintMethod = "printImmHex";
692 // imm0_255 predicate - True if the immediate is in the range [0,255].
693 def Imm0_255Operand : AsmImmRange<0,255>;
695 def imm0_255 : Operand<i32>, ImmLeaf<i32, [{
696 return ((uint32_t)Imm) < 256;
698 let ParserMatchClass = Imm0_255Operand;
699 let PrintMethod = "printImm";
702 // imm0_127 predicate - True if the immediate is in the range [0,127]
703 def Imm0_127Operand : AsmImmRange<0, 127>;
704 def imm0_127 : Operand<i32>, ImmLeaf<i32, [{
705 return ((uint32_t)Imm) < 128;
707 let ParserMatchClass = Imm0_127Operand;
708 let PrintMethod = "printImm";
711 // NOTE: These imm0_N operands have to be of type i64 because i64 is the size
712 // for all shift-amounts.
714 // imm0_63 predicate - True if the immediate is in the range [0,63]
715 def imm0_63 : Operand<i64>, ImmLeaf<i64, [{
716 return ((uint64_t)Imm) < 64;
718 let ParserMatchClass = Imm0_63Operand;
721 // imm0_31 predicate - True if the immediate is in the range [0,31]
722 def imm0_31 : Operand<i64>, ImmLeaf<i64, [{
723 return ((uint64_t)Imm) < 32;
725 let ParserMatchClass = Imm0_31Operand;
728 // True if the 32-bit immediate is in the range [0,31]
729 def imm32_0_31 : Operand<i32>, ImmLeaf<i32, [{
730 return ((uint64_t)Imm) < 32;
732 let ParserMatchClass = Imm0_31Operand;
735 // imm0_1 predicate - True if the immediate is in the range [0,1]
736 def imm0_1 : Operand<i64>, ImmLeaf<i64, [{
737 return ((uint64_t)Imm) < 2;
739 let ParserMatchClass = Imm0_1Operand;
742 // imm0_15 predicate - True if the immediate is in the range [0,15]
743 def imm0_15 : Operand<i64>, ImmLeaf<i64, [{
744 return ((uint64_t)Imm) < 16;
746 let ParserMatchClass = Imm0_15Operand;
749 // imm0_7 predicate - True if the immediate is in the range [0,7]
750 def imm0_7 : Operand<i64>, ImmLeaf<i64, [{
751 return ((uint64_t)Imm) < 8;
753 let ParserMatchClass = Imm0_7Operand;
756 // imm32_0_15 predicate - True if the 32-bit immediate is in the range [0,15]
757 def imm32_0_15 : Operand<i32>, ImmLeaf<i32, [{
758 return ((uint32_t)Imm) < 16;
760 let ParserMatchClass = Imm0_15Operand;
763 // An arithmetic shifter operand:
764 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr
766 class arith_shift<ValueType Ty, int width> : Operand<Ty> {
767 let PrintMethod = "printShifter";
768 let ParserMatchClass = !cast<AsmOperandClass>(
769 "ArithmeticShifterOperand" # width);
772 def arith_shift32 : arith_shift<i32, 32>;
773 def arith_shift64 : arith_shift<i64, 64>;
775 class arith_shifted_reg<ValueType Ty, RegisterClass regclass, int width>
777 ComplexPattern<Ty, 2, "SelectArithShiftedRegister", []> {
778 let PrintMethod = "printShiftedRegister";
779 let MIOperandInfo = (ops regclass, !cast<Operand>("arith_shift" # width));
782 def arith_shifted_reg32 : arith_shifted_reg<i32, GPR32, 32>;
783 def arith_shifted_reg64 : arith_shifted_reg<i64, GPR64, 64>;
785 // An arithmetic shifter operand:
786 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr, 11 = ror
788 class logical_shift<int width> : Operand<i32> {
789 let PrintMethod = "printShifter";
790 let ParserMatchClass = !cast<AsmOperandClass>(
791 "LogicalShifterOperand" # width);
794 def logical_shift32 : logical_shift<32>;
795 def logical_shift64 : logical_shift<64>;
797 class logical_shifted_reg<ValueType Ty, RegisterClass regclass, Operand shiftop>
799 ComplexPattern<Ty, 2, "SelectLogicalShiftedRegister", []> {
800 let PrintMethod = "printShiftedRegister";
801 let MIOperandInfo = (ops regclass, shiftop);
804 def logical_shifted_reg32 : logical_shifted_reg<i32, GPR32, logical_shift32>;
805 def logical_shifted_reg64 : logical_shifted_reg<i64, GPR64, logical_shift64>;
807 // A logical vector shifter operand:
808 // {7-6} - shift type: 00 = lsl
809 // {5-0} - imm6: #0, #8, #16, or #24
810 def logical_vec_shift : Operand<i32> {
811 let PrintMethod = "printShifter";
812 let EncoderMethod = "getVecShifterOpValue";
813 let ParserMatchClass = LogicalVecShifterOperand;
816 // A logical vector half-word shifter operand:
817 // {7-6} - shift type: 00 = lsl
818 // {5-0} - imm6: #0 or #8
819 def logical_vec_hw_shift : Operand<i32> {
820 let PrintMethod = "printShifter";
821 let EncoderMethod = "getVecShifterOpValue";
822 let ParserMatchClass = LogicalVecHalfWordShifterOperand;
825 // A vector move shifter operand:
826 // {0} - imm1: #8 or #16
827 def move_vec_shift : Operand<i32> {
828 let PrintMethod = "printShifter";
829 let EncoderMethod = "getMoveVecShifterOpValue";
830 let ParserMatchClass = MoveVecShifterOperand;
833 let DiagnosticType = "AddSubSecondSource" in {
834 def AddSubImmOperand : AsmOperandClass {
835 let Name = "AddSubImm";
836 let ParserMethod = "tryParseImmWithOptionalShift";
837 let RenderMethod = "addImmWithOptionalShiftOperands<12>";
839 def AddSubImmNegOperand : AsmOperandClass {
840 let Name = "AddSubImmNeg";
841 let ParserMethod = "tryParseImmWithOptionalShift";
842 let RenderMethod = "addImmNegWithOptionalShiftOperands<12>";
845 // An ADD/SUB immediate shifter operand:
847 // {7-6} - shift type: 00 = lsl
848 // {5-0} - imm6: #0 or #12
849 class addsub_shifted_imm<ValueType Ty>
850 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectArithImmed", [imm]> {
851 let PrintMethod = "printAddSubImm";
852 let EncoderMethod = "getAddSubImmOpValue";
853 let ParserMatchClass = AddSubImmOperand;
854 let MIOperandInfo = (ops i32imm, i32imm);
857 class addsub_shifted_imm_neg<ValueType Ty>
859 let EncoderMethod = "getAddSubImmOpValue";
860 let ParserMatchClass = AddSubImmNegOperand;
861 let MIOperandInfo = (ops i32imm, i32imm);
864 def addsub_shifted_imm32 : addsub_shifted_imm<i32>;
865 def addsub_shifted_imm64 : addsub_shifted_imm<i64>;
866 def addsub_shifted_imm32_neg : addsub_shifted_imm_neg<i32>;
867 def addsub_shifted_imm64_neg : addsub_shifted_imm_neg<i64>;
869 def gi_addsub_shifted_imm32 :
870 GIComplexOperandMatcher<s32, "selectArithImmed">,
871 GIComplexPatternEquiv<addsub_shifted_imm32>;
873 def gi_addsub_shifted_imm64 :
874 GIComplexOperandMatcher<s64, "selectArithImmed">,
875 GIComplexPatternEquiv<addsub_shifted_imm64>;
877 class neg_addsub_shifted_imm<ValueType Ty>
878 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectNegArithImmed", [imm]> {
879 let PrintMethod = "printAddSubImm";
880 let EncoderMethod = "getAddSubImmOpValue";
881 let ParserMatchClass = AddSubImmOperand;
882 let MIOperandInfo = (ops i32imm, i32imm);
885 def neg_addsub_shifted_imm32 : neg_addsub_shifted_imm<i32>;
886 def neg_addsub_shifted_imm64 : neg_addsub_shifted_imm<i64>;
888 // An extend operand:
889 // {5-3} - extend type
891 def arith_extend : Operand<i32> {
892 let PrintMethod = "printArithExtend";
893 let ParserMatchClass = ExtendOperand;
895 def arith_extend64 : Operand<i32> {
896 let PrintMethod = "printArithExtend";
897 let ParserMatchClass = ExtendOperand64;
900 // 'extend' that's a lsl of a 64-bit register.
901 def arith_extendlsl64 : Operand<i32> {
902 let PrintMethod = "printArithExtend";
903 let ParserMatchClass = ExtendOperandLSL64;
906 class arith_extended_reg32<ValueType Ty> : Operand<Ty>,
907 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
908 let PrintMethod = "printExtendedRegister";
909 let MIOperandInfo = (ops GPR32, arith_extend);
912 class arith_extended_reg32to64<ValueType Ty> : Operand<Ty>,
913 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
914 let PrintMethod = "printExtendedRegister";
915 let MIOperandInfo = (ops GPR32, arith_extend64);
918 // Floating-point immediate.
919 def fpimm16 : Operand<f16>,
921 return AArch64_AM::getFP16Imm(Imm) != -1;
922 }], SDNodeXForm<fpimm, [{
923 APFloat InVal = N->getValueAPF();
924 uint32_t enc = AArch64_AM::getFP16Imm(InVal);
925 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
927 let ParserMatchClass = FPImmOperand;
928 let PrintMethod = "printFPImmOperand";
930 def fpimm32 : Operand<f32>,
932 return AArch64_AM::getFP32Imm(Imm) != -1;
933 }], SDNodeXForm<fpimm, [{
934 APFloat InVal = N->getValueAPF();
935 uint32_t enc = AArch64_AM::getFP32Imm(InVal);
936 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
938 let ParserMatchClass = FPImmOperand;
939 let PrintMethod = "printFPImmOperand";
941 def fpimm64 : Operand<f64>,
943 return AArch64_AM::getFP64Imm(Imm) != -1;
944 }], SDNodeXForm<fpimm, [{
945 APFloat InVal = N->getValueAPF();
946 uint32_t enc = AArch64_AM::getFP64Imm(InVal);
947 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
949 let ParserMatchClass = FPImmOperand;
950 let PrintMethod = "printFPImmOperand";
953 def fpimm8 : Operand<i32> {
954 let ParserMatchClass = FPImmOperand;
955 let PrintMethod = "printFPImmOperand";
958 def fpimm0 : FPImmLeaf<fAny, [{
959 return Imm.isExactlyValue(+0.0);
962 // Vector lane operands
963 class AsmVectorIndex<int Min, int Max, string NamePrefix=""> : AsmOperandClass {
964 let Name = NamePrefix # "IndexRange" # Min # "_" # Max;
965 let DiagnosticType = "Invalid" # Name;
966 let PredicateMethod = "isVectorIndex<" # Min # ", " # Max # ">";
967 let RenderMethod = "addVectorIndexOperands";
970 class AsmVectorIndexOpnd<AsmOperandClass mc, code pred>
971 : Operand<i64>, ImmLeaf<i64, pred> {
972 let ParserMatchClass = mc;
973 let PrintMethod = "printVectorIndex";
976 def VectorIndex1Operand : AsmVectorIndex<1, 1>;
977 def VectorIndexBOperand : AsmVectorIndex<0, 15>;
978 def VectorIndexHOperand : AsmVectorIndex<0, 7>;
979 def VectorIndexSOperand : AsmVectorIndex<0, 3>;
980 def VectorIndexDOperand : AsmVectorIndex<0, 1>;
982 def VectorIndex1 : AsmVectorIndexOpnd<VectorIndex1Operand, [{ return ((uint64_t)Imm) == 1; }]>;
983 def VectorIndexB : AsmVectorIndexOpnd<VectorIndexBOperand, [{ return ((uint64_t)Imm) < 16; }]>;
984 def VectorIndexH : AsmVectorIndexOpnd<VectorIndexHOperand, [{ return ((uint64_t)Imm) < 8; }]>;
985 def VectorIndexS : AsmVectorIndexOpnd<VectorIndexSOperand, [{ return ((uint64_t)Imm) < 4; }]>;
986 def VectorIndexD : AsmVectorIndexOpnd<VectorIndexDOperand, [{ return ((uint64_t)Imm) < 2; }]>;
988 def SVEVectorIndexExtDupBOperand : AsmVectorIndex<0, 63, "SVE">;
989 def SVEVectorIndexExtDupHOperand : AsmVectorIndex<0, 31, "SVE">;
990 def SVEVectorIndexExtDupSOperand : AsmVectorIndex<0, 15, "SVE">;
991 def SVEVectorIndexExtDupDOperand : AsmVectorIndex<0, 7, "SVE">;
992 def SVEVectorIndexExtDupQOperand : AsmVectorIndex<0, 3, "SVE">;
994 def sve_elm_idx_extdup_b
995 : AsmVectorIndexOpnd<SVEVectorIndexExtDupBOperand, [{ return ((uint64_t)Imm) < 64; }]>;
996 def sve_elm_idx_extdup_h
997 : AsmVectorIndexOpnd<SVEVectorIndexExtDupHOperand, [{ return ((uint64_t)Imm) < 32; }]>;
998 def sve_elm_idx_extdup_s
999 : AsmVectorIndexOpnd<SVEVectorIndexExtDupSOperand, [{ return ((uint64_t)Imm) < 16; }]>;
1000 def sve_elm_idx_extdup_d
1001 : AsmVectorIndexOpnd<SVEVectorIndexExtDupDOperand, [{ return ((uint64_t)Imm) < 8; }]>;
1002 def sve_elm_idx_extdup_q
1003 : AsmVectorIndexOpnd<SVEVectorIndexExtDupQOperand, [{ return ((uint64_t)Imm) < 4; }]>;
1005 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
1006 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
1007 // are encoded as the eight bit value 'abcdefgh'.
1008 def simdimmtype10 : Operand<i32>,
1010 return AArch64_AM::isAdvSIMDModImmType10(
1011 Imm.bitcastToAPInt().getZExtValue());
1012 }], SDNodeXForm<fpimm, [{
1013 APFloat InVal = N->getValueAPF();
1014 uint32_t enc = AArch64_AM::encodeAdvSIMDModImmType10(N->getValueAPF()
1017 return CurDAG->getTargetConstant(enc, SDLoc(N), MVT::i32);
1019 let ParserMatchClass = SIMDImmType10Operand;
1020 let PrintMethod = "printSIMDType10Operand";
1025 // System management
1028 // Base encoding for system instruction operands.
1029 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
1030 class BaseSystemI<bit L, dag oops, dag iops, string asm, string operands,
1031 list<dag> pattern = []>
1032 : I<oops, iops, asm, operands, "", pattern> {
1033 let Inst{31-22} = 0b1101010100;
1037 // System instructions which do not have an Rt register.
1038 class SimpleSystemI<bit L, dag iops, string asm, string operands,
1039 list<dag> pattern = []>
1040 : BaseSystemI<L, (outs), iops, asm, operands, pattern> {
1041 let Inst{4-0} = 0b11111;
1044 // System instructions which have an Rt register.
1045 class RtSystemI<bit L, dag oops, dag iops, string asm, string operands>
1046 : BaseSystemI<L, oops, iops, asm, operands>,
1052 // Hint instructions that take both a CRm and a 3-bit immediate.
1053 // NOTE: ideally, this would have mayStore = 0, mayLoad = 0, but we cannot
1054 // model patterns with sufficiently fine granularity
1055 let mayStore = 1, mayLoad = 1, hasSideEffects = 1 in
1056 class HintI<string mnemonic>
1057 : SimpleSystemI<0, (ins imm0_127:$imm), mnemonic#"\t$imm", "",
1058 [(int_aarch64_hint imm0_127:$imm)]>,
1059 Sched<[WriteHint]> {
1061 let Inst{20-12} = 0b000110010;
1062 let Inst{11-5} = imm;
1065 // System instructions taking a single literal operand which encodes into
1066 // CRm. op2 differentiates the opcodes.
1067 def BarrierAsmOperand : AsmOperandClass {
1068 let Name = "Barrier";
1069 let ParserMethod = "tryParseBarrierOperand";
1071 def barrier_op : Operand<i32> {
1072 let PrintMethod = "printBarrierOption";
1073 let ParserMatchClass = BarrierAsmOperand;
1075 class CRmSystemI<Operand crmtype, bits<3> opc, string asm,
1076 list<dag> pattern = []>
1077 : SimpleSystemI<0, (ins crmtype:$CRm), asm, "\t$CRm", pattern>,
1078 Sched<[WriteBarrier]> {
1080 let Inst{20-12} = 0b000110011;
1081 let Inst{11-8} = CRm;
1082 let Inst{7-5} = opc;
1085 class SystemNoOperands<bits<3> op2, string asm, list<dag> pattern = []>
1086 : SimpleSystemI<0, (ins), asm, "", pattern>,
1090 let Inst{31-12} = 0b11010101000000110010;
1091 let Inst{11-8} = CRm;
1092 let Inst{7-5} = op2;
1093 let Inst{4-0} = 0b11111;
1096 // MRS/MSR system instructions. These have different operand classes because
1097 // a different subset of registers can be accessed through each instruction.
1098 def MRSSystemRegisterOperand : AsmOperandClass {
1099 let Name = "MRSSystemRegister";
1100 let ParserMethod = "tryParseSysReg";
1101 let DiagnosticType = "MRS";
1103 // concatenation of op0, op1, CRn, CRm, op2. 16-bit immediate.
1104 def mrs_sysreg_op : Operand<i32> {
1105 let ParserMatchClass = MRSSystemRegisterOperand;
1106 let DecoderMethod = "DecodeMRSSystemRegister";
1107 let PrintMethod = "printMRSSystemRegister";
1110 def MSRSystemRegisterOperand : AsmOperandClass {
1111 let Name = "MSRSystemRegister";
1112 let ParserMethod = "tryParseSysReg";
1113 let DiagnosticType = "MSR";
1115 def msr_sysreg_op : Operand<i32> {
1116 let ParserMatchClass = MSRSystemRegisterOperand;
1117 let DecoderMethod = "DecodeMSRSystemRegister";
1118 let PrintMethod = "printMSRSystemRegister";
1121 def PSBHintOperand : AsmOperandClass {
1122 let Name = "PSBHint";
1123 let ParserMethod = "tryParsePSBHint";
1125 def psbhint_op : Operand<i32> {
1126 let ParserMatchClass = PSBHintOperand;
1127 let PrintMethod = "printPSBHintOp";
1128 let MCOperandPredicate = [{
1129 // Check, if operand is valid, to fix exhaustive aliasing in disassembly.
1130 // "psb" is an alias to "hint" only for certain values of CRm:Op2 fields.
1133 return AArch64PSBHint::lookupPSBByEncoding(MCOp.getImm()) != nullptr;
1137 class MRSI : RtSystemI<1, (outs GPR64:$Rt), (ins mrs_sysreg_op:$systemreg),
1138 "mrs", "\t$Rt, $systemreg"> {
1140 let Inst{20-5} = systemreg;
1143 // FIXME: Some of these def NZCV, others don't. Best way to model that?
1144 // Explicitly modeling each of the system register as a register class
1145 // would do it, but feels like overkill at this point.
1146 class MSRI : RtSystemI<0, (outs), (ins msr_sysreg_op:$systemreg, GPR64:$Rt),
1147 "msr", "\t$systemreg, $Rt"> {
1149 let Inst{20-5} = systemreg;
1152 def SystemPStateFieldWithImm0_15Operand : AsmOperandClass {
1153 let Name = "SystemPStateFieldWithImm0_15";
1154 let ParserMethod = "tryParseSysReg";
1156 def pstatefield4_op : Operand<i32> {
1157 let ParserMatchClass = SystemPStateFieldWithImm0_15Operand;
1158 let PrintMethod = "printSystemPStateField";
1161 let Defs = [NZCV] in
1162 class MSRpstateImm0_15
1163 : SimpleSystemI<0, (ins pstatefield4_op:$pstatefield, imm0_15:$imm),
1164 "msr", "\t$pstatefield, $imm">,
1166 bits<6> pstatefield;
1168 let Inst{20-19} = 0b00;
1169 let Inst{18-16} = pstatefield{5-3};
1170 let Inst{15-12} = 0b0100;
1171 let Inst{11-8} = imm;
1172 let Inst{7-5} = pstatefield{2-0};
1174 let DecoderMethod = "DecodeSystemPStateInstruction";
1175 // MSRpstateI aliases with MSRI. When the MSRpstateI decoder method returns
1176 // Fail the decoder should attempt to decode the instruction as MSRI.
1177 let hasCompleteDecoder = 0;
1180 def SystemPStateFieldWithImm0_1Operand : AsmOperandClass {
1181 let Name = "SystemPStateFieldWithImm0_1";
1182 let ParserMethod = "tryParseSysReg";
1184 def pstatefield1_op : Operand<i32> {
1185 let ParserMatchClass = SystemPStateFieldWithImm0_1Operand;
1186 let PrintMethod = "printSystemPStateField";
1189 let Defs = [NZCV] in
1190 class MSRpstateImm0_1
1191 : SimpleSystemI<0, (ins pstatefield1_op:$pstatefield, imm0_1:$imm),
1192 "msr", "\t$pstatefield, $imm">,
1194 bits<6> pstatefield;
1196 let Inst{20-19} = 0b00;
1197 let Inst{18-16} = pstatefield{5-3};
1198 let Inst{15-9} = 0b0100000;
1200 let Inst{7-5} = pstatefield{2-0};
1202 let DecoderMethod = "DecodeSystemPStateInstruction";
1203 // MSRpstateI aliases with MSRI. When the MSRpstateI decoder method returns
1204 // Fail the decoder should attempt to decode the instruction as MSRI.
1205 let hasCompleteDecoder = 0;
1208 // SYS and SYSL generic system instructions.
1209 def SysCRAsmOperand : AsmOperandClass {
1211 let ParserMethod = "tryParseSysCROperand";
1214 def sys_cr_op : Operand<i32> {
1215 let PrintMethod = "printSysCROperand";
1216 let ParserMatchClass = SysCRAsmOperand;
1219 class SystemXtI<bit L, string asm>
1220 : RtSystemI<L, (outs),
1221 (ins imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, GPR64:$Rt),
1222 asm, "\t$op1, $Cn, $Cm, $op2, $Rt"> {
1227 let Inst{20-19} = 0b01;
1228 let Inst{18-16} = op1;
1229 let Inst{15-12} = Cn;
1230 let Inst{11-8} = Cm;
1231 let Inst{7-5} = op2;
1234 class SystemLXtI<bit L, string asm>
1235 : RtSystemI<L, (outs),
1236 (ins GPR64:$Rt, imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2),
1237 asm, "\t$Rt, $op1, $Cn, $Cm, $op2"> {
1242 let Inst{20-19} = 0b01;
1243 let Inst{18-16} = op1;
1244 let Inst{15-12} = Cn;
1245 let Inst{11-8} = Cm;
1246 let Inst{7-5} = op2;
1250 // Branch (register) instructions:
1258 // otherwise UNDEFINED
1259 class BaseBranchReg<bits<4> opc, dag oops, dag iops, string asm,
1260 string operands, list<dag> pattern>
1261 : I<oops, iops, asm, operands, "", pattern>, Sched<[WriteBrReg]> {
1262 let Inst{31-25} = 0b1101011;
1263 let Inst{24-21} = opc;
1264 let Inst{20-16} = 0b11111;
1265 let Inst{15-10} = 0b000000;
1266 let Inst{4-0} = 0b00000;
1269 class BranchReg<bits<4> opc, string asm, list<dag> pattern>
1270 : BaseBranchReg<opc, (outs), (ins GPR64:$Rn), asm, "\t$Rn", pattern> {
1275 let mayLoad = 0, mayStore = 0, hasSideEffects = 1, isReturn = 1 in
1276 class SpecialReturn<bits<4> opc, string asm>
1277 : BaseBranchReg<opc, (outs), (ins), asm, "", []> {
1278 let Inst{9-5} = 0b11111;
1282 class RCPCLoad<bits<2> sz, string asm, RegisterClass RC>
1283 : I<(outs RC:$Rt), (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]", "", []>,
1287 let Inst{31-30} = sz;
1288 let Inst{29-10} = 0b11100010111111110000;
1293 class AuthBase<bits<1> M, dag oops, dag iops, string asm, string operands,
1295 : I<oops, iops, asm, operands, "", pattern>, Sched<[]> {
1296 let Inst{31-25} = 0b1101011;
1297 let Inst{20-11} = 0b1111100001;
1299 let Inst{4-0} = 0b11111;
1302 class AuthBranchTwoOperands<bits<1> op, bits<1> M, string asm>
1303 : AuthBase<M, (outs), (ins GPR64:$Rn, GPR64sp:$Rm), asm, "\t$Rn, $Rm", []> {
1306 let Inst{24-22} = 0b100;
1312 class AuthOneOperand<bits<3> opc, bits<1> M, string asm>
1313 : AuthBase<M, (outs), (ins GPR64:$Rn), asm, "\t$Rn", []> {
1316 let Inst{23-21} = opc;
1320 class AuthReturn<bits<3> op, bits<1> M, string asm>
1321 : AuthBase<M, (outs), (ins), asm, "", []> {
1323 let Inst{23-21} = op;
1324 let Inst{9-0} = 0b1111111111;
1328 class BaseAuthLoad<bit M, bit W, dag oops, dag iops, string asm,
1329 string operands, string cstr, Operand opr>
1330 : I<oops, iops, asm, operands, cstr, []>, Sched<[]> {
1334 let Inst{31-24} = 0b11111000;
1336 let Inst{22} = offset{9};
1338 let Inst{20-12} = offset{8-0};
1345 multiclass AuthLoad<bit M, string asm, Operand opr> {
1346 def indexed : BaseAuthLoad<M, 0, (outs GPR64:$Rt),
1347 (ins GPR64sp:$Rn, opr:$offset),
1348 asm, "\t$Rt, [$Rn, $offset]", "", opr>;
1349 def writeback : BaseAuthLoad<M, 1, (outs GPR64sp:$wback, GPR64:$Rt),
1350 (ins GPR64sp:$Rn, opr:$offset),
1351 asm, "\t$Rt, [$Rn, $offset]!",
1352 "$Rn = $wback,@earlyclobber $wback", opr>;
1354 def : InstAlias<asm # "\t$Rt, [$Rn]",
1355 (!cast<Instruction>(NAME # "indexed") GPR64:$Rt, GPR64sp:$Rn, 0)>;
1359 // Conditional branch instruction.
1363 // 4-bit immediate. Pretty-printed as <cc>
1364 def ccode : Operand<i32> {
1365 let PrintMethod = "printCondCode";
1366 let ParserMatchClass = CondCode;
1368 def inv_ccode : Operand<i32> {
1369 // AL and NV are invalid in the aliases which use inv_ccode
1370 let PrintMethod = "printInverseCondCode";
1371 let ParserMatchClass = CondCode;
1372 let MCOperandPredicate = [{
1373 return MCOp.isImm() &&
1374 MCOp.getImm() != AArch64CC::AL &&
1375 MCOp.getImm() != AArch64CC::NV;
1379 // Conditional branch target. 19-bit immediate. The low two bits of the target
1380 // offset are implied zero and so are not part of the immediate.
1381 def am_brcond : Operand<OtherVT> {
1382 let EncoderMethod = "getCondBranchTargetOpValue";
1383 let DecoderMethod = "DecodePCRelLabel19";
1384 let PrintMethod = "printAlignedLabel";
1385 let ParserMatchClass = PCRelLabel19Operand;
1386 let OperandType = "OPERAND_PCREL";
1389 class BranchCond : I<(outs), (ins ccode:$cond, am_brcond:$target),
1390 "b", ".$cond\t$target", "",
1391 [(AArch64brcond bb:$target, imm:$cond, NZCV)]>,
1394 let isTerminator = 1;
1399 let Inst{31-24} = 0b01010100;
1400 let Inst{23-5} = target;
1402 let Inst{3-0} = cond;
1406 // Compare-and-branch instructions.
1408 class BaseCmpBranch<RegisterClass regtype, bit op, string asm, SDNode node>
1409 : I<(outs), (ins regtype:$Rt, am_brcond:$target),
1410 asm, "\t$Rt, $target", "",
1411 [(node regtype:$Rt, bb:$target)]>,
1414 let isTerminator = 1;
1418 let Inst{30-25} = 0b011010;
1420 let Inst{23-5} = target;
1424 multiclass CmpBranch<bit op, string asm, SDNode node> {
1425 def W : BaseCmpBranch<GPR32, op, asm, node> {
1428 def X : BaseCmpBranch<GPR64, op, asm, node> {
1434 // Test-bit-and-branch instructions.
1436 // Test-and-branch target. 14-bit sign-extended immediate. The low two bits of
1437 // the target offset are implied zero and so are not part of the immediate.
1438 def am_tbrcond : Operand<OtherVT> {
1439 let EncoderMethod = "getTestBranchTargetOpValue";
1440 let PrintMethod = "printAlignedLabel";
1441 let ParserMatchClass = BranchTarget14Operand;
1442 let OperandType = "OPERAND_PCREL";
1445 // AsmOperand classes to emit (or not) special diagnostics
1446 def TBZImm0_31Operand : AsmOperandClass {
1447 let Name = "TBZImm0_31";
1448 let PredicateMethod = "isImmInRange<0,31>";
1449 let RenderMethod = "addImmOperands";
1451 def TBZImm32_63Operand : AsmOperandClass {
1452 let Name = "Imm32_63";
1453 let PredicateMethod = "isImmInRange<32,63>";
1454 let DiagnosticType = "InvalidImm0_63";
1455 let RenderMethod = "addImmOperands";
1458 class tbz_imm0_31<AsmOperandClass matcher> : Operand<i64>, ImmLeaf<i64, [{
1459 return (((uint32_t)Imm) < 32);
1461 let ParserMatchClass = matcher;
1464 def tbz_imm0_31_diag : tbz_imm0_31<Imm0_31Operand>;
1465 def tbz_imm0_31_nodiag : tbz_imm0_31<TBZImm0_31Operand>;
1467 def tbz_imm32_63 : Operand<i64>, ImmLeaf<i64, [{
1468 return (((uint32_t)Imm) > 31) && (((uint32_t)Imm) < 64);
1470 let ParserMatchClass = TBZImm32_63Operand;
1473 class BaseTestBranch<RegisterClass regtype, Operand immtype,
1474 bit op, string asm, SDNode node>
1475 : I<(outs), (ins regtype:$Rt, immtype:$bit_off, am_tbrcond:$target),
1476 asm, "\t$Rt, $bit_off, $target", "",
1477 [(node regtype:$Rt, immtype:$bit_off, bb:$target)]>,
1480 let isTerminator = 1;
1486 let Inst{30-25} = 0b011011;
1488 let Inst{23-19} = bit_off{4-0};
1489 let Inst{18-5} = target;
1492 let DecoderMethod = "DecodeTestAndBranch";
1495 multiclass TestBranch<bit op, string asm, SDNode node> {
1496 def W : BaseTestBranch<GPR32, tbz_imm0_31_diag, op, asm, node> {
1500 def X : BaseTestBranch<GPR64, tbz_imm32_63, op, asm, node> {
1504 // Alias X-reg with 0-31 imm to W-Reg.
1505 def : InstAlias<asm # "\t$Rd, $imm, $target",
1506 (!cast<Instruction>(NAME#"W") GPR32as64:$Rd,
1507 tbz_imm0_31_nodiag:$imm, am_tbrcond:$target), 0>;
1508 def : Pat<(node GPR64:$Rn, tbz_imm0_31_diag:$imm, bb:$target),
1509 (!cast<Instruction>(NAME#"W") (EXTRACT_SUBREG GPR64:$Rn, sub_32),
1510 tbz_imm0_31_diag:$imm, bb:$target)>;
1514 // Unconditional branch (immediate) instructions.
1516 def am_b_target : Operand<OtherVT> {
1517 let EncoderMethod = "getBranchTargetOpValue";
1518 let PrintMethod = "printAlignedLabel";
1519 let ParserMatchClass = BranchTarget26Operand;
1520 let OperandType = "OPERAND_PCREL";
1522 def am_bl_target : Operand<i64> {
1523 let EncoderMethod = "getBranchTargetOpValue";
1524 let PrintMethod = "printAlignedLabel";
1525 let ParserMatchClass = BranchTarget26Operand;
1526 let OperandType = "OPERAND_PCREL";
1529 class BImm<bit op, dag iops, string asm, list<dag> pattern>
1530 : I<(outs), iops, asm, "\t$addr", "", pattern>, Sched<[WriteBr]> {
1533 let Inst{30-26} = 0b00101;
1534 let Inst{25-0} = addr;
1536 let DecoderMethod = "DecodeUnconditionalBranch";
1539 class BranchImm<bit op, string asm, list<dag> pattern>
1540 : BImm<op, (ins am_b_target:$addr), asm, pattern>;
1541 class CallImm<bit op, string asm, list<dag> pattern>
1542 : BImm<op, (ins am_bl_target:$addr), asm, pattern>;
1545 // Basic one-operand data processing instructions.
1548 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1549 class BaseOneOperandData<bits<3> opc, RegisterClass regtype, string asm,
1550 SDPatternOperator node>
1551 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
1552 [(set regtype:$Rd, (node regtype:$Rn))]>,
1553 Sched<[WriteI, ReadI]> {
1557 let Inst{30-13} = 0b101101011000000000;
1558 let Inst{12-10} = opc;
1563 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1564 multiclass OneOperandData<bits<3> opc, string asm,
1565 SDPatternOperator node = null_frag> {
1566 def Wr : BaseOneOperandData<opc, GPR32, asm, node> {
1570 def Xr : BaseOneOperandData<opc, GPR64, asm, node> {
1575 class OneWRegData<bits<3> opc, string asm, SDPatternOperator node>
1576 : BaseOneOperandData<opc, GPR32, asm, node> {
1580 class OneXRegData<bits<3> opc, string asm, SDPatternOperator node>
1581 : BaseOneOperandData<opc, GPR64, asm, node> {
1585 class SignAuthOneData<bits<3> opcode_prefix, bits<2> opcode, string asm>
1586 : I<(outs GPR64:$Rd), (ins GPR64sp:$Rn), asm, "\t$Rd, $Rn", "",
1588 Sched<[WriteI, ReadI]> {
1591 let Inst{31-15} = 0b11011010110000010;
1592 let Inst{14-12} = opcode_prefix;
1593 let Inst{11-10} = opcode;
1598 class SignAuthZero<bits<3> opcode_prefix, bits<2> opcode, string asm>
1599 : I<(outs GPR64:$Rd), (ins), asm, "\t$Rd", "", []>, Sched<[]> {
1601 let Inst{31-15} = 0b11011010110000010;
1602 let Inst{14-12} = opcode_prefix;
1603 let Inst{11-10} = opcode;
1604 let Inst{9-5} = 0b11111;
1608 class SignAuthTwoOperand<bits<4> opc, string asm,
1609 SDPatternOperator OpNode>
1610 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64sp:$Rm),
1611 asm, "\t$Rd, $Rn, $Rm", "",
1612 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64sp:$Rm))]>,
1613 Sched<[WriteI, ReadI, ReadI]> {
1617 let Inst{31-21} = 0b10011010110;
1618 let Inst{20-16} = Rm;
1619 let Inst{15-14} = 0b00;
1620 let Inst{13-10} = opc;
1625 // Base class for the Armv8.4-A 8 and 16-bit flag manipulation instructions
1626 class BaseFlagManipulation<bit sf, bit sz, dag iops, string asm, string ops>
1627 : I<(outs), iops, asm, ops, "", []>,
1628 Sched<[WriteI, ReadI, ReadI]> {
1632 let Inst{30-15} = 0b0111010000000000;
1634 let Inst{13-10} = 0b0010;
1636 let Inst{4-0} = 0b01101;
1639 class FlagRotate<dag iops, string asm, string ops>
1640 : BaseFlagManipulation<0b1, 0b0, iops, asm, ops> {
1643 let Inst{20-15} = imm;
1644 let Inst{13-10} = 0b0001;
1646 let Inst{3-0} = mask;
1650 // Basic two-operand data processing instructions.
1652 class BaseBaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1654 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1655 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
1656 Sched<[WriteI, ReadI, ReadI]> {
1661 let Inst{30} = isSub;
1662 let Inst{28-21} = 0b11010000;
1663 let Inst{20-16} = Rm;
1664 let Inst{15-10} = 0;
1669 class BaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1671 : BaseBaseAddSubCarry<isSub, regtype, asm,
1672 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, NZCV))]>;
1674 class BaseAddSubCarrySetFlags<bit isSub, RegisterClass regtype, string asm,
1676 : BaseBaseAddSubCarry<isSub, regtype, asm,
1677 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, NZCV)),
1682 multiclass AddSubCarry<bit isSub, string asm, string asm_setflags,
1683 SDNode OpNode, SDNode OpNode_setflags> {
1684 def Wr : BaseAddSubCarry<isSub, GPR32, asm, OpNode> {
1688 def Xr : BaseAddSubCarry<isSub, GPR64, asm, OpNode> {
1694 def SWr : BaseAddSubCarrySetFlags<isSub, GPR32, asm_setflags,
1699 def SXr : BaseAddSubCarrySetFlags<isSub, GPR64, asm_setflags,
1706 class BaseTwoOperand<bits<4> opc, RegisterClass regtype, string asm,
1707 SDPatternOperator OpNode>
1708 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1709 asm, "\t$Rd, $Rn, $Rm", "",
1710 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]> {
1714 let Inst{30-21} = 0b0011010110;
1715 let Inst{20-16} = Rm;
1716 let Inst{15-14} = 0b00;
1717 let Inst{13-10} = opc;
1722 class BaseDiv<bit isSigned, RegisterClass regtype, string asm,
1723 SDPatternOperator OpNode>
1724 : BaseTwoOperand<{0,0,1,?}, regtype, asm, OpNode> {
1725 let Inst{10} = isSigned;
1728 multiclass Div<bit isSigned, string asm, SDPatternOperator OpNode> {
1729 def Wr : BaseDiv<isSigned, GPR32, asm, OpNode>,
1730 Sched<[WriteID32, ReadID, ReadID]> {
1733 def Xr : BaseDiv<isSigned, GPR64, asm, OpNode>,
1734 Sched<[WriteID64, ReadID, ReadID]> {
1739 class BaseShift<bits<2> shift_type, RegisterClass regtype, string asm,
1740 SDPatternOperator OpNode = null_frag>
1741 : BaseTwoOperand<{1,0,?,?}, regtype, asm, OpNode>,
1742 Sched<[WriteIS, ReadI]> {
1743 let Inst{11-10} = shift_type;
1746 multiclass Shift<bits<2> shift_type, string asm, SDNode OpNode> {
1747 def Wr : BaseShift<shift_type, GPR32, asm> {
1751 def Xr : BaseShift<shift_type, GPR64, asm, OpNode> {
1755 def : Pat<(i32 (OpNode GPR32:$Rn, i64:$Rm)),
1756 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn,
1757 (EXTRACT_SUBREG i64:$Rm, sub_32))>;
1759 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (zext GPR32:$Rm)))),
1760 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1762 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (anyext GPR32:$Rm)))),
1763 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1765 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (sext GPR32:$Rm)))),
1766 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1769 class ShiftAlias<string asm, Instruction inst, RegisterClass regtype>
1770 : InstAlias<asm#"\t$dst, $src1, $src2",
1771 (inst regtype:$dst, regtype:$src1, regtype:$src2), 0>;
1773 class BaseMulAccum<bit isSub, bits<3> opc, RegisterClass multype,
1774 RegisterClass addtype, string asm,
1776 : I<(outs addtype:$Rd), (ins multype:$Rn, multype:$Rm, addtype:$Ra),
1777 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pattern> {
1782 let Inst{30-24} = 0b0011011;
1783 let Inst{23-21} = opc;
1784 let Inst{20-16} = Rm;
1785 let Inst{15} = isSub;
1786 let Inst{14-10} = Ra;
1791 multiclass MulAccum<bit isSub, string asm, SDNode AccNode> {
1792 // MADD/MSUB generation is decided by MachineCombiner.cpp
1793 def Wrrr : BaseMulAccum<isSub, 0b000, GPR32, GPR32, asm,
1794 [/*(set GPR32:$Rd, (AccNode GPR32:$Ra, (mul GPR32:$Rn, GPR32:$Rm)))*/]>,
1795 Sched<[WriteIM32, ReadIM, ReadIM, ReadIMA]> {
1799 def Xrrr : BaseMulAccum<isSub, 0b000, GPR64, GPR64, asm,
1800 [/*(set GPR64:$Rd, (AccNode GPR64:$Ra, (mul GPR64:$Rn, GPR64:$Rm)))*/]>,
1801 Sched<[WriteIM64, ReadIM, ReadIM, ReadIMA]> {
1806 class WideMulAccum<bit isSub, bits<3> opc, string asm,
1807 SDNode AccNode, SDNode ExtNode>
1808 : BaseMulAccum<isSub, opc, GPR32, GPR64, asm,
1809 [(set GPR64:$Rd, (AccNode GPR64:$Ra,
1810 (mul (ExtNode GPR32:$Rn), (ExtNode GPR32:$Rm))))]>,
1811 Sched<[WriteIM32, ReadIM, ReadIM, ReadIMA]> {
1815 class MulHi<bits<3> opc, string asm, SDNode OpNode>
1816 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64:$Rm),
1817 asm, "\t$Rd, $Rn, $Rm", "",
1818 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64:$Rm))]>,
1819 Sched<[WriteIM64, ReadIM, ReadIM]> {
1823 let Inst{31-24} = 0b10011011;
1824 let Inst{23-21} = opc;
1825 let Inst{20-16} = Rm;
1830 // The Ra field of SMULH and UMULH is unused: it should be assembled as 31
1831 // (i.e. all bits 1) but is ignored by the processor.
1832 let PostEncoderMethod = "fixMulHigh";
1835 class MulAccumWAlias<string asm, Instruction inst>
1836 : InstAlias<asm#"\t$dst, $src1, $src2",
1837 (inst GPR32:$dst, GPR32:$src1, GPR32:$src2, WZR)>;
1838 class MulAccumXAlias<string asm, Instruction inst>
1839 : InstAlias<asm#"\t$dst, $src1, $src2",
1840 (inst GPR64:$dst, GPR64:$src1, GPR64:$src2, XZR)>;
1841 class WideMulAccumAlias<string asm, Instruction inst>
1842 : InstAlias<asm#"\t$dst, $src1, $src2",
1843 (inst GPR64:$dst, GPR32:$src1, GPR32:$src2, XZR)>;
1845 class BaseCRC32<bit sf, bits<2> sz, bit C, RegisterClass StreamReg,
1846 SDPatternOperator OpNode, string asm>
1847 : I<(outs GPR32:$Rd), (ins GPR32:$Rn, StreamReg:$Rm),
1848 asm, "\t$Rd, $Rn, $Rm", "",
1849 [(set GPR32:$Rd, (OpNode GPR32:$Rn, StreamReg:$Rm))]>,
1850 Sched<[WriteISReg, ReadI, ReadISReg]> {
1856 let Inst{30-21} = 0b0011010110;
1857 let Inst{20-16} = Rm;
1858 let Inst{15-13} = 0b010;
1860 let Inst{11-10} = sz;
1863 let Predicates = [HasCRC];
1867 // Address generation.
1870 class ADRI<bit page, string asm, Operand adr, list<dag> pattern>
1871 : I<(outs GPR64:$Xd), (ins adr:$label), asm, "\t$Xd, $label", "",
1876 let Inst{31} = page;
1877 let Inst{30-29} = label{1-0};
1878 let Inst{28-24} = 0b10000;
1879 let Inst{23-5} = label{20-2};
1882 let DecoderMethod = "DecodeAdrInstruction";
1889 def movimm32_imm : Operand<i32> {
1890 let ParserMatchClass = Imm0_65535Operand;
1891 let EncoderMethod = "getMoveWideImmOpValue";
1892 let PrintMethod = "printImm";
1894 def movimm32_shift : Operand<i32> {
1895 let PrintMethod = "printShifter";
1896 let ParserMatchClass = MovImm32ShifterOperand;
1898 def movimm64_shift : Operand<i32> {
1899 let PrintMethod = "printShifter";
1900 let ParserMatchClass = MovImm64ShifterOperand;
1903 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1904 class BaseMoveImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1906 : I<(outs regtype:$Rd), (ins movimm32_imm:$imm, shifter:$shift),
1907 asm, "\t$Rd, $imm$shift", "", []>,
1912 let Inst{30-29} = opc;
1913 let Inst{28-23} = 0b100101;
1914 let Inst{22-21} = shift{5-4};
1915 let Inst{20-5} = imm;
1918 let DecoderMethod = "DecodeMoveImmInstruction";
1921 multiclass MoveImmediate<bits<2> opc, string asm> {
1922 def Wi : BaseMoveImmediate<opc, GPR32, movimm32_shift, asm> {
1926 def Xi : BaseMoveImmediate<opc, GPR64, movimm64_shift, asm> {
1931 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1932 class BaseInsertImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1934 : I<(outs regtype:$Rd),
1935 (ins regtype:$src, movimm32_imm:$imm, shifter:$shift),
1936 asm, "\t$Rd, $imm$shift", "$src = $Rd", []>,
1937 Sched<[WriteI, ReadI]> {
1941 let Inst{30-29} = opc;
1942 let Inst{28-23} = 0b100101;
1943 let Inst{22-21} = shift{5-4};
1944 let Inst{20-5} = imm;
1947 let DecoderMethod = "DecodeMoveImmInstruction";
1950 multiclass InsertImmediate<bits<2> opc, string asm> {
1951 def Wi : BaseInsertImmediate<opc, GPR32, movimm32_shift, asm> {
1955 def Xi : BaseInsertImmediate<opc, GPR64, movimm64_shift, asm> {
1964 class BaseAddSubImm<bit isSub, bit setFlags, RegisterClass dstRegtype,
1965 RegisterClass srcRegtype, addsub_shifted_imm immtype,
1966 string asm, SDPatternOperator OpNode>
1967 : I<(outs dstRegtype:$Rd), (ins srcRegtype:$Rn, immtype:$imm),
1968 asm, "\t$Rd, $Rn, $imm", "",
1969 [(set dstRegtype:$Rd, (OpNode srcRegtype:$Rn, immtype:$imm))]>,
1970 Sched<[WriteI, ReadI]> {
1974 let Inst{30} = isSub;
1975 let Inst{29} = setFlags;
1976 let Inst{28-24} = 0b10001;
1977 let Inst{23-22} = imm{13-12}; // '00' => lsl #0, '01' => lsl #12
1978 let Inst{21-10} = imm{11-0};
1981 let DecoderMethod = "DecodeBaseAddSubImm";
1984 class BaseAddSubRegPseudo<RegisterClass regtype,
1985 SDPatternOperator OpNode>
1986 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1987 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
1988 Sched<[WriteI, ReadI, ReadI]>;
1990 class BaseAddSubSReg<bit isSub, bit setFlags, RegisterClass regtype,
1991 arith_shifted_reg shifted_regtype, string asm,
1992 SDPatternOperator OpNode>
1993 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
1994 asm, "\t$Rd, $Rn, $Rm", "",
1995 [(set regtype:$Rd, (OpNode regtype:$Rn, shifted_regtype:$Rm))]>,
1996 Sched<[WriteISReg, ReadI, ReadISReg]> {
1997 // The operands are in order to match the 'addr' MI operands, so we
1998 // don't need an encoder method and by-name matching. Just use the default
1999 // in-order handling. Since we're using by-order, make sure the names
2005 let Inst{30} = isSub;
2006 let Inst{29} = setFlags;
2007 let Inst{28-24} = 0b01011;
2008 let Inst{23-22} = shift{7-6};
2010 let Inst{20-16} = src2;
2011 let Inst{15-10} = shift{5-0};
2012 let Inst{9-5} = src1;
2013 let Inst{4-0} = dst;
2015 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
2018 class BaseAddSubEReg<bit isSub, bit setFlags, RegisterClass dstRegtype,
2019 RegisterClass src1Regtype, Operand src2Regtype,
2020 string asm, SDPatternOperator OpNode>
2021 : I<(outs dstRegtype:$R1),
2022 (ins src1Regtype:$R2, src2Regtype:$R3),
2023 asm, "\t$R1, $R2, $R3", "",
2024 [(set dstRegtype:$R1, (OpNode src1Regtype:$R2, src2Regtype:$R3))]>,
2025 Sched<[WriteIEReg, ReadI, ReadIEReg]> {
2030 let Inst{30} = isSub;
2031 let Inst{29} = setFlags;
2032 let Inst{28-24} = 0b01011;
2033 let Inst{23-21} = 0b001;
2034 let Inst{20-16} = Rm;
2035 let Inst{15-13} = ext{5-3};
2036 let Inst{12-10} = ext{2-0};
2040 let DecoderMethod = "DecodeAddSubERegInstruction";
2043 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2044 class BaseAddSubEReg64<bit isSub, bit setFlags, RegisterClass dstRegtype,
2045 RegisterClass src1Regtype, RegisterClass src2Regtype,
2046 Operand ext_op, string asm>
2047 : I<(outs dstRegtype:$Rd),
2048 (ins src1Regtype:$Rn, src2Regtype:$Rm, ext_op:$ext),
2049 asm, "\t$Rd, $Rn, $Rm$ext", "", []>,
2050 Sched<[WriteIEReg, ReadI, ReadIEReg]> {
2055 let Inst{30} = isSub;
2056 let Inst{29} = setFlags;
2057 let Inst{28-24} = 0b01011;
2058 let Inst{23-21} = 0b001;
2059 let Inst{20-16} = Rm;
2060 let Inst{15} = ext{5};
2061 let Inst{12-10} = ext{2-0};
2065 let DecoderMethod = "DecodeAddSubERegInstruction";
2068 // Aliases for register+register add/subtract.
2069 class AddSubRegAlias<string asm, Instruction inst, RegisterClass dstRegtype,
2070 RegisterClass src1Regtype, RegisterClass src2Regtype,
2072 : InstAlias<asm#"\t$dst, $src1, $src2",
2073 (inst dstRegtype:$dst, src1Regtype:$src1, src2Regtype:$src2,
2076 multiclass AddSub<bit isSub, string mnemonic, string alias,
2077 SDPatternOperator OpNode = null_frag> {
2078 let hasSideEffects = 0, isReMaterializable = 1, isAsCheapAsAMove = 1 in {
2079 // Add/Subtract immediate
2080 // Increase the weight of the immediate variant to try to match it before
2081 // the extended register variant.
2082 // We used to match the register variant before the immediate when the
2083 // register argument could be implicitly zero-extended.
2084 let AddedComplexity = 6 in
2085 def Wri : BaseAddSubImm<isSub, 0, GPR32sp, GPR32sp, addsub_shifted_imm32,
2089 let AddedComplexity = 6 in
2090 def Xri : BaseAddSubImm<isSub, 0, GPR64sp, GPR64sp, addsub_shifted_imm64,
2095 // Add/Subtract register - Only used for CodeGen
2096 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
2097 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
2099 // Add/Subtract shifted register
2100 def Wrs : BaseAddSubSReg<isSub, 0, GPR32, arith_shifted_reg32, mnemonic,
2104 def Xrs : BaseAddSubSReg<isSub, 0, GPR64, arith_shifted_reg64, mnemonic,
2110 // Add/Subtract extended register
2111 let AddedComplexity = 1, hasSideEffects = 0 in {
2112 def Wrx : BaseAddSubEReg<isSub, 0, GPR32sp, GPR32sp,
2113 arith_extended_reg32<i32>, mnemonic, OpNode> {
2116 def Xrx : BaseAddSubEReg<isSub, 0, GPR64sp, GPR64sp,
2117 arith_extended_reg32to64<i64>, mnemonic, OpNode> {
2122 def Xrx64 : BaseAddSubEReg64<isSub, 0, GPR64sp, GPR64sp, GPR64,
2123 arith_extendlsl64, mnemonic> {
2124 // UXTX and SXTX only.
2125 let Inst{14-13} = 0b11;
2129 // add Rd, Rb, -imm -> sub Rd, Rn, imm
2130 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2131 (!cast<Instruction>(NAME # "Wri") GPR32sp:$Rd, GPR32sp:$Rn,
2132 addsub_shifted_imm32_neg:$imm), 0>;
2133 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2134 (!cast<Instruction>(NAME # "Xri") GPR64sp:$Rd, GPR64sp:$Rn,
2135 addsub_shifted_imm64_neg:$imm), 0>;
2137 // Register/register aliases with no shift when SP is not used.
2138 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
2139 GPR32, GPR32, GPR32, 0>;
2140 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
2141 GPR64, GPR64, GPR64, 0>;
2143 // Register/register aliases with no shift when either the destination or
2144 // first source register is SP.
2145 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2146 GPR32sponly, GPR32sp, GPR32, 16>; // UXTW #0
2147 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2148 GPR32sp, GPR32sponly, GPR32, 16>; // UXTW #0
2149 def : AddSubRegAlias<mnemonic,
2150 !cast<Instruction>(NAME#"Xrx64"),
2151 GPR64sponly, GPR64sp, GPR64, 24>; // UXTX #0
2152 def : AddSubRegAlias<mnemonic,
2153 !cast<Instruction>(NAME#"Xrx64"),
2154 GPR64sp, GPR64sponly, GPR64, 24>; // UXTX #0
2157 multiclass AddSubS<bit isSub, string mnemonic, SDNode OpNode, string cmp,
2158 string alias, string cmpAlias> {
2159 let isCompare = 1, Defs = [NZCV] in {
2160 // Add/Subtract immediate
2161 def Wri : BaseAddSubImm<isSub, 1, GPR32, GPR32sp, addsub_shifted_imm32,
2165 def Xri : BaseAddSubImm<isSub, 1, GPR64, GPR64sp, addsub_shifted_imm64,
2170 // Add/Subtract register
2171 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
2172 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
2174 // Add/Subtract shifted register
2175 def Wrs : BaseAddSubSReg<isSub, 1, GPR32, arith_shifted_reg32, mnemonic,
2179 def Xrs : BaseAddSubSReg<isSub, 1, GPR64, arith_shifted_reg64, mnemonic,
2184 // Add/Subtract extended register
2185 let AddedComplexity = 1 in {
2186 def Wrx : BaseAddSubEReg<isSub, 1, GPR32, GPR32sp,
2187 arith_extended_reg32<i32>, mnemonic, OpNode> {
2190 def Xrx : BaseAddSubEReg<isSub, 1, GPR64, GPR64sp,
2191 arith_extended_reg32<i64>, mnemonic, OpNode> {
2196 def Xrx64 : BaseAddSubEReg64<isSub, 1, GPR64, GPR64sp, GPR64,
2197 arith_extendlsl64, mnemonic> {
2198 // UXTX and SXTX only.
2199 let Inst{14-13} = 0b11;
2204 // Support negative immediates, e.g. adds Rd, Rn, -imm -> subs Rd, Rn, imm
2205 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2206 (!cast<Instruction>(NAME # "Wri") GPR32:$Rd, GPR32sp:$Rn,
2207 addsub_shifted_imm32_neg:$imm), 0>;
2208 def : InstSubst<alias#"\t$Rd, $Rn, $imm",
2209 (!cast<Instruction>(NAME # "Xri") GPR64:$Rd, GPR64sp:$Rn,
2210 addsub_shifted_imm64_neg:$imm), 0>;
2213 def : InstAlias<cmp#"\t$src, $imm", (!cast<Instruction>(NAME#"Wri")
2214 WZR, GPR32sp:$src, addsub_shifted_imm32:$imm), 5>;
2215 def : InstAlias<cmp#"\t$src, $imm", (!cast<Instruction>(NAME#"Xri")
2216 XZR, GPR64sp:$src, addsub_shifted_imm64:$imm), 5>;
2217 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Wrx")
2218 WZR, GPR32sp:$src1, GPR32:$src2, arith_extend:$sh), 4>;
2219 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrx")
2220 XZR, GPR64sp:$src1, GPR32:$src2, arith_extend:$sh), 4>;
2221 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrx64")
2222 XZR, GPR64sp:$src1, GPR64:$src2, arith_extendlsl64:$sh), 4>;
2223 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Wrs")
2224 WZR, GPR32:$src1, GPR32:$src2, arith_shift32:$sh), 4>;
2225 def : InstAlias<cmp#"\t$src1, $src2$sh", (!cast<Instruction>(NAME#"Xrs")
2226 XZR, GPR64:$src1, GPR64:$src2, arith_shift64:$sh), 4>;
2228 // Support negative immediates, e.g. cmp Rn, -imm -> cmn Rn, imm
2229 def : InstSubst<cmpAlias#"\t$src, $imm", (!cast<Instruction>(NAME#"Wri")
2230 WZR, GPR32sp:$src, addsub_shifted_imm32_neg:$imm), 0>;
2231 def : InstSubst<cmpAlias#"\t$src, $imm", (!cast<Instruction>(NAME#"Xri")
2232 XZR, GPR64sp:$src, addsub_shifted_imm64_neg:$imm), 0>;
2234 // Compare shorthands
2235 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Wrs")
2236 WZR, GPR32:$src1, GPR32:$src2, 0), 5>;
2237 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Xrs")
2238 XZR, GPR64:$src1, GPR64:$src2, 0), 5>;
2239 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Wrx")
2240 WZR, GPR32sponly:$src1, GPR32:$src2, 16), 5>;
2241 def : InstAlias<cmp#"\t$src1, $src2", (!cast<Instruction>(NAME#"Xrx64")
2242 XZR, GPR64sponly:$src1, GPR64:$src2, 24), 5>;
2244 // Register/register aliases with no shift when SP is not used.
2245 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
2246 GPR32, GPR32, GPR32, 0>;
2247 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
2248 GPR64, GPR64, GPR64, 0>;
2250 // Register/register aliases with no shift when the first source register
2252 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
2253 GPR32, GPR32sponly, GPR32, 16>; // UXTW #0
2254 def : AddSubRegAlias<mnemonic,
2255 !cast<Instruction>(NAME#"Xrx64"),
2256 GPR64, GPR64sponly, GPR64, 24>; // UXTX #0
2262 def SDTA64EXTR : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>,
2264 def AArch64Extr : SDNode<"AArch64ISD::EXTR", SDTA64EXTR>;
2266 class BaseExtractImm<RegisterClass regtype, Operand imm_type, string asm,
2268 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, imm_type:$imm),
2269 asm, "\t$Rd, $Rn, $Rm, $imm", "", patterns>,
2270 Sched<[WriteExtr, ReadExtrHi]> {
2276 let Inst{30-23} = 0b00100111;
2278 let Inst{20-16} = Rm;
2279 let Inst{15-10} = imm;
2284 multiclass ExtractImm<string asm> {
2285 def Wrri : BaseExtractImm<GPR32, imm0_31, asm,
2287 (AArch64Extr GPR32:$Rn, GPR32:$Rm, imm0_31:$imm))]> {
2290 // imm<5> must be zero.
2293 def Xrri : BaseExtractImm<GPR64, imm0_63, asm,
2295 (AArch64Extr GPR64:$Rn, GPR64:$Rm, imm0_63:$imm))]> {
2306 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2307 class BaseBitfieldImm<bits<2> opc,
2308 RegisterClass regtype, Operand imm_type, string asm>
2309 : I<(outs regtype:$Rd), (ins regtype:$Rn, imm_type:$immr, imm_type:$imms),
2310 asm, "\t$Rd, $Rn, $immr, $imms", "", []>,
2311 Sched<[WriteIS, ReadI]> {
2317 let Inst{30-29} = opc;
2318 let Inst{28-23} = 0b100110;
2319 let Inst{21-16} = immr;
2320 let Inst{15-10} = imms;
2325 multiclass BitfieldImm<bits<2> opc, string asm> {
2326 def Wri : BaseBitfieldImm<opc, GPR32, imm0_31, asm> {
2329 // imms<5> and immr<5> must be zero, else ReservedValue().
2333 def Xri : BaseBitfieldImm<opc, GPR64, imm0_63, asm> {
2339 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2340 class BaseBitfieldImmWith2RegArgs<bits<2> opc,
2341 RegisterClass regtype, Operand imm_type, string asm>
2342 : I<(outs regtype:$Rd), (ins regtype:$src, regtype:$Rn, imm_type:$immr,
2344 asm, "\t$Rd, $Rn, $immr, $imms", "$src = $Rd", []>,
2345 Sched<[WriteIS, ReadI]> {
2351 let Inst{30-29} = opc;
2352 let Inst{28-23} = 0b100110;
2353 let Inst{21-16} = immr;
2354 let Inst{15-10} = imms;
2359 multiclass BitfieldImmWith2RegArgs<bits<2> opc, string asm> {
2360 def Wri : BaseBitfieldImmWith2RegArgs<opc, GPR32, imm0_31, asm> {
2363 // imms<5> and immr<5> must be zero, else ReservedValue().
2367 def Xri : BaseBitfieldImmWith2RegArgs<opc, GPR64, imm0_63, asm> {
2377 // Logical (immediate)
2378 class BaseLogicalImm<bits<2> opc, RegisterClass dregtype,
2379 RegisterClass sregtype, Operand imm_type, string asm,
2381 : I<(outs dregtype:$Rd), (ins sregtype:$Rn, imm_type:$imm),
2382 asm, "\t$Rd, $Rn, $imm", "", pattern>,
2383 Sched<[WriteI, ReadI]> {
2387 let Inst{30-29} = opc;
2388 let Inst{28-23} = 0b100100;
2389 let Inst{22} = imm{12};
2390 let Inst{21-16} = imm{11-6};
2391 let Inst{15-10} = imm{5-0};
2395 let DecoderMethod = "DecodeLogicalImmInstruction";
2398 // Logical (shifted register)
2399 class BaseLogicalSReg<bits<2> opc, bit N, RegisterClass regtype,
2400 logical_shifted_reg shifted_regtype, string asm,
2402 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
2403 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
2404 Sched<[WriteISReg, ReadI, ReadISReg]> {
2405 // The operands are in order to match the 'addr' MI operands, so we
2406 // don't need an encoder method and by-name matching. Just use the default
2407 // in-order handling. Since we're using by-order, make sure the names
2413 let Inst{30-29} = opc;
2414 let Inst{28-24} = 0b01010;
2415 let Inst{23-22} = shift{7-6};
2417 let Inst{20-16} = src2;
2418 let Inst{15-10} = shift{5-0};
2419 let Inst{9-5} = src1;
2420 let Inst{4-0} = dst;
2422 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
2425 // Aliases for register+register logical instructions.
2426 class LogicalRegAlias<string asm, Instruction inst, RegisterClass regtype>
2427 : InstAlias<asm#"\t$dst, $src1, $src2",
2428 (inst regtype:$dst, regtype:$src1, regtype:$src2, 0)>;
2430 multiclass LogicalImm<bits<2> opc, string mnemonic, SDNode OpNode,
2432 let AddedComplexity = 6, isReMaterializable = 1, isAsCheapAsAMove = 1 in
2433 def Wri : BaseLogicalImm<opc, GPR32sp, GPR32, logical_imm32, mnemonic,
2434 [(set GPR32sp:$Rd, (OpNode GPR32:$Rn,
2435 logical_imm32:$imm))]> {
2437 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
2439 let AddedComplexity = 6, isReMaterializable = 1, isAsCheapAsAMove = 1 in
2440 def Xri : BaseLogicalImm<opc, GPR64sp, GPR64, logical_imm64, mnemonic,
2441 [(set GPR64sp:$Rd, (OpNode GPR64:$Rn,
2442 logical_imm64:$imm))]> {
2446 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2447 (!cast<Instruction>(NAME # "Wri") GPR32sp:$Rd, GPR32:$Rn,
2448 logical_imm32_not:$imm), 0>;
2449 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2450 (!cast<Instruction>(NAME # "Xri") GPR64sp:$Rd, GPR64:$Rn,
2451 logical_imm64_not:$imm), 0>;
2454 multiclass LogicalImmS<bits<2> opc, string mnemonic, SDNode OpNode,
2456 let isCompare = 1, Defs = [NZCV] in {
2457 def Wri : BaseLogicalImm<opc, GPR32, GPR32, logical_imm32, mnemonic,
2458 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_imm32:$imm))]> {
2460 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
2462 def Xri : BaseLogicalImm<opc, GPR64, GPR64, logical_imm64, mnemonic,
2463 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_imm64:$imm))]> {
2466 } // end Defs = [NZCV]
2468 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2469 (!cast<Instruction>(NAME # "Wri") GPR32:$Rd, GPR32:$Rn,
2470 logical_imm32_not:$imm), 0>;
2471 def : InstSubst<Alias # "\t$Rd, $Rn, $imm",
2472 (!cast<Instruction>(NAME # "Xri") GPR64:$Rd, GPR64:$Rn,
2473 logical_imm64_not:$imm), 0>;
2476 class BaseLogicalRegPseudo<RegisterClass regtype, SDPatternOperator OpNode>
2477 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
2478 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
2479 Sched<[WriteI, ReadI, ReadI]>;
2481 // Split from LogicalImm as not all instructions have both.
2482 multiclass LogicalReg<bits<2> opc, bit N, string mnemonic,
2483 SDPatternOperator OpNode> {
2484 let isReMaterializable = 1, isAsCheapAsAMove = 1 in {
2485 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
2486 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
2489 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
2490 [(set GPR32:$Rd, (OpNode GPR32:$Rn,
2491 logical_shifted_reg32:$Rm))]> {
2494 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
2495 [(set GPR64:$Rd, (OpNode GPR64:$Rn,
2496 logical_shifted_reg64:$Rm))]> {
2500 def : LogicalRegAlias<mnemonic,
2501 !cast<Instruction>(NAME#"Wrs"), GPR32>;
2502 def : LogicalRegAlias<mnemonic,
2503 !cast<Instruction>(NAME#"Xrs"), GPR64>;
2506 // Split from LogicalReg to allow setting NZCV Defs
2507 multiclass LogicalRegS<bits<2> opc, bit N, string mnemonic,
2508 SDPatternOperator OpNode = null_frag> {
2509 let Defs = [NZCV], mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
2510 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
2511 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
2513 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
2514 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_shifted_reg32:$Rm))]> {
2517 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
2518 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_shifted_reg64:$Rm))]> {
2523 def : LogicalRegAlias<mnemonic,
2524 !cast<Instruction>(NAME#"Wrs"), GPR32>;
2525 def : LogicalRegAlias<mnemonic,
2526 !cast<Instruction>(NAME#"Xrs"), GPR64>;
2530 // Conditionally set flags
2533 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2534 class BaseCondComparisonImm<bit op, RegisterClass regtype, ImmLeaf immtype,
2535 string mnemonic, SDNode OpNode>
2536 : I<(outs), (ins regtype:$Rn, immtype:$imm, imm32_0_15:$nzcv, ccode:$cond),
2537 mnemonic, "\t$Rn, $imm, $nzcv, $cond", "",
2538 [(set NZCV, (OpNode regtype:$Rn, immtype:$imm, (i32 imm:$nzcv),
2539 (i32 imm:$cond), NZCV))]>,
2540 Sched<[WriteI, ReadI]> {
2550 let Inst{29-21} = 0b111010010;
2551 let Inst{20-16} = imm;
2552 let Inst{15-12} = cond;
2553 let Inst{11-10} = 0b10;
2556 let Inst{3-0} = nzcv;
2559 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
2560 class BaseCondComparisonReg<bit op, RegisterClass regtype, string mnemonic,
2562 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm32_0_15:$nzcv, ccode:$cond),
2563 mnemonic, "\t$Rn, $Rm, $nzcv, $cond", "",
2564 [(set NZCV, (OpNode regtype:$Rn, regtype:$Rm, (i32 imm:$nzcv),
2565 (i32 imm:$cond), NZCV))]>,
2566 Sched<[WriteI, ReadI, ReadI]> {
2576 let Inst{29-21} = 0b111010010;
2577 let Inst{20-16} = Rm;
2578 let Inst{15-12} = cond;
2579 let Inst{11-10} = 0b00;
2582 let Inst{3-0} = nzcv;
2585 multiclass CondComparison<bit op, string mnemonic, SDNode OpNode> {
2586 // immediate operand variants
2587 def Wi : BaseCondComparisonImm<op, GPR32, imm32_0_31, mnemonic, OpNode> {
2590 def Xi : BaseCondComparisonImm<op, GPR64, imm0_31, mnemonic, OpNode> {
2593 // register operand variants
2594 def Wr : BaseCondComparisonReg<op, GPR32, mnemonic, OpNode> {
2597 def Xr : BaseCondComparisonReg<op, GPR64, mnemonic, OpNode> {
2603 // Conditional select
2606 class BaseCondSelect<bit op, bits<2> op2, RegisterClass regtype, string asm>
2607 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
2608 asm, "\t$Rd, $Rn, $Rm, $cond", "",
2610 (AArch64csel regtype:$Rn, regtype:$Rm, (i32 imm:$cond), NZCV))]>,
2611 Sched<[WriteI, ReadI, ReadI]> {
2620 let Inst{29-21} = 0b011010100;
2621 let Inst{20-16} = Rm;
2622 let Inst{15-12} = cond;
2623 let Inst{11-10} = op2;
2628 multiclass CondSelect<bit op, bits<2> op2, string asm> {
2629 def Wr : BaseCondSelect<op, op2, GPR32, asm> {
2632 def Xr : BaseCondSelect<op, op2, GPR64, asm> {
2637 class BaseCondSelectOp<bit op, bits<2> op2, RegisterClass regtype, string asm,
2639 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
2640 asm, "\t$Rd, $Rn, $Rm, $cond", "",
2642 (AArch64csel regtype:$Rn, (frag regtype:$Rm),
2643 (i32 imm:$cond), NZCV))]>,
2644 Sched<[WriteI, ReadI, ReadI]> {
2653 let Inst{29-21} = 0b011010100;
2654 let Inst{20-16} = Rm;
2655 let Inst{15-12} = cond;
2656 let Inst{11-10} = op2;
2661 def inv_cond_XFORM : SDNodeXForm<imm, [{
2662 AArch64CC::CondCode CC = static_cast<AArch64CC::CondCode>(N->getZExtValue());
2663 return CurDAG->getTargetConstant(AArch64CC::getInvertedCondCode(CC), SDLoc(N),
2667 multiclass CondSelectOp<bit op, bits<2> op2, string asm, PatFrag frag> {
2668 def Wr : BaseCondSelectOp<op, op2, GPR32, asm, frag> {
2671 def Xr : BaseCondSelectOp<op, op2, GPR64, asm, frag> {
2675 def : Pat<(AArch64csel (frag GPR32:$Rm), GPR32:$Rn, (i32 imm:$cond), NZCV),
2676 (!cast<Instruction>(NAME # Wr) GPR32:$Rn, GPR32:$Rm,
2677 (inv_cond_XFORM imm:$cond))>;
2679 def : Pat<(AArch64csel (frag GPR64:$Rm), GPR64:$Rn, (i32 imm:$cond), NZCV),
2680 (!cast<Instruction>(NAME # Xr) GPR64:$Rn, GPR64:$Rm,
2681 (inv_cond_XFORM imm:$cond))>;
2685 // Special Mask Value
2687 def maski8_or_more : Operand<i32>,
2688 ImmLeaf<i32, [{ return (Imm & 0xff) == 0xff; }]> {
2690 def maski16_or_more : Operand<i32>,
2691 ImmLeaf<i32, [{ return (Imm & 0xffff) == 0xffff; }]> {
2699 // (unsigned immediate)
2700 // Indexed for 8-bit registers. offset is in range [0,4095].
2701 def am_indexed8 : ComplexPattern<i64, 2, "SelectAddrModeIndexed8", []>;
2702 def am_indexed16 : ComplexPattern<i64, 2, "SelectAddrModeIndexed16", []>;
2703 def am_indexed32 : ComplexPattern<i64, 2, "SelectAddrModeIndexed32", []>;
2704 def am_indexed64 : ComplexPattern<i64, 2, "SelectAddrModeIndexed64", []>;
2705 def am_indexed128 : ComplexPattern<i64, 2, "SelectAddrModeIndexed128", []>;
2707 def gi_am_indexed8 :
2708 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<8>">,
2709 GIComplexPatternEquiv<am_indexed8>;
2710 def gi_am_indexed16 :
2711 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<16>">,
2712 GIComplexPatternEquiv<am_indexed16>;
2713 def gi_am_indexed32 :
2714 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<32>">,
2715 GIComplexPatternEquiv<am_indexed32>;
2716 def gi_am_indexed64 :
2717 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<64>">,
2718 GIComplexPatternEquiv<am_indexed64>;
2719 def gi_am_indexed128 :
2720 GIComplexOperandMatcher<s64, "selectAddrModeIndexed<128>">,
2721 GIComplexPatternEquiv<am_indexed128>;
2723 class UImm12OffsetOperand<int Scale> : AsmOperandClass {
2724 let Name = "UImm12Offset" # Scale;
2725 let RenderMethod = "addUImm12OffsetOperands<" # Scale # ">";
2726 let PredicateMethod = "isUImm12Offset<" # Scale # ">";
2727 let DiagnosticType = "InvalidMemoryIndexed" # Scale;
2730 def UImm12OffsetScale1Operand : UImm12OffsetOperand<1>;
2731 def UImm12OffsetScale2Operand : UImm12OffsetOperand<2>;
2732 def UImm12OffsetScale4Operand : UImm12OffsetOperand<4>;
2733 def UImm12OffsetScale8Operand : UImm12OffsetOperand<8>;
2734 def UImm12OffsetScale16Operand : UImm12OffsetOperand<16>;
2736 class uimm12_scaled<int Scale> : Operand<i64> {
2737 let ParserMatchClass
2738 = !cast<AsmOperandClass>("UImm12OffsetScale" # Scale # "Operand");
2740 = "getLdStUImm12OpValue<AArch64::fixup_aarch64_ldst_imm12_scale" # Scale # ">";
2741 let PrintMethod = "printUImm12Offset<" # Scale # ">";
2744 def uimm12s1 : uimm12_scaled<1>;
2745 def uimm12s2 : uimm12_scaled<2>;
2746 def uimm12s4 : uimm12_scaled<4>;
2747 def uimm12s8 : uimm12_scaled<8>;
2748 def uimm12s16 : uimm12_scaled<16>;
2750 class BaseLoadStoreUI<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2751 string asm, list<dag> pattern>
2752 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", pattern> {
2758 let Inst{31-30} = sz;
2759 let Inst{29-27} = 0b111;
2761 let Inst{25-24} = 0b01;
2762 let Inst{23-22} = opc;
2763 let Inst{21-10} = offset;
2767 let DecoderMethod = "DecodeUnsignedLdStInstruction";
2770 multiclass LoadUI<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2771 Operand indextype, string asm, list<dag> pattern> {
2772 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2773 def ui : BaseLoadStoreUI<sz, V, opc, (outs regtype:$Rt),
2774 (ins GPR64sp:$Rn, indextype:$offset),
2778 def : InstAlias<asm # "\t$Rt, [$Rn]",
2779 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
2782 multiclass StoreUI<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2783 Operand indextype, string asm, list<dag> pattern> {
2784 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2785 def ui : BaseLoadStoreUI<sz, V, opc, (outs),
2786 (ins regtype:$Rt, GPR64sp:$Rn, indextype:$offset),
2790 def : InstAlias<asm # "\t$Rt, [$Rn]",
2791 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
2794 // Same as StoreUI, but take a RegisterOperand. This is used by GlobalISel to
2795 // substitute zero-registers automatically.
2797 // TODO: Roll out zero-register subtitution to GPR32/GPR64 and fold this back
2799 multiclass StoreUIz<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2800 Operand indextype, string asm, list<dag> pattern> {
2801 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2802 def ui : BaseLoadStoreUI<sz, V, opc, (outs),
2803 (ins regtype:$Rt, GPR64sp:$Rn, indextype:$offset),
2807 def : InstAlias<asm # "\t$Rt, [$Rn]",
2808 (!cast<Instruction>(NAME # "ui") regtype:$Rt, GPR64sp:$Rn, 0)>;
2811 def PrefetchOperand : AsmOperandClass {
2812 let Name = "Prefetch";
2813 let ParserMethod = "tryParsePrefetch";
2815 def prfop : Operand<i32> {
2816 let PrintMethod = "printPrefetchOp";
2817 let ParserMatchClass = PrefetchOperand;
2820 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2821 class PrefetchUI<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2822 : BaseLoadStoreUI<sz, V, opc,
2823 (outs), (ins prfop:$Rt, GPR64sp:$Rn, uimm12s8:$offset),
2831 // Load literal address: 19-bit immediate. The low two bits of the target
2832 // offset are implied zero and so are not part of the immediate.
2833 def am_ldrlit : Operand<iPTR> {
2834 let EncoderMethod = "getLoadLiteralOpValue";
2835 let DecoderMethod = "DecodePCRelLabel19";
2836 let PrintMethod = "printAlignedLabel";
2837 let ParserMatchClass = PCRelLabel19Operand;
2838 let OperandType = "OPERAND_PCREL";
2841 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2842 class LoadLiteral<bits<2> opc, bit V, RegisterOperand regtype, string asm>
2843 : I<(outs regtype:$Rt), (ins am_ldrlit:$label),
2844 asm, "\t$Rt, $label", "", []>,
2848 let Inst{31-30} = opc;
2849 let Inst{29-27} = 0b011;
2851 let Inst{25-24} = 0b00;
2852 let Inst{23-5} = label;
2856 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2857 class PrefetchLiteral<bits<2> opc, bit V, string asm, list<dag> pat>
2858 : I<(outs), (ins prfop:$Rt, am_ldrlit:$label),
2859 asm, "\t$Rt, $label", "", pat>,
2863 let Inst{31-30} = opc;
2864 let Inst{29-27} = 0b011;
2866 let Inst{25-24} = 0b00;
2867 let Inst{23-5} = label;
2872 // Load/store register offset
2875 def ro_Xindexed8 : ComplexPattern<i64, 4, "SelectAddrModeXRO<8>", []>;
2876 def ro_Xindexed16 : ComplexPattern<i64, 4, "SelectAddrModeXRO<16>", []>;
2877 def ro_Xindexed32 : ComplexPattern<i64, 4, "SelectAddrModeXRO<32>", []>;
2878 def ro_Xindexed64 : ComplexPattern<i64, 4, "SelectAddrModeXRO<64>", []>;
2879 def ro_Xindexed128 : ComplexPattern<i64, 4, "SelectAddrModeXRO<128>", []>;
2881 def ro_Windexed8 : ComplexPattern<i64, 4, "SelectAddrModeWRO<8>", []>;
2882 def ro_Windexed16 : ComplexPattern<i64, 4, "SelectAddrModeWRO<16>", []>;
2883 def ro_Windexed32 : ComplexPattern<i64, 4, "SelectAddrModeWRO<32>", []>;
2884 def ro_Windexed64 : ComplexPattern<i64, 4, "SelectAddrModeWRO<64>", []>;
2885 def ro_Windexed128 : ComplexPattern<i64, 4, "SelectAddrModeWRO<128>", []>;
2887 class MemExtendOperand<string Reg, int Width> : AsmOperandClass {
2888 let Name = "Mem" # Reg # "Extend" # Width;
2889 let PredicateMethod = "isMem" # Reg # "Extend<" # Width # ">";
2890 let RenderMethod = "addMemExtendOperands";
2891 let DiagnosticType = "InvalidMemory" # Reg # "Extend" # Width;
2894 def MemWExtend8Operand : MemExtendOperand<"W", 8> {
2895 // The address "[x0, x1, lsl #0]" actually maps to the variant which performs
2896 // the trivial shift.
2897 let RenderMethod = "addMemExtend8Operands";
2899 def MemWExtend16Operand : MemExtendOperand<"W", 16>;
2900 def MemWExtend32Operand : MemExtendOperand<"W", 32>;
2901 def MemWExtend64Operand : MemExtendOperand<"W", 64>;
2902 def MemWExtend128Operand : MemExtendOperand<"W", 128>;
2904 def MemXExtend8Operand : MemExtendOperand<"X", 8> {
2905 // The address "[x0, x1, lsl #0]" actually maps to the variant which performs
2906 // the trivial shift.
2907 let RenderMethod = "addMemExtend8Operands";
2909 def MemXExtend16Operand : MemExtendOperand<"X", 16>;
2910 def MemXExtend32Operand : MemExtendOperand<"X", 32>;
2911 def MemXExtend64Operand : MemExtendOperand<"X", 64>;
2912 def MemXExtend128Operand : MemExtendOperand<"X", 128>;
2914 class ro_extend<AsmOperandClass ParserClass, string Reg, int Width>
2916 let ParserMatchClass = ParserClass;
2917 let PrintMethod = "printMemExtend<'" # Reg # "', " # Width # ">";
2918 let DecoderMethod = "DecodeMemExtend";
2919 let EncoderMethod = "getMemExtendOpValue";
2920 let MIOperandInfo = (ops i32imm:$signed, i32imm:$doshift);
2923 def ro_Wextend8 : ro_extend<MemWExtend8Operand, "w", 8>;
2924 def ro_Wextend16 : ro_extend<MemWExtend16Operand, "w", 16>;
2925 def ro_Wextend32 : ro_extend<MemWExtend32Operand, "w", 32>;
2926 def ro_Wextend64 : ro_extend<MemWExtend64Operand, "w", 64>;
2927 def ro_Wextend128 : ro_extend<MemWExtend128Operand, "w", 128>;
2929 def ro_Xextend8 : ro_extend<MemXExtend8Operand, "x", 8>;
2930 def ro_Xextend16 : ro_extend<MemXExtend16Operand, "x", 16>;
2931 def ro_Xextend32 : ro_extend<MemXExtend32Operand, "x", 32>;
2932 def ro_Xextend64 : ro_extend<MemXExtend64Operand, "x", 64>;
2933 def ro_Xextend128 : ro_extend<MemXExtend128Operand, "x", 128>;
2935 class ROAddrMode<ComplexPattern windex, ComplexPattern xindex,
2936 Operand wextend, Operand xextend> {
2937 // CodeGen-level pattern covering the entire addressing mode.
2938 ComplexPattern Wpat = windex;
2939 ComplexPattern Xpat = xindex;
2941 // Asm-level Operand covering the valid "uxtw #3" style syntax.
2942 Operand Wext = wextend;
2943 Operand Xext = xextend;
2946 def ro8 : ROAddrMode<ro_Windexed8, ro_Xindexed8, ro_Wextend8, ro_Xextend8>;
2947 def ro16 : ROAddrMode<ro_Windexed16, ro_Xindexed16, ro_Wextend16, ro_Xextend16>;
2948 def ro32 : ROAddrMode<ro_Windexed32, ro_Xindexed32, ro_Wextend32, ro_Xextend32>;
2949 def ro64 : ROAddrMode<ro_Windexed64, ro_Xindexed64, ro_Wextend64, ro_Xextend64>;
2950 def ro128 : ROAddrMode<ro_Windexed128, ro_Xindexed128, ro_Wextend128,
2953 class LoadStore8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2954 string asm, dag ins, dag outs, list<dag> pat>
2955 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
2960 let Inst{31-30} = sz;
2961 let Inst{29-27} = 0b111;
2963 let Inst{25-24} = 0b00;
2964 let Inst{23-22} = opc;
2966 let Inst{20-16} = Rm;
2967 let Inst{15} = extend{1}; // sign extend Rm?
2969 let Inst{12} = extend{0}; // do shift?
2970 let Inst{11-10} = 0b10;
2975 class ROInstAlias<string asm, RegisterOperand regtype, Instruction INST>
2976 : InstAlias<asm # "\t$Rt, [$Rn, $Rm]",
2977 (INST regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, 0, 0)>;
2979 multiclass Load8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
2980 string asm, ValueType Ty, SDPatternOperator loadop> {
2981 let AddedComplexity = 10 in
2982 def roW : LoadStore8RO<sz, V, opc, regtype, asm,
2984 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$extend),
2985 [(set (Ty regtype:$Rt),
2986 (loadop (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm,
2987 ro_Wextend8:$extend)))]>,
2988 Sched<[WriteLDIdx, ReadAdrBase]> {
2992 let AddedComplexity = 10 in
2993 def roX : LoadStore8RO<sz, V, opc, regtype, asm,
2995 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend8:$extend),
2996 [(set (Ty regtype:$Rt),
2997 (loadop (ro_Xindexed8 GPR64sp:$Rn, GPR64:$Rm,
2998 ro_Xextend8:$extend)))]>,
2999 Sched<[WriteLDIdx, ReadAdrBase]> {
3003 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3006 multiclass Store8RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3007 string asm, ValueType Ty, SDPatternOperator storeop> {
3008 let AddedComplexity = 10 in
3009 def roW : LoadStore8RO<sz, V, opc, regtype, asm, (outs),
3010 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$extend),
3011 [(storeop (Ty regtype:$Rt),
3012 (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm,
3013 ro_Wextend8:$extend))]>,
3014 Sched<[WriteSTIdx, ReadAdrBase]> {
3018 let AddedComplexity = 10 in
3019 def roX : LoadStore8RO<sz, V, opc, regtype, asm, (outs),
3020 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend8:$extend),
3021 [(storeop (Ty regtype:$Rt),
3022 (ro_Xindexed8 GPR64sp:$Rn, GPR64:$Rm,
3023 ro_Xextend8:$extend))]>,
3024 Sched<[WriteSTIdx, ReadAdrBase]> {
3028 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3031 class LoadStore16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3032 string asm, dag ins, dag outs, list<dag> pat>
3033 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3038 let Inst{31-30} = sz;
3039 let Inst{29-27} = 0b111;
3041 let Inst{25-24} = 0b00;
3042 let Inst{23-22} = opc;
3044 let Inst{20-16} = Rm;
3045 let Inst{15} = extend{1}; // sign extend Rm?
3047 let Inst{12} = extend{0}; // do shift?
3048 let Inst{11-10} = 0b10;
3053 multiclass Load16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3054 string asm, ValueType Ty, SDPatternOperator loadop> {
3055 let AddedComplexity = 10 in
3056 def roW : LoadStore16RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3057 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend16:$extend),
3058 [(set (Ty regtype:$Rt),
3059 (loadop (ro_Windexed16 GPR64sp:$Rn, GPR32:$Rm,
3060 ro_Wextend16:$extend)))]>,
3061 Sched<[WriteLDIdx, ReadAdrBase]> {
3065 let AddedComplexity = 10 in
3066 def roX : LoadStore16RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3067 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend16:$extend),
3068 [(set (Ty regtype:$Rt),
3069 (loadop (ro_Xindexed16 GPR64sp:$Rn, GPR64:$Rm,
3070 ro_Xextend16:$extend)))]>,
3071 Sched<[WriteLDIdx, ReadAdrBase]> {
3075 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3078 multiclass Store16RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3079 string asm, ValueType Ty, SDPatternOperator storeop> {
3080 let AddedComplexity = 10 in
3081 def roW : LoadStore16RO<sz, V, opc, regtype, asm, (outs),
3082 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend16:$extend),
3083 [(storeop (Ty regtype:$Rt),
3084 (ro_Windexed16 GPR64sp:$Rn, GPR32:$Rm,
3085 ro_Wextend16:$extend))]>,
3086 Sched<[WriteSTIdx, ReadAdrBase]> {
3090 let AddedComplexity = 10 in
3091 def roX : LoadStore16RO<sz, V, opc, regtype, asm, (outs),
3092 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend16:$extend),
3093 [(storeop (Ty regtype:$Rt),
3094 (ro_Xindexed16 GPR64sp:$Rn, GPR64:$Rm,
3095 ro_Xextend16:$extend))]>,
3096 Sched<[WriteSTIdx, ReadAdrBase]> {
3100 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3103 class LoadStore32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3104 string asm, dag ins, dag outs, list<dag> pat>
3105 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3110 let Inst{31-30} = sz;
3111 let Inst{29-27} = 0b111;
3113 let Inst{25-24} = 0b00;
3114 let Inst{23-22} = opc;
3116 let Inst{20-16} = Rm;
3117 let Inst{15} = extend{1}; // sign extend Rm?
3119 let Inst{12} = extend{0}; // do shift?
3120 let Inst{11-10} = 0b10;
3125 multiclass Load32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3126 string asm, ValueType Ty, SDPatternOperator loadop> {
3127 let AddedComplexity = 10 in
3128 def roW : LoadStore32RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3129 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend32:$extend),
3130 [(set (Ty regtype:$Rt),
3131 (loadop (ro_Windexed32 GPR64sp:$Rn, GPR32:$Rm,
3132 ro_Wextend32:$extend)))]>,
3133 Sched<[WriteLDIdx, ReadAdrBase]> {
3137 let AddedComplexity = 10 in
3138 def roX : LoadStore32RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3139 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend32:$extend),
3140 [(set (Ty regtype:$Rt),
3141 (loadop (ro_Xindexed32 GPR64sp:$Rn, GPR64:$Rm,
3142 ro_Xextend32:$extend)))]>,
3143 Sched<[WriteLDIdx, ReadAdrBase]> {
3147 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3150 multiclass Store32RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3151 string asm, ValueType Ty, SDPatternOperator storeop> {
3152 let AddedComplexity = 10 in
3153 def roW : LoadStore32RO<sz, V, opc, regtype, asm, (outs),
3154 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend32:$extend),
3155 [(storeop (Ty regtype:$Rt),
3156 (ro_Windexed32 GPR64sp:$Rn, GPR32:$Rm,
3157 ro_Wextend32:$extend))]>,
3158 Sched<[WriteSTIdx, ReadAdrBase]> {
3162 let AddedComplexity = 10 in
3163 def roX : LoadStore32RO<sz, V, opc, regtype, asm, (outs),
3164 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend32:$extend),
3165 [(storeop (Ty regtype:$Rt),
3166 (ro_Xindexed32 GPR64sp:$Rn, GPR64:$Rm,
3167 ro_Xextend32:$extend))]>,
3168 Sched<[WriteSTIdx, ReadAdrBase]> {
3172 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3175 class LoadStore64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3176 string asm, dag ins, dag outs, list<dag> pat>
3177 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3182 let Inst{31-30} = sz;
3183 let Inst{29-27} = 0b111;
3185 let Inst{25-24} = 0b00;
3186 let Inst{23-22} = opc;
3188 let Inst{20-16} = Rm;
3189 let Inst{15} = extend{1}; // sign extend Rm?
3191 let Inst{12} = extend{0}; // do shift?
3192 let Inst{11-10} = 0b10;
3197 multiclass Load64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3198 string asm, ValueType Ty, SDPatternOperator loadop> {
3199 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3200 def roW : LoadStore64RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3201 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3202 [(set (Ty regtype:$Rt),
3203 (loadop (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3204 ro_Wextend64:$extend)))]>,
3205 Sched<[WriteLDIdx, ReadAdrBase]> {
3209 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3210 def roX : LoadStore64RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3211 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3212 [(set (Ty regtype:$Rt),
3213 (loadop (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3214 ro_Xextend64:$extend)))]>,
3215 Sched<[WriteLDIdx, ReadAdrBase]> {
3219 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3222 multiclass Store64RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3223 string asm, ValueType Ty, SDPatternOperator storeop> {
3224 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3225 def roW : LoadStore64RO<sz, V, opc, regtype, asm, (outs),
3226 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3227 [(storeop (Ty regtype:$Rt),
3228 (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3229 ro_Wextend64:$extend))]>,
3230 Sched<[WriteSTIdx, ReadAdrBase]> {
3234 let AddedComplexity = 10, mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3235 def roX : LoadStore64RO<sz, V, opc, regtype, asm, (outs),
3236 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3237 [(storeop (Ty regtype:$Rt),
3238 (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3239 ro_Xextend64:$extend))]>,
3240 Sched<[WriteSTIdx, ReadAdrBase]> {
3244 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3247 class LoadStore128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3248 string asm, dag ins, dag outs, list<dag> pat>
3249 : I<ins, outs, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat> {
3254 let Inst{31-30} = sz;
3255 let Inst{29-27} = 0b111;
3257 let Inst{25-24} = 0b00;
3258 let Inst{23-22} = opc;
3260 let Inst{20-16} = Rm;
3261 let Inst{15} = extend{1}; // sign extend Rm?
3263 let Inst{12} = extend{0}; // do shift?
3264 let Inst{11-10} = 0b10;
3269 multiclass Load128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3270 string asm, ValueType Ty, SDPatternOperator loadop> {
3271 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3272 def roW : LoadStore128RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3273 (ins GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend),
3274 [(set (Ty regtype:$Rt),
3275 (loadop (ro_Windexed128 GPR64sp:$Rn, GPR32:$Rm,
3276 ro_Wextend128:$extend)))]>,
3277 Sched<[WriteLDIdx, ReadAdrBase]> {
3281 let AddedComplexity = 10, mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
3282 def roX : LoadStore128RO<sz, V, opc, regtype, asm, (outs regtype:$Rt),
3283 (ins GPR64sp:$Rn, GPR64:$Rm, ro_Xextend128:$extend),
3284 [(set (Ty regtype:$Rt),
3285 (loadop (ro_Xindexed128 GPR64sp:$Rn, GPR64:$Rm,
3286 ro_Xextend128:$extend)))]>,
3287 Sched<[WriteLDIdx, ReadAdrBase]> {
3291 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3294 multiclass Store128RO<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3295 string asm, ValueType Ty, SDPatternOperator storeop> {
3296 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3297 def roW : LoadStore128RO<sz, V, opc, regtype, asm, (outs),
3298 (ins regtype:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend128:$extend),
3300 Sched<[WriteSTIdx, ReadAdrBase]> {
3304 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
3305 def roX : LoadStore128RO<sz, V, opc, regtype, asm, (outs),
3306 (ins regtype:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend128:$extend),
3308 Sched<[WriteSTIdx, ReadAdrBase]> {
3312 def : ROInstAlias<asm, regtype, !cast<Instruction>(NAME # "roX")>;
3315 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3316 class BasePrefetchRO<bits<2> sz, bit V, bits<2> opc, dag outs, dag ins,
3317 string asm, list<dag> pat>
3318 : I<outs, ins, asm, "\t$Rt, [$Rn, $Rm, $extend]", "", pat>,
3324 let Inst{31-30} = sz;
3325 let Inst{29-27} = 0b111;
3327 let Inst{25-24} = 0b00;
3328 let Inst{23-22} = opc;
3330 let Inst{20-16} = Rm;
3331 let Inst{15} = extend{1}; // sign extend Rm?
3333 let Inst{12} = extend{0}; // do shift?
3334 let Inst{11-10} = 0b10;
3339 multiclass PrefetchRO<bits<2> sz, bit V, bits<2> opc, string asm> {
3340 def roW : BasePrefetchRO<sz, V, opc, (outs),
3341 (ins prfop:$Rt, GPR64sp:$Rn, GPR32:$Rm, ro_Wextend64:$extend),
3342 asm, [(AArch64Prefetch imm:$Rt,
3343 (ro_Windexed64 GPR64sp:$Rn, GPR32:$Rm,
3344 ro_Wextend64:$extend))]> {
3348 def roX : BasePrefetchRO<sz, V, opc, (outs),
3349 (ins prfop:$Rt, GPR64sp:$Rn, GPR64:$Rm, ro_Xextend64:$extend),
3350 asm, [(AArch64Prefetch imm:$Rt,
3351 (ro_Xindexed64 GPR64sp:$Rn, GPR64:$Rm,
3352 ro_Xextend64:$extend))]> {
3356 def : InstAlias<"prfm $Rt, [$Rn, $Rm]",
3357 (!cast<Instruction>(NAME # "roX") prfop:$Rt,
3358 GPR64sp:$Rn, GPR64:$Rm, 0, 0)>;
3362 // Load/store unscaled immediate
3365 def am_unscaled8 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled8", []>;
3366 def am_unscaled16 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled16", []>;
3367 def am_unscaled32 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled32", []>;
3368 def am_unscaled64 : ComplexPattern<i64, 2, "SelectAddrModeUnscaled64", []>;
3369 def am_unscaled128 :ComplexPattern<i64, 2, "SelectAddrModeUnscaled128", []>;
3371 def gi_am_unscaled8 :
3372 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled8">,
3373 GIComplexPatternEquiv<am_unscaled8>;
3374 def gi_am_unscaled16 :
3375 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled16">,
3376 GIComplexPatternEquiv<am_unscaled16>;
3377 def gi_am_unscaled32 :
3378 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled32">,
3379 GIComplexPatternEquiv<am_unscaled32>;
3380 def gi_am_unscaled64 :
3381 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled64">,
3382 GIComplexPatternEquiv<am_unscaled64>;
3383 def gi_am_unscaled128 :
3384 GIComplexOperandMatcher<s64, "selectAddrModeUnscaled128">,
3385 GIComplexPatternEquiv<am_unscaled128>;
3388 class BaseLoadStoreUnscale<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3389 string asm, list<dag> pattern>
3390 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", pattern> {
3394 let Inst{31-30} = sz;
3395 let Inst{29-27} = 0b111;
3397 let Inst{25-24} = 0b00;
3398 let Inst{23-22} = opc;
3400 let Inst{20-12} = offset;
3401 let Inst{11-10} = 0b00;
3405 let DecoderMethod = "DecodeSignedLdStInstruction";
3408 // Armv8.4 LDAPR & STLR with Immediate Offset instruction
3409 multiclass BaseLoadUnscaleV84<string asm, bits<2> sz, bits<2> opc,
3410 RegisterOperand regtype > {
3411 def i : BaseLoadStoreUnscale<sz, 0, opc, (outs regtype:$Rt),
3412 (ins GPR64sp:$Rn, simm9:$offset), asm, []>,
3417 def : InstAlias<asm # "\t$Rt, [$Rn]",
3418 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3421 multiclass BaseStoreUnscaleV84<string asm, bits<2> sz, bits<2> opc,
3422 RegisterOperand regtype > {
3423 def i : BaseLoadStoreUnscale<sz, 0, opc, (outs),
3424 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3430 def : InstAlias<asm # "\t$Rt, [$Rn]",
3431 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3434 multiclass LoadUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3435 string asm, list<dag> pattern> {
3436 let AddedComplexity = 1 in // try this before LoadUI
3437 def i : BaseLoadStoreUnscale<sz, V, opc, (outs regtype:$Rt),
3438 (ins GPR64sp:$Rn, simm9:$offset), asm, pattern>,
3441 def : InstAlias<asm # "\t$Rt, [$Rn]",
3442 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3445 multiclass StoreUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3446 string asm, list<dag> pattern> {
3447 let AddedComplexity = 1 in // try this before StoreUI
3448 def i : BaseLoadStoreUnscale<sz, V, opc, (outs),
3449 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3453 def : InstAlias<asm # "\t$Rt, [$Rn]",
3454 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3457 multiclass PrefetchUnscaled<bits<2> sz, bit V, bits<2> opc, string asm,
3459 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3460 def i : BaseLoadStoreUnscale<sz, V, opc, (outs),
3461 (ins prfop:$Rt, GPR64sp:$Rn, simm9:$offset),
3465 def : InstAlias<asm # "\t$Rt, [$Rn]",
3466 (!cast<Instruction>(NAME # "i") prfop:$Rt, GPR64sp:$Rn, 0)>;
3470 // Load/store unscaled immediate, unprivileged
3473 class BaseLoadStoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
3474 dag oops, dag iops, string asm>
3475 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]", "", []> {
3479 let Inst{31-30} = sz;
3480 let Inst{29-27} = 0b111;
3482 let Inst{25-24} = 0b00;
3483 let Inst{23-22} = opc;
3485 let Inst{20-12} = offset;
3486 let Inst{11-10} = 0b10;
3490 let DecoderMethod = "DecodeSignedLdStInstruction";
3493 multiclass LoadUnprivileged<bits<2> sz, bit V, bits<2> opc,
3494 RegisterClass regtype, string asm> {
3495 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in
3496 def i : BaseLoadStoreUnprivileged<sz, V, opc, (outs regtype:$Rt),
3497 (ins GPR64sp:$Rn, simm9:$offset), asm>,
3500 def : InstAlias<asm # "\t$Rt, [$Rn]",
3501 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3504 multiclass StoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
3505 RegisterClass regtype, string asm> {
3506 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
3507 def i : BaseLoadStoreUnprivileged<sz, V, opc, (outs),
3508 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3512 def : InstAlias<asm # "\t$Rt, [$Rn]",
3513 (!cast<Instruction>(NAME # "i") regtype:$Rt, GPR64sp:$Rn, 0)>;
3517 // Load/store pre-indexed
3520 class BaseLoadStorePreIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3521 string asm, string cstr, list<dag> pat>
3522 : I<oops, iops, asm, "\t$Rt, [$Rn, $offset]!", cstr, pat> {
3526 let Inst{31-30} = sz;
3527 let Inst{29-27} = 0b111;
3529 let Inst{25-24} = 0;
3530 let Inst{23-22} = opc;
3532 let Inst{20-12} = offset;
3533 let Inst{11-10} = 0b11;
3537 let DecoderMethod = "DecodeSignedLdStInstruction";
3540 let hasSideEffects = 0 in {
3541 let mayStore = 0, mayLoad = 1 in
3542 class LoadPreIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3544 : BaseLoadStorePreIdx<sz, V, opc,
3545 (outs GPR64sp:$wback, regtype:$Rt),
3546 (ins GPR64sp:$Rn, simm9:$offset), asm,
3547 "$Rn = $wback,@earlyclobber $wback", []>,
3548 Sched<[WriteLD, WriteAdr]>;
3550 let mayStore = 1, mayLoad = 0 in
3551 class StorePreIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3552 string asm, SDPatternOperator storeop, ValueType Ty>
3553 : BaseLoadStorePreIdx<sz, V, opc,
3554 (outs GPR64sp:$wback),
3555 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3556 asm, "$Rn = $wback,@earlyclobber $wback",
3557 [(set GPR64sp:$wback,
3558 (storeop (Ty regtype:$Rt), GPR64sp:$Rn, simm9:$offset))]>,
3559 Sched<[WriteAdr, WriteST]>;
3560 } // hasSideEffects = 0
3563 // Load/store post-indexed
3566 class BaseLoadStorePostIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
3567 string asm, string cstr, list<dag> pat>
3568 : I<oops, iops, asm, "\t$Rt, [$Rn], $offset", cstr, pat> {
3572 let Inst{31-30} = sz;
3573 let Inst{29-27} = 0b111;
3575 let Inst{25-24} = 0b00;
3576 let Inst{23-22} = opc;
3578 let Inst{20-12} = offset;
3579 let Inst{11-10} = 0b01;
3583 let DecoderMethod = "DecodeSignedLdStInstruction";
3586 let hasSideEffects = 0 in {
3587 let mayStore = 0, mayLoad = 1 in
3588 class LoadPostIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3590 : BaseLoadStorePostIdx<sz, V, opc,
3591 (outs GPR64sp:$wback, regtype:$Rt),
3592 (ins GPR64sp:$Rn, simm9:$offset),
3593 asm, "$Rn = $wback,@earlyclobber $wback", []>,
3594 Sched<[WriteLD, WriteAdr]>;
3596 let mayStore = 1, mayLoad = 0 in
3597 class StorePostIdx<bits<2> sz, bit V, bits<2> opc, RegisterOperand regtype,
3598 string asm, SDPatternOperator storeop, ValueType Ty>
3599 : BaseLoadStorePostIdx<sz, V, opc,
3600 (outs GPR64sp:$wback),
3601 (ins regtype:$Rt, GPR64sp:$Rn, simm9:$offset),
3602 asm, "$Rn = $wback,@earlyclobber $wback",
3603 [(set GPR64sp:$wback,
3604 (storeop (Ty regtype:$Rt), GPR64sp:$Rn, simm9:$offset))]>,
3605 Sched<[WriteAdr, WriteST]>;
3606 } // hasSideEffects = 0
3613 // (indexed, offset)
3615 class BaseLoadStorePairOffset<bits<2> opc, bit V, bit L, dag oops, dag iops,
3617 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]", "", []> {
3622 let Inst{31-30} = opc;
3623 let Inst{29-27} = 0b101;
3625 let Inst{25-23} = 0b010;
3627 let Inst{21-15} = offset;
3628 let Inst{14-10} = Rt2;
3632 let DecoderMethod = "DecodePairLdStInstruction";
3635 multiclass LoadPairOffset<bits<2> opc, bit V, RegisterOperand regtype,
3636 Operand indextype, string asm> {
3637 let hasSideEffects = 0, mayStore = 0, mayLoad = 1 in
3638 def i : BaseLoadStorePairOffset<opc, V, 1,
3639 (outs regtype:$Rt, regtype:$Rt2),
3640 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3641 Sched<[WriteLD, WriteLDHi]>;
3643 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3644 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3649 multiclass StorePairOffset<bits<2> opc, bit V, RegisterOperand regtype,
3650 Operand indextype, string asm> {
3651 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in
3652 def i : BaseLoadStorePairOffset<opc, V, 0, (outs),
3653 (ins regtype:$Rt, regtype:$Rt2,
3654 GPR64sp:$Rn, indextype:$offset),
3658 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3659 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3664 class BaseLoadStorePairPreIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
3666 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]!", "$Rn = $wback,@earlyclobber $wback", []> {
3671 let Inst{31-30} = opc;
3672 let Inst{29-27} = 0b101;
3674 let Inst{25-23} = 0b011;
3676 let Inst{21-15} = offset;
3677 let Inst{14-10} = Rt2;
3681 let DecoderMethod = "DecodePairLdStInstruction";
3684 let hasSideEffects = 0 in {
3685 let mayStore = 0, mayLoad = 1 in
3686 class LoadPairPreIdx<bits<2> opc, bit V, RegisterOperand regtype,
3687 Operand indextype, string asm>
3688 : BaseLoadStorePairPreIdx<opc, V, 1,
3689 (outs GPR64sp:$wback, regtype:$Rt, regtype:$Rt2),
3690 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3691 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
3693 let mayStore = 1, mayLoad = 0 in
3694 class StorePairPreIdx<bits<2> opc, bit V, RegisterOperand regtype,
3695 Operand indextype, string asm>
3696 : BaseLoadStorePairPreIdx<opc, V, 0, (outs GPR64sp:$wback),
3697 (ins regtype:$Rt, regtype:$Rt2,
3698 GPR64sp:$Rn, indextype:$offset),
3700 Sched<[WriteAdr, WriteSTP]>;
3701 } // hasSideEffects = 0
3705 class BaseLoadStorePairPostIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
3707 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn], $offset", "$Rn = $wback,@earlyclobber $wback", []> {
3712 let Inst{31-30} = opc;
3713 let Inst{29-27} = 0b101;
3715 let Inst{25-23} = 0b001;
3717 let Inst{21-15} = offset;
3718 let Inst{14-10} = Rt2;
3722 let DecoderMethod = "DecodePairLdStInstruction";
3725 let hasSideEffects = 0 in {
3726 let mayStore = 0, mayLoad = 1 in
3727 class LoadPairPostIdx<bits<2> opc, bit V, RegisterOperand regtype,
3728 Operand idxtype, string asm>
3729 : BaseLoadStorePairPostIdx<opc, V, 1,
3730 (outs GPR64sp:$wback, regtype:$Rt, regtype:$Rt2),
3731 (ins GPR64sp:$Rn, idxtype:$offset), asm>,
3732 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
3734 let mayStore = 1, mayLoad = 0 in
3735 class StorePairPostIdx<bits<2> opc, bit V, RegisterOperand regtype,
3736 Operand idxtype, string asm>
3737 : BaseLoadStorePairPostIdx<opc, V, 0, (outs GPR64sp:$wback),
3738 (ins regtype:$Rt, regtype:$Rt2,
3739 GPR64sp:$Rn, idxtype:$offset),
3741 Sched<[WriteAdr, WriteSTP]>;
3742 } // hasSideEffects = 0
3746 class BaseLoadStorePairNoAlloc<bits<2> opc, bit V, bit L, dag oops, dag iops,
3748 : I<oops, iops, asm, "\t$Rt, $Rt2, [$Rn, $offset]", "", []> {
3753 let Inst{31-30} = opc;
3754 let Inst{29-27} = 0b101;
3756 let Inst{25-23} = 0b000;
3758 let Inst{21-15} = offset;
3759 let Inst{14-10} = Rt2;
3763 let DecoderMethod = "DecodePairLdStInstruction";
3766 multiclass LoadPairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
3767 Operand indextype, string asm> {
3768 let hasSideEffects = 0, mayStore = 0, mayLoad = 1 in
3769 def i : BaseLoadStorePairNoAlloc<opc, V, 1,
3770 (outs regtype:$Rt, regtype:$Rt2),
3771 (ins GPR64sp:$Rn, indextype:$offset), asm>,
3772 Sched<[WriteLD, WriteLDHi]>;
3775 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3776 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3780 multiclass StorePairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
3781 Operand indextype, string asm> {
3782 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in
3783 def i : BaseLoadStorePairNoAlloc<opc, V, 0, (outs),
3784 (ins regtype:$Rt, regtype:$Rt2,
3785 GPR64sp:$Rn, indextype:$offset),
3789 def : InstAlias<asm # "\t$Rt, $Rt2, [$Rn]",
3790 (!cast<Instruction>(NAME # "i") regtype:$Rt, regtype:$Rt2,
3795 // Load/store exclusive
3798 // True exclusive operations write to and/or read from the system's exclusive
3799 // monitors, which as far as a compiler is concerned can be modelled as a
3800 // random shared memory address. Hence LoadExclusive mayStore.
3802 // Since these instructions have the undefined register bits set to 1 in
3803 // their canonical form, we need a post encoder method to set those bits
3804 // to 1 when encoding these instructions. We do this using the
3805 // fixLoadStoreExclusive function. This function has template parameters:
3807 // fixLoadStoreExclusive<int hasRs, int hasRt2>
3809 // hasRs indicates that the instruction uses the Rs field, so we won't set
3810 // it to 1 (and the same for Rt2). We don't need template parameters for
3811 // the other register fields since Rt and Rn are always used.
3813 let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
3814 class BaseLoadStoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3815 dag oops, dag iops, string asm, string operands>
3816 : I<oops, iops, asm, operands, "", []> {
3817 let Inst{31-30} = sz;
3818 let Inst{29-24} = 0b001000;
3824 let DecoderMethod = "DecodeExclusiveLdStInstruction";
3827 // Neither Rs nor Rt2 operands.
3828 class LoadStoreExclusiveSimple<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3829 dag oops, dag iops, string asm, string operands>
3830 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, oops, iops, asm, operands> {
3833 let Inst{20-16} = 0b11111;
3834 let Unpredictable{20-16} = 0b11111;
3835 let Inst{14-10} = 0b11111;
3836 let Unpredictable{14-10} = 0b11111;
3840 let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
3843 // Simple load acquires don't set the exclusive monitor
3844 let mayLoad = 1, mayStore = 0 in
3845 class LoadAcquire<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3846 RegisterClass regtype, string asm>
3847 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
3848 (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]">,
3851 class LoadExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3852 RegisterClass regtype, string asm>
3853 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
3854 (ins GPR64sp0:$Rn), asm, "\t$Rt, [$Rn]">,
3857 class LoadExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3858 RegisterClass regtype, string asm>
3859 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3860 (outs regtype:$Rt, regtype:$Rt2),
3861 (ins GPR64sp0:$Rn), asm,
3862 "\t$Rt, $Rt2, [$Rn]">,
3863 Sched<[WriteLD, WriteLDHi]> {
3867 let Inst{14-10} = Rt2;
3871 let PostEncoderMethod = "fixLoadStoreExclusive<0,1>";
3874 // Simple store release operations do not check the exclusive monitor.
3875 let mayLoad = 0, mayStore = 1 in
3876 class StoreRelease<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3877 RegisterClass regtype, string asm>
3878 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs),
3879 (ins regtype:$Rt, GPR64sp0:$Rn),
3880 asm, "\t$Rt, [$Rn]">,
3883 let mayLoad = 1, mayStore = 1 in
3884 class StoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3885 RegisterClass regtype, string asm>
3886 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, (outs GPR32:$Ws),
3887 (ins regtype:$Rt, GPR64sp0:$Rn),
3888 asm, "\t$Ws, $Rt, [$Rn]">,
3893 let Inst{20-16} = Ws;
3897 let Constraints = "@earlyclobber $Ws";
3898 let PostEncoderMethod = "fixLoadStoreExclusive<1,0>";
3901 class StoreExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3902 RegisterClass regtype, string asm>
3903 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3905 (ins regtype:$Rt, regtype:$Rt2, GPR64sp0:$Rn),
3906 asm, "\t$Ws, $Rt, $Rt2, [$Rn]">,
3912 let Inst{20-16} = Ws;
3913 let Inst{14-10} = Rt2;
3917 let Constraints = "@earlyclobber $Ws";
3921 // Exception generation
3924 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3925 class ExceptionGeneration<bits<3> op1, bits<2> ll, string asm>
3926 : I<(outs), (ins imm0_65535:$imm), asm, "\t$imm", "", []>,
3929 let Inst{31-24} = 0b11010100;
3930 let Inst{23-21} = op1;
3931 let Inst{20-5} = imm;
3932 let Inst{4-2} = 0b000;
3936 let Predicates = [HasFPARMv8] in {
3939 // Floating point to integer conversion
3942 class BaseFPToIntegerUnscaled<bits<2> type, bits<2> rmode, bits<3> opcode,
3943 RegisterClass srcType, RegisterClass dstType,
3944 string asm, list<dag> pattern>
3945 : I<(outs dstType:$Rd), (ins srcType:$Rn),
3946 asm, "\t$Rd, $Rn", "", pattern>,
3947 Sched<[WriteFCvt]> {
3950 let Inst{30-29} = 0b00;
3951 let Inst{28-24} = 0b11110;
3952 let Inst{23-22} = type;
3954 let Inst{20-19} = rmode;
3955 let Inst{18-16} = opcode;
3956 let Inst{15-10} = 0;
3961 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3962 class BaseFPToInteger<bits<2> type, bits<2> rmode, bits<3> opcode,
3963 RegisterClass srcType, RegisterClass dstType,
3964 Operand immType, string asm, list<dag> pattern>
3965 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
3966 asm, "\t$Rd, $Rn, $scale", "", pattern>,
3967 Sched<[WriteFCvt]> {
3971 let Inst{30-29} = 0b00;
3972 let Inst{28-24} = 0b11110;
3973 let Inst{23-22} = type;
3975 let Inst{20-19} = rmode;
3976 let Inst{18-16} = opcode;
3977 let Inst{15-10} = scale;
3982 multiclass FPToIntegerUnscaled<bits<2> rmode, bits<3> opcode, string asm,
3983 SDPatternOperator OpN> {
3984 // Unscaled half-precision to 32-bit
3985 def UWHr : BaseFPToIntegerUnscaled<0b11, rmode, opcode, FPR16, GPR32, asm,
3986 [(set GPR32:$Rd, (OpN FPR16:$Rn))]> {
3987 let Inst{31} = 0; // 32-bit GPR flag
3988 let Predicates = [HasFullFP16];
3991 // Unscaled half-precision to 64-bit
3992 def UXHr : BaseFPToIntegerUnscaled<0b11, rmode, opcode, FPR16, GPR64, asm,
3993 [(set GPR64:$Rd, (OpN FPR16:$Rn))]> {
3994 let Inst{31} = 1; // 64-bit GPR flag
3995 let Predicates = [HasFullFP16];
3998 // Unscaled single-precision to 32-bit
3999 def UWSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR32, asm,
4000 [(set GPR32:$Rd, (OpN FPR32:$Rn))]> {
4001 let Inst{31} = 0; // 32-bit GPR flag
4004 // Unscaled single-precision to 64-bit
4005 def UXSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR64, asm,
4006 [(set GPR64:$Rd, (OpN FPR32:$Rn))]> {
4007 let Inst{31} = 1; // 64-bit GPR flag
4010 // Unscaled double-precision to 32-bit
4011 def UWDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR32, asm,
4012 [(set GPR32:$Rd, (OpN (f64 FPR64:$Rn)))]> {
4013 let Inst{31} = 0; // 32-bit GPR flag
4016 // Unscaled double-precision to 64-bit
4017 def UXDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR64, asm,
4018 [(set GPR64:$Rd, (OpN (f64 FPR64:$Rn)))]> {
4019 let Inst{31} = 1; // 64-bit GPR flag
4023 multiclass FPToIntegerScaled<bits<2> rmode, bits<3> opcode, string asm,
4024 SDPatternOperator OpN> {
4025 // Scaled half-precision to 32-bit
4026 def SWHri : BaseFPToInteger<0b11, rmode, opcode, FPR16, GPR32,
4027 fixedpoint_f16_i32, asm,
4028 [(set GPR32:$Rd, (OpN (fmul FPR16:$Rn,
4029 fixedpoint_f16_i32:$scale)))]> {
4030 let Inst{31} = 0; // 32-bit GPR flag
4032 let Predicates = [HasFullFP16];
4035 // Scaled half-precision to 64-bit
4036 def SXHri : BaseFPToInteger<0b11, rmode, opcode, FPR16, GPR64,
4037 fixedpoint_f16_i64, asm,
4038 [(set GPR64:$Rd, (OpN (fmul FPR16:$Rn,
4039 fixedpoint_f16_i64:$scale)))]> {
4040 let Inst{31} = 1; // 64-bit GPR flag
4041 let Predicates = [HasFullFP16];
4044 // Scaled single-precision to 32-bit
4045 def SWSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR32,
4046 fixedpoint_f32_i32, asm,
4047 [(set GPR32:$Rd, (OpN (fmul FPR32:$Rn,
4048 fixedpoint_f32_i32:$scale)))]> {
4049 let Inst{31} = 0; // 32-bit GPR flag
4053 // Scaled single-precision to 64-bit
4054 def SXSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR64,
4055 fixedpoint_f32_i64, asm,
4056 [(set GPR64:$Rd, (OpN (fmul FPR32:$Rn,
4057 fixedpoint_f32_i64:$scale)))]> {
4058 let Inst{31} = 1; // 64-bit GPR flag
4061 // Scaled double-precision to 32-bit
4062 def SWDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR32,
4063 fixedpoint_f64_i32, asm,
4064 [(set GPR32:$Rd, (OpN (fmul FPR64:$Rn,
4065 fixedpoint_f64_i32:$scale)))]> {
4066 let Inst{31} = 0; // 32-bit GPR flag
4070 // Scaled double-precision to 64-bit
4071 def SXDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR64,
4072 fixedpoint_f64_i64, asm,
4073 [(set GPR64:$Rd, (OpN (fmul FPR64:$Rn,
4074 fixedpoint_f64_i64:$scale)))]> {
4075 let Inst{31} = 1; // 64-bit GPR flag
4080 // Integer to floating point conversion
4083 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
4084 class BaseIntegerToFP<bit isUnsigned,
4085 RegisterClass srcType, RegisterClass dstType,
4086 Operand immType, string asm, list<dag> pattern>
4087 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
4088 asm, "\t$Rd, $Rn, $scale", "", pattern>,
4089 Sched<[WriteFCvt]> {
4093 let Inst{30-24} = 0b0011110;
4094 let Inst{21-17} = 0b00001;
4095 let Inst{16} = isUnsigned;
4096 let Inst{15-10} = scale;
4101 class BaseIntegerToFPUnscaled<bit isUnsigned,
4102 RegisterClass srcType, RegisterClass dstType,
4103 ValueType dvt, string asm, SDNode node>
4104 : I<(outs dstType:$Rd), (ins srcType:$Rn),
4105 asm, "\t$Rd, $Rn", "", [(set (dvt dstType:$Rd), (node srcType:$Rn))]>,
4106 Sched<[WriteFCvt]> {
4110 let Inst{30-24} = 0b0011110;
4111 let Inst{21-17} = 0b10001;
4112 let Inst{16} = isUnsigned;
4113 let Inst{15-10} = 0b000000;
4118 multiclass IntegerToFP<bit isUnsigned, string asm, SDNode node> {
4120 def UWHri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR16, f16, asm, node> {
4121 let Inst{31} = 0; // 32-bit GPR flag
4122 let Inst{23-22} = 0b11; // 16-bit FPR flag
4123 let Predicates = [HasFullFP16];
4126 def UWSri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR32, f32, asm, node> {
4127 let Inst{31} = 0; // 32-bit GPR flag
4128 let Inst{23-22} = 0b00; // 32-bit FPR flag
4131 def UWDri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR64, f64, asm, node> {
4132 let Inst{31} = 0; // 32-bit GPR flag
4133 let Inst{23-22} = 0b01; // 64-bit FPR flag
4136 def UXHri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR16, f16, asm, node> {
4137 let Inst{31} = 1; // 64-bit GPR flag
4138 let Inst{23-22} = 0b11; // 16-bit FPR flag
4139 let Predicates = [HasFullFP16];
4142 def UXSri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR32, f32, asm, node> {
4143 let Inst{31} = 1; // 64-bit GPR flag
4144 let Inst{23-22} = 0b00; // 32-bit FPR flag
4147 def UXDri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR64, f64, asm, node> {
4148 let Inst{31} = 1; // 64-bit GPR flag
4149 let Inst{23-22} = 0b01; // 64-bit FPR flag
4153 def SWHri: BaseIntegerToFP<isUnsigned, GPR32, FPR16, fixedpoint_f16_i32, asm,
4155 (fdiv (node GPR32:$Rn),
4156 fixedpoint_f16_i32:$scale))]> {
4157 let Inst{31} = 0; // 32-bit GPR flag
4158 let Inst{23-22} = 0b11; // 16-bit FPR flag
4160 let Predicates = [HasFullFP16];
4163 def SWSri: BaseIntegerToFP<isUnsigned, GPR32, FPR32, fixedpoint_f32_i32, asm,
4165 (fdiv (node GPR32:$Rn),
4166 fixedpoint_f32_i32:$scale))]> {
4167 let Inst{31} = 0; // 32-bit GPR flag
4168 let Inst{23-22} = 0b00; // 32-bit FPR flag
4172 def SWDri: BaseIntegerToFP<isUnsigned, GPR32, FPR64, fixedpoint_f64_i32, asm,
4174 (fdiv (node GPR32:$Rn),
4175 fixedpoint_f64_i32:$scale))]> {
4176 let Inst{31} = 0; // 32-bit GPR flag
4177 let Inst{23-22} = 0b01; // 64-bit FPR flag
4181 def SXHri: BaseIntegerToFP<isUnsigned, GPR64, FPR16, fixedpoint_f16_i64, asm,
4183 (fdiv (node GPR64:$Rn),
4184 fixedpoint_f16_i64:$scale))]> {
4185 let Inst{31} = 1; // 64-bit GPR flag
4186 let Inst{23-22} = 0b11; // 16-bit FPR flag
4187 let Predicates = [HasFullFP16];
4190 def SXSri: BaseIntegerToFP<isUnsigned, GPR64, FPR32, fixedpoint_f32_i64, asm,
4192 (fdiv (node GPR64:$Rn),
4193 fixedpoint_f32_i64:$scale))]> {
4194 let Inst{31} = 1; // 64-bit GPR flag
4195 let Inst{23-22} = 0b00; // 32-bit FPR flag
4198 def SXDri: BaseIntegerToFP<isUnsigned, GPR64, FPR64, fixedpoint_f64_i64, asm,
4200 (fdiv (node GPR64:$Rn),
4201 fixedpoint_f64_i64:$scale))]> {
4202 let Inst{31} = 1; // 64-bit GPR flag
4203 let Inst{23-22} = 0b01; // 64-bit FPR flag
4208 // Unscaled integer <-> floating point conversion (i.e. FMOV)
4211 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4212 class BaseUnscaledConversion<bits<2> rmode, bits<3> opcode,
4213 RegisterClass srcType, RegisterClass dstType,
4215 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "",
4216 // We use COPY_TO_REGCLASS for these bitconvert operations.
4217 // copyPhysReg() expands the resultant COPY instructions after
4218 // regalloc is done. This gives greater freedom for the allocator
4219 // and related passes (coalescing, copy propagation, et. al.) to
4220 // be more effective.
4221 [/*(set (dvt dstType:$Rd), (bitconvert (svt srcType:$Rn)))*/]>,
4222 Sched<[WriteFCopy]> {
4225 let Inst{30-24} = 0b0011110;
4227 let Inst{20-19} = rmode;
4228 let Inst{18-16} = opcode;
4229 let Inst{15-10} = 0b000000;
4234 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4235 class BaseUnscaledConversionToHigh<bits<2> rmode, bits<3> opcode,
4236 RegisterClass srcType, RegisterOperand dstType, string asm,
4238 : I<(outs dstType:$Rd), (ins srcType:$Rn, VectorIndex1:$idx), asm,
4239 "{\t$Rd"#kind#"$idx, $Rn|"#kind#"\t$Rd$idx, $Rn}", "", []>,
4240 Sched<[WriteFCopy]> {
4243 let Inst{30-23} = 0b00111101;
4245 let Inst{20-19} = rmode;
4246 let Inst{18-16} = opcode;
4247 let Inst{15-10} = 0b000000;
4251 let DecoderMethod = "DecodeFMOVLaneInstruction";
4254 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4255 class BaseUnscaledConversionFromHigh<bits<2> rmode, bits<3> opcode,
4256 RegisterOperand srcType, RegisterClass dstType, string asm,
4258 : I<(outs dstType:$Rd), (ins srcType:$Rn, VectorIndex1:$idx), asm,
4259 "{\t$Rd, $Rn"#kind#"$idx|"#kind#"\t$Rd, $Rn$idx}", "", []>,
4260 Sched<[WriteFCopy]> {
4263 let Inst{30-23} = 0b00111101;
4265 let Inst{20-19} = rmode;
4266 let Inst{18-16} = opcode;
4267 let Inst{15-10} = 0b000000;
4271 let DecoderMethod = "DecodeFMOVLaneInstruction";
4275 multiclass UnscaledConversion<string asm> {
4276 def WHr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR16, asm> {
4277 let Inst{31} = 0; // 32-bit GPR flag
4278 let Inst{23-22} = 0b11; // 16-bit FPR flag
4279 let Predicates = [HasFullFP16];
4282 def XHr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR16, asm> {
4283 let Inst{31} = 1; // 64-bit GPR flag
4284 let Inst{23-22} = 0b11; // 16-bit FPR flag
4285 let Predicates = [HasFullFP16];
4288 def WSr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR32, asm> {
4289 let Inst{31} = 0; // 32-bit GPR flag
4290 let Inst{23-22} = 0b00; // 32-bit FPR flag
4293 def XDr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR64, asm> {
4294 let Inst{31} = 1; // 64-bit GPR flag
4295 let Inst{23-22} = 0b01; // 64-bit FPR flag
4298 def HWr : BaseUnscaledConversion<0b00, 0b110, FPR16, GPR32, asm> {
4299 let Inst{31} = 0; // 32-bit GPR flag
4300 let Inst{23-22} = 0b11; // 16-bit FPR flag
4301 let Predicates = [HasFullFP16];
4304 def HXr : BaseUnscaledConversion<0b00, 0b110, FPR16, GPR64, asm> {
4305 let Inst{31} = 1; // 64-bit GPR flag
4306 let Inst{23-22} = 0b11; // 16-bit FPR flag
4307 let Predicates = [HasFullFP16];
4310 def SWr : BaseUnscaledConversion<0b00, 0b110, FPR32, GPR32, asm> {
4311 let Inst{31} = 0; // 32-bit GPR flag
4312 let Inst{23-22} = 0b00; // 32-bit FPR flag
4315 def DXr : BaseUnscaledConversion<0b00, 0b110, FPR64, GPR64, asm> {
4316 let Inst{31} = 1; // 64-bit GPR flag
4317 let Inst{23-22} = 0b01; // 64-bit FPR flag
4320 def XDHighr : BaseUnscaledConversionToHigh<0b01, 0b111, GPR64, V128,
4326 def DXHighr : BaseUnscaledConversionFromHigh<0b01, 0b110, V128, GPR64,
4334 // Floating point conversion
4337 class BaseFPConversion<bits<2> type, bits<2> opcode, RegisterClass dstType,
4338 RegisterClass srcType, string asm, list<dag> pattern>
4339 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "", pattern>,
4340 Sched<[WriteFCvt]> {
4343 let Inst{31-24} = 0b00011110;
4344 let Inst{23-22} = type;
4345 let Inst{21-17} = 0b10001;
4346 let Inst{16-15} = opcode;
4347 let Inst{14-10} = 0b10000;
4352 multiclass FPConversion<string asm> {
4353 // Double-precision to Half-precision
4354 def HDr : BaseFPConversion<0b01, 0b11, FPR16, FPR64, asm,
4355 [(set FPR16:$Rd, (fpround FPR64:$Rn))]>;
4357 // Double-precision to Single-precision
4358 def SDr : BaseFPConversion<0b01, 0b00, FPR32, FPR64, asm,
4359 [(set FPR32:$Rd, (fpround FPR64:$Rn))]>;
4361 // Half-precision to Double-precision
4362 def DHr : BaseFPConversion<0b11, 0b01, FPR64, FPR16, asm,
4363 [(set FPR64:$Rd, (fpextend FPR16:$Rn))]>;
4365 // Half-precision to Single-precision
4366 def SHr : BaseFPConversion<0b11, 0b00, FPR32, FPR16, asm,
4367 [(set FPR32:$Rd, (fpextend FPR16:$Rn))]>;
4369 // Single-precision to Double-precision
4370 def DSr : BaseFPConversion<0b00, 0b01, FPR64, FPR32, asm,
4371 [(set FPR64:$Rd, (fpextend FPR32:$Rn))]>;
4373 // Single-precision to Half-precision
4374 def HSr : BaseFPConversion<0b00, 0b11, FPR16, FPR32, asm,
4375 [(set FPR16:$Rd, (fpround FPR32:$Rn))]>;
4379 // Single operand floating point data processing
4382 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4383 class BaseSingleOperandFPData<bits<4> opcode, RegisterClass regtype,
4384 ValueType vt, string asm, SDPatternOperator node>
4385 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
4386 [(set (vt regtype:$Rd), (node (vt regtype:$Rn)))]>,
4390 let Inst{31-24} = 0b00011110;
4391 let Inst{21-19} = 0b100;
4392 let Inst{18-15} = opcode;
4393 let Inst{14-10} = 0b10000;
4398 multiclass SingleOperandFPData<bits<4> opcode, string asm,
4399 SDPatternOperator node = null_frag> {
4400 def Hr : BaseSingleOperandFPData<opcode, FPR16, f16, asm, node> {
4401 let Inst{23-22} = 0b11; // 16-bit size flag
4402 let Predicates = [HasFullFP16];
4405 def Sr : BaseSingleOperandFPData<opcode, FPR32, f32, asm, node> {
4406 let Inst{23-22} = 0b00; // 32-bit size flag
4409 def Dr : BaseSingleOperandFPData<opcode, FPR64, f64, asm, node> {
4410 let Inst{23-22} = 0b01; // 64-bit size flag
4415 // Two operand floating point data processing
4418 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4419 class BaseTwoOperandFPData<bits<4> opcode, RegisterClass regtype,
4420 string asm, list<dag> pat>
4421 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
4422 asm, "\t$Rd, $Rn, $Rm", "", pat>,
4427 let Inst{31-24} = 0b00011110;
4429 let Inst{20-16} = Rm;
4430 let Inst{15-12} = opcode;
4431 let Inst{11-10} = 0b10;
4436 multiclass TwoOperandFPData<bits<4> opcode, string asm,
4437 SDPatternOperator node = null_frag> {
4438 def Hrr : BaseTwoOperandFPData<opcode, FPR16, asm,
4439 [(set (f16 FPR16:$Rd),
4440 (node (f16 FPR16:$Rn), (f16 FPR16:$Rm)))]> {
4441 let Inst{23-22} = 0b11; // 16-bit size flag
4442 let Predicates = [HasFullFP16];
4445 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
4446 [(set (f32 FPR32:$Rd),
4447 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]> {
4448 let Inst{23-22} = 0b00; // 32-bit size flag
4451 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
4452 [(set (f64 FPR64:$Rd),
4453 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]> {
4454 let Inst{23-22} = 0b01; // 64-bit size flag
4458 multiclass TwoOperandFPDataNeg<bits<4> opcode, string asm, SDNode node> {
4459 def Hrr : BaseTwoOperandFPData<opcode, FPR16, asm,
4460 [(set FPR16:$Rd, (fneg (node FPR16:$Rn, (f16 FPR16:$Rm))))]> {
4461 let Inst{23-22} = 0b11; // 16-bit size flag
4462 let Predicates = [HasFullFP16];
4465 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
4466 [(set FPR32:$Rd, (fneg (node FPR32:$Rn, (f32 FPR32:$Rm))))]> {
4467 let Inst{23-22} = 0b00; // 32-bit size flag
4470 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
4471 [(set FPR64:$Rd, (fneg (node FPR64:$Rn, (f64 FPR64:$Rm))))]> {
4472 let Inst{23-22} = 0b01; // 64-bit size flag
4478 // Three operand floating point data processing
4481 class BaseThreeOperandFPData<bit isNegated, bit isSub,
4482 RegisterClass regtype, string asm, list<dag> pat>
4483 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, regtype: $Ra),
4484 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pat>,
4485 Sched<[WriteFMul]> {
4490 let Inst{31-24} = 0b00011111;
4491 let Inst{21} = isNegated;
4492 let Inst{20-16} = Rm;
4493 let Inst{15} = isSub;
4494 let Inst{14-10} = Ra;
4499 multiclass ThreeOperandFPData<bit isNegated, bit isSub,string asm,
4500 SDPatternOperator node> {
4501 def Hrrr : BaseThreeOperandFPData<isNegated, isSub, FPR16, asm,
4503 (node (f16 FPR16:$Rn), (f16 FPR16:$Rm), (f16 FPR16:$Ra)))]> {
4504 let Inst{23-22} = 0b11; // 16-bit size flag
4505 let Predicates = [HasFullFP16];
4508 def Srrr : BaseThreeOperandFPData<isNegated, isSub, FPR32, asm,
4510 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm), (f32 FPR32:$Ra)))]> {
4511 let Inst{23-22} = 0b00; // 32-bit size flag
4514 def Drrr : BaseThreeOperandFPData<isNegated, isSub, FPR64, asm,
4516 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm), (f64 FPR64:$Ra)))]> {
4517 let Inst{23-22} = 0b01; // 64-bit size flag
4522 // Floating point data comparisons
4525 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4526 class BaseOneOperandFPComparison<bit signalAllNans,
4527 RegisterClass regtype, string asm,
4529 : I<(outs), (ins regtype:$Rn), asm, "\t$Rn, #0.0", "", pat>,
4530 Sched<[WriteFCmp]> {
4532 let Inst{31-24} = 0b00011110;
4535 let Inst{15-10} = 0b001000;
4537 let Inst{4} = signalAllNans;
4538 let Inst{3-0} = 0b1000;
4540 // Rm should be 0b00000 canonically, but we need to accept any value.
4541 let PostEncoderMethod = "fixOneOperandFPComparison";
4544 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4545 class BaseTwoOperandFPComparison<bit signalAllNans, RegisterClass regtype,
4546 string asm, list<dag> pat>
4547 : I<(outs), (ins regtype:$Rn, regtype:$Rm), asm, "\t$Rn, $Rm", "", pat>,
4548 Sched<[WriteFCmp]> {
4551 let Inst{31-24} = 0b00011110;
4553 let Inst{20-16} = Rm;
4554 let Inst{15-10} = 0b001000;
4556 let Inst{4} = signalAllNans;
4557 let Inst{3-0} = 0b0000;
4560 multiclass FPComparison<bit signalAllNans, string asm,
4561 SDPatternOperator OpNode = null_frag> {
4562 let Defs = [NZCV] in {
4563 def Hrr : BaseTwoOperandFPComparison<signalAllNans, FPR16, asm,
4564 [(OpNode FPR16:$Rn, (f16 FPR16:$Rm)), (implicit NZCV)]> {
4565 let Inst{23-22} = 0b11;
4566 let Predicates = [HasFullFP16];
4569 def Hri : BaseOneOperandFPComparison<signalAllNans, FPR16, asm,
4570 [(OpNode (f16 FPR16:$Rn), fpimm0), (implicit NZCV)]> {
4571 let Inst{23-22} = 0b11;
4572 let Predicates = [HasFullFP16];
4575 def Srr : BaseTwoOperandFPComparison<signalAllNans, FPR32, asm,
4576 [(OpNode FPR32:$Rn, (f32 FPR32:$Rm)), (implicit NZCV)]> {
4577 let Inst{23-22} = 0b00;
4580 def Sri : BaseOneOperandFPComparison<signalAllNans, FPR32, asm,
4581 [(OpNode (f32 FPR32:$Rn), fpimm0), (implicit NZCV)]> {
4582 let Inst{23-22} = 0b00;
4585 def Drr : BaseTwoOperandFPComparison<signalAllNans, FPR64, asm,
4586 [(OpNode FPR64:$Rn, (f64 FPR64:$Rm)), (implicit NZCV)]> {
4587 let Inst{23-22} = 0b01;
4590 def Dri : BaseOneOperandFPComparison<signalAllNans, FPR64, asm,
4591 [(OpNode (f64 FPR64:$Rn), fpimm0), (implicit NZCV)]> {
4592 let Inst{23-22} = 0b01;
4598 // Floating point conditional comparisons
4601 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4602 class BaseFPCondComparison<bit signalAllNans, RegisterClass regtype,
4603 string mnemonic, list<dag> pat>
4604 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm32_0_15:$nzcv, ccode:$cond),
4605 mnemonic, "\t$Rn, $Rm, $nzcv, $cond", "", pat>,
4606 Sched<[WriteFCmp]> {
4615 let Inst{31-24} = 0b00011110;
4617 let Inst{20-16} = Rm;
4618 let Inst{15-12} = cond;
4619 let Inst{11-10} = 0b01;
4621 let Inst{4} = signalAllNans;
4622 let Inst{3-0} = nzcv;
4625 multiclass FPCondComparison<bit signalAllNans, string mnemonic,
4626 SDPatternOperator OpNode = null_frag> {
4627 def Hrr : BaseFPCondComparison<signalAllNans, FPR16, mnemonic, []> {
4628 let Inst{23-22} = 0b11;
4629 let Predicates = [HasFullFP16];
4632 def Srr : BaseFPCondComparison<signalAllNans, FPR32, mnemonic,
4633 [(set NZCV, (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm), (i32 imm:$nzcv),
4634 (i32 imm:$cond), NZCV))]> {
4635 let Inst{23-22} = 0b00;
4638 def Drr : BaseFPCondComparison<signalAllNans, FPR64, mnemonic,
4639 [(set NZCV, (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm), (i32 imm:$nzcv),
4640 (i32 imm:$cond), NZCV))]> {
4641 let Inst{23-22} = 0b01;
4646 // Floating point conditional select
4649 class BaseFPCondSelect<RegisterClass regtype, ValueType vt, string asm>
4650 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
4651 asm, "\t$Rd, $Rn, $Rm, $cond", "",
4653 (AArch64csel (vt regtype:$Rn), regtype:$Rm,
4654 (i32 imm:$cond), NZCV))]>,
4661 let Inst{31-24} = 0b00011110;
4663 let Inst{20-16} = Rm;
4664 let Inst{15-12} = cond;
4665 let Inst{11-10} = 0b11;
4670 multiclass FPCondSelect<string asm> {
4671 let Uses = [NZCV] in {
4672 def Hrrr : BaseFPCondSelect<FPR16, f16, asm> {
4673 let Inst{23-22} = 0b11;
4674 let Predicates = [HasFullFP16];
4677 def Srrr : BaseFPCondSelect<FPR32, f32, asm> {
4678 let Inst{23-22} = 0b00;
4681 def Drrr : BaseFPCondSelect<FPR64, f64, asm> {
4682 let Inst{23-22} = 0b01;
4688 // Floating move immediate
4691 class BaseFPMoveImmediate<RegisterClass regtype, Operand fpimmtype, string asm>
4692 : I<(outs regtype:$Rd), (ins fpimmtype:$imm), asm, "\t$Rd, $imm", "",
4693 [(set regtype:$Rd, fpimmtype:$imm)]>,
4694 Sched<[WriteFImm]> {
4697 let Inst{31-24} = 0b00011110;
4699 let Inst{20-13} = imm;
4700 let Inst{12-5} = 0b10000000;
4704 multiclass FPMoveImmediate<string asm> {
4705 def Hi : BaseFPMoveImmediate<FPR16, fpimm16, asm> {
4706 let Inst{23-22} = 0b11;
4707 let Predicates = [HasFullFP16];
4710 def Si : BaseFPMoveImmediate<FPR32, fpimm32, asm> {
4711 let Inst{23-22} = 0b00;
4714 def Di : BaseFPMoveImmediate<FPR64, fpimm64, asm> {
4715 let Inst{23-22} = 0b01;
4718 } // end of 'let Predicates = [HasFPARMv8]'
4720 //----------------------------------------------------------------------------
4722 //----------------------------------------------------------------------------
4724 let Predicates = [HasNEON] in {
4726 //----------------------------------------------------------------------------
4727 // AdvSIMD three register vector instructions
4728 //----------------------------------------------------------------------------
4730 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4731 class BaseSIMDThreeSameVector<bit Q, bit U, bits<3> size, bits<5> opcode,
4732 RegisterOperand regtype, string asm, string kind,
4734 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
4735 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
4736 "|" # kind # "\t$Rd, $Rn, $Rm|}", "", pattern>,
4744 let Inst{28-24} = 0b01110;
4745 let Inst{23-21} = size;
4746 let Inst{20-16} = Rm;
4747 let Inst{15-11} = opcode;
4753 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4754 class BaseSIMDThreeSameVectorTied<bit Q, bit U, bits<3> size, bits<5> opcode,
4755 RegisterOperand regtype, string asm, string kind,
4757 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn, regtype:$Rm), asm,
4758 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
4759 "|" # kind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
4767 let Inst{28-24} = 0b01110;
4768 let Inst{23-21} = size;
4769 let Inst{20-16} = Rm;
4770 let Inst{15-11} = opcode;
4776 class BaseSIMDThreeSameVectorDot<bit Q, bit U, string asm, string kind1,
4777 string kind2, RegisterOperand RegType,
4778 ValueType AccumType, ValueType InputType,
4779 SDPatternOperator OpNode> :
4780 BaseSIMDThreeSameVectorTied<Q, U, 0b100, 0b10010, RegType, asm, kind1,
4781 [(set (AccumType RegType:$dst),
4782 (OpNode (AccumType RegType:$Rd),
4783 (InputType RegType:$Rn),
4784 (InputType RegType:$Rm)))]> {
4785 let AsmString = !strconcat(asm, "{\t$Rd" # kind1 # ", $Rn" # kind2 # ", $Rm" # kind2 # "}");
4788 multiclass SIMDThreeSameVectorDot<bit U, string asm, SDPatternOperator OpNode> {
4789 def v8i8 : BaseSIMDThreeSameVectorDot<0, U, asm, ".2s", ".8b", V64,
4790 v2i32, v8i8, OpNode>;
4791 def v16i8 : BaseSIMDThreeSameVectorDot<1, U, asm, ".4s", ".16b", V128,
4792 v4i32, v16i8, OpNode>;
4795 // All operand sizes distinguished in the encoding.
4796 multiclass SIMDThreeSameVector<bit U, bits<5> opc, string asm,
4797 SDPatternOperator OpNode> {
4798 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
4800 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4801 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
4803 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
4804 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
4806 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4807 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
4809 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
4810 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
4812 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4813 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
4815 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
4816 def v2i64 : BaseSIMDThreeSameVector<1, U, 0b111, opc, V128,
4818 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
4821 // As above, but D sized elements unsupported.
4822 multiclass SIMDThreeSameVectorBHS<bit U, bits<5> opc, string asm,
4823 SDPatternOperator OpNode> {
4824 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
4826 [(set V64:$Rd, (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))]>;
4827 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
4829 [(set V128:$Rd, (v16i8 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm))))]>;
4830 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
4832 [(set V64:$Rd, (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))]>;
4833 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
4835 [(set V128:$Rd, (v8i16 (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm))))]>;
4836 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
4838 [(set V64:$Rd, (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))]>;
4839 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
4841 [(set V128:$Rd, (v4i32 (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm))))]>;
4844 multiclass SIMDThreeSameVectorBHSTied<bit U, bits<5> opc, string asm,
4845 SDPatternOperator OpNode> {
4846 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, 0b001, opc, V64,
4848 [(set (v8i8 V64:$dst),
4849 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4850 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, 0b001, opc, V128,
4852 [(set (v16i8 V128:$dst),
4853 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
4854 def v4i16 : BaseSIMDThreeSameVectorTied<0, U, 0b011, opc, V64,
4856 [(set (v4i16 V64:$dst),
4857 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4858 def v8i16 : BaseSIMDThreeSameVectorTied<1, U, 0b011, opc, V128,
4860 [(set (v8i16 V128:$dst),
4861 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
4862 def v2i32 : BaseSIMDThreeSameVectorTied<0, U, 0b101, opc, V64,
4864 [(set (v2i32 V64:$dst),
4865 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4866 def v4i32 : BaseSIMDThreeSameVectorTied<1, U, 0b101, opc, V128,
4868 [(set (v4i32 V128:$dst),
4869 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
4872 // As above, but only B sized elements supported.
4873 multiclass SIMDThreeSameVectorB<bit U, bits<5> opc, string asm,
4874 SDPatternOperator OpNode> {
4875 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b001, opc, V64,
4877 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4878 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b001, opc, V128,
4880 [(set (v16i8 V128:$Rd),
4881 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
4884 // As above, but only floating point elements supported.
4885 multiclass SIMDThreeSameVectorFP<bit U, bit S, bits<3> opc,
4886 string asm, SDPatternOperator OpNode> {
4887 let Predicates = [HasNEON, HasFullFP16] in {
4888 def v4f16 : BaseSIMDThreeSameVector<0, U, {S,0b10}, {0b00,opc}, V64,
4890 [(set (v4f16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
4891 def v8f16 : BaseSIMDThreeSameVector<1, U, {S,0b10}, {0b00,opc}, V128,
4893 [(set (v8f16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
4894 } // Predicates = [HasNEON, HasFullFP16]
4895 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0b01}, {0b11,opc}, V64,
4897 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
4898 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0b01}, {0b11,opc}, V128,
4900 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
4901 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,0b11}, {0b11,opc}, V128,
4903 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
4906 multiclass SIMDThreeSameVectorFPCmp<bit U, bit S, bits<3> opc,
4908 SDPatternOperator OpNode> {
4909 let Predicates = [HasNEON, HasFullFP16] in {
4910 def v4f16 : BaseSIMDThreeSameVector<0, U, {S,0b10}, {0b00,opc}, V64,
4912 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
4913 def v8f16 : BaseSIMDThreeSameVector<1, U, {S,0b10}, {0b00,opc}, V128,
4915 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
4916 } // Predicates = [HasNEON, HasFullFP16]
4917 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0b01}, {0b11,opc}, V64,
4919 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
4920 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0b01}, {0b11,opc}, V128,
4922 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
4923 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,0b11}, {0b11,opc}, V128,
4925 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
4928 multiclass SIMDThreeSameVectorFPTied<bit U, bit S, bits<3> opc,
4929 string asm, SDPatternOperator OpNode> {
4930 let Predicates = [HasNEON, HasFullFP16] in {
4931 def v4f16 : BaseSIMDThreeSameVectorTied<0, U, {S,0b10}, {0b00,opc}, V64,
4933 [(set (v4f16 V64:$dst),
4934 (OpNode (v4f16 V64:$Rd), (v4f16 V64:$Rn), (v4f16 V64:$Rm)))]>;
4935 def v8f16 : BaseSIMDThreeSameVectorTied<1, U, {S,0b10}, {0b00,opc}, V128,
4937 [(set (v8f16 V128:$dst),
4938 (OpNode (v8f16 V128:$Rd), (v8f16 V128:$Rn), (v8f16 V128:$Rm)))]>;
4939 } // Predicates = [HasNEON, HasFullFP16]
4940 def v2f32 : BaseSIMDThreeSameVectorTied<0, U, {S,0b01}, {0b11,opc}, V64,
4942 [(set (v2f32 V64:$dst),
4943 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
4944 def v4f32 : BaseSIMDThreeSameVectorTied<1, U, {S,0b01}, {0b11,opc}, V128,
4946 [(set (v4f32 V128:$dst),
4947 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
4948 def v2f64 : BaseSIMDThreeSameVectorTied<1, U, {S,0b11}, {0b11,opc}, V128,
4950 [(set (v2f64 V128:$dst),
4951 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
4954 // As above, but D and B sized elements unsupported.
4955 multiclass SIMDThreeSameVectorHS<bit U, bits<5> opc, string asm,
4956 SDPatternOperator OpNode> {
4957 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b011, opc, V64,
4959 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4960 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b011, opc, V128,
4962 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
4963 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b101, opc, V64,
4965 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4966 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b101, opc, V128,
4968 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
4971 // Logical three vector ops share opcode bits, and only use B sized elements.
4972 multiclass SIMDLogicalThreeVector<bit U, bits<2> size, string asm,
4973 SDPatternOperator OpNode = null_frag> {
4974 def v8i8 : BaseSIMDThreeSameVector<0, U, {size,1}, 0b00011, V64,
4976 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn, V64:$Rm))]>;
4977 def v16i8 : BaseSIMDThreeSameVector<1, U, {size,1}, 0b00011, V128,
4979 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn, V128:$Rm))]>;
4981 def : Pat<(v4i16 (OpNode V64:$LHS, V64:$RHS)),
4982 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
4983 def : Pat<(v2i32 (OpNode V64:$LHS, V64:$RHS)),
4984 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
4985 def : Pat<(v1i64 (OpNode V64:$LHS, V64:$RHS)),
4986 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
4988 def : Pat<(v8i16 (OpNode V128:$LHS, V128:$RHS)),
4989 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
4990 def : Pat<(v4i32 (OpNode V128:$LHS, V128:$RHS)),
4991 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
4992 def : Pat<(v2i64 (OpNode V128:$LHS, V128:$RHS)),
4993 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
4996 multiclass SIMDLogicalThreeVectorTied<bit U, bits<2> size,
4997 string asm, SDPatternOperator OpNode> {
4998 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, {size,1}, 0b00011, V64,
5000 [(set (v8i8 V64:$dst),
5001 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5002 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, {size,1}, 0b00011, V128,
5004 [(set (v16i8 V128:$dst),
5005 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
5006 (v16i8 V128:$Rm)))]>;
5008 def : Pat<(v4i16 (OpNode (v4i16 V64:$LHS), (v4i16 V64:$MHS),
5010 (!cast<Instruction>(NAME#"v8i8")
5011 V64:$LHS, V64:$MHS, V64:$RHS)>;
5012 def : Pat<(v2i32 (OpNode (v2i32 V64:$LHS), (v2i32 V64:$MHS),
5014 (!cast<Instruction>(NAME#"v8i8")
5015 V64:$LHS, V64:$MHS, V64:$RHS)>;
5016 def : Pat<(v1i64 (OpNode (v1i64 V64:$LHS), (v1i64 V64:$MHS),
5018 (!cast<Instruction>(NAME#"v8i8")
5019 V64:$LHS, V64:$MHS, V64:$RHS)>;
5021 def : Pat<(v8i16 (OpNode (v8i16 V128:$LHS), (v8i16 V128:$MHS),
5022 (v8i16 V128:$RHS))),
5023 (!cast<Instruction>(NAME#"v16i8")
5024 V128:$LHS, V128:$MHS, V128:$RHS)>;
5025 def : Pat<(v4i32 (OpNode (v4i32 V128:$LHS), (v4i32 V128:$MHS),
5026 (v4i32 V128:$RHS))),
5027 (!cast<Instruction>(NAME#"v16i8")
5028 V128:$LHS, V128:$MHS, V128:$RHS)>;
5029 def : Pat<(v2i64 (OpNode (v2i64 V128:$LHS), (v2i64 V128:$MHS),
5030 (v2i64 V128:$RHS))),
5031 (!cast<Instruction>(NAME#"v16i8")
5032 V128:$LHS, V128:$MHS, V128:$RHS)>;
5036 //----------------------------------------------------------------------------
5037 // AdvSIMD two register vector instructions.
5038 //----------------------------------------------------------------------------
5040 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5041 class BaseSIMDTwoSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5042 bits<2> size2, RegisterOperand regtype, string asm,
5043 string dstkind, string srckind, list<dag> pattern>
5044 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5045 "{\t$Rd" # dstkind # ", $Rn" # srckind #
5046 "|" # dstkind # "\t$Rd, $Rn}", "", pattern>,
5053 let Inst{28-24} = 0b01110;
5054 let Inst{23-22} = size;
5056 let Inst{20-19} = size2;
5057 let Inst{18-17} = 0b00;
5058 let Inst{16-12} = opcode;
5059 let Inst{11-10} = 0b10;
5064 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5065 class BaseSIMDTwoSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5066 bits<2> size2, RegisterOperand regtype,
5067 string asm, string dstkind, string srckind,
5069 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn), asm,
5070 "{\t$Rd" # dstkind # ", $Rn" # srckind #
5071 "|" # dstkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
5078 let Inst{28-24} = 0b01110;
5079 let Inst{23-22} = size;
5081 let Inst{20-19} = size2;
5082 let Inst{18-17} = 0b00;
5083 let Inst{16-12} = opcode;
5084 let Inst{11-10} = 0b10;
5089 // Supports B, H, and S element sizes.
5090 multiclass SIMDTwoVectorBHS<bit U, bits<5> opc, string asm,
5091 SDPatternOperator OpNode> {
5092 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5094 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5095 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5096 asm, ".16b", ".16b",
5097 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5098 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5100 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5101 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5103 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5104 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5106 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5107 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5109 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5112 class BaseSIMDVectorLShiftLongBySize<bit Q, bits<2> size,
5113 RegisterOperand regtype, string asm, string dstkind,
5114 string srckind, string amount>
5115 : I<(outs V128:$Rd), (ins regtype:$Rn), asm,
5116 "{\t$Rd" # dstkind # ", $Rn" # srckind # ", #" # amount #
5117 "|" # dstkind # "\t$Rd, $Rn, #" # amount # "}", "", []>,
5123 let Inst{29-24} = 0b101110;
5124 let Inst{23-22} = size;
5125 let Inst{21-10} = 0b100001001110;
5130 multiclass SIMDVectorLShiftLongBySizeBHS {
5131 let hasSideEffects = 0 in {
5132 def v8i8 : BaseSIMDVectorLShiftLongBySize<0, 0b00, V64,
5133 "shll", ".8h", ".8b", "8">;
5134 def v16i8 : BaseSIMDVectorLShiftLongBySize<1, 0b00, V128,
5135 "shll2", ".8h", ".16b", "8">;
5136 def v4i16 : BaseSIMDVectorLShiftLongBySize<0, 0b01, V64,
5137 "shll", ".4s", ".4h", "16">;
5138 def v8i16 : BaseSIMDVectorLShiftLongBySize<1, 0b01, V128,
5139 "shll2", ".4s", ".8h", "16">;
5140 def v2i32 : BaseSIMDVectorLShiftLongBySize<0, 0b10, V64,
5141 "shll", ".2d", ".2s", "32">;
5142 def v4i32 : BaseSIMDVectorLShiftLongBySize<1, 0b10, V128,
5143 "shll2", ".2d", ".4s", "32">;
5147 // Supports all element sizes.
5148 multiclass SIMDLongTwoVector<bit U, bits<5> opc, string asm,
5149 SDPatternOperator OpNode> {
5150 def v8i8_v4i16 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5152 [(set (v4i16 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5153 def v16i8_v8i16 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5155 [(set (v8i16 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5156 def v4i16_v2i32 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5158 [(set (v2i32 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5159 def v8i16_v4i32 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5161 [(set (v4i32 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5162 def v2i32_v1i64 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5164 [(set (v1i64 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5165 def v4i32_v2i64 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5167 [(set (v2i64 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5170 multiclass SIMDLongTwoVectorTied<bit U, bits<5> opc, string asm,
5171 SDPatternOperator OpNode> {
5172 def v8i8_v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, 0b00, V64,
5174 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd),
5176 def v16i8_v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, 0b00, V128,
5178 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd),
5179 (v16i8 V128:$Rn)))]>;
5180 def v4i16_v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, 0b00, V64,
5182 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd),
5183 (v4i16 V64:$Rn)))]>;
5184 def v8i16_v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, 0b00, V128,
5186 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd),
5187 (v8i16 V128:$Rn)))]>;
5188 def v2i32_v1i64 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, 0b00, V64,
5190 [(set (v1i64 V64:$dst), (OpNode (v1i64 V64:$Rd),
5191 (v2i32 V64:$Rn)))]>;
5192 def v4i32_v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, 0b00, V128,
5194 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd),
5195 (v4i32 V128:$Rn)))]>;
5198 // Supports all element sizes, except 1xD.
5199 multiclass SIMDTwoVectorBHSDTied<bit U, bits<5> opc, string asm,
5200 SDPatternOperator OpNode> {
5201 def v8i8 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, 0b00, V64,
5203 [(set (v8i8 V64:$dst), (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn)))]>;
5204 def v16i8 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, 0b00, V128,
5205 asm, ".16b", ".16b",
5206 [(set (v16i8 V128:$dst), (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
5207 def v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, 0b00, V64,
5209 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn)))]>;
5210 def v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, 0b00, V128,
5212 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn)))]>;
5213 def v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, 0b00, V64,
5215 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn)))]>;
5216 def v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, 0b00, V128,
5218 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
5219 def v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b11, opc, 0b00, V128,
5221 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn)))]>;
5224 multiclass SIMDTwoVectorBHSD<bit U, bits<5> opc, string asm,
5225 SDPatternOperator OpNode = null_frag> {
5226 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5228 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5229 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5230 asm, ".16b", ".16b",
5231 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5232 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5234 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5235 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5237 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5238 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, 0b00, V64,
5240 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5241 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, 0b00, V128,
5243 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5244 def v2i64 : BaseSIMDTwoSameVector<1, U, 0b11, opc, 0b00, V128,
5246 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5250 // Supports only B element sizes.
5251 multiclass SIMDTwoVectorB<bit U, bits<2> size, bits<5> opc, string asm,
5252 SDPatternOperator OpNode> {
5253 def v8i8 : BaseSIMDTwoSameVector<0, U, size, opc, 0b00, V64,
5255 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
5256 def v16i8 : BaseSIMDTwoSameVector<1, U, size, opc, 0b00, V128,
5257 asm, ".16b", ".16b",
5258 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
5262 // Supports only B and H element sizes.
5263 multiclass SIMDTwoVectorBH<bit U, bits<5> opc, string asm,
5264 SDPatternOperator OpNode> {
5265 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, 0b00, V64,
5267 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn))]>;
5268 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, 0b00, V128,
5269 asm, ".16b", ".16b",
5270 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn))]>;
5271 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, 0b00, V64,
5273 [(set (v4i16 V64:$Rd), (OpNode V64:$Rn))]>;
5274 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, 0b00, V128,
5276 [(set (v8i16 V128:$Rd), (OpNode V128:$Rn))]>;
5279 // Supports only S and D element sizes, uses high bit of the size field
5280 // as an extra opcode bit.
5281 multiclass SIMDTwoVectorFP<bit U, bit S, bits<5> opc, string asm,
5282 SDPatternOperator OpNode> {
5283 let Predicates = [HasNEON, HasFullFP16] in {
5284 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5286 [(set (v4f16 V64:$Rd), (OpNode (v4f16 V64:$Rn)))]>;
5287 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5289 [(set (v8f16 V128:$Rd), (OpNode (v8f16 V128:$Rn)))]>;
5290 } // Predicates = [HasNEON, HasFullFP16]
5291 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5293 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5294 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5296 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5297 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5299 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5302 // Supports only S element size.
5303 multiclass SIMDTwoVectorS<bit U, bit S, bits<5> opc, string asm,
5304 SDPatternOperator OpNode> {
5305 def v2i32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5307 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5308 def v4i32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5310 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5314 multiclass SIMDTwoVectorFPToInt<bit U, bit S, bits<5> opc, string asm,
5315 SDPatternOperator OpNode> {
5316 let Predicates = [HasNEON, HasFullFP16] in {
5317 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5319 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn)))]>;
5320 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5322 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn)))]>;
5323 } // Predicates = [HasNEON, HasFullFP16]
5324 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5326 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
5327 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5329 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
5330 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5332 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5335 multiclass SIMDTwoVectorIntToFP<bit U, bit S, bits<5> opc, string asm,
5336 SDPatternOperator OpNode> {
5337 let Predicates = [HasNEON, HasFullFP16] in {
5338 def v4f16 : BaseSIMDTwoSameVector<0, U, {S,1}, opc, 0b11, V64,
5340 [(set (v4f16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
5341 def v8f16 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b11, V128,
5343 [(set (v8f16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5344 } // Predicates = [HasNEON, HasFullFP16]
5345 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, 0b00, V64,
5347 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
5348 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, 0b00, V128,
5350 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5351 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, 0b00, V128,
5353 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5357 class BaseSIMDMixedTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5358 RegisterOperand inreg, RegisterOperand outreg,
5359 string asm, string outkind, string inkind,
5361 : I<(outs outreg:$Rd), (ins inreg:$Rn), asm,
5362 "{\t$Rd" # outkind # ", $Rn" # inkind #
5363 "|" # outkind # "\t$Rd, $Rn}", "", pattern>,
5370 let Inst{28-24} = 0b01110;
5371 let Inst{23-22} = size;
5372 let Inst{21-17} = 0b10000;
5373 let Inst{16-12} = opcode;
5374 let Inst{11-10} = 0b10;
5379 class BaseSIMDMixedTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5380 RegisterOperand inreg, RegisterOperand outreg,
5381 string asm, string outkind, string inkind,
5383 : I<(outs outreg:$dst), (ins outreg:$Rd, inreg:$Rn), asm,
5384 "{\t$Rd" # outkind # ", $Rn" # inkind #
5385 "|" # outkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
5392 let Inst{28-24} = 0b01110;
5393 let Inst{23-22} = size;
5394 let Inst{21-17} = 0b10000;
5395 let Inst{16-12} = opcode;
5396 let Inst{11-10} = 0b10;
5401 multiclass SIMDMixedTwoVector<bit U, bits<5> opc, string asm,
5402 SDPatternOperator OpNode> {
5403 def v8i8 : BaseSIMDMixedTwoVector<0, U, 0b00, opc, V128, V64,
5405 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
5406 def v16i8 : BaseSIMDMixedTwoVectorTied<1, U, 0b00, opc, V128, V128,
5407 asm#"2", ".16b", ".8h", []>;
5408 def v4i16 : BaseSIMDMixedTwoVector<0, U, 0b01, opc, V128, V64,
5410 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
5411 def v8i16 : BaseSIMDMixedTwoVectorTied<1, U, 0b01, opc, V128, V128,
5412 asm#"2", ".8h", ".4s", []>;
5413 def v2i32 : BaseSIMDMixedTwoVector<0, U, 0b10, opc, V128, V64,
5415 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
5416 def v4i32 : BaseSIMDMixedTwoVectorTied<1, U, 0b10, opc, V128, V128,
5417 asm#"2", ".4s", ".2d", []>;
5419 def : Pat<(concat_vectors (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn))),
5420 (!cast<Instruction>(NAME # "v16i8")
5421 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5422 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn))),
5423 (!cast<Instruction>(NAME # "v8i16")
5424 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5425 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn))),
5426 (!cast<Instruction>(NAME # "v4i32")
5427 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5430 class BaseSIMDCmpTwoVector<bit Q, bit U, bits<2> size, bits<2> size2,
5431 bits<5> opcode, RegisterOperand regtype, string asm,
5432 string kind, string zero, ValueType dty,
5433 ValueType sty, SDNode OpNode>
5434 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5435 "{\t$Rd" # kind # ", $Rn" # kind # ", #" # zero #
5436 "|" # kind # "\t$Rd, $Rn, #" # zero # "}", "",
5437 [(set (dty regtype:$Rd), (OpNode (sty regtype:$Rn)))]>,
5444 let Inst{28-24} = 0b01110;
5445 let Inst{23-22} = size;
5447 let Inst{20-19} = size2;
5448 let Inst{18-17} = 0b00;
5449 let Inst{16-12} = opcode;
5450 let Inst{11-10} = 0b10;
5455 // Comparisons support all element sizes, except 1xD.
5456 multiclass SIMDCmpTwoVector<bit U, bits<5> opc, string asm,
5458 def v8i8rz : BaseSIMDCmpTwoVector<0, U, 0b00, 0b00, opc, V64,
5460 v8i8, v8i8, OpNode>;
5461 def v16i8rz : BaseSIMDCmpTwoVector<1, U, 0b00, 0b00, opc, V128,
5463 v16i8, v16i8, OpNode>;
5464 def v4i16rz : BaseSIMDCmpTwoVector<0, U, 0b01, 0b00, opc, V64,
5466 v4i16, v4i16, OpNode>;
5467 def v8i16rz : BaseSIMDCmpTwoVector<1, U, 0b01, 0b00, opc, V128,
5469 v8i16, v8i16, OpNode>;
5470 def v2i32rz : BaseSIMDCmpTwoVector<0, U, 0b10, 0b00, opc, V64,
5472 v2i32, v2i32, OpNode>;
5473 def v4i32rz : BaseSIMDCmpTwoVector<1, U, 0b10, 0b00, opc, V128,
5475 v4i32, v4i32, OpNode>;
5476 def v2i64rz : BaseSIMDCmpTwoVector<1, U, 0b11, 0b00, opc, V128,
5478 v2i64, v2i64, OpNode>;
5481 // FP Comparisons support only S and D element sizes (and H for v8.2a).
5482 multiclass SIMDFPCmpTwoVector<bit U, bit S, bits<5> opc,
5483 string asm, SDNode OpNode> {
5485 let Predicates = [HasNEON, HasFullFP16] in {
5486 def v4i16rz : BaseSIMDCmpTwoVector<0, U, {S,1}, 0b11, opc, V64,
5488 v4i16, v4f16, OpNode>;
5489 def v8i16rz : BaseSIMDCmpTwoVector<1, U, {S,1}, 0b11, opc, V128,
5491 v8i16, v8f16, OpNode>;
5492 } // Predicates = [HasNEON, HasFullFP16]
5493 def v2i32rz : BaseSIMDCmpTwoVector<0, U, {S,0}, 0b00, opc, V64,
5495 v2i32, v2f32, OpNode>;
5496 def v4i32rz : BaseSIMDCmpTwoVector<1, U, {S,0}, 0b00, opc, V128,
5498 v4i32, v4f32, OpNode>;
5499 def v2i64rz : BaseSIMDCmpTwoVector<1, U, {S,1}, 0b00, opc, V128,
5501 v2i64, v2f64, OpNode>;
5503 let Predicates = [HasNEON, HasFullFP16] in {
5504 def : InstAlias<asm # "\t$Vd.4h, $Vn.4h, #0",
5505 (!cast<Instruction>(NAME # v4i16rz) V64:$Vd, V64:$Vn), 0>;
5506 def : InstAlias<asm # "\t$Vd.8h, $Vn.8h, #0",
5507 (!cast<Instruction>(NAME # v8i16rz) V128:$Vd, V128:$Vn), 0>;
5509 def : InstAlias<asm # "\t$Vd.2s, $Vn.2s, #0",
5510 (!cast<Instruction>(NAME # v2i32rz) V64:$Vd, V64:$Vn), 0>;
5511 def : InstAlias<asm # "\t$Vd.4s, $Vn.4s, #0",
5512 (!cast<Instruction>(NAME # v4i32rz) V128:$Vd, V128:$Vn), 0>;
5513 def : InstAlias<asm # "\t$Vd.2d, $Vn.2d, #0",
5514 (!cast<Instruction>(NAME # v2i64rz) V128:$Vd, V128:$Vn), 0>;
5515 let Predicates = [HasNEON, HasFullFP16] in {
5516 def : InstAlias<asm # ".4h\t$Vd, $Vn, #0",
5517 (!cast<Instruction>(NAME # v4i16rz) V64:$Vd, V64:$Vn), 0>;
5518 def : InstAlias<asm # ".8h\t$Vd, $Vn, #0",
5519 (!cast<Instruction>(NAME # v8i16rz) V128:$Vd, V128:$Vn), 0>;
5521 def : InstAlias<asm # ".2s\t$Vd, $Vn, #0",
5522 (!cast<Instruction>(NAME # v2i32rz) V64:$Vd, V64:$Vn), 0>;
5523 def : InstAlias<asm # ".4s\t$Vd, $Vn, #0",
5524 (!cast<Instruction>(NAME # v4i32rz) V128:$Vd, V128:$Vn), 0>;
5525 def : InstAlias<asm # ".2d\t$Vd, $Vn, #0",
5526 (!cast<Instruction>(NAME # v2i64rz) V128:$Vd, V128:$Vn), 0>;
5529 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5530 class BaseSIMDFPCvtTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
5531 RegisterOperand outtype, RegisterOperand intype,
5532 string asm, string VdTy, string VnTy,
5534 : I<(outs outtype:$Rd), (ins intype:$Rn), asm,
5535 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "", pattern>,
5542 let Inst{28-24} = 0b01110;
5543 let Inst{23-22} = size;
5544 let Inst{21-17} = 0b10000;
5545 let Inst{16-12} = opcode;
5546 let Inst{11-10} = 0b10;
5551 class BaseSIMDFPCvtTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
5552 RegisterOperand outtype, RegisterOperand intype,
5553 string asm, string VdTy, string VnTy,
5555 : I<(outs outtype:$dst), (ins outtype:$Rd, intype:$Rn), asm,
5556 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "$Rd = $dst", pattern>,
5563 let Inst{28-24} = 0b01110;
5564 let Inst{23-22} = size;
5565 let Inst{21-17} = 0b10000;
5566 let Inst{16-12} = opcode;
5567 let Inst{11-10} = 0b10;
5572 multiclass SIMDFPWidenTwoVector<bit U, bit S, bits<5> opc, string asm> {
5573 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V128, V64,
5574 asm, ".4s", ".4h", []>;
5575 def v8i16 : BaseSIMDFPCvtTwoVector<1, U, {S,0}, opc, V128, V128,
5576 asm#"2", ".4s", ".8h", []>;
5577 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V128, V64,
5578 asm, ".2d", ".2s", []>;
5579 def v4i32 : BaseSIMDFPCvtTwoVector<1, U, {S,1}, opc, V128, V128,
5580 asm#"2", ".2d", ".4s", []>;
5583 multiclass SIMDFPNarrowTwoVector<bit U, bit S, bits<5> opc, string asm> {
5584 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V64, V128,
5585 asm, ".4h", ".4s", []>;
5586 def v8i16 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,0}, opc, V128, V128,
5587 asm#"2", ".8h", ".4s", []>;
5588 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
5589 asm, ".2s", ".2d", []>;
5590 def v4i32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
5591 asm#"2", ".4s", ".2d", []>;
5594 multiclass SIMDFPInexactCvtTwoVector<bit U, bit S, bits<5> opc, string asm,
5596 def v2f32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
5598 [(set (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
5599 def v4f32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
5600 asm#"2", ".4s", ".2d", []>;
5602 def : Pat<(concat_vectors (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn))),
5603 (!cast<Instruction>(NAME # "v4f32")
5604 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
5607 //----------------------------------------------------------------------------
5608 // AdvSIMD three register different-size vector instructions.
5609 //----------------------------------------------------------------------------
5611 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5612 class BaseSIMDDifferentThreeVector<bit U, bits<3> size, bits<4> opcode,
5613 RegisterOperand outtype, RegisterOperand intype1,
5614 RegisterOperand intype2, string asm,
5615 string outkind, string inkind1, string inkind2,
5617 : I<(outs outtype:$Rd), (ins intype1:$Rn, intype2:$Rm), asm,
5618 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
5619 "|" # outkind # "\t$Rd, $Rn, $Rm}", "", pattern>,
5625 let Inst{30} = size{0};
5627 let Inst{28-24} = 0b01110;
5628 let Inst{23-22} = size{2-1};
5630 let Inst{20-16} = Rm;
5631 let Inst{15-12} = opcode;
5632 let Inst{11-10} = 0b00;
5637 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5638 class BaseSIMDDifferentThreeVectorTied<bit U, bits<3> size, bits<4> opcode,
5639 RegisterOperand outtype, RegisterOperand intype1,
5640 RegisterOperand intype2, string asm,
5641 string outkind, string inkind1, string inkind2,
5643 : I<(outs outtype:$dst), (ins outtype:$Rd, intype1:$Rn, intype2:$Rm), asm,
5644 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
5645 "|" # outkind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
5651 let Inst{30} = size{0};
5653 let Inst{28-24} = 0b01110;
5654 let Inst{23-22} = size{2-1};
5656 let Inst{20-16} = Rm;
5657 let Inst{15-12} = opcode;
5658 let Inst{11-10} = 0b00;
5663 // FIXME: TableGen doesn't know how to deal with expanded types that also
5664 // change the element count (in this case, placing the results in
5665 // the high elements of the result register rather than the low
5666 // elements). Until that's fixed, we can't code-gen those.
5667 multiclass SIMDNarrowThreeVectorBHS<bit U, bits<4> opc, string asm,
5669 def v8i16_v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
5671 asm, ".8b", ".8h", ".8h",
5672 [(set (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
5673 def v8i16_v16i8 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
5675 asm#"2", ".16b", ".8h", ".8h",
5677 def v4i32_v4i16 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
5679 asm, ".4h", ".4s", ".4s",
5680 [(set (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
5681 def v4i32_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
5683 asm#"2", ".8h", ".4s", ".4s",
5685 def v2i64_v2i32 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
5687 asm, ".2s", ".2d", ".2d",
5688 [(set (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
5689 def v2i64_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
5691 asm#"2", ".4s", ".2d", ".2d",
5695 // Patterns for the '2' variants involve INSERT_SUBREG, which you can't put in
5696 // a version attached to an instruction.
5697 def : Pat<(concat_vectors (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn),
5699 (!cast<Instruction>(NAME # "v8i16_v16i8")
5700 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5701 V128:$Rn, V128:$Rm)>;
5702 def : Pat<(concat_vectors (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn),
5704 (!cast<Instruction>(NAME # "v4i32_v8i16")
5705 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5706 V128:$Rn, V128:$Rm)>;
5707 def : Pat<(concat_vectors (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn),
5709 (!cast<Instruction>(NAME # "v2i64_v4i32")
5710 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
5711 V128:$Rn, V128:$Rm)>;
5714 multiclass SIMDDifferentThreeVectorBD<bit U, bits<4> opc, string asm,
5716 def v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
5718 asm, ".8h", ".8b", ".8b",
5719 [(set (v8i16 V128:$Rd), (IntOp (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5720 def v16i8 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
5722 asm#"2", ".8h", ".16b", ".16b", []>;
5723 let Predicates = [HasAES] in {
5724 def v1i64 : BaseSIMDDifferentThreeVector<U, 0b110, opc,
5726 asm, ".1q", ".1d", ".1d", []>;
5727 def v2i64 : BaseSIMDDifferentThreeVector<U, 0b111, opc,
5729 asm#"2", ".1q", ".2d", ".2d", []>;
5732 def : Pat<(v8i16 (IntOp (v8i8 (extract_high_v16i8 V128:$Rn)),
5733 (v8i8 (extract_high_v16i8 V128:$Rm)))),
5734 (!cast<Instruction>(NAME#"v16i8") V128:$Rn, V128:$Rm)>;
5737 multiclass SIMDLongThreeVectorHS<bit U, bits<4> opc, string asm,
5738 SDPatternOperator OpNode> {
5739 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
5741 asm, ".4s", ".4h", ".4h",
5742 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5743 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
5745 asm#"2", ".4s", ".8h", ".8h",
5746 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
5747 (extract_high_v8i16 V128:$Rm)))]>;
5748 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
5750 asm, ".2d", ".2s", ".2s",
5751 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5752 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
5754 asm#"2", ".2d", ".4s", ".4s",
5755 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
5756 (extract_high_v4i32 V128:$Rm)))]>;
5759 multiclass SIMDLongThreeVectorBHSabdl<bit U, bits<4> opc, string asm,
5760 SDPatternOperator OpNode = null_frag> {
5761 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
5763 asm, ".8h", ".8b", ".8b",
5764 [(set (v8i16 V128:$Rd),
5765 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))))]>;
5766 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
5768 asm#"2", ".8h", ".16b", ".16b",
5769 [(set (v8i16 V128:$Rd),
5770 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
5771 (extract_high_v16i8 V128:$Rm)))))]>;
5772 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
5774 asm, ".4s", ".4h", ".4h",
5775 [(set (v4i32 V128:$Rd),
5776 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))))]>;
5777 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
5779 asm#"2", ".4s", ".8h", ".8h",
5780 [(set (v4i32 V128:$Rd),
5781 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
5782 (extract_high_v8i16 V128:$Rm)))))]>;
5783 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
5785 asm, ".2d", ".2s", ".2s",
5786 [(set (v2i64 V128:$Rd),
5787 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))))]>;
5788 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
5790 asm#"2", ".2d", ".4s", ".4s",
5791 [(set (v2i64 V128:$Rd),
5792 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
5793 (extract_high_v4i32 V128:$Rm)))))]>;
5796 multiclass SIMDLongThreeVectorTiedBHSabal<bit U, bits<4> opc,
5798 SDPatternOperator OpNode> {
5799 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
5801 asm, ".8h", ".8b", ".8b",
5802 [(set (v8i16 V128:$dst),
5803 (add (v8i16 V128:$Rd),
5804 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))))]>;
5805 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
5807 asm#"2", ".8h", ".16b", ".16b",
5808 [(set (v8i16 V128:$dst),
5809 (add (v8i16 V128:$Rd),
5810 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
5811 (extract_high_v16i8 V128:$Rm))))))]>;
5812 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
5814 asm, ".4s", ".4h", ".4h",
5815 [(set (v4i32 V128:$dst),
5816 (add (v4i32 V128:$Rd),
5817 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))))]>;
5818 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
5820 asm#"2", ".4s", ".8h", ".8h",
5821 [(set (v4i32 V128:$dst),
5822 (add (v4i32 V128:$Rd),
5823 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
5824 (extract_high_v8i16 V128:$Rm))))))]>;
5825 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
5827 asm, ".2d", ".2s", ".2s",
5828 [(set (v2i64 V128:$dst),
5829 (add (v2i64 V128:$Rd),
5830 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))))]>;
5831 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
5833 asm#"2", ".2d", ".4s", ".4s",
5834 [(set (v2i64 V128:$dst),
5835 (add (v2i64 V128:$Rd),
5836 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
5837 (extract_high_v4i32 V128:$Rm))))))]>;
5840 multiclass SIMDLongThreeVectorBHS<bit U, bits<4> opc, string asm,
5841 SDPatternOperator OpNode = null_frag> {
5842 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
5844 asm, ".8h", ".8b", ".8b",
5845 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5846 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
5848 asm#"2", ".8h", ".16b", ".16b",
5849 [(set (v8i16 V128:$Rd), (OpNode (extract_high_v16i8 V128:$Rn),
5850 (extract_high_v16i8 V128:$Rm)))]>;
5851 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
5853 asm, ".4s", ".4h", ".4h",
5854 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5855 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
5857 asm#"2", ".4s", ".8h", ".8h",
5858 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
5859 (extract_high_v8i16 V128:$Rm)))]>;
5860 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
5862 asm, ".2d", ".2s", ".2s",
5863 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5864 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
5866 asm#"2", ".2d", ".4s", ".4s",
5867 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
5868 (extract_high_v4i32 V128:$Rm)))]>;
5871 multiclass SIMDLongThreeVectorTiedBHS<bit U, bits<4> opc,
5873 SDPatternOperator OpNode> {
5874 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
5876 asm, ".8h", ".8b", ".8b",
5877 [(set (v8i16 V128:$dst),
5878 (OpNode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
5879 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
5881 asm#"2", ".8h", ".16b", ".16b",
5882 [(set (v8i16 V128:$dst),
5883 (OpNode (v8i16 V128:$Rd),
5884 (extract_high_v16i8 V128:$Rn),
5885 (extract_high_v16i8 V128:$Rm)))]>;
5886 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
5888 asm, ".4s", ".4h", ".4h",
5889 [(set (v4i32 V128:$dst),
5890 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
5891 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
5893 asm#"2", ".4s", ".8h", ".8h",
5894 [(set (v4i32 V128:$dst),
5895 (OpNode (v4i32 V128:$Rd),
5896 (extract_high_v8i16 V128:$Rn),
5897 (extract_high_v8i16 V128:$Rm)))]>;
5898 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
5900 asm, ".2d", ".2s", ".2s",
5901 [(set (v2i64 V128:$dst),
5902 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
5903 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
5905 asm#"2", ".2d", ".4s", ".4s",
5906 [(set (v2i64 V128:$dst),
5907 (OpNode (v2i64 V128:$Rd),
5908 (extract_high_v4i32 V128:$Rn),
5909 (extract_high_v4i32 V128:$Rm)))]>;
5912 multiclass SIMDLongThreeVectorSQDMLXTiedHS<bit U, bits<4> opc, string asm,
5913 SDPatternOperator Accum> {
5914 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
5916 asm, ".4s", ".4h", ".4h",
5917 [(set (v4i32 V128:$dst),
5918 (Accum (v4i32 V128:$Rd),
5919 (v4i32 (int_aarch64_neon_sqdmull (v4i16 V64:$Rn),
5920 (v4i16 V64:$Rm)))))]>;
5921 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
5923 asm#"2", ".4s", ".8h", ".8h",
5924 [(set (v4i32 V128:$dst),
5925 (Accum (v4i32 V128:$Rd),
5926 (v4i32 (int_aarch64_neon_sqdmull (extract_high_v8i16 V128:$Rn),
5927 (extract_high_v8i16 V128:$Rm)))))]>;
5928 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
5930 asm, ".2d", ".2s", ".2s",
5931 [(set (v2i64 V128:$dst),
5932 (Accum (v2i64 V128:$Rd),
5933 (v2i64 (int_aarch64_neon_sqdmull (v2i32 V64:$Rn),
5934 (v2i32 V64:$Rm)))))]>;
5935 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
5937 asm#"2", ".2d", ".4s", ".4s",
5938 [(set (v2i64 V128:$dst),
5939 (Accum (v2i64 V128:$Rd),
5940 (v2i64 (int_aarch64_neon_sqdmull (extract_high_v4i32 V128:$Rn),
5941 (extract_high_v4i32 V128:$Rm)))))]>;
5944 multiclass SIMDWideThreeVectorBHS<bit U, bits<4> opc, string asm,
5945 SDPatternOperator OpNode> {
5946 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
5948 asm, ".8h", ".8h", ".8b",
5949 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i8 V64:$Rm)))]>;
5950 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
5952 asm#"2", ".8h", ".8h", ".16b",
5953 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
5954 (extract_high_v16i8 V128:$Rm)))]>;
5955 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
5957 asm, ".4s", ".4s", ".4h",
5958 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i16 V64:$Rm)))]>;
5959 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
5961 asm#"2", ".4s", ".4s", ".8h",
5962 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
5963 (extract_high_v8i16 V128:$Rm)))]>;
5964 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
5966 asm, ".2d", ".2d", ".2s",
5967 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i32 V64:$Rm)))]>;
5968 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
5970 asm#"2", ".2d", ".2d", ".4s",
5971 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
5972 (extract_high_v4i32 V128:$Rm)))]>;
5975 //----------------------------------------------------------------------------
5976 // AdvSIMD bitwise extract from vector
5977 //----------------------------------------------------------------------------
5979 class BaseSIMDBitwiseExtract<bit size, RegisterOperand regtype, ValueType vty,
5980 string asm, string kind>
5981 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, i32imm:$imm), asm,
5982 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $imm" #
5983 "|" # kind # "\t$Rd, $Rn, $Rm, $imm}", "",
5984 [(set (vty regtype:$Rd),
5985 (AArch64ext regtype:$Rn, regtype:$Rm, (i32 imm:$imm)))]>,
5992 let Inst{30} = size;
5993 let Inst{29-21} = 0b101110000;
5994 let Inst{20-16} = Rm;
5996 let Inst{14-11} = imm;
6003 multiclass SIMDBitwiseExtract<string asm> {
6004 def v8i8 : BaseSIMDBitwiseExtract<0, V64, v8i8, asm, ".8b"> {
6007 def v16i8 : BaseSIMDBitwiseExtract<1, V128, v16i8, asm, ".16b">;
6010 //----------------------------------------------------------------------------
6011 // AdvSIMD zip vector
6012 //----------------------------------------------------------------------------
6014 class BaseSIMDZipVector<bits<3> size, bits<3> opc, RegisterOperand regtype,
6015 string asm, string kind, SDNode OpNode, ValueType valty>
6016 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
6017 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
6018 "|" # kind # "\t$Rd, $Rn, $Rm}", "",
6019 [(set (valty regtype:$Rd), (OpNode regtype:$Rn, regtype:$Rm))]>,
6025 let Inst{30} = size{0};
6026 let Inst{29-24} = 0b001110;
6027 let Inst{23-22} = size{2-1};
6029 let Inst{20-16} = Rm;
6031 let Inst{14-12} = opc;
6032 let Inst{11-10} = 0b10;
6037 multiclass SIMDZipVector<bits<3>opc, string asm,
6039 def v8i8 : BaseSIMDZipVector<0b000, opc, V64,
6040 asm, ".8b", OpNode, v8i8>;
6041 def v16i8 : BaseSIMDZipVector<0b001, opc, V128,
6042 asm, ".16b", OpNode, v16i8>;
6043 def v4i16 : BaseSIMDZipVector<0b010, opc, V64,
6044 asm, ".4h", OpNode, v4i16>;
6045 def v8i16 : BaseSIMDZipVector<0b011, opc, V128,
6046 asm, ".8h", OpNode, v8i16>;
6047 def v2i32 : BaseSIMDZipVector<0b100, opc, V64,
6048 asm, ".2s", OpNode, v2i32>;
6049 def v4i32 : BaseSIMDZipVector<0b101, opc, V128,
6050 asm, ".4s", OpNode, v4i32>;
6051 def v2i64 : BaseSIMDZipVector<0b111, opc, V128,
6052 asm, ".2d", OpNode, v2i64>;
6054 def : Pat<(v4f16 (OpNode V64:$Rn, V64:$Rm)),
6055 (!cast<Instruction>(NAME#"v4i16") V64:$Rn, V64:$Rm)>;
6056 def : Pat<(v8f16 (OpNode V128:$Rn, V128:$Rm)),
6057 (!cast<Instruction>(NAME#"v8i16") V128:$Rn, V128:$Rm)>;
6058 def : Pat<(v2f32 (OpNode V64:$Rn, V64:$Rm)),
6059 (!cast<Instruction>(NAME#"v2i32") V64:$Rn, V64:$Rm)>;
6060 def : Pat<(v4f32 (OpNode V128:$Rn, V128:$Rm)),
6061 (!cast<Instruction>(NAME#"v4i32") V128:$Rn, V128:$Rm)>;
6062 def : Pat<(v2f64 (OpNode V128:$Rn, V128:$Rm)),
6063 (!cast<Instruction>(NAME#"v2i64") V128:$Rn, V128:$Rm)>;
6066 //----------------------------------------------------------------------------
6067 // AdvSIMD three register scalar instructions
6068 //----------------------------------------------------------------------------
6070 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6071 class BaseSIMDThreeScalar<bit U, bits<3> size, bits<5> opcode,
6072 RegisterClass regtype, string asm,
6074 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
6075 "\t$Rd, $Rn, $Rm", "", pattern>,
6080 let Inst{31-30} = 0b01;
6082 let Inst{28-24} = 0b11110;
6083 let Inst{23-21} = size;
6084 let Inst{20-16} = Rm;
6085 let Inst{15-11} = opcode;
6091 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6092 class BaseSIMDThreeScalarTied<bit U, bits<2> size, bit R, bits<5> opcode,
6093 dag oops, dag iops, string asm,
6095 : I<oops, iops, asm, "\t$Rd, $Rn, $Rm", "$Rd = $dst", pattern>,
6100 let Inst{31-30} = 0b01;
6102 let Inst{28-24} = 0b11110;
6103 let Inst{23-22} = size;
6105 let Inst{20-16} = Rm;
6106 let Inst{15-11} = opcode;
6112 multiclass SIMDThreeScalarD<bit U, bits<5> opc, string asm,
6113 SDPatternOperator OpNode> {
6114 def v1i64 : BaseSIMDThreeScalar<U, 0b111, opc, FPR64, asm,
6115 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
6118 multiclass SIMDThreeScalarBHSD<bit U, bits<5> opc, string asm,
6119 SDPatternOperator OpNode> {
6120 def v1i64 : BaseSIMDThreeScalar<U, 0b111, opc, FPR64, asm,
6121 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
6122 def v1i32 : BaseSIMDThreeScalar<U, 0b101, opc, FPR32, asm, []>;
6123 def v1i16 : BaseSIMDThreeScalar<U, 0b011, opc, FPR16, asm, []>;
6124 def v1i8 : BaseSIMDThreeScalar<U, 0b001, opc, FPR8 , asm, []>;
6126 def : Pat<(i64 (OpNode (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
6127 (!cast<Instruction>(NAME#"v1i64") FPR64:$Rn, FPR64:$Rm)>;
6128 def : Pat<(i32 (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm))),
6129 (!cast<Instruction>(NAME#"v1i32") FPR32:$Rn, FPR32:$Rm)>;
6132 multiclass SIMDThreeScalarHS<bit U, bits<5> opc, string asm,
6133 SDPatternOperator OpNode> {
6134 def v1i32 : BaseSIMDThreeScalar<U, 0b101, opc, FPR32, asm,
6135 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
6136 def v1i16 : BaseSIMDThreeScalar<U, 0b011, opc, FPR16, asm, []>;
6139 multiclass SIMDThreeScalarHSTied<bit U, bit R, bits<5> opc, string asm,
6140 SDPatternOperator OpNode = null_frag> {
6141 def v1i32: BaseSIMDThreeScalarTied<U, 0b10, R, opc, (outs FPR32:$dst),
6142 (ins FPR32:$Rd, FPR32:$Rn, FPR32:$Rm),
6144 def v1i16: BaseSIMDThreeScalarTied<U, 0b01, R, opc, (outs FPR16:$dst),
6145 (ins FPR16:$Rd, FPR16:$Rn, FPR16:$Rm),
6149 multiclass SIMDFPThreeScalar<bit U, bit S, bits<3> opc, string asm,
6150 SDPatternOperator OpNode = null_frag> {
6151 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6152 def #NAME#64 : BaseSIMDThreeScalar<U, {S,0b11}, {0b11,opc}, FPR64, asm,
6153 [(set (f64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
6154 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0b01}, {0b11,opc}, FPR32, asm,
6155 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
6156 let Predicates = [HasNEON, HasFullFP16] in {
6157 def #NAME#16 : BaseSIMDThreeScalar<U, {S,0b10}, {0b00,opc}, FPR16, asm,
6158 [(set FPR16:$Rd, (OpNode FPR16:$Rn, FPR16:$Rm))]>;
6159 } // Predicates = [HasNEON, HasFullFP16]
6162 def : Pat<(v1f64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
6163 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
6166 multiclass SIMDThreeScalarFPCmp<bit U, bit S, bits<3> opc, string asm,
6167 SDPatternOperator OpNode = null_frag> {
6168 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6169 def #NAME#64 : BaseSIMDThreeScalar<U, {S,0b11}, {0b11,opc}, FPR64, asm,
6170 [(set (i64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
6171 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0b01}, {0b11,opc}, FPR32, asm,
6172 [(set (i32 FPR32:$Rd), (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]>;
6173 let Predicates = [HasNEON, HasFullFP16] in {
6174 def #NAME#16 : BaseSIMDThreeScalar<U, {S,0b10}, {0b00,opc}, FPR16, asm,
6176 } // Predicates = [HasNEON, HasFullFP16]
6179 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
6180 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
6183 class BaseSIMDThreeScalarMixed<bit U, bits<2> size, bits<5> opcode,
6184 dag oops, dag iops, string asm, string cstr, list<dag> pat>
6185 : I<oops, iops, asm,
6186 "\t$Rd, $Rn, $Rm", cstr, pat>,
6191 let Inst{31-30} = 0b01;
6193 let Inst{28-24} = 0b11110;
6194 let Inst{23-22} = size;
6196 let Inst{20-16} = Rm;
6197 let Inst{15-11} = opcode;
6203 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6204 multiclass SIMDThreeScalarMixedHS<bit U, bits<5> opc, string asm,
6205 SDPatternOperator OpNode = null_frag> {
6206 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
6208 (ins FPR16:$Rn, FPR16:$Rm), asm, "", []>;
6209 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
6211 (ins FPR32:$Rn, FPR32:$Rm), asm, "",
6212 [(set (i64 FPR64:$Rd), (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
6215 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6216 multiclass SIMDThreeScalarMixedTiedHS<bit U, bits<5> opc, string asm,
6217 SDPatternOperator OpNode = null_frag> {
6218 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
6220 (ins FPR32:$Rd, FPR16:$Rn, FPR16:$Rm),
6221 asm, "$Rd = $dst", []>;
6222 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
6224 (ins FPR64:$Rd, FPR32:$Rn, FPR32:$Rm),
6226 [(set (i64 FPR64:$dst),
6227 (OpNode (i64 FPR64:$Rd), (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
6230 //----------------------------------------------------------------------------
6231 // AdvSIMD two register scalar instructions
6232 //----------------------------------------------------------------------------
6234 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6235 class BaseSIMDTwoScalar<bit U, bits<2> size, bits<2> size2, bits<5> opcode,
6236 RegisterClass regtype, RegisterClass regtype2,
6237 string asm, list<dag> pat>
6238 : I<(outs regtype:$Rd), (ins regtype2:$Rn), asm,
6239 "\t$Rd, $Rn", "", pat>,
6243 let Inst{31-30} = 0b01;
6245 let Inst{28-24} = 0b11110;
6246 let Inst{23-22} = size;
6248 let Inst{20-19} = size2;
6249 let Inst{18-17} = 0b00;
6250 let Inst{16-12} = opcode;
6251 let Inst{11-10} = 0b10;
6256 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6257 class BaseSIMDTwoScalarTied<bit U, bits<2> size, bits<5> opcode,
6258 RegisterClass regtype, RegisterClass regtype2,
6259 string asm, list<dag> pat>
6260 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype2:$Rn), asm,
6261 "\t$Rd, $Rn", "$Rd = $dst", pat>,
6265 let Inst{31-30} = 0b01;
6267 let Inst{28-24} = 0b11110;
6268 let Inst{23-22} = size;
6269 let Inst{21-17} = 0b10000;
6270 let Inst{16-12} = opcode;
6271 let Inst{11-10} = 0b10;
6277 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6278 class BaseSIMDCmpTwoScalar<bit U, bits<2> size, bits<2> size2, bits<5> opcode,
6279 RegisterClass regtype, string asm, string zero>
6280 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
6281 "\t$Rd, $Rn, #" # zero, "", []>,
6285 let Inst{31-30} = 0b01;
6287 let Inst{28-24} = 0b11110;
6288 let Inst{23-22} = size;
6290 let Inst{20-19} = size2;
6291 let Inst{18-17} = 0b00;
6292 let Inst{16-12} = opcode;
6293 let Inst{11-10} = 0b10;
6298 class SIMDInexactCvtTwoScalar<bits<5> opcode, string asm>
6299 : I<(outs FPR32:$Rd), (ins FPR64:$Rn), asm, "\t$Rd, $Rn", "",
6300 [(set (f32 FPR32:$Rd), (int_aarch64_sisd_fcvtxn (f64 FPR64:$Rn)))]>,
6304 let Inst{31-17} = 0b011111100110000;
6305 let Inst{16-12} = opcode;
6306 let Inst{11-10} = 0b10;
6311 multiclass SIMDCmpTwoScalarD<bit U, bits<5> opc, string asm,
6312 SDPatternOperator OpNode> {
6313 def v1i64rz : BaseSIMDCmpTwoScalar<U, 0b11, 0b00, opc, FPR64, asm, "0">;
6315 def : Pat<(v1i64 (OpNode FPR64:$Rn)),
6316 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
6319 multiclass SIMDFPCmpTwoScalar<bit U, bit S, bits<5> opc, string asm,
6320 SDPatternOperator OpNode> {
6321 def v1i64rz : BaseSIMDCmpTwoScalar<U, {S,1}, 0b00, opc, FPR64, asm, "0.0">;
6322 def v1i32rz : BaseSIMDCmpTwoScalar<U, {S,0}, 0b00, opc, FPR32, asm, "0.0">;
6323 let Predicates = [HasNEON, HasFullFP16] in {
6324 def v1i16rz : BaseSIMDCmpTwoScalar<U, {S,1}, 0b11, opc, FPR16, asm, "0.0">;
6327 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6328 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rd, FPR64:$Rn), 0>;
6329 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6330 (!cast<Instruction>(NAME # v1i32rz) FPR32:$Rd, FPR32:$Rn), 0>;
6331 let Predicates = [HasNEON, HasFullFP16] in {
6332 def : InstAlias<asm # "\t$Rd, $Rn, #0",
6333 (!cast<Instruction>(NAME # v1i16rz) FPR16:$Rd, FPR16:$Rn), 0>;
6336 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn))),
6337 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
6340 multiclass SIMDTwoScalarD<bit U, bits<5> opc, string asm,
6341 SDPatternOperator OpNode = null_frag> {
6342 def v1i64 : BaseSIMDTwoScalar<U, 0b11, 0b00, opc, FPR64, FPR64, asm,
6343 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn)))]>;
6345 def : Pat<(i64 (OpNode (i64 FPR64:$Rn))),
6346 (!cast<Instruction>(NAME # "v1i64") FPR64:$Rn)>;
6349 multiclass SIMDFPTwoScalar<bit U, bit S, bits<5> opc, string asm> {
6350 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, 0b00, opc, FPR64, FPR64, asm,[]>;
6351 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, 0b00, opc, FPR32, FPR32, asm,[]>;
6352 let Predicates = [HasNEON, HasFullFP16] in {
6353 def v1f16 : BaseSIMDTwoScalar<U, {S,1}, 0b11, opc, FPR16, FPR16, asm,[]>;
6357 multiclass SIMDFPTwoScalarCVT<bit U, bit S, bits<5> opc, string asm,
6358 SDPatternOperator OpNode> {
6359 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, 0b00, opc, FPR64, FPR64, asm,
6360 [(set FPR64:$Rd, (OpNode (f64 FPR64:$Rn)))]>;
6361 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, 0b00, opc, FPR32, FPR32, asm,
6362 [(set FPR32:$Rd, (OpNode (f32 FPR32:$Rn)))]>;
6363 let Predicates = [HasNEON, HasFullFP16] in {
6364 def v1i16 : BaseSIMDTwoScalar<U, {S,1}, 0b11, opc, FPR16, FPR16, asm,
6365 [(set FPR16:$Rd, (OpNode (f16 FPR16:$Rn)))]>;
6369 multiclass SIMDTwoScalarBHSD<bit U, bits<5> opc, string asm,
6370 SDPatternOperator OpNode = null_frag> {
6371 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6372 def v1i64 : BaseSIMDTwoScalar<U, 0b11, 0b00, opc, FPR64, FPR64, asm,
6373 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
6374 def v1i32 : BaseSIMDTwoScalar<U, 0b10, 0b00, opc, FPR32, FPR32, asm,
6375 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
6376 def v1i16 : BaseSIMDTwoScalar<U, 0b01, 0b00, opc, FPR16, FPR16, asm, []>;
6377 def v1i8 : BaseSIMDTwoScalar<U, 0b00, 0b00, opc, FPR8 , FPR8 , asm, []>;
6380 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn))),
6381 (!cast<Instruction>(NAME # v1i64) FPR64:$Rn)>;
6384 multiclass SIMDTwoScalarBHSDTied<bit U, bits<5> opc, string asm,
6386 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6387 def v1i64 : BaseSIMDTwoScalarTied<U, 0b11, opc, FPR64, FPR64, asm,
6388 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn)))]>;
6389 def v1i32 : BaseSIMDTwoScalarTied<U, 0b10, opc, FPR32, FPR32, asm,
6390 [(set (i32 FPR32:$dst), (OpNode (i32 FPR32:$Rd), (i32 FPR32:$Rn)))]>;
6391 def v1i16 : BaseSIMDTwoScalarTied<U, 0b01, opc, FPR16, FPR16, asm, []>;
6392 def v1i8 : BaseSIMDTwoScalarTied<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
6395 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn))),
6396 (!cast<Instruction>(NAME # v1i64) FPR64:$Rd, FPR64:$Rn)>;
6401 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6402 multiclass SIMDTwoScalarMixedBHS<bit U, bits<5> opc, string asm,
6403 SDPatternOperator OpNode = null_frag> {
6404 def v1i32 : BaseSIMDTwoScalar<U, 0b10, 0b00, opc, FPR32, FPR64, asm,
6405 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
6406 def v1i16 : BaseSIMDTwoScalar<U, 0b01, 0b00, opc, FPR16, FPR32, asm, []>;
6407 def v1i8 : BaseSIMDTwoScalar<U, 0b00, 0b00, opc, FPR8 , FPR16, asm, []>;
6410 //----------------------------------------------------------------------------
6411 // AdvSIMD scalar pairwise instructions
6412 //----------------------------------------------------------------------------
6414 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6415 class BaseSIMDPairwiseScalar<bit U, bits<2> size, bits<5> opcode,
6416 RegisterOperand regtype, RegisterOperand vectype,
6417 string asm, string kind>
6418 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
6419 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", []>,
6423 let Inst{31-30} = 0b01;
6425 let Inst{28-24} = 0b11110;
6426 let Inst{23-22} = size;
6427 let Inst{21-17} = 0b11000;
6428 let Inst{16-12} = opcode;
6429 let Inst{11-10} = 0b10;
6434 multiclass SIMDPairwiseScalarD<bit U, bits<5> opc, string asm> {
6435 def v2i64p : BaseSIMDPairwiseScalar<U, 0b11, opc, FPR64Op, V128,
6439 multiclass SIMDFPPairwiseScalar<bit S, bits<5> opc, string asm> {
6440 let Predicates = [HasNEON, HasFullFP16] in {
6441 def v2i16p : BaseSIMDPairwiseScalar<0, {S,0}, opc, FPR16Op, V64,
6444 def v2i32p : BaseSIMDPairwiseScalar<1, {S,0}, opc, FPR32Op, V64,
6446 def v2i64p : BaseSIMDPairwiseScalar<1, {S,1}, opc, FPR64Op, V128,
6450 //----------------------------------------------------------------------------
6451 // AdvSIMD across lanes instructions
6452 //----------------------------------------------------------------------------
6454 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6455 class BaseSIMDAcrossLanes<bit Q, bit U, bits<2> size, bits<5> opcode,
6456 RegisterClass regtype, RegisterOperand vectype,
6457 string asm, string kind, list<dag> pattern>
6458 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
6459 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", pattern>,
6466 let Inst{28-24} = 0b01110;
6467 let Inst{23-22} = size;
6468 let Inst{21-17} = 0b11000;
6469 let Inst{16-12} = opcode;
6470 let Inst{11-10} = 0b10;
6475 multiclass SIMDAcrossLanesBHS<bit U, bits<5> opcode,
6477 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR8, V64,
6479 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR8, V128,
6481 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR16, V64,
6483 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR16, V128,
6485 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR32, V128,
6489 multiclass SIMDAcrossLanesHSD<bit U, bits<5> opcode, string asm> {
6490 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR16, V64,
6492 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR16, V128,
6494 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR32, V64,
6496 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR32, V128,
6498 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR64, V128,
6502 multiclass SIMDFPAcrossLanes<bits<5> opcode, bit sz1, string asm,
6504 let Predicates = [HasNEON, HasFullFP16] in {
6505 def v4i16v : BaseSIMDAcrossLanes<0, 0, {sz1, 0}, opcode, FPR16, V64,
6507 [(set FPR16:$Rd, (intOp (v4f16 V64:$Rn)))]>;
6508 def v8i16v : BaseSIMDAcrossLanes<1, 0, {sz1, 0}, opcode, FPR16, V128,
6510 [(set FPR16:$Rd, (intOp (v8f16 V128:$Rn)))]>;
6511 } // Predicates = [HasNEON, HasFullFP16]
6512 def v4i32v : BaseSIMDAcrossLanes<1, 1, {sz1, 0}, opcode, FPR32, V128,
6514 [(set FPR32:$Rd, (intOp (v4f32 V128:$Rn)))]>;
6517 //----------------------------------------------------------------------------
6518 // AdvSIMD INS/DUP instructions
6519 //----------------------------------------------------------------------------
6521 // FIXME: There has got to be a better way to factor these. ugh.
6523 class BaseSIMDInsDup<bit Q, bit op, dag outs, dag ins, string asm,
6524 string operands, string constraints, list<dag> pattern>
6525 : I<outs, ins, asm, operands, constraints, pattern>,
6532 let Inst{28-21} = 0b01110000;
6539 class SIMDDupFromMain<bit Q, bits<5> imm5, string size, ValueType vectype,
6540 RegisterOperand vecreg, RegisterClass regtype>
6541 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins regtype:$Rn), "dup",
6542 "{\t$Rd" # size # ", $Rn" #
6543 "|" # size # "\t$Rd, $Rn}", "",
6544 [(set (vectype vecreg:$Rd), (AArch64dup regtype:$Rn))]> {
6545 let Inst{20-16} = imm5;
6546 let Inst{14-11} = 0b0001;
6549 class SIMDDupFromElement<bit Q, string dstkind, string srckind,
6550 ValueType vectype, ValueType insreg,
6551 RegisterOperand vecreg, Operand idxtype,
6552 ValueType elttype, SDNode OpNode>
6553 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins V128:$Rn, idxtype:$idx), "dup",
6554 "{\t$Rd" # dstkind # ", $Rn" # srckind # "$idx" #
6555 "|" # dstkind # "\t$Rd, $Rn$idx}", "",
6556 [(set (vectype vecreg:$Rd),
6557 (OpNode (insreg V128:$Rn), idxtype:$idx))]> {
6558 let Inst{14-11} = 0b0000;
6561 class SIMDDup64FromElement
6562 : SIMDDupFromElement<1, ".2d", ".d", v2i64, v2i64, V128,
6563 VectorIndexD, i64, AArch64duplane64> {
6566 let Inst{19-16} = 0b1000;
6569 class SIMDDup32FromElement<bit Q, string size, ValueType vectype,
6570 RegisterOperand vecreg>
6571 : SIMDDupFromElement<Q, size, ".s", vectype, v4i32, vecreg,
6572 VectorIndexS, i64, AArch64duplane32> {
6574 let Inst{20-19} = idx;
6575 let Inst{18-16} = 0b100;
6578 class SIMDDup16FromElement<bit Q, string size, ValueType vectype,
6579 RegisterOperand vecreg>
6580 : SIMDDupFromElement<Q, size, ".h", vectype, v8i16, vecreg,
6581 VectorIndexH, i64, AArch64duplane16> {
6583 let Inst{20-18} = idx;
6584 let Inst{17-16} = 0b10;
6587 class SIMDDup8FromElement<bit Q, string size, ValueType vectype,
6588 RegisterOperand vecreg>
6589 : SIMDDupFromElement<Q, size, ".b", vectype, v16i8, vecreg,
6590 VectorIndexB, i64, AArch64duplane8> {
6592 let Inst{20-17} = idx;
6596 class BaseSIMDMov<bit Q, string size, bits<4> imm4, RegisterClass regtype,
6597 Operand idxtype, string asm, list<dag> pattern>
6598 : BaseSIMDInsDup<Q, 0, (outs regtype:$Rd), (ins V128:$Rn, idxtype:$idx), asm,
6599 "{\t$Rd, $Rn" # size # "$idx" #
6600 "|" # size # "\t$Rd, $Rn$idx}", "", pattern> {
6601 let Inst{14-11} = imm4;
6604 class SIMDSMov<bit Q, string size, RegisterClass regtype,
6606 : BaseSIMDMov<Q, size, 0b0101, regtype, idxtype, "smov", []>;
6607 class SIMDUMov<bit Q, string size, ValueType vectype, RegisterClass regtype,
6609 : BaseSIMDMov<Q, size, 0b0111, regtype, idxtype, "umov",
6610 [(set regtype:$Rd, (vector_extract (vectype V128:$Rn), idxtype:$idx))]>;
6612 class SIMDMovAlias<string asm, string size, Instruction inst,
6613 RegisterClass regtype, Operand idxtype>
6614 : InstAlias<asm#"{\t$dst, $src"#size#"$idx" #
6615 "|" # size # "\t$dst, $src$idx}",
6616 (inst regtype:$dst, V128:$src, idxtype:$idx)>;
6619 def vi8to32 : SIMDSMov<0, ".b", GPR32, VectorIndexB> {
6621 let Inst{20-17} = idx;
6624 def vi8to64 : SIMDSMov<1, ".b", GPR64, VectorIndexB> {
6626 let Inst{20-17} = idx;
6629 def vi16to32 : SIMDSMov<0, ".h", GPR32, VectorIndexH> {
6631 let Inst{20-18} = idx;
6632 let Inst{17-16} = 0b10;
6634 def vi16to64 : SIMDSMov<1, ".h", GPR64, VectorIndexH> {
6636 let Inst{20-18} = idx;
6637 let Inst{17-16} = 0b10;
6639 def vi32to64 : SIMDSMov<1, ".s", GPR64, VectorIndexS> {
6641 let Inst{20-19} = idx;
6642 let Inst{18-16} = 0b100;
6647 def vi8 : SIMDUMov<0, ".b", v16i8, GPR32, VectorIndexB> {
6649 let Inst{20-17} = idx;
6652 def vi16 : SIMDUMov<0, ".h", v8i16, GPR32, VectorIndexH> {
6654 let Inst{20-18} = idx;
6655 let Inst{17-16} = 0b10;
6657 def vi32 : SIMDUMov<0, ".s", v4i32, GPR32, VectorIndexS> {
6659 let Inst{20-19} = idx;
6660 let Inst{18-16} = 0b100;
6662 def vi64 : SIMDUMov<1, ".d", v2i64, GPR64, VectorIndexD> {
6665 let Inst{19-16} = 0b1000;
6667 def : SIMDMovAlias<"mov", ".s",
6668 !cast<Instruction>(NAME#"vi32"),
6669 GPR32, VectorIndexS>;
6670 def : SIMDMovAlias<"mov", ".d",
6671 !cast<Instruction>(NAME#"vi64"),
6672 GPR64, VectorIndexD>;
6675 class SIMDInsFromMain<string size, ValueType vectype,
6676 RegisterClass regtype, Operand idxtype>
6677 : BaseSIMDInsDup<1, 0, (outs V128:$dst),
6678 (ins V128:$Rd, idxtype:$idx, regtype:$Rn), "ins",
6679 "{\t$Rd" # size # "$idx, $Rn" #
6680 "|" # size # "\t$Rd$idx, $Rn}",
6683 (vector_insert (vectype V128:$Rd), regtype:$Rn, idxtype:$idx))]> {
6684 let Inst{14-11} = 0b0011;
6687 class SIMDInsFromElement<string size, ValueType vectype,
6688 ValueType elttype, Operand idxtype>
6689 : BaseSIMDInsDup<1, 1, (outs V128:$dst),
6690 (ins V128:$Rd, idxtype:$idx, V128:$Rn, idxtype:$idx2), "ins",
6691 "{\t$Rd" # size # "$idx, $Rn" # size # "$idx2" #
6692 "|" # size # "\t$Rd$idx, $Rn$idx2}",
6697 (elttype (vector_extract (vectype V128:$Rn), idxtype:$idx2)),
6700 class SIMDInsMainMovAlias<string size, Instruction inst,
6701 RegisterClass regtype, Operand idxtype>
6702 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" #
6703 "|" # size #"\t$dst$idx, $src}",
6704 (inst V128:$dst, idxtype:$idx, regtype:$src)>;
6705 class SIMDInsElementMovAlias<string size, Instruction inst,
6707 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" # size # "$idx2" #
6708 # "|" # size #"\t$dst$idx, $src$idx2}",
6709 (inst V128:$dst, idxtype:$idx, V128:$src, idxtype:$idx2)>;
6712 multiclass SIMDIns {
6713 def vi8gpr : SIMDInsFromMain<".b", v16i8, GPR32, VectorIndexB> {
6715 let Inst{20-17} = idx;
6718 def vi16gpr : SIMDInsFromMain<".h", v8i16, GPR32, VectorIndexH> {
6720 let Inst{20-18} = idx;
6721 let Inst{17-16} = 0b10;
6723 def vi32gpr : SIMDInsFromMain<".s", v4i32, GPR32, VectorIndexS> {
6725 let Inst{20-19} = idx;
6726 let Inst{18-16} = 0b100;
6728 def vi64gpr : SIMDInsFromMain<".d", v2i64, GPR64, VectorIndexD> {
6731 let Inst{19-16} = 0b1000;
6734 def vi8lane : SIMDInsFromElement<".b", v16i8, i32, VectorIndexB> {
6737 let Inst{20-17} = idx;
6739 let Inst{14-11} = idx2;
6741 def vi16lane : SIMDInsFromElement<".h", v8i16, i32, VectorIndexH> {
6744 let Inst{20-18} = idx;
6745 let Inst{17-16} = 0b10;
6746 let Inst{14-12} = idx2;
6749 def vi32lane : SIMDInsFromElement<".s", v4i32, i32, VectorIndexS> {
6752 let Inst{20-19} = idx;
6753 let Inst{18-16} = 0b100;
6754 let Inst{14-13} = idx2;
6755 let Inst{12-11} = {?,?};
6757 def vi64lane : SIMDInsFromElement<".d", v2i64, i64, VectorIndexD> {
6761 let Inst{19-16} = 0b1000;
6762 let Inst{14} = idx2;
6763 let Inst{13-11} = {?,?,?};
6766 // For all forms of the INS instruction, the "mov" mnemonic is the
6767 // preferred alias. Why they didn't just call the instruction "mov" in
6768 // the first place is a very good question indeed...
6769 def : SIMDInsMainMovAlias<".b", !cast<Instruction>(NAME#"vi8gpr"),
6770 GPR32, VectorIndexB>;
6771 def : SIMDInsMainMovAlias<".h", !cast<Instruction>(NAME#"vi16gpr"),
6772 GPR32, VectorIndexH>;
6773 def : SIMDInsMainMovAlias<".s", !cast<Instruction>(NAME#"vi32gpr"),
6774 GPR32, VectorIndexS>;
6775 def : SIMDInsMainMovAlias<".d", !cast<Instruction>(NAME#"vi64gpr"),
6776 GPR64, VectorIndexD>;
6778 def : SIMDInsElementMovAlias<".b", !cast<Instruction>(NAME#"vi8lane"),
6780 def : SIMDInsElementMovAlias<".h", !cast<Instruction>(NAME#"vi16lane"),
6782 def : SIMDInsElementMovAlias<".s", !cast<Instruction>(NAME#"vi32lane"),
6784 def : SIMDInsElementMovAlias<".d", !cast<Instruction>(NAME#"vi64lane"),
6788 //----------------------------------------------------------------------------
6790 //----------------------------------------------------------------------------
6792 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6793 class BaseSIMDTableLookup<bit Q, bits<2> len, bit op, RegisterOperand vectype,
6794 RegisterOperand listtype, string asm, string kind>
6795 : I<(outs vectype:$Vd), (ins listtype:$Vn, vectype:$Vm), asm,
6796 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "", []>,
6803 let Inst{29-21} = 0b001110000;
6804 let Inst{20-16} = Vm;
6806 let Inst{14-13} = len;
6808 let Inst{11-10} = 0b00;
6813 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6814 class BaseSIMDTableLookupTied<bit Q, bits<2> len, bit op, RegisterOperand vectype,
6815 RegisterOperand listtype, string asm, string kind>
6816 : I<(outs vectype:$dst), (ins vectype:$Vd, listtype:$Vn, vectype:$Vm), asm,
6817 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "$Vd = $dst", []>,
6824 let Inst{29-21} = 0b001110000;
6825 let Inst{20-16} = Vm;
6827 let Inst{14-13} = len;
6829 let Inst{11-10} = 0b00;
6834 class SIMDTableLookupAlias<string asm, Instruction inst,
6835 RegisterOperand vectype, RegisterOperand listtype>
6836 : InstAlias<!strconcat(asm, "\t$dst, $lst, $index"),
6837 (inst vectype:$dst, listtype:$lst, vectype:$index), 0>;
6839 multiclass SIMDTableLookup<bit op, string asm> {
6840 def v8i8One : BaseSIMDTableLookup<0, 0b00, op, V64, VecListOne16b,
6842 def v8i8Two : BaseSIMDTableLookup<0, 0b01, op, V64, VecListTwo16b,
6844 def v8i8Three : BaseSIMDTableLookup<0, 0b10, op, V64, VecListThree16b,
6846 def v8i8Four : BaseSIMDTableLookup<0, 0b11, op, V64, VecListFour16b,
6848 def v16i8One : BaseSIMDTableLookup<1, 0b00, op, V128, VecListOne16b,
6850 def v16i8Two : BaseSIMDTableLookup<1, 0b01, op, V128, VecListTwo16b,
6852 def v16i8Three: BaseSIMDTableLookup<1, 0b10, op, V128, VecListThree16b,
6854 def v16i8Four : BaseSIMDTableLookup<1, 0b11, op, V128, VecListFour16b,
6857 def : SIMDTableLookupAlias<asm # ".8b",
6858 !cast<Instruction>(NAME#"v8i8One"),
6859 V64, VecListOne128>;
6860 def : SIMDTableLookupAlias<asm # ".8b",
6861 !cast<Instruction>(NAME#"v8i8Two"),
6862 V64, VecListTwo128>;
6863 def : SIMDTableLookupAlias<asm # ".8b",
6864 !cast<Instruction>(NAME#"v8i8Three"),
6865 V64, VecListThree128>;
6866 def : SIMDTableLookupAlias<asm # ".8b",
6867 !cast<Instruction>(NAME#"v8i8Four"),
6868 V64, VecListFour128>;
6869 def : SIMDTableLookupAlias<asm # ".16b",
6870 !cast<Instruction>(NAME#"v16i8One"),
6871 V128, VecListOne128>;
6872 def : SIMDTableLookupAlias<asm # ".16b",
6873 !cast<Instruction>(NAME#"v16i8Two"),
6874 V128, VecListTwo128>;
6875 def : SIMDTableLookupAlias<asm # ".16b",
6876 !cast<Instruction>(NAME#"v16i8Three"),
6877 V128, VecListThree128>;
6878 def : SIMDTableLookupAlias<asm # ".16b",
6879 !cast<Instruction>(NAME#"v16i8Four"),
6880 V128, VecListFour128>;
6883 multiclass SIMDTableLookupTied<bit op, string asm> {
6884 def v8i8One : BaseSIMDTableLookupTied<0, 0b00, op, V64, VecListOne16b,
6886 def v8i8Two : BaseSIMDTableLookupTied<0, 0b01, op, V64, VecListTwo16b,
6888 def v8i8Three : BaseSIMDTableLookupTied<0, 0b10, op, V64, VecListThree16b,
6890 def v8i8Four : BaseSIMDTableLookupTied<0, 0b11, op, V64, VecListFour16b,
6892 def v16i8One : BaseSIMDTableLookupTied<1, 0b00, op, V128, VecListOne16b,
6894 def v16i8Two : BaseSIMDTableLookupTied<1, 0b01, op, V128, VecListTwo16b,
6896 def v16i8Three: BaseSIMDTableLookupTied<1, 0b10, op, V128, VecListThree16b,
6898 def v16i8Four : BaseSIMDTableLookupTied<1, 0b11, op, V128, VecListFour16b,
6901 def : SIMDTableLookupAlias<asm # ".8b",
6902 !cast<Instruction>(NAME#"v8i8One"),
6903 V64, VecListOne128>;
6904 def : SIMDTableLookupAlias<asm # ".8b",
6905 !cast<Instruction>(NAME#"v8i8Two"),
6906 V64, VecListTwo128>;
6907 def : SIMDTableLookupAlias<asm # ".8b",
6908 !cast<Instruction>(NAME#"v8i8Three"),
6909 V64, VecListThree128>;
6910 def : SIMDTableLookupAlias<asm # ".8b",
6911 !cast<Instruction>(NAME#"v8i8Four"),
6912 V64, VecListFour128>;
6913 def : SIMDTableLookupAlias<asm # ".16b",
6914 !cast<Instruction>(NAME#"v16i8One"),
6915 V128, VecListOne128>;
6916 def : SIMDTableLookupAlias<asm # ".16b",
6917 !cast<Instruction>(NAME#"v16i8Two"),
6918 V128, VecListTwo128>;
6919 def : SIMDTableLookupAlias<asm # ".16b",
6920 !cast<Instruction>(NAME#"v16i8Three"),
6921 V128, VecListThree128>;
6922 def : SIMDTableLookupAlias<asm # ".16b",
6923 !cast<Instruction>(NAME#"v16i8Four"),
6924 V128, VecListFour128>;
6928 //----------------------------------------------------------------------------
6929 // AdvSIMD scalar CPY
6930 //----------------------------------------------------------------------------
6931 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6932 class BaseSIMDScalarCPY<RegisterClass regtype, RegisterOperand vectype,
6933 string kind, Operand idxtype>
6934 : I<(outs regtype:$dst), (ins vectype:$src, idxtype:$idx), "mov",
6935 "{\t$dst, $src" # kind # "$idx" #
6936 "|\t$dst, $src$idx}", "", []>,
6940 let Inst{31-21} = 0b01011110000;
6941 let Inst{15-10} = 0b000001;
6942 let Inst{9-5} = src;
6943 let Inst{4-0} = dst;
6946 class SIMDScalarCPYAlias<string asm, string size, Instruction inst,
6947 RegisterClass regtype, RegisterOperand vectype, Operand idxtype>
6948 : InstAlias<asm # "{\t$dst, $src" # size # "$index" #
6949 # "|\t$dst, $src$index}",
6950 (inst regtype:$dst, vectype:$src, idxtype:$index), 0>;
6953 multiclass SIMDScalarCPY<string asm> {
6954 def i8 : BaseSIMDScalarCPY<FPR8, V128, ".b", VectorIndexB> {
6956 let Inst{20-17} = idx;
6959 def i16 : BaseSIMDScalarCPY<FPR16, V128, ".h", VectorIndexH> {
6961 let Inst{20-18} = idx;
6962 let Inst{17-16} = 0b10;
6964 def i32 : BaseSIMDScalarCPY<FPR32, V128, ".s", VectorIndexS> {
6966 let Inst{20-19} = idx;
6967 let Inst{18-16} = 0b100;
6969 def i64 : BaseSIMDScalarCPY<FPR64, V128, ".d", VectorIndexD> {
6972 let Inst{19-16} = 0b1000;
6975 def : Pat<(v1i64 (scalar_to_vector (i64 (vector_extract (v2i64 V128:$src),
6976 VectorIndexD:$idx)))),
6977 (!cast<Instruction>(NAME # i64) V128:$src, VectorIndexD:$idx)>;
6979 // 'DUP' mnemonic aliases.
6980 def : SIMDScalarCPYAlias<"dup", ".b",
6981 !cast<Instruction>(NAME#"i8"),
6982 FPR8, V128, VectorIndexB>;
6983 def : SIMDScalarCPYAlias<"dup", ".h",
6984 !cast<Instruction>(NAME#"i16"),
6985 FPR16, V128, VectorIndexH>;
6986 def : SIMDScalarCPYAlias<"dup", ".s",
6987 !cast<Instruction>(NAME#"i32"),
6988 FPR32, V128, VectorIndexS>;
6989 def : SIMDScalarCPYAlias<"dup", ".d",
6990 !cast<Instruction>(NAME#"i64"),
6991 FPR64, V128, VectorIndexD>;
6994 //----------------------------------------------------------------------------
6995 // AdvSIMD modified immediate instructions
6996 //----------------------------------------------------------------------------
6998 class BaseSIMDModifiedImm<bit Q, bit op, bit op2, dag oops, dag iops,
6999 string asm, string op_string,
7000 string cstr, list<dag> pattern>
7001 : I<oops, iops, asm, op_string, cstr, pattern>,
7008 let Inst{28-19} = 0b0111100000;
7009 let Inst{18-16} = imm8{7-5};
7012 let Inst{9-5} = imm8{4-0};
7016 class BaseSIMDModifiedImmVector<bit Q, bit op, bit op2, RegisterOperand vectype,
7017 Operand immtype, dag opt_shift_iop,
7018 string opt_shift, string asm, string kind,
7020 : BaseSIMDModifiedImm<Q, op, op2, (outs vectype:$Rd),
7021 !con((ins immtype:$imm8), opt_shift_iop), asm,
7022 "{\t$Rd" # kind # ", $imm8" # opt_shift #
7023 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
7025 let DecoderMethod = "DecodeModImmInstruction";
7028 class BaseSIMDModifiedImmVectorTied<bit Q, bit op, RegisterOperand vectype,
7029 Operand immtype, dag opt_shift_iop,
7030 string opt_shift, string asm, string kind,
7032 : BaseSIMDModifiedImm<Q, op, 0, (outs vectype:$dst),
7033 !con((ins vectype:$Rd, immtype:$imm8), opt_shift_iop),
7034 asm, "{\t$Rd" # kind # ", $imm8" # opt_shift #
7035 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
7036 "$Rd = $dst", pattern> {
7037 let DecoderMethod = "DecodeModImmTiedInstruction";
7040 class BaseSIMDModifiedImmVectorShift<bit Q, bit op, bits<2> b15_b12,
7041 RegisterOperand vectype, string asm,
7042 string kind, list<dag> pattern>
7043 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7044 (ins logical_vec_shift:$shift),
7045 "$shift", asm, kind, pattern> {
7047 let Inst{15} = b15_b12{1};
7048 let Inst{14-13} = shift;
7049 let Inst{12} = b15_b12{0};
7052 class BaseSIMDModifiedImmVectorShiftTied<bit Q, bit op, bits<2> b15_b12,
7053 RegisterOperand vectype, string asm,
7054 string kind, list<dag> pattern>
7055 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
7056 (ins logical_vec_shift:$shift),
7057 "$shift", asm, kind, pattern> {
7059 let Inst{15} = b15_b12{1};
7060 let Inst{14-13} = shift;
7061 let Inst{12} = b15_b12{0};
7065 class BaseSIMDModifiedImmVectorShiftHalf<bit Q, bit op, bits<2> b15_b12,
7066 RegisterOperand vectype, string asm,
7067 string kind, list<dag> pattern>
7068 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7069 (ins logical_vec_hw_shift:$shift),
7070 "$shift", asm, kind, pattern> {
7072 let Inst{15} = b15_b12{1};
7074 let Inst{13} = shift{0};
7075 let Inst{12} = b15_b12{0};
7078 class BaseSIMDModifiedImmVectorShiftHalfTied<bit Q, bit op, bits<2> b15_b12,
7079 RegisterOperand vectype, string asm,
7080 string kind, list<dag> pattern>
7081 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
7082 (ins logical_vec_hw_shift:$shift),
7083 "$shift", asm, kind, pattern> {
7085 let Inst{15} = b15_b12{1};
7087 let Inst{13} = shift{0};
7088 let Inst{12} = b15_b12{0};
7091 multiclass SIMDModifiedImmVectorShift<bit op, bits<2> hw_cmode, bits<2> w_cmode,
7093 def v4i16 : BaseSIMDModifiedImmVectorShiftHalf<0, op, hw_cmode, V64,
7095 def v8i16 : BaseSIMDModifiedImmVectorShiftHalf<1, op, hw_cmode, V128,
7098 def v2i32 : BaseSIMDModifiedImmVectorShift<0, op, w_cmode, V64,
7100 def v4i32 : BaseSIMDModifiedImmVectorShift<1, op, w_cmode, V128,
7104 multiclass SIMDModifiedImmVectorShiftTied<bit op, bits<2> hw_cmode,
7105 bits<2> w_cmode, string asm,
7107 def v4i16 : BaseSIMDModifiedImmVectorShiftHalfTied<0, op, hw_cmode, V64,
7109 [(set (v4i16 V64:$dst), (OpNode V64:$Rd,
7111 (i32 imm:$shift)))]>;
7112 def v8i16 : BaseSIMDModifiedImmVectorShiftHalfTied<1, op, hw_cmode, V128,
7114 [(set (v8i16 V128:$dst), (OpNode V128:$Rd,
7116 (i32 imm:$shift)))]>;
7118 def v2i32 : BaseSIMDModifiedImmVectorShiftTied<0, op, w_cmode, V64,
7120 [(set (v2i32 V64:$dst), (OpNode V64:$Rd,
7122 (i32 imm:$shift)))]>;
7123 def v4i32 : BaseSIMDModifiedImmVectorShiftTied<1, op, w_cmode, V128,
7125 [(set (v4i32 V128:$dst), (OpNode V128:$Rd,
7127 (i32 imm:$shift)))]>;
7130 class SIMDModifiedImmMoveMSL<bit Q, bit op, bits<4> cmode,
7131 RegisterOperand vectype, string asm,
7132 string kind, list<dag> pattern>
7133 : BaseSIMDModifiedImmVector<Q, op, 0, vectype, imm0_255,
7134 (ins move_vec_shift:$shift),
7135 "$shift", asm, kind, pattern> {
7137 let Inst{15-13} = cmode{3-1};
7138 let Inst{12} = shift;
7141 class SIMDModifiedImmVectorNoShift<bit Q, bit op, bit op2, bits<4> cmode,
7142 RegisterOperand vectype,
7143 Operand imm_type, string asm,
7144 string kind, list<dag> pattern>
7145 : BaseSIMDModifiedImmVector<Q, op, op2, vectype, imm_type, (ins), "",
7146 asm, kind, pattern> {
7147 let Inst{15-12} = cmode;
7150 class SIMDModifiedImmScalarNoShift<bit Q, bit op, bits<4> cmode, string asm,
7152 : BaseSIMDModifiedImm<Q, op, 0, (outs FPR64:$Rd), (ins simdimmtype10:$imm8), asm,
7153 "\t$Rd, $imm8", "", pattern> {
7154 let Inst{15-12} = cmode;
7155 let DecoderMethod = "DecodeModImmInstruction";
7158 //----------------------------------------------------------------------------
7159 // AdvSIMD indexed element
7160 //----------------------------------------------------------------------------
7162 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7163 class BaseSIMDIndexed<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
7164 RegisterOperand dst_reg, RegisterOperand lhs_reg,
7165 RegisterOperand rhs_reg, Operand vec_idx, string asm,
7166 string apple_kind, string dst_kind, string lhs_kind,
7167 string rhs_kind, list<dag> pattern>
7168 : I<(outs dst_reg:$Rd), (ins lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx),
7170 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
7171 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "", pattern>,
7180 let Inst{28} = Scalar;
7181 let Inst{27-24} = 0b1111;
7182 let Inst{23-22} = size;
7183 // Bit 21 must be set by the derived class.
7184 let Inst{20-16} = Rm;
7185 let Inst{15-12} = opc;
7186 // Bit 11 must be set by the derived class.
7192 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7193 class BaseSIMDIndexedTied<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
7194 RegisterOperand dst_reg, RegisterOperand lhs_reg,
7195 RegisterOperand rhs_reg, Operand vec_idx, string asm,
7196 string apple_kind, string dst_kind, string lhs_kind,
7197 string rhs_kind, list<dag> pattern>
7198 : I<(outs dst_reg:$dst),
7199 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx), asm,
7200 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
7201 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "$Rd = $dst", pattern>,
7210 let Inst{28} = Scalar;
7211 let Inst{27-24} = 0b1111;
7212 let Inst{23-22} = size;
7213 // Bit 21 must be set by the derived class.
7214 let Inst{20-16} = Rm;
7215 let Inst{15-12} = opc;
7216 // Bit 11 must be set by the derived class.
7222 // ARMv8.2 Index Dot product instructions
7223 class BaseSIMDThreeSameVectorDotIndex<bit Q, bit U, string asm, string dst_kind,
7224 string lhs_kind, string rhs_kind,
7225 RegisterOperand RegType,
7226 ValueType AccumType, ValueType InputType,
7227 SDPatternOperator OpNode> :
7228 BaseSIMDIndexedTied<Q, U, 0b0, 0b10, 0b1110, RegType, RegType, V128,
7229 VectorIndexS, asm, "", dst_kind, lhs_kind, rhs_kind,
7230 [(set (AccumType RegType:$dst),
7231 (AccumType (OpNode (AccumType RegType:$Rd),
7232 (InputType RegType:$Rn),
7233 (InputType (bitconvert (AccumType
7234 (AArch64duplane32 (v4i32 V128:$Rm),
7235 VectorIndexS:$idx)))))))]> {
7237 let Inst{21} = idx{0}; // L
7238 let Inst{11} = idx{1}; // H
7241 multiclass SIMDThreeSameVectorDotIndex<bit U, string asm,
7242 SDPatternOperator OpNode> {
7243 def v8i8 : BaseSIMDThreeSameVectorDotIndex<0, U, asm, ".2s", ".8b", ".4b", V64,
7244 v2i32, v8i8, OpNode>;
7245 def v16i8 : BaseSIMDThreeSameVectorDotIndex<1, U, asm, ".4s", ".16b", ".4b", V128,
7246 v4i32, v16i8, OpNode>;
7249 multiclass SIMDFPIndexed<bit U, bits<4> opc, string asm,
7250 SDPatternOperator OpNode> {
7251 let Predicates = [HasNEON, HasFullFP16] in {
7252 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b00, opc,
7254 V128_lo, VectorIndexH,
7255 asm, ".4h", ".4h", ".4h", ".h",
7256 [(set (v4f16 V64:$Rd),
7257 (OpNode (v4f16 V64:$Rn),
7258 (v4f16 (AArch64duplane16 (v8f16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7260 let Inst{11} = idx{2};
7261 let Inst{21} = idx{1};
7262 let Inst{20} = idx{0};
7265 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b00, opc,
7267 V128_lo, VectorIndexH,
7268 asm, ".8h", ".8h", ".8h", ".h",
7269 [(set (v8f16 V128:$Rd),
7270 (OpNode (v8f16 V128:$Rn),
7271 (v8f16 (AArch64duplane16 (v8f16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7273 let Inst{11} = idx{2};
7274 let Inst{21} = idx{1};
7275 let Inst{20} = idx{0};
7277 } // Predicates = [HasNEON, HasFullFP16]
7279 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7282 asm, ".2s", ".2s", ".2s", ".s",
7283 [(set (v2f32 V64:$Rd),
7284 (OpNode (v2f32 V64:$Rn),
7285 (v2f32 (AArch64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
7287 let Inst{11} = idx{1};
7288 let Inst{21} = idx{0};
7291 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7294 asm, ".4s", ".4s", ".4s", ".s",
7295 [(set (v4f32 V128:$Rd),
7296 (OpNode (v4f32 V128:$Rn),
7297 (v4f32 (AArch64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
7299 let Inst{11} = idx{1};
7300 let Inst{21} = idx{0};
7303 def v2i64_indexed : BaseSIMDIndexed<1, U, 0, 0b11, opc,
7306 asm, ".2d", ".2d", ".2d", ".d",
7307 [(set (v2f64 V128:$Rd),
7308 (OpNode (v2f64 V128:$Rn),
7309 (v2f64 (AArch64duplane64 (v2f64 V128:$Rm), VectorIndexD:$idx))))]> {
7311 let Inst{11} = idx{0};
7315 let Predicates = [HasNEON, HasFullFP16] in {
7316 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b00, opc,
7317 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7318 asm, ".h", "", "", ".h",
7319 [(set (f16 FPR16Op:$Rd),
7320 (OpNode (f16 FPR16Op:$Rn),
7321 (f16 (vector_extract (v8f16 V128_lo:$Rm),
7322 VectorIndexH:$idx))))]> {
7324 let Inst{11} = idx{2};
7325 let Inst{21} = idx{1};
7326 let Inst{20} = idx{0};
7328 } // Predicates = [HasNEON, HasFullFP16]
7330 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
7331 FPR32Op, FPR32Op, V128, VectorIndexS,
7332 asm, ".s", "", "", ".s",
7333 [(set (f32 FPR32Op:$Rd),
7334 (OpNode (f32 FPR32Op:$Rn),
7335 (f32 (vector_extract (v4f32 V128:$Rm),
7336 VectorIndexS:$idx))))]> {
7338 let Inst{11} = idx{1};
7339 let Inst{21} = idx{0};
7342 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b11, opc,
7343 FPR64Op, FPR64Op, V128, VectorIndexD,
7344 asm, ".d", "", "", ".d",
7345 [(set (f64 FPR64Op:$Rd),
7346 (OpNode (f64 FPR64Op:$Rn),
7347 (f64 (vector_extract (v2f64 V128:$Rm),
7348 VectorIndexD:$idx))))]> {
7350 let Inst{11} = idx{0};
7355 multiclass SIMDFPIndexedTiedPatterns<string INST, SDPatternOperator OpNode> {
7356 // 2 variants for the .2s version: DUPLANE from 128-bit and DUP scalar.
7357 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7358 (AArch64duplane32 (v4f32 V128:$Rm),
7359 VectorIndexS:$idx))),
7360 (!cast<Instruction>(INST # v2i32_indexed)
7361 V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7362 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
7363 (AArch64dup (f32 FPR32Op:$Rm)))),
7364 (!cast<Instruction>(INST # "v2i32_indexed") V64:$Rd, V64:$Rn,
7365 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7368 // 2 variants for the .4s version: DUPLANE from 128-bit and DUP scalar.
7369 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7370 (AArch64duplane32 (v4f32 V128:$Rm),
7371 VectorIndexS:$idx))),
7372 (!cast<Instruction>(INST # "v4i32_indexed")
7373 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7374 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
7375 (AArch64dup (f32 FPR32Op:$Rm)))),
7376 (!cast<Instruction>(INST # "v4i32_indexed") V128:$Rd, V128:$Rn,
7377 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
7379 // 2 variants for the .2d version: DUPLANE from 128-bit and DUP scalar.
7380 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7381 (AArch64duplane64 (v2f64 V128:$Rm),
7382 VectorIndexD:$idx))),
7383 (!cast<Instruction>(INST # "v2i64_indexed")
7384 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
7385 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
7386 (AArch64dup (f64 FPR64Op:$Rm)))),
7387 (!cast<Instruction>(INST # "v2i64_indexed") V128:$Rd, V128:$Rn,
7388 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
7390 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
7391 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7392 (vector_extract (v4f32 V128:$Rm), VectorIndexS:$idx))),
7393 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
7394 V128:$Rm, VectorIndexS:$idx)>;
7395 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
7396 (vector_extract (v2f32 V64:$Rm), VectorIndexS:$idx))),
7397 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
7398 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
7400 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
7401 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
7402 (vector_extract (v2f64 V128:$Rm), VectorIndexD:$idx))),
7403 (!cast<Instruction>(INST # "v1i64_indexed") FPR64:$Rd, FPR64:$Rn,
7404 V128:$Rm, VectorIndexD:$idx)>;
7407 multiclass SIMDFPIndexedTied<bit U, bits<4> opc, string asm> {
7408 let Predicates = [HasNEON, HasFullFP16] in {
7409 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b00, opc, V64, V64,
7410 V128_lo, VectorIndexH,
7411 asm, ".4h", ".4h", ".4h", ".h", []> {
7413 let Inst{11} = idx{2};
7414 let Inst{21} = idx{1};
7415 let Inst{20} = idx{0};
7418 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b00, opc,
7420 V128_lo, VectorIndexH,
7421 asm, ".8h", ".8h", ".8h", ".h", []> {
7423 let Inst{11} = idx{2};
7424 let Inst{21} = idx{1};
7425 let Inst{20} = idx{0};
7427 } // Predicates = [HasNEON, HasFullFP16]
7429 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc, V64, V64,
7431 asm, ".2s", ".2s", ".2s", ".s", []> {
7433 let Inst{11} = idx{1};
7434 let Inst{21} = idx{0};
7437 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
7440 asm, ".4s", ".4s", ".4s", ".s", []> {
7442 let Inst{11} = idx{1};
7443 let Inst{21} = idx{0};
7446 def v2i64_indexed : BaseSIMDIndexedTied<1, U, 0, 0b11, opc,
7449 asm, ".2d", ".2d", ".2d", ".d", []> {
7451 let Inst{11} = idx{0};
7455 let Predicates = [HasNEON, HasFullFP16] in {
7456 def v1i16_indexed : BaseSIMDIndexedTied<1, U, 1, 0b00, opc,
7457 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7458 asm, ".h", "", "", ".h", []> {
7460 let Inst{11} = idx{2};
7461 let Inst{21} = idx{1};
7462 let Inst{20} = idx{0};
7464 } // Predicates = [HasNEON, HasFullFP16]
7466 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
7467 FPR32Op, FPR32Op, V128, VectorIndexS,
7468 asm, ".s", "", "", ".s", []> {
7470 let Inst{11} = idx{1};
7471 let Inst{21} = idx{0};
7474 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b11, opc,
7475 FPR64Op, FPR64Op, V128, VectorIndexD,
7476 asm, ".d", "", "", ".d", []> {
7478 let Inst{11} = idx{0};
7483 multiclass SIMDIndexedHS<bit U, bits<4> opc, string asm,
7484 SDPatternOperator OpNode> {
7485 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc, V64, V64,
7486 V128_lo, VectorIndexH,
7487 asm, ".4h", ".4h", ".4h", ".h",
7488 [(set (v4i16 V64:$Rd),
7489 (OpNode (v4i16 V64:$Rn),
7490 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7492 let Inst{11} = idx{2};
7493 let Inst{21} = idx{1};
7494 let Inst{20} = idx{0};
7497 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7499 V128_lo, VectorIndexH,
7500 asm, ".8h", ".8h", ".8h", ".h",
7501 [(set (v8i16 V128:$Rd),
7502 (OpNode (v8i16 V128:$Rn),
7503 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7505 let Inst{11} = idx{2};
7506 let Inst{21} = idx{1};
7507 let Inst{20} = idx{0};
7510 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7513 asm, ".2s", ".2s", ".2s", ".s",
7514 [(set (v2i32 V64:$Rd),
7515 (OpNode (v2i32 V64:$Rn),
7516 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7518 let Inst{11} = idx{1};
7519 let Inst{21} = idx{0};
7522 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7525 asm, ".4s", ".4s", ".4s", ".s",
7526 [(set (v4i32 V128:$Rd),
7527 (OpNode (v4i32 V128:$Rn),
7528 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7530 let Inst{11} = idx{1};
7531 let Inst{21} = idx{0};
7534 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
7535 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
7536 asm, ".h", "", "", ".h", []> {
7538 let Inst{11} = idx{2};
7539 let Inst{21} = idx{1};
7540 let Inst{20} = idx{0};
7543 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
7544 FPR32Op, FPR32Op, V128, VectorIndexS,
7545 asm, ".s", "", "", ".s",
7546 [(set (i32 FPR32Op:$Rd),
7547 (OpNode FPR32Op:$Rn,
7548 (i32 (vector_extract (v4i32 V128:$Rm),
7549 VectorIndexS:$idx))))]> {
7551 let Inst{11} = idx{1};
7552 let Inst{21} = idx{0};
7556 multiclass SIMDVectorIndexedHS<bit U, bits<4> opc, string asm,
7557 SDPatternOperator OpNode> {
7558 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
7560 V128_lo, VectorIndexH,
7561 asm, ".4h", ".4h", ".4h", ".h",
7562 [(set (v4i16 V64:$Rd),
7563 (OpNode (v4i16 V64:$Rn),
7564 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7566 let Inst{11} = idx{2};
7567 let Inst{21} = idx{1};
7568 let Inst{20} = idx{0};
7571 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7573 V128_lo, VectorIndexH,
7574 asm, ".8h", ".8h", ".8h", ".h",
7575 [(set (v8i16 V128:$Rd),
7576 (OpNode (v8i16 V128:$Rn),
7577 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7579 let Inst{11} = idx{2};
7580 let Inst{21} = idx{1};
7581 let Inst{20} = idx{0};
7584 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7587 asm, ".2s", ".2s", ".2s", ".s",
7588 [(set (v2i32 V64:$Rd),
7589 (OpNode (v2i32 V64:$Rn),
7590 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7592 let Inst{11} = idx{1};
7593 let Inst{21} = idx{0};
7596 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7599 asm, ".4s", ".4s", ".4s", ".s",
7600 [(set (v4i32 V128:$Rd),
7601 (OpNode (v4i32 V128:$Rn),
7602 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7604 let Inst{11} = idx{1};
7605 let Inst{21} = idx{0};
7609 multiclass SIMDVectorIndexedHSTied<bit U, bits<4> opc, string asm,
7610 SDPatternOperator OpNode> {
7611 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc, V64, V64,
7612 V128_lo, VectorIndexH,
7613 asm, ".4h", ".4h", ".4h", ".h",
7614 [(set (v4i16 V64:$dst),
7615 (OpNode (v4i16 V64:$Rd),(v4i16 V64:$Rn),
7616 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7618 let Inst{11} = idx{2};
7619 let Inst{21} = idx{1};
7620 let Inst{20} = idx{0};
7623 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
7625 V128_lo, VectorIndexH,
7626 asm, ".8h", ".8h", ".8h", ".h",
7627 [(set (v8i16 V128:$dst),
7628 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7629 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7631 let Inst{11} = idx{2};
7632 let Inst{21} = idx{1};
7633 let Inst{20} = idx{0};
7636 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
7639 asm, ".2s", ".2s", ".2s", ".s",
7640 [(set (v2i32 V64:$dst),
7641 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7642 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7644 let Inst{11} = idx{1};
7645 let Inst{21} = idx{0};
7648 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
7651 asm, ".4s", ".4s", ".4s", ".s",
7652 [(set (v4i32 V128:$dst),
7653 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7654 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7656 let Inst{11} = idx{1};
7657 let Inst{21} = idx{0};
7661 multiclass SIMDIndexedLongSD<bit U, bits<4> opc, string asm,
7662 SDPatternOperator OpNode> {
7663 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
7665 V128_lo, VectorIndexH,
7666 asm, ".4s", ".4s", ".4h", ".h",
7667 [(set (v4i32 V128:$Rd),
7668 (OpNode (v4i16 V64:$Rn),
7669 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7671 let Inst{11} = idx{2};
7672 let Inst{21} = idx{1};
7673 let Inst{20} = idx{0};
7676 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7678 V128_lo, VectorIndexH,
7679 asm#"2", ".4s", ".4s", ".8h", ".h",
7680 [(set (v4i32 V128:$Rd),
7681 (OpNode (extract_high_v8i16 V128:$Rn),
7682 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
7683 VectorIndexH:$idx))))]> {
7686 let Inst{11} = idx{2};
7687 let Inst{21} = idx{1};
7688 let Inst{20} = idx{0};
7691 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7694 asm, ".2d", ".2d", ".2s", ".s",
7695 [(set (v2i64 V128:$Rd),
7696 (OpNode (v2i32 V64:$Rn),
7697 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7699 let Inst{11} = idx{1};
7700 let Inst{21} = idx{0};
7703 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7706 asm#"2", ".2d", ".2d", ".4s", ".s",
7707 [(set (v2i64 V128:$Rd),
7708 (OpNode (extract_high_v4i32 V128:$Rn),
7709 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
7710 VectorIndexS:$idx))))]> {
7712 let Inst{11} = idx{1};
7713 let Inst{21} = idx{0};
7716 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
7717 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
7718 asm, ".h", "", "", ".h", []> {
7720 let Inst{11} = idx{2};
7721 let Inst{21} = idx{1};
7722 let Inst{20} = idx{0};
7725 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
7726 FPR64Op, FPR32Op, V128, VectorIndexS,
7727 asm, ".s", "", "", ".s", []> {
7729 let Inst{11} = idx{1};
7730 let Inst{21} = idx{0};
7734 multiclass SIMDIndexedLongSQDMLXSDTied<bit U, bits<4> opc, string asm,
7735 SDPatternOperator Accum> {
7736 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
7738 V128_lo, VectorIndexH,
7739 asm, ".4s", ".4s", ".4h", ".h",
7740 [(set (v4i32 V128:$dst),
7741 (Accum (v4i32 V128:$Rd),
7742 (v4i32 (int_aarch64_neon_sqdmull
7744 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
7745 VectorIndexH:$idx))))))]> {
7747 let Inst{11} = idx{2};
7748 let Inst{21} = idx{1};
7749 let Inst{20} = idx{0};
7752 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but an
7753 // intermediate EXTRACT_SUBREG would be untyped.
7754 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
7755 (i32 (vector_extract (v4i32
7756 (int_aarch64_neon_sqdmull (v4i16 V64:$Rn),
7757 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
7758 VectorIndexH:$idx)))),
7761 (!cast<Instruction>(NAME # v4i16_indexed)
7762 (SUBREG_TO_REG (i32 0), FPR32Op:$Rd, ssub), V64:$Rn,
7763 V128_lo:$Rm, VectorIndexH:$idx),
7766 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
7768 V128_lo, VectorIndexH,
7769 asm#"2", ".4s", ".4s", ".8h", ".h",
7770 [(set (v4i32 V128:$dst),
7771 (Accum (v4i32 V128:$Rd),
7772 (v4i32 (int_aarch64_neon_sqdmull
7773 (extract_high_v8i16 V128:$Rn),
7775 (AArch64duplane16 (v8i16 V128_lo:$Rm),
7776 VectorIndexH:$idx))))))]> {
7778 let Inst{11} = idx{2};
7779 let Inst{21} = idx{1};
7780 let Inst{20} = idx{0};
7783 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
7786 asm, ".2d", ".2d", ".2s", ".s",
7787 [(set (v2i64 V128:$dst),
7788 (Accum (v2i64 V128:$Rd),
7789 (v2i64 (int_aarch64_neon_sqdmull
7791 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm),
7792 VectorIndexS:$idx))))))]> {
7794 let Inst{11} = idx{1};
7795 let Inst{21} = idx{0};
7798 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
7801 asm#"2", ".2d", ".2d", ".4s", ".s",
7802 [(set (v2i64 V128:$dst),
7803 (Accum (v2i64 V128:$Rd),
7804 (v2i64 (int_aarch64_neon_sqdmull
7805 (extract_high_v4i32 V128:$Rn),
7807 (AArch64duplane32 (v4i32 V128:$Rm),
7808 VectorIndexS:$idx))))))]> {
7810 let Inst{11} = idx{1};
7811 let Inst{21} = idx{0};
7814 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
7815 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
7816 asm, ".h", "", "", ".h", []> {
7818 let Inst{11} = idx{2};
7819 let Inst{21} = idx{1};
7820 let Inst{20} = idx{0};
7824 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
7825 FPR64Op, FPR32Op, V128, VectorIndexS,
7826 asm, ".s", "", "", ".s",
7827 [(set (i64 FPR64Op:$dst),
7828 (Accum (i64 FPR64Op:$Rd),
7829 (i64 (int_aarch64_neon_sqdmulls_scalar
7831 (i32 (vector_extract (v4i32 V128:$Rm),
7832 VectorIndexS:$idx))))))]> {
7835 let Inst{11} = idx{1};
7836 let Inst{21} = idx{0};
7840 multiclass SIMDVectorIndexedLongSD<bit U, bits<4> opc, string asm,
7841 SDPatternOperator OpNode> {
7842 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
7843 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
7845 V128_lo, VectorIndexH,
7846 asm, ".4s", ".4s", ".4h", ".h",
7847 [(set (v4i32 V128:$Rd),
7848 (OpNode (v4i16 V64:$Rn),
7849 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7851 let Inst{11} = idx{2};
7852 let Inst{21} = idx{1};
7853 let Inst{20} = idx{0};
7856 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
7858 V128_lo, VectorIndexH,
7859 asm#"2", ".4s", ".4s", ".8h", ".h",
7860 [(set (v4i32 V128:$Rd),
7861 (OpNode (extract_high_v8i16 V128:$Rn),
7862 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
7863 VectorIndexH:$idx))))]> {
7866 let Inst{11} = idx{2};
7867 let Inst{21} = idx{1};
7868 let Inst{20} = idx{0};
7871 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
7874 asm, ".2d", ".2d", ".2s", ".s",
7875 [(set (v2i64 V128:$Rd),
7876 (OpNode (v2i32 V64:$Rn),
7877 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7879 let Inst{11} = idx{1};
7880 let Inst{21} = idx{0};
7883 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
7886 asm#"2", ".2d", ".2d", ".4s", ".s",
7887 [(set (v2i64 V128:$Rd),
7888 (OpNode (extract_high_v4i32 V128:$Rn),
7889 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
7890 VectorIndexS:$idx))))]> {
7892 let Inst{11} = idx{1};
7893 let Inst{21} = idx{0};
7898 multiclass SIMDVectorIndexedLongSDTied<bit U, bits<4> opc, string asm,
7899 SDPatternOperator OpNode> {
7900 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
7901 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
7903 V128_lo, VectorIndexH,
7904 asm, ".4s", ".4s", ".4h", ".h",
7905 [(set (v4i32 V128:$dst),
7906 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn),
7907 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
7909 let Inst{11} = idx{2};
7910 let Inst{21} = idx{1};
7911 let Inst{20} = idx{0};
7914 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
7916 V128_lo, VectorIndexH,
7917 asm#"2", ".4s", ".4s", ".8h", ".h",
7918 [(set (v4i32 V128:$dst),
7919 (OpNode (v4i32 V128:$Rd),
7920 (extract_high_v8i16 V128:$Rn),
7921 (extract_high_v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
7922 VectorIndexH:$idx))))]> {
7924 let Inst{11} = idx{2};
7925 let Inst{21} = idx{1};
7926 let Inst{20} = idx{0};
7929 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
7932 asm, ".2d", ".2d", ".2s", ".s",
7933 [(set (v2i64 V128:$dst),
7934 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn),
7935 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
7937 let Inst{11} = idx{1};
7938 let Inst{21} = idx{0};
7941 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
7944 asm#"2", ".2d", ".2d", ".4s", ".s",
7945 [(set (v2i64 V128:$dst),
7946 (OpNode (v2i64 V128:$Rd),
7947 (extract_high_v4i32 V128:$Rn),
7948 (extract_high_v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
7949 VectorIndexS:$idx))))]> {
7951 let Inst{11} = idx{1};
7952 let Inst{21} = idx{0};
7957 //----------------------------------------------------------------------------
7958 // AdvSIMD scalar shift by immediate
7959 //----------------------------------------------------------------------------
7961 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
7962 class BaseSIMDScalarShift<bit U, bits<5> opc, bits<7> fixed_imm,
7963 RegisterClass regtype1, RegisterClass regtype2,
7964 Operand immtype, string asm, list<dag> pattern>
7965 : I<(outs regtype1:$Rd), (ins regtype2:$Rn, immtype:$imm),
7966 asm, "\t$Rd, $Rn, $imm", "", pattern>,
7971 let Inst{31-30} = 0b01;
7973 let Inst{28-23} = 0b111110;
7974 let Inst{22-16} = fixed_imm;
7975 let Inst{15-11} = opc;
7981 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
7982 class BaseSIMDScalarShiftTied<bit U, bits<5> opc, bits<7> fixed_imm,
7983 RegisterClass regtype1, RegisterClass regtype2,
7984 Operand immtype, string asm, list<dag> pattern>
7985 : I<(outs regtype1:$dst), (ins regtype1:$Rd, regtype2:$Rn, immtype:$imm),
7986 asm, "\t$Rd, $Rn, $imm", "$Rd = $dst", pattern>,
7991 let Inst{31-30} = 0b01;
7993 let Inst{28-23} = 0b111110;
7994 let Inst{22-16} = fixed_imm;
7995 let Inst{15-11} = opc;
8002 multiclass SIMDFPScalarRShift<bit U, bits<5> opc, string asm> {
8003 let Predicates = [HasNEON, HasFullFP16] in {
8004 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8005 FPR16, FPR16, vecshiftR16, asm, []> {
8006 let Inst{19-16} = imm{3-0};
8008 } // Predicates = [HasNEON, HasFullFP16]
8009 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8010 FPR32, FPR32, vecshiftR32, asm, []> {
8011 let Inst{20-16} = imm{4-0};
8013 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8014 FPR64, FPR64, vecshiftR64, asm, []> {
8015 let Inst{21-16} = imm{5-0};
8019 multiclass SIMDScalarRShiftD<bit U, bits<5> opc, string asm,
8020 SDPatternOperator OpNode> {
8021 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8022 FPR64, FPR64, vecshiftR64, asm,
8023 [(set (i64 FPR64:$Rd),
8024 (OpNode (i64 FPR64:$Rn), (i32 vecshiftR64:$imm)))]> {
8025 let Inst{21-16} = imm{5-0};
8028 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftR64:$imm))),
8029 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftR64:$imm)>;
8032 multiclass SIMDScalarRShiftDTied<bit U, bits<5> opc, string asm,
8033 SDPatternOperator OpNode = null_frag> {
8034 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
8035 FPR64, FPR64, vecshiftR64, asm,
8036 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn),
8037 (i32 vecshiftR64:$imm)))]> {
8038 let Inst{21-16} = imm{5-0};
8041 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
8042 (i32 vecshiftR64:$imm))),
8043 (!cast<Instruction>(NAME # "d") FPR64:$Rd, FPR64:$Rn,
8047 multiclass SIMDScalarLShiftD<bit U, bits<5> opc, string asm,
8048 SDPatternOperator OpNode> {
8049 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8050 FPR64, FPR64, vecshiftL64, asm,
8051 [(set (v1i64 FPR64:$Rd),
8052 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
8053 let Inst{21-16} = imm{5-0};
8057 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8058 multiclass SIMDScalarLShiftDTied<bit U, bits<5> opc, string asm> {
8059 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
8060 FPR64, FPR64, vecshiftL64, asm, []> {
8061 let Inst{21-16} = imm{5-0};
8065 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8066 multiclass SIMDScalarRShiftBHS<bit U, bits<5> opc, string asm,
8067 SDPatternOperator OpNode = null_frag> {
8068 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8069 FPR8, FPR16, vecshiftR8, asm, []> {
8070 let Inst{18-16} = imm{2-0};
8073 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8074 FPR16, FPR32, vecshiftR16, asm, []> {
8075 let Inst{19-16} = imm{3-0};
8078 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8079 FPR32, FPR64, vecshiftR32, asm,
8080 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn), vecshiftR32:$imm))]> {
8081 let Inst{20-16} = imm{4-0};
8085 multiclass SIMDScalarLShiftBHSD<bit U, bits<5> opc, string asm,
8086 SDPatternOperator OpNode> {
8087 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8088 FPR8, FPR8, vecshiftL8, asm, []> {
8089 let Inst{18-16} = imm{2-0};
8092 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8093 FPR16, FPR16, vecshiftL16, asm, []> {
8094 let Inst{19-16} = imm{3-0};
8097 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8098 FPR32, FPR32, vecshiftL32, asm,
8099 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn), (i32 vecshiftL32:$imm)))]> {
8100 let Inst{20-16} = imm{4-0};
8103 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8104 FPR64, FPR64, vecshiftL64, asm,
8105 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
8106 let Inst{21-16} = imm{5-0};
8109 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm))),
8110 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftL64:$imm)>;
8113 multiclass SIMDScalarRShiftBHSD<bit U, bits<5> opc, string asm> {
8114 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
8115 FPR8, FPR8, vecshiftR8, asm, []> {
8116 let Inst{18-16} = imm{2-0};
8119 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
8120 FPR16, FPR16, vecshiftR16, asm, []> {
8121 let Inst{19-16} = imm{3-0};
8124 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
8125 FPR32, FPR32, vecshiftR32, asm, []> {
8126 let Inst{20-16} = imm{4-0};
8129 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
8130 FPR64, FPR64, vecshiftR64, asm, []> {
8131 let Inst{21-16} = imm{5-0};
8135 //----------------------------------------------------------------------------
8136 // AdvSIMD vector x indexed element
8137 //----------------------------------------------------------------------------
8139 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8140 class BaseSIMDVectorShift<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
8141 RegisterOperand dst_reg, RegisterOperand src_reg,
8143 string asm, string dst_kind, string src_kind,
8145 : I<(outs dst_reg:$Rd), (ins src_reg:$Rn, immtype:$imm),
8146 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
8147 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "", pattern>,
8154 let Inst{28-23} = 0b011110;
8155 let Inst{22-16} = fixed_imm;
8156 let Inst{15-11} = opc;
8162 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
8163 class BaseSIMDVectorShiftTied<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
8164 RegisterOperand vectype1, RegisterOperand vectype2,
8166 string asm, string dst_kind, string src_kind,
8168 : I<(outs vectype1:$dst), (ins vectype1:$Rd, vectype2:$Rn, immtype:$imm),
8169 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
8170 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "$Rd = $dst", pattern>,
8177 let Inst{28-23} = 0b011110;
8178 let Inst{22-16} = fixed_imm;
8179 let Inst{15-11} = opc;
8185 multiclass SIMDVectorRShiftSD<bit U, bits<5> opc, string asm,
8187 let Predicates = [HasNEON, HasFullFP16] in {
8188 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8189 V64, V64, vecshiftR16,
8191 [(set (v4i16 V64:$Rd), (OpNode (v4f16 V64:$Rn), (i32 imm:$imm)))]> {
8193 let Inst{19-16} = imm;
8196 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8197 V128, V128, vecshiftR16,
8199 [(set (v8i16 V128:$Rd), (OpNode (v8f16 V128:$Rn), (i32 imm:$imm)))]> {
8201 let Inst{19-16} = imm;
8203 } // Predicates = [HasNEON, HasFullFP16]
8204 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8205 V64, V64, vecshiftR32,
8207 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (i32 imm:$imm)))]> {
8209 let Inst{20-16} = imm;
8212 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8213 V128, V128, vecshiftR32,
8215 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (i32 imm:$imm)))]> {
8217 let Inst{20-16} = imm;
8220 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8221 V128, V128, vecshiftR64,
8223 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (i32 imm:$imm)))]> {
8225 let Inst{21-16} = imm;
8229 multiclass SIMDVectorRShiftToFP<bit U, bits<5> opc, string asm,
8231 let Predicates = [HasNEON, HasFullFP16] in {
8232 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8233 V64, V64, vecshiftR16,
8235 [(set (v4f16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (i32 imm:$imm)))]> {
8237 let Inst{19-16} = imm;
8240 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8241 V128, V128, vecshiftR16,
8243 [(set (v8f16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (i32 imm:$imm)))]> {
8245 let Inst{19-16} = imm;
8247 } // Predicates = [HasNEON, HasFullFP16]
8249 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8250 V64, V64, vecshiftR32,
8252 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (i32 imm:$imm)))]> {
8254 let Inst{20-16} = imm;
8257 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8258 V128, V128, vecshiftR32,
8260 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (i32 imm:$imm)))]> {
8262 let Inst{20-16} = imm;
8265 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8266 V128, V128, vecshiftR64,
8268 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (i32 imm:$imm)))]> {
8270 let Inst{21-16} = imm;
8274 multiclass SIMDVectorRShiftNarrowBHS<bit U, bits<5> opc, string asm,
8275 SDPatternOperator OpNode> {
8276 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8277 V64, V128, vecshiftR16Narrow,
8279 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))]> {
8281 let Inst{18-16} = imm;
8284 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8285 V128, V128, vecshiftR16Narrow,
8286 asm#"2", ".16b", ".8h", []> {
8288 let Inst{18-16} = imm;
8289 let hasSideEffects = 0;
8292 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8293 V64, V128, vecshiftR32Narrow,
8295 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))]> {
8297 let Inst{19-16} = imm;
8300 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8301 V128, V128, vecshiftR32Narrow,
8302 asm#"2", ".8h", ".4s", []> {
8304 let Inst{19-16} = imm;
8305 let hasSideEffects = 0;
8308 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8309 V64, V128, vecshiftR64Narrow,
8311 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))]> {
8313 let Inst{20-16} = imm;
8316 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8317 V128, V128, vecshiftR64Narrow,
8318 asm#"2", ".4s", ".2d", []> {
8320 let Inst{20-16} = imm;
8321 let hasSideEffects = 0;
8324 // TableGen doesn't like patters w/ INSERT_SUBREG on the instructions
8325 // themselves, so put them here instead.
8327 // Patterns involving what's effectively an insert high and a normal
8328 // intrinsic, represented by CONCAT_VECTORS.
8329 def : Pat<(concat_vectors (v8i8 V64:$Rd),(OpNode (v8i16 V128:$Rn),
8330 vecshiftR16Narrow:$imm)),
8331 (!cast<Instruction>(NAME # "v16i8_shift")
8332 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8333 V128:$Rn, vecshiftR16Narrow:$imm)>;
8334 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn),
8335 vecshiftR32Narrow:$imm)),
8336 (!cast<Instruction>(NAME # "v8i16_shift")
8337 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8338 V128:$Rn, vecshiftR32Narrow:$imm)>;
8339 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn),
8340 vecshiftR64Narrow:$imm)),
8341 (!cast<Instruction>(NAME # "v4i32_shift")
8342 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
8343 V128:$Rn, vecshiftR64Narrow:$imm)>;
8346 multiclass SIMDVectorLShiftBHSD<bit U, bits<5> opc, string asm,
8347 SDPatternOperator OpNode> {
8348 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8349 V64, V64, vecshiftL8,
8351 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
8352 (i32 vecshiftL8:$imm)))]> {
8354 let Inst{18-16} = imm;
8357 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8358 V128, V128, vecshiftL8,
8359 asm, ".16b", ".16b",
8360 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
8361 (i32 vecshiftL8:$imm)))]> {
8363 let Inst{18-16} = imm;
8366 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8367 V64, V64, vecshiftL16,
8369 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
8370 (i32 vecshiftL16:$imm)))]> {
8372 let Inst{19-16} = imm;
8375 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8376 V128, V128, vecshiftL16,
8378 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
8379 (i32 vecshiftL16:$imm)))]> {
8381 let Inst{19-16} = imm;
8384 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8385 V64, V64, vecshiftL32,
8387 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
8388 (i32 vecshiftL32:$imm)))]> {
8390 let Inst{20-16} = imm;
8393 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8394 V128, V128, vecshiftL32,
8396 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
8397 (i32 vecshiftL32:$imm)))]> {
8399 let Inst{20-16} = imm;
8402 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8403 V128, V128, vecshiftL64,
8405 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
8406 (i32 vecshiftL64:$imm)))]> {
8408 let Inst{21-16} = imm;
8412 multiclass SIMDVectorRShiftBHSD<bit U, bits<5> opc, string asm,
8413 SDPatternOperator OpNode> {
8414 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8415 V64, V64, vecshiftR8,
8417 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
8418 (i32 vecshiftR8:$imm)))]> {
8420 let Inst{18-16} = imm;
8423 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8424 V128, V128, vecshiftR8,
8425 asm, ".16b", ".16b",
8426 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
8427 (i32 vecshiftR8:$imm)))]> {
8429 let Inst{18-16} = imm;
8432 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8433 V64, V64, vecshiftR16,
8435 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
8436 (i32 vecshiftR16:$imm)))]> {
8438 let Inst{19-16} = imm;
8441 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8442 V128, V128, vecshiftR16,
8444 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
8445 (i32 vecshiftR16:$imm)))]> {
8447 let Inst{19-16} = imm;
8450 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8451 V64, V64, vecshiftR32,
8453 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
8454 (i32 vecshiftR32:$imm)))]> {
8456 let Inst{20-16} = imm;
8459 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8460 V128, V128, vecshiftR32,
8462 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
8463 (i32 vecshiftR32:$imm)))]> {
8465 let Inst{20-16} = imm;
8468 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
8469 V128, V128, vecshiftR64,
8471 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
8472 (i32 vecshiftR64:$imm)))]> {
8474 let Inst{21-16} = imm;
8478 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8479 multiclass SIMDVectorRShiftBHSDTied<bit U, bits<5> opc, string asm,
8480 SDPatternOperator OpNode = null_frag> {
8481 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
8482 V64, V64, vecshiftR8, asm, ".8b", ".8b",
8483 [(set (v8i8 V64:$dst),
8484 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
8485 (i32 vecshiftR8:$imm)))]> {
8487 let Inst{18-16} = imm;
8490 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8491 V128, V128, vecshiftR8, asm, ".16b", ".16b",
8492 [(set (v16i8 V128:$dst),
8493 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
8494 (i32 vecshiftR8:$imm)))]> {
8496 let Inst{18-16} = imm;
8499 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
8500 V64, V64, vecshiftR16, asm, ".4h", ".4h",
8501 [(set (v4i16 V64:$dst),
8502 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
8503 (i32 vecshiftR16:$imm)))]> {
8505 let Inst{19-16} = imm;
8508 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8509 V128, V128, vecshiftR16, asm, ".8h", ".8h",
8510 [(set (v8i16 V128:$dst),
8511 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
8512 (i32 vecshiftR16:$imm)))]> {
8514 let Inst{19-16} = imm;
8517 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
8518 V64, V64, vecshiftR32, asm, ".2s", ".2s",
8519 [(set (v2i32 V64:$dst),
8520 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
8521 (i32 vecshiftR32:$imm)))]> {
8523 let Inst{20-16} = imm;
8526 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8527 V128, V128, vecshiftR32, asm, ".4s", ".4s",
8528 [(set (v4i32 V128:$dst),
8529 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8530 (i32 vecshiftR32:$imm)))]> {
8532 let Inst{20-16} = imm;
8535 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
8536 V128, V128, vecshiftR64,
8537 asm, ".2d", ".2d", [(set (v2i64 V128:$dst),
8538 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
8539 (i32 vecshiftR64:$imm)))]> {
8541 let Inst{21-16} = imm;
8545 multiclass SIMDVectorLShiftBHSDTied<bit U, bits<5> opc, string asm,
8546 SDPatternOperator OpNode = null_frag> {
8547 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
8548 V64, V64, vecshiftL8,
8550 [(set (v8i8 V64:$dst),
8551 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
8552 (i32 vecshiftL8:$imm)))]> {
8554 let Inst{18-16} = imm;
8557 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
8558 V128, V128, vecshiftL8,
8559 asm, ".16b", ".16b",
8560 [(set (v16i8 V128:$dst),
8561 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
8562 (i32 vecshiftL8:$imm)))]> {
8564 let Inst{18-16} = imm;
8567 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
8568 V64, V64, vecshiftL16,
8570 [(set (v4i16 V64:$dst),
8571 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
8572 (i32 vecshiftL16:$imm)))]> {
8574 let Inst{19-16} = imm;
8577 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
8578 V128, V128, vecshiftL16,
8580 [(set (v8i16 V128:$dst),
8581 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
8582 (i32 vecshiftL16:$imm)))]> {
8584 let Inst{19-16} = imm;
8587 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
8588 V64, V64, vecshiftL32,
8590 [(set (v2i32 V64:$dst),
8591 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
8592 (i32 vecshiftL32:$imm)))]> {
8594 let Inst{20-16} = imm;
8597 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
8598 V128, V128, vecshiftL32,
8600 [(set (v4i32 V128:$dst),
8601 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8602 (i32 vecshiftL32:$imm)))]> {
8604 let Inst{20-16} = imm;
8607 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
8608 V128, V128, vecshiftL64,
8610 [(set (v2i64 V128:$dst),
8611 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
8612 (i32 vecshiftL64:$imm)))]> {
8614 let Inst{21-16} = imm;
8618 multiclass SIMDVectorLShiftLongBHSD<bit U, bits<5> opc, string asm,
8619 SDPatternOperator OpNode> {
8620 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
8621 V128, V64, vecshiftL8, asm, ".8h", ".8b",
8622 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), vecshiftL8:$imm))]> {
8624 let Inst{18-16} = imm;
8627 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
8628 V128, V128, vecshiftL8,
8629 asm#"2", ".8h", ".16b",
8630 [(set (v8i16 V128:$Rd),
8631 (OpNode (extract_high_v16i8 V128:$Rn), vecshiftL8:$imm))]> {
8633 let Inst{18-16} = imm;
8636 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
8637 V128, V64, vecshiftL16, asm, ".4s", ".4h",
8638 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), vecshiftL16:$imm))]> {
8640 let Inst{19-16} = imm;
8643 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
8644 V128, V128, vecshiftL16,
8645 asm#"2", ".4s", ".8h",
8646 [(set (v4i32 V128:$Rd),
8647 (OpNode (extract_high_v8i16 V128:$Rn), vecshiftL16:$imm))]> {
8650 let Inst{19-16} = imm;
8653 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
8654 V128, V64, vecshiftL32, asm, ".2d", ".2s",
8655 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), vecshiftL32:$imm))]> {
8657 let Inst{20-16} = imm;
8660 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
8661 V128, V128, vecshiftL32,
8662 asm#"2", ".2d", ".4s",
8663 [(set (v2i64 V128:$Rd),
8664 (OpNode (extract_high_v4i32 V128:$Rn), vecshiftL32:$imm))]> {
8666 let Inst{20-16} = imm;
8672 // Vector load/store
8674 // SIMD ldX/stX no-index memory references don't allow the optional
8675 // ", #0" constant and handle post-indexing explicitly, so we use
8676 // a more specialized parse method for them. Otherwise, it's the same as
8677 // the general GPR64sp handling.
8679 class BaseSIMDLdSt<bit Q, bit L, bits<4> opcode, bits<2> size,
8680 string asm, dag oops, dag iops, list<dag> pattern>
8681 : I<oops, iops, asm, "\t$Vt, [$Rn]", "", pattern> {
8686 let Inst{29-23} = 0b0011000;
8688 let Inst{21-16} = 0b000000;
8689 let Inst{15-12} = opcode;
8690 let Inst{11-10} = size;
8695 class BaseSIMDLdStPost<bit Q, bit L, bits<4> opcode, bits<2> size,
8696 string asm, dag oops, dag iops>
8697 : I<oops, iops, asm, "\t$Vt, [$Rn], $Xm", "$Rn = $wback", []> {
8703 let Inst{29-23} = 0b0011001;
8706 let Inst{20-16} = Xm;
8707 let Inst{15-12} = opcode;
8708 let Inst{11-10} = size;
8713 // The immediate form of AdvSIMD post-indexed addressing is encoded with
8714 // register post-index addressing from the zero register.
8715 multiclass SIMDLdStAliases<string BaseName, string asm, string layout, string Count,
8716 int Offset, int Size> {
8717 // E.g. "ld1 { v0.8b, v1.8b }, [x1], #16"
8718 // "ld1\t$Vt, [$Rn], #16"
8719 // may get mapped to
8720 // (LD1Twov8b_POST VecListTwo8b:$Vt, GPR64sp:$Rn, XZR)
8721 def : InstAlias<asm # "\t$Vt, [$Rn], #" # Offset,
8722 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
8724 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
8727 // E.g. "ld1.8b { v0, v1 }, [x1], #16"
8728 // "ld1.8b\t$Vt, [$Rn], #16"
8729 // may get mapped to
8730 // (LD1Twov8b_POST VecListTwo64:$Vt, GPR64sp:$Rn, XZR)
8731 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], #" # Offset,
8732 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
8734 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
8737 // E.g. "ld1.8b { v0, v1 }, [x1]"
8738 // "ld1\t$Vt, [$Rn]"
8739 // may get mapped to
8740 // (LD1Twov8b VecListTwo64:$Vt, GPR64sp:$Rn)
8741 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn]",
8742 (!cast<Instruction>(BaseName # Count # "v" # layout)
8743 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
8746 // E.g. "ld1.8b { v0, v1 }, [x1], x2"
8747 // "ld1\t$Vt, [$Rn], $Xm"
8748 // may get mapped to
8749 // (LD1Twov8b_POST VecListTwo64:$Vt, GPR64sp:$Rn, GPR64pi8:$Xm)
8750 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], $Xm",
8751 (!cast<Instruction>(BaseName # Count # "v" # layout # "_POST")
8753 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
8754 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
8757 multiclass BaseSIMDLdN<string BaseName, string Count, string asm, string veclist,
8758 int Offset128, int Offset64, bits<4> opcode> {
8759 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
8760 def v16b: BaseSIMDLdSt<1, 1, opcode, 0b00, asm,
8761 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
8762 (ins GPR64sp:$Rn), []>;
8763 def v8h : BaseSIMDLdSt<1, 1, opcode, 0b01, asm,
8764 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
8765 (ins GPR64sp:$Rn), []>;
8766 def v4s : BaseSIMDLdSt<1, 1, opcode, 0b10, asm,
8767 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
8768 (ins GPR64sp:$Rn), []>;
8769 def v2d : BaseSIMDLdSt<1, 1, opcode, 0b11, asm,
8770 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
8771 (ins GPR64sp:$Rn), []>;
8772 def v8b : BaseSIMDLdSt<0, 1, opcode, 0b00, asm,
8773 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
8774 (ins GPR64sp:$Rn), []>;
8775 def v4h : BaseSIMDLdSt<0, 1, opcode, 0b01, asm,
8776 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
8777 (ins GPR64sp:$Rn), []>;
8778 def v2s : BaseSIMDLdSt<0, 1, opcode, 0b10, asm,
8779 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
8780 (ins GPR64sp:$Rn), []>;
8783 def v16b_POST: BaseSIMDLdStPost<1, 1, opcode, 0b00, asm,
8784 (outs GPR64sp:$wback,
8785 !cast<RegisterOperand>(veclist # "16b"):$Vt),
8787 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8788 def v8h_POST : BaseSIMDLdStPost<1, 1, opcode, 0b01, asm,
8789 (outs GPR64sp:$wback,
8790 !cast<RegisterOperand>(veclist # "8h"):$Vt),
8792 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8793 def v4s_POST : BaseSIMDLdStPost<1, 1, opcode, 0b10, asm,
8794 (outs GPR64sp:$wback,
8795 !cast<RegisterOperand>(veclist # "4s"):$Vt),
8797 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8798 def v2d_POST : BaseSIMDLdStPost<1, 1, opcode, 0b11, asm,
8799 (outs GPR64sp:$wback,
8800 !cast<RegisterOperand>(veclist # "2d"):$Vt),
8802 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8803 def v8b_POST : BaseSIMDLdStPost<0, 1, opcode, 0b00, asm,
8804 (outs GPR64sp:$wback,
8805 !cast<RegisterOperand>(veclist # "8b"):$Vt),
8807 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8808 def v4h_POST : BaseSIMDLdStPost<0, 1, opcode, 0b01, asm,
8809 (outs GPR64sp:$wback,
8810 !cast<RegisterOperand>(veclist # "4h"):$Vt),
8812 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8813 def v2s_POST : BaseSIMDLdStPost<0, 1, opcode, 0b10, asm,
8814 (outs GPR64sp:$wback,
8815 !cast<RegisterOperand>(veclist # "2s"):$Vt),
8817 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8820 defm : SIMDLdStAliases<BaseName, asm, "16b", Count, Offset128, 128>;
8821 defm : SIMDLdStAliases<BaseName, asm, "8h", Count, Offset128, 128>;
8822 defm : SIMDLdStAliases<BaseName, asm, "4s", Count, Offset128, 128>;
8823 defm : SIMDLdStAliases<BaseName, asm, "2d", Count, Offset128, 128>;
8824 defm : SIMDLdStAliases<BaseName, asm, "8b", Count, Offset64, 64>;
8825 defm : SIMDLdStAliases<BaseName, asm, "4h", Count, Offset64, 64>;
8826 defm : SIMDLdStAliases<BaseName, asm, "2s", Count, Offset64, 64>;
8829 // Only ld1/st1 has a v1d version.
8830 multiclass BaseSIMDStN<string BaseName, string Count, string asm, string veclist,
8831 int Offset128, int Offset64, bits<4> opcode> {
8832 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in {
8833 def v16b : BaseSIMDLdSt<1, 0, opcode, 0b00, asm, (outs),
8834 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
8836 def v8h : BaseSIMDLdSt<1, 0, opcode, 0b01, asm, (outs),
8837 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
8839 def v4s : BaseSIMDLdSt<1, 0, opcode, 0b10, asm, (outs),
8840 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
8842 def v2d : BaseSIMDLdSt<1, 0, opcode, 0b11, asm, (outs),
8843 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
8845 def v8b : BaseSIMDLdSt<0, 0, opcode, 0b00, asm, (outs),
8846 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
8848 def v4h : BaseSIMDLdSt<0, 0, opcode, 0b01, asm, (outs),
8849 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
8851 def v2s : BaseSIMDLdSt<0, 0, opcode, 0b10, asm, (outs),
8852 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
8855 def v16b_POST : BaseSIMDLdStPost<1, 0, opcode, 0b00, asm,
8856 (outs GPR64sp:$wback),
8857 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
8859 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8860 def v8h_POST : BaseSIMDLdStPost<1, 0, opcode, 0b01, asm,
8861 (outs GPR64sp:$wback),
8862 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
8864 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8865 def v4s_POST : BaseSIMDLdStPost<1, 0, opcode, 0b10, asm,
8866 (outs GPR64sp:$wback),
8867 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
8869 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8870 def v2d_POST : BaseSIMDLdStPost<1, 0, opcode, 0b11, asm,
8871 (outs GPR64sp:$wback),
8872 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
8874 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
8875 def v8b_POST : BaseSIMDLdStPost<0, 0, opcode, 0b00, asm,
8876 (outs GPR64sp:$wback),
8877 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
8879 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8880 def v4h_POST : BaseSIMDLdStPost<0, 0, opcode, 0b01, asm,
8881 (outs GPR64sp:$wback),
8882 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
8884 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8885 def v2s_POST : BaseSIMDLdStPost<0, 0, opcode, 0b10, asm,
8886 (outs GPR64sp:$wback),
8887 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
8889 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8892 defm : SIMDLdStAliases<BaseName, asm, "16b", Count, Offset128, 128>;
8893 defm : SIMDLdStAliases<BaseName, asm, "8h", Count, Offset128, 128>;
8894 defm : SIMDLdStAliases<BaseName, asm, "4s", Count, Offset128, 128>;
8895 defm : SIMDLdStAliases<BaseName, asm, "2d", Count, Offset128, 128>;
8896 defm : SIMDLdStAliases<BaseName, asm, "8b", Count, Offset64, 64>;
8897 defm : SIMDLdStAliases<BaseName, asm, "4h", Count, Offset64, 64>;
8898 defm : SIMDLdStAliases<BaseName, asm, "2s", Count, Offset64, 64>;
8901 multiclass BaseSIMDLd1<string BaseName, string Count, string asm, string veclist,
8902 int Offset128, int Offset64, bits<4> opcode>
8903 : BaseSIMDLdN<BaseName, Count, asm, veclist, Offset128, Offset64, opcode> {
8905 // LD1 instructions have extra "1d" variants.
8906 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
8907 def v1d : BaseSIMDLdSt<0, 1, opcode, 0b11, asm,
8908 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
8909 (ins GPR64sp:$Rn), []>;
8911 def v1d_POST : BaseSIMDLdStPost<0, 1, opcode, 0b11, asm,
8912 (outs GPR64sp:$wback,
8913 !cast<RegisterOperand>(veclist # "1d"):$Vt),
8915 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8918 defm : SIMDLdStAliases<BaseName, asm, "1d", Count, Offset64, 64>;
8921 multiclass BaseSIMDSt1<string BaseName, string Count, string asm, string veclist,
8922 int Offset128, int Offset64, bits<4> opcode>
8923 : BaseSIMDStN<BaseName, Count, asm, veclist, Offset128, Offset64, opcode> {
8925 // ST1 instructions have extra "1d" variants.
8926 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in {
8927 def v1d : BaseSIMDLdSt<0, 0, opcode, 0b11, asm, (outs),
8928 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
8931 def v1d_POST : BaseSIMDLdStPost<0, 0, opcode, 0b11, asm,
8932 (outs GPR64sp:$wback),
8933 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
8935 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
8938 defm : SIMDLdStAliases<BaseName, asm, "1d", Count, Offset64, 64>;
8941 multiclass SIMDLd1Multiple<string asm> {
8942 defm One : BaseSIMDLd1<NAME, "One", asm, "VecListOne", 16, 8, 0b0111>;
8943 defm Two : BaseSIMDLd1<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1010>;
8944 defm Three : BaseSIMDLd1<NAME, "Three", asm, "VecListThree", 48, 24, 0b0110>;
8945 defm Four : BaseSIMDLd1<NAME, "Four", asm, "VecListFour", 64, 32, 0b0010>;
8948 multiclass SIMDSt1Multiple<string asm> {
8949 defm One : BaseSIMDSt1<NAME, "One", asm, "VecListOne", 16, 8, 0b0111>;
8950 defm Two : BaseSIMDSt1<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1010>;
8951 defm Three : BaseSIMDSt1<NAME, "Three", asm, "VecListThree", 48, 24, 0b0110>;
8952 defm Four : BaseSIMDSt1<NAME, "Four", asm, "VecListFour", 64, 32, 0b0010>;
8955 multiclass SIMDLd2Multiple<string asm> {
8956 defm Two : BaseSIMDLdN<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1000>;
8959 multiclass SIMDSt2Multiple<string asm> {
8960 defm Two : BaseSIMDStN<NAME, "Two", asm, "VecListTwo", 32, 16, 0b1000>;
8963 multiclass SIMDLd3Multiple<string asm> {
8964 defm Three : BaseSIMDLdN<NAME, "Three", asm, "VecListThree", 48, 24, 0b0100>;
8967 multiclass SIMDSt3Multiple<string asm> {
8968 defm Three : BaseSIMDStN<NAME, "Three", asm, "VecListThree", 48, 24, 0b0100>;
8971 multiclass SIMDLd4Multiple<string asm> {
8972 defm Four : BaseSIMDLdN<NAME, "Four", asm, "VecListFour", 64, 32, 0b0000>;
8975 multiclass SIMDSt4Multiple<string asm> {
8976 defm Four : BaseSIMDStN<NAME, "Four", asm, "VecListFour", 64, 32, 0b0000>;
8980 // AdvSIMD Load/store single-element
8983 class BaseSIMDLdStSingle<bit L, bit R, bits<3> opcode,
8984 string asm, string operands, string cst,
8985 dag oops, dag iops, list<dag> pattern>
8986 : I<oops, iops, asm, operands, cst, pattern> {
8990 let Inst{29-24} = 0b001101;
8993 let Inst{15-13} = opcode;
8998 class BaseSIMDLdStSingleTied<bit L, bit R, bits<3> opcode,
8999 string asm, string operands, string cst,
9000 dag oops, dag iops, list<dag> pattern>
9001 : I<oops, iops, asm, operands, "$Vt = $dst," # cst, pattern> {
9005 let Inst{29-24} = 0b001101;
9008 let Inst{15-13} = opcode;
9014 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9015 class BaseSIMDLdR<bit Q, bit R, bits<3> opcode, bit S, bits<2> size, string asm,
9016 DAGOperand listtype>
9017 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, [$Rn]", "",
9018 (outs listtype:$Vt), (ins GPR64sp:$Rn),
9022 let Inst{20-16} = 0b00000;
9024 let Inst{11-10} = size;
9026 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9027 class BaseSIMDLdRPost<bit Q, bit R, bits<3> opcode, bit S, bits<2> size,
9028 string asm, DAGOperand listtype, DAGOperand GPR64pi>
9029 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, [$Rn], $Xm",
9031 (outs GPR64sp:$wback, listtype:$Vt),
9032 (ins GPR64sp:$Rn, GPR64pi:$Xm), []> {
9036 let Inst{20-16} = Xm;
9038 let Inst{11-10} = size;
9041 multiclass SIMDLdrAliases<string BaseName, string asm, string layout, string Count,
9042 int Offset, int Size> {
9043 // E.g. "ld1r { v0.8b }, [x1], #1"
9044 // "ld1r.8b\t$Vt, [$Rn], #1"
9045 // may get mapped to
9046 // (LD1Rv8b_POST VecListOne8b:$Vt, GPR64sp:$Rn, XZR)
9047 def : InstAlias<asm # "\t$Vt, [$Rn], #" # Offset,
9048 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9050 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
9053 // E.g. "ld1r.8b { v0 }, [x1], #1"
9054 // "ld1r.8b\t$Vt, [$Rn], #1"
9055 // may get mapped to
9056 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, XZR)
9057 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], #" # Offset,
9058 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9060 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9063 // E.g. "ld1r.8b { v0 }, [x1]"
9064 // "ld1r.8b\t$Vt, [$Rn]"
9065 // may get mapped to
9066 // (LD1Rv8b VecListOne64:$Vt, GPR64sp:$Rn)
9067 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn]",
9068 (!cast<Instruction>(BaseName # "v" # layout)
9069 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9072 // E.g. "ld1r.8b { v0 }, [x1], x2"
9073 // "ld1r.8b\t$Vt, [$Rn], $Xm"
9074 // may get mapped to
9075 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, GPR64pi1:$Xm)
9076 def : InstAlias<asm # "." # layout # "\t$Vt, [$Rn], $Xm",
9077 (!cast<Instruction>(BaseName # "v" # layout # "_POST")
9079 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
9080 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9083 multiclass SIMDLdR<bit R, bits<3> opcode, bit S, string asm, string Count,
9084 int Offset1, int Offset2, int Offset4, int Offset8> {
9085 def v8b : BaseSIMDLdR<0, R, opcode, S, 0b00, asm,
9086 !cast<DAGOperand>("VecList" # Count # "8b")>;
9087 def v16b: BaseSIMDLdR<1, R, opcode, S, 0b00, asm,
9088 !cast<DAGOperand>("VecList" # Count #"16b")>;
9089 def v4h : BaseSIMDLdR<0, R, opcode, S, 0b01, asm,
9090 !cast<DAGOperand>("VecList" # Count #"4h")>;
9091 def v8h : BaseSIMDLdR<1, R, opcode, S, 0b01, asm,
9092 !cast<DAGOperand>("VecList" # Count #"8h")>;
9093 def v2s : BaseSIMDLdR<0, R, opcode, S, 0b10, asm,
9094 !cast<DAGOperand>("VecList" # Count #"2s")>;
9095 def v4s : BaseSIMDLdR<1, R, opcode, S, 0b10, asm,
9096 !cast<DAGOperand>("VecList" # Count #"4s")>;
9097 def v1d : BaseSIMDLdR<0, R, opcode, S, 0b11, asm,
9098 !cast<DAGOperand>("VecList" # Count #"1d")>;
9099 def v2d : BaseSIMDLdR<1, R, opcode, S, 0b11, asm,
9100 !cast<DAGOperand>("VecList" # Count #"2d")>;
9102 def v8b_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b00, asm,
9103 !cast<DAGOperand>("VecList" # Count # "8b"),
9104 !cast<DAGOperand>("GPR64pi" # Offset1)>;
9105 def v16b_POST: BaseSIMDLdRPost<1, R, opcode, S, 0b00, asm,
9106 !cast<DAGOperand>("VecList" # Count # "16b"),
9107 !cast<DAGOperand>("GPR64pi" # Offset1)>;
9108 def v4h_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b01, asm,
9109 !cast<DAGOperand>("VecList" # Count # "4h"),
9110 !cast<DAGOperand>("GPR64pi" # Offset2)>;
9111 def v8h_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b01, asm,
9112 !cast<DAGOperand>("VecList" # Count # "8h"),
9113 !cast<DAGOperand>("GPR64pi" # Offset2)>;
9114 def v2s_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b10, asm,
9115 !cast<DAGOperand>("VecList" # Count # "2s"),
9116 !cast<DAGOperand>("GPR64pi" # Offset4)>;
9117 def v4s_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b10, asm,
9118 !cast<DAGOperand>("VecList" # Count # "4s"),
9119 !cast<DAGOperand>("GPR64pi" # Offset4)>;
9120 def v1d_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b11, asm,
9121 !cast<DAGOperand>("VecList" # Count # "1d"),
9122 !cast<DAGOperand>("GPR64pi" # Offset8)>;
9123 def v2d_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b11, asm,
9124 !cast<DAGOperand>("VecList" # Count # "2d"),
9125 !cast<DAGOperand>("GPR64pi" # Offset8)>;
9127 defm : SIMDLdrAliases<NAME, asm, "8b", Count, Offset1, 64>;
9128 defm : SIMDLdrAliases<NAME, asm, "16b", Count, Offset1, 128>;
9129 defm : SIMDLdrAliases<NAME, asm, "4h", Count, Offset2, 64>;
9130 defm : SIMDLdrAliases<NAME, asm, "8h", Count, Offset2, 128>;
9131 defm : SIMDLdrAliases<NAME, asm, "2s", Count, Offset4, 64>;
9132 defm : SIMDLdrAliases<NAME, asm, "4s", Count, Offset4, 128>;
9133 defm : SIMDLdrAliases<NAME, asm, "1d", Count, Offset8, 64>;
9134 defm : SIMDLdrAliases<NAME, asm, "2d", Count, Offset8, 128>;
9137 class SIMDLdStSingleB<bit L, bit R, bits<3> opcode, string asm,
9138 dag oops, dag iops, list<dag> pattern>
9139 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9141 // idx encoded in Q:S:size fields.
9143 let Inst{30} = idx{3};
9145 let Inst{20-16} = 0b00000;
9146 let Inst{12} = idx{2};
9147 let Inst{11-10} = idx{1-0};
9149 class SIMDLdStSingleBTied<bit L, bit R, bits<3> opcode, string asm,
9150 dag oops, dag iops, list<dag> pattern>
9151 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9152 oops, iops, pattern> {
9153 // idx encoded in Q:S:size fields.
9155 let Inst{30} = idx{3};
9157 let Inst{20-16} = 0b00000;
9158 let Inst{12} = idx{2};
9159 let Inst{11-10} = idx{1-0};
9161 class SIMDLdStSingleBPost<bit L, bit R, bits<3> opcode, string asm,
9163 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9164 "$Rn = $wback", oops, iops, []> {
9165 // idx encoded in Q:S:size fields.
9168 let Inst{30} = idx{3};
9170 let Inst{20-16} = Xm;
9171 let Inst{12} = idx{2};
9172 let Inst{11-10} = idx{1-0};
9174 class SIMDLdStSingleBTiedPost<bit L, bit R, bits<3> opcode, string asm,
9176 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9177 "$Rn = $wback", oops, iops, []> {
9178 // idx encoded in Q:S:size fields.
9181 let Inst{30} = idx{3};
9183 let Inst{20-16} = Xm;
9184 let Inst{12} = idx{2};
9185 let Inst{11-10} = idx{1-0};
9188 class SIMDLdStSingleH<bit L, bit R, bits<3> opcode, bit size, string asm,
9189 dag oops, dag iops, list<dag> pattern>
9190 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9192 // idx encoded in Q:S:size<1> fields.
9194 let Inst{30} = idx{2};
9196 let Inst{20-16} = 0b00000;
9197 let Inst{12} = idx{1};
9198 let Inst{11} = idx{0};
9199 let Inst{10} = size;
9201 class SIMDLdStSingleHTied<bit L, bit R, bits<3> opcode, bit size, string asm,
9202 dag oops, dag iops, list<dag> pattern>
9203 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9204 oops, iops, pattern> {
9205 // idx encoded in Q:S:size<1> fields.
9207 let Inst{30} = idx{2};
9209 let Inst{20-16} = 0b00000;
9210 let Inst{12} = idx{1};
9211 let Inst{11} = idx{0};
9212 let Inst{10} = size;
9215 class SIMDLdStSingleHPost<bit L, bit R, bits<3> opcode, bit size, string asm,
9217 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9218 "$Rn = $wback", oops, iops, []> {
9219 // idx encoded in Q:S:size<1> fields.
9222 let Inst{30} = idx{2};
9224 let Inst{20-16} = Xm;
9225 let Inst{12} = idx{1};
9226 let Inst{11} = idx{0};
9227 let Inst{10} = size;
9229 class SIMDLdStSingleHTiedPost<bit L, bit R, bits<3> opcode, bit size, string asm,
9231 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9232 "$Rn = $wback", oops, iops, []> {
9233 // idx encoded in Q:S:size<1> fields.
9236 let Inst{30} = idx{2};
9238 let Inst{20-16} = Xm;
9239 let Inst{12} = idx{1};
9240 let Inst{11} = idx{0};
9241 let Inst{10} = size;
9243 class SIMDLdStSingleS<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9244 dag oops, dag iops, list<dag> pattern>
9245 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9247 // idx encoded in Q:S fields.
9249 let Inst{30} = idx{1};
9251 let Inst{20-16} = 0b00000;
9252 let Inst{12} = idx{0};
9253 let Inst{11-10} = size;
9255 class SIMDLdStSingleSTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9256 dag oops, dag iops, list<dag> pattern>
9257 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9258 oops, iops, pattern> {
9259 // idx encoded in Q:S fields.
9261 let Inst{30} = idx{1};
9263 let Inst{20-16} = 0b00000;
9264 let Inst{12} = idx{0};
9265 let Inst{11-10} = size;
9267 class SIMDLdStSingleSPost<bit L, bit R, bits<3> opcode, bits<2> size,
9268 string asm, dag oops, dag iops>
9269 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9270 "$Rn = $wback", oops, iops, []> {
9271 // idx encoded in Q:S fields.
9274 let Inst{30} = idx{1};
9276 let Inst{20-16} = Xm;
9277 let Inst{12} = idx{0};
9278 let Inst{11-10} = size;
9280 class SIMDLdStSingleSTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
9281 string asm, dag oops, dag iops>
9282 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9283 "$Rn = $wback", oops, iops, []> {
9284 // idx encoded in Q:S fields.
9287 let Inst{30} = idx{1};
9289 let Inst{20-16} = Xm;
9290 let Inst{12} = idx{0};
9291 let Inst{11-10} = size;
9293 class SIMDLdStSingleD<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9294 dag oops, dag iops, list<dag> pattern>
9295 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "", oops, iops,
9297 // idx encoded in Q field.
9301 let Inst{20-16} = 0b00000;
9303 let Inst{11-10} = size;
9305 class SIMDLdStSingleDTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
9306 dag oops, dag iops, list<dag> pattern>
9307 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn]", "",
9308 oops, iops, pattern> {
9309 // idx encoded in Q field.
9313 let Inst{20-16} = 0b00000;
9315 let Inst{11-10} = size;
9317 class SIMDLdStSingleDPost<bit L, bit R, bits<3> opcode, bits<2> size,
9318 string asm, dag oops, dag iops>
9319 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9320 "$Rn = $wback", oops, iops, []> {
9321 // idx encoded in Q field.
9326 let Inst{20-16} = Xm;
9328 let Inst{11-10} = size;
9330 class SIMDLdStSingleDTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
9331 string asm, dag oops, dag iops>
9332 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, [$Rn], $Xm",
9333 "$Rn = $wback", oops, iops, []> {
9334 // idx encoded in Q field.
9339 let Inst{20-16} = Xm;
9341 let Inst{11-10} = size;
9344 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9345 multiclass SIMDLdSingleBTied<bit R, bits<3> opcode, string asm,
9346 RegisterOperand listtype,
9347 RegisterOperand GPR64pi> {
9348 def i8 : SIMDLdStSingleBTied<1, R, opcode, asm,
9349 (outs listtype:$dst),
9350 (ins listtype:$Vt, VectorIndexB:$idx,
9353 def i8_POST : SIMDLdStSingleBTiedPost<1, R, opcode, asm,
9354 (outs GPR64sp:$wback, listtype:$dst),
9355 (ins listtype:$Vt, VectorIndexB:$idx,
9356 GPR64sp:$Rn, GPR64pi:$Xm)>;
9358 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9359 multiclass SIMDLdSingleHTied<bit R, bits<3> opcode, bit size, string asm,
9360 RegisterOperand listtype,
9361 RegisterOperand GPR64pi> {
9362 def i16 : SIMDLdStSingleHTied<1, R, opcode, size, asm,
9363 (outs listtype:$dst),
9364 (ins listtype:$Vt, VectorIndexH:$idx,
9367 def i16_POST : SIMDLdStSingleHTiedPost<1, R, opcode, size, asm,
9368 (outs GPR64sp:$wback, listtype:$dst),
9369 (ins listtype:$Vt, VectorIndexH:$idx,
9370 GPR64sp:$Rn, GPR64pi:$Xm)>;
9372 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9373 multiclass SIMDLdSingleSTied<bit R, bits<3> opcode, bits<2> size,string asm,
9374 RegisterOperand listtype,
9375 RegisterOperand GPR64pi> {
9376 def i32 : SIMDLdStSingleSTied<1, R, opcode, size, asm,
9377 (outs listtype:$dst),
9378 (ins listtype:$Vt, VectorIndexS:$idx,
9381 def i32_POST : SIMDLdStSingleSTiedPost<1, R, opcode, size, asm,
9382 (outs GPR64sp:$wback, listtype:$dst),
9383 (ins listtype:$Vt, VectorIndexS:$idx,
9384 GPR64sp:$Rn, GPR64pi:$Xm)>;
9386 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
9387 multiclass SIMDLdSingleDTied<bit R, bits<3> opcode, bits<2> size, string asm,
9388 RegisterOperand listtype, RegisterOperand GPR64pi> {
9389 def i64 : SIMDLdStSingleDTied<1, R, opcode, size, asm,
9390 (outs listtype:$dst),
9391 (ins listtype:$Vt, VectorIndexD:$idx,
9394 def i64_POST : SIMDLdStSingleDTiedPost<1, R, opcode, size, asm,
9395 (outs GPR64sp:$wback, listtype:$dst),
9396 (ins listtype:$Vt, VectorIndexD:$idx,
9397 GPR64sp:$Rn, GPR64pi:$Xm)>;
9399 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9400 multiclass SIMDStSingleB<bit R, bits<3> opcode, string asm,
9401 RegisterOperand listtype, RegisterOperand GPR64pi> {
9402 def i8 : SIMDLdStSingleB<0, R, opcode, asm,
9403 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
9406 def i8_POST : SIMDLdStSingleBPost<0, R, opcode, asm,
9407 (outs GPR64sp:$wback),
9408 (ins listtype:$Vt, VectorIndexB:$idx,
9409 GPR64sp:$Rn, GPR64pi:$Xm)>;
9411 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9412 multiclass SIMDStSingleH<bit R, bits<3> opcode, bit size, string asm,
9413 RegisterOperand listtype, RegisterOperand GPR64pi> {
9414 def i16 : SIMDLdStSingleH<0, R, opcode, size, asm,
9415 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
9418 def i16_POST : SIMDLdStSingleHPost<0, R, opcode, size, asm,
9419 (outs GPR64sp:$wback),
9420 (ins listtype:$Vt, VectorIndexH:$idx,
9421 GPR64sp:$Rn, GPR64pi:$Xm)>;
9423 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9424 multiclass SIMDStSingleS<bit R, bits<3> opcode, bits<2> size,string asm,
9425 RegisterOperand listtype, RegisterOperand GPR64pi> {
9426 def i32 : SIMDLdStSingleS<0, R, opcode, size, asm,
9427 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
9430 def i32_POST : SIMDLdStSingleSPost<0, R, opcode, size, asm,
9431 (outs GPR64sp:$wback),
9432 (ins listtype:$Vt, VectorIndexS:$idx,
9433 GPR64sp:$Rn, GPR64pi:$Xm)>;
9435 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
9436 multiclass SIMDStSingleD<bit R, bits<3> opcode, bits<2> size, string asm,
9437 RegisterOperand listtype, RegisterOperand GPR64pi> {
9438 def i64 : SIMDLdStSingleD<0, R, opcode, size, asm,
9439 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
9442 def i64_POST : SIMDLdStSingleDPost<0, R, opcode, size, asm,
9443 (outs GPR64sp:$wback),
9444 (ins listtype:$Vt, VectorIndexD:$idx,
9445 GPR64sp:$Rn, GPR64pi:$Xm)>;
9448 multiclass SIMDLdStSingleAliases<string asm, string layout, string Type,
9449 string Count, int Offset, Operand idxtype> {
9450 // E.g. "ld1 { v0.8b }[0], [x1], #1"
9451 // "ld1\t$Vt, [$Rn], #1"
9452 // may get mapped to
9453 // (LD1Rv8b_POST VecListOne8b:$Vt, GPR64sp:$Rn, XZR)
9454 def : InstAlias<asm # "\t$Vt$idx, [$Rn], #" # Offset,
9455 (!cast<Instruction>(NAME # Type # "_POST")
9457 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
9458 idxtype:$idx, XZR), 1>;
9460 // E.g. "ld1.8b { v0 }[0], [x1], #1"
9461 // "ld1.8b\t$Vt, [$Rn], #1"
9462 // may get mapped to
9463 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, XZR)
9464 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn], #" # Offset,
9465 (!cast<Instruction>(NAME # Type # "_POST")
9467 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9468 idxtype:$idx, XZR), 0>;
9470 // E.g. "ld1.8b { v0 }[0], [x1]"
9471 // "ld1.8b\t$Vt, [$Rn]"
9472 // may get mapped to
9473 // (LD1Rv8b VecListOne64:$Vt, GPR64sp:$Rn)
9474 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn]",
9475 (!cast<Instruction>(NAME # Type)
9476 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9477 idxtype:$idx, GPR64sp:$Rn), 0>;
9479 // E.g. "ld1.8b { v0 }[0], [x1], x2"
9480 // "ld1.8b\t$Vt, [$Rn], $Xm"
9481 // may get mapped to
9482 // (LD1Rv8b_POST VecListOne64:$Vt, GPR64sp:$Rn, GPR64pi1:$Xm)
9483 def : InstAlias<asm # "." # layout # "\t$Vt$idx, [$Rn], $Xm",
9484 (!cast<Instruction>(NAME # Type # "_POST")
9486 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
9488 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
9491 multiclass SIMDLdSt1SingleAliases<string asm> {
9492 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "One", 1, VectorIndexB>;
9493 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "One", 2, VectorIndexH>;
9494 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "One", 4, VectorIndexS>;
9495 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "One", 8, VectorIndexD>;
9498 multiclass SIMDLdSt2SingleAliases<string asm> {
9499 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Two", 2, VectorIndexB>;
9500 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Two", 4, VectorIndexH>;
9501 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Two", 8, VectorIndexS>;
9502 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Two", 16, VectorIndexD>;
9505 multiclass SIMDLdSt3SingleAliases<string asm> {
9506 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Three", 3, VectorIndexB>;
9507 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Three", 6, VectorIndexH>;
9508 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Three", 12, VectorIndexS>;
9509 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Three", 24, VectorIndexD>;
9512 multiclass SIMDLdSt4SingleAliases<string asm> {
9513 defm "" : SIMDLdStSingleAliases<asm, "b", "i8", "Four", 4, VectorIndexB>;
9514 defm "" : SIMDLdStSingleAliases<asm, "h", "i16", "Four", 8, VectorIndexH>;
9515 defm "" : SIMDLdStSingleAliases<asm, "s", "i32", "Four", 16, VectorIndexS>;
9516 defm "" : SIMDLdStSingleAliases<asm, "d", "i64", "Four", 32, VectorIndexD>;
9518 } // end of 'let Predicates = [HasNEON]'
9520 //----------------------------------------------------------------------------
9521 // AdvSIMD v8.1 Rounding Double Multiply Add/Subtract
9522 //----------------------------------------------------------------------------
9524 let Predicates = [HasNEON, HasRDM] in {
9526 class BaseSIMDThreeSameVectorTiedR0<bit Q, bit U, bits<2> size, bits<5> opcode,
9527 RegisterOperand regtype, string asm,
9528 string kind, list<dag> pattern>
9529 : BaseSIMDThreeSameVectorTied<Q, U, {size,0}, opcode, regtype, asm, kind,
9532 multiclass SIMDThreeSameVectorSQRDMLxHTiedHS<bit U, bits<5> opc, string asm,
9533 SDPatternOperator Accum> {
9534 def v4i16 : BaseSIMDThreeSameVectorTiedR0<0, U, 0b01, opc, V64, asm, ".4h",
9535 [(set (v4i16 V64:$dst),
9536 (Accum (v4i16 V64:$Rd),
9537 (v4i16 (int_aarch64_neon_sqrdmulh (v4i16 V64:$Rn),
9538 (v4i16 V64:$Rm)))))]>;
9539 def v8i16 : BaseSIMDThreeSameVectorTiedR0<1, U, 0b01, opc, V128, asm, ".8h",
9540 [(set (v8i16 V128:$dst),
9541 (Accum (v8i16 V128:$Rd),
9542 (v8i16 (int_aarch64_neon_sqrdmulh (v8i16 V128:$Rn),
9543 (v8i16 V128:$Rm)))))]>;
9544 def v2i32 : BaseSIMDThreeSameVectorTiedR0<0, U, 0b10, opc, V64, asm, ".2s",
9545 [(set (v2i32 V64:$dst),
9546 (Accum (v2i32 V64:$Rd),
9547 (v2i32 (int_aarch64_neon_sqrdmulh (v2i32 V64:$Rn),
9548 (v2i32 V64:$Rm)))))]>;
9549 def v4i32 : BaseSIMDThreeSameVectorTiedR0<1, U, 0b10, opc, V128, asm, ".4s",
9550 [(set (v4i32 V128:$dst),
9551 (Accum (v4i32 V128:$Rd),
9552 (v4i32 (int_aarch64_neon_sqrdmulh (v4i32 V128:$Rn),
9553 (v4i32 V128:$Rm)))))]>;
9556 multiclass SIMDIndexedSQRDMLxHSDTied<bit U, bits<4> opc, string asm,
9557 SDPatternOperator Accum> {
9558 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
9559 V64, V64, V128_lo, VectorIndexH,
9560 asm, ".4h", ".4h", ".4h", ".h",
9561 [(set (v4i16 V64:$dst),
9562 (Accum (v4i16 V64:$Rd),
9563 (v4i16 (int_aarch64_neon_sqrdmulh
9565 (v4i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
9566 VectorIndexH:$idx))))))]> {
9568 let Inst{11} = idx{2};
9569 let Inst{21} = idx{1};
9570 let Inst{20} = idx{0};
9573 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
9574 V128, V128, V128_lo, VectorIndexH,
9575 asm, ".8h", ".8h", ".8h", ".h",
9576 [(set (v8i16 V128:$dst),
9577 (Accum (v8i16 V128:$Rd),
9578 (v8i16 (int_aarch64_neon_sqrdmulh
9580 (v8i16 (AArch64duplane16 (v8i16 V128_lo:$Rm),
9581 VectorIndexH:$idx))))))]> {
9583 let Inst{11} = idx{2};
9584 let Inst{21} = idx{1};
9585 let Inst{20} = idx{0};
9588 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
9589 V64, V64, V128, VectorIndexS,
9590 asm, ".2s", ".2s", ".2s", ".s",
9591 [(set (v2i32 V64:$dst),
9592 (Accum (v2i32 V64:$Rd),
9593 (v2i32 (int_aarch64_neon_sqrdmulh
9595 (v2i32 (AArch64duplane32 (v4i32 V128:$Rm),
9596 VectorIndexS:$idx))))))]> {
9598 let Inst{11} = idx{1};
9599 let Inst{21} = idx{0};
9602 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but
9603 // an intermediate EXTRACT_SUBREG would be untyped.
9604 // FIXME: direct EXTRACT_SUBREG from v2i32 to i32 is illegal, that's why we
9605 // got it lowered here as (i32 vector_extract (v4i32 insert_subvector(..)))
9606 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
9607 (i32 (vector_extract
9608 (v4i32 (insert_subvector
9610 (v2i32 (int_aarch64_neon_sqrdmulh
9612 (v2i32 (AArch64duplane32
9614 VectorIndexS:$idx)))),
9618 (v2i32 (!cast<Instruction>(NAME # v2i32_indexed)
9619 (v2i32 (INSERT_SUBREG (v2i32 (IMPLICIT_DEF)),
9624 VectorIndexS:$idx)),
9627 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
9628 V128, V128, V128, VectorIndexS,
9629 asm, ".4s", ".4s", ".4s", ".s",
9630 [(set (v4i32 V128:$dst),
9631 (Accum (v4i32 V128:$Rd),
9632 (v4i32 (int_aarch64_neon_sqrdmulh
9634 (v4i32 (AArch64duplane32 (v4i32 V128:$Rm),
9635 VectorIndexS:$idx))))))]> {
9637 let Inst{11} = idx{1};
9638 let Inst{21} = idx{0};
9641 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but
9642 // an intermediate EXTRACT_SUBREG would be untyped.
9643 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
9644 (i32 (vector_extract
9645 (v4i32 (int_aarch64_neon_sqrdmulh
9647 (v4i32 (AArch64duplane32
9649 VectorIndexS:$idx)))),
9652 (v4i32 (!cast<Instruction>(NAME # v4i32_indexed)
9653 (v4i32 (INSERT_SUBREG (v4i32 (IMPLICIT_DEF)),
9658 VectorIndexS:$idx)),
9661 def i16_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
9662 FPR16Op, FPR16Op, V128_lo,
9663 VectorIndexH, asm, ".h", "", "", ".h",
9666 let Inst{11} = idx{2};
9667 let Inst{21} = idx{1};
9668 let Inst{20} = idx{0};
9671 def i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
9672 FPR32Op, FPR32Op, V128, VectorIndexS,
9673 asm, ".s", "", "", ".s",
9674 [(set (i32 FPR32Op:$dst),
9675 (Accum (i32 FPR32Op:$Rd),
9676 (i32 (int_aarch64_neon_sqrdmulh
9678 (i32 (vector_extract (v4i32 V128:$Rm),
9679 VectorIndexS:$idx))))))]> {
9681 let Inst{11} = idx{1};
9682 let Inst{21} = idx{0};
9685 } // let Predicates = [HasNeon, HasRDM]
9687 //----------------------------------------------------------------------------
9688 // ARMv8.3 Complex ADD/MLA instructions
9689 //----------------------------------------------------------------------------
9691 class ComplexRotationOperand<int Angle, int Remainder, string Type>
9693 let PredicateMethod = "isComplexRotation<" # Angle # ", " # Remainder # ">";
9694 let DiagnosticType = "InvalidComplexRotation" # Type;
9695 let Name = "ComplexRotation" # Type;
9697 def complexrotateop : Operand<i32> {
9698 let ParserMatchClass = ComplexRotationOperand<90, 0, "Even">;
9699 let PrintMethod = "printComplexRotationOp<90, 0>";
9701 def complexrotateopodd : Operand<i32> {
9702 let ParserMatchClass = ComplexRotationOperand<180, 90, "Odd">;
9703 let PrintMethod = "printComplexRotationOp<180, 90>";
9706 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
9707 class BaseSIMDThreeSameVectorComplex<bit Q, bit U, bits<2> size, bits<3> opcode,
9708 RegisterOperand regtype, Operand rottype,
9709 string asm, string kind, list<dag> pattern>
9710 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, rottype:$rot), asm,
9711 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $rot"
9712 "|" # kind # "\t$Rd, $Rn, $Rm, $rot}", "", pattern>,
9721 let Inst{28-24} = 0b01110;
9722 let Inst{23-22} = size;
9724 let Inst{20-16} = Rm;
9725 let Inst{15-13} = opcode;
9726 // Non-tied version (FCADD) only has one rotation bit
9734 multiclass SIMDThreeSameVectorComplexHSD<bit U, bits<3> opcode, Operand rottype,
9735 string asm, SDPatternOperator OpNode>{
9736 let Predicates = [HasV8_3a, HasNEON, HasFullFP16] in {
9737 def v4f16 : BaseSIMDThreeSameVectorComplex<0, U, 0b01, opcode, V64, rottype,
9739 [(set (v4f16 V64:$dst), (OpNode (v4f16 V64:$Rd),
9742 (rottype i32:$rot)))]>;
9744 def v8f16 : BaseSIMDThreeSameVectorComplex<1, U, 0b01, opcode, V128, rottype,
9746 [(set (v8f16 V128:$dst), (OpNode (v8f16 V128:$Rd),
9749 (rottype i32:$rot)))]>;
9752 let Predicates = [HasV8_3a, HasNEON] in {
9753 def v2f32 : BaseSIMDThreeSameVectorComplex<0, U, 0b10, opcode, V64, rottype,
9755 [(set (v2f32 V64:$dst), (OpNode (v2f32 V64:$Rd),
9758 (rottype i32:$rot)))]>;
9760 def v4f32 : BaseSIMDThreeSameVectorComplex<1, U, 0b10, opcode, V128, rottype,
9762 [(set (v4f32 V128:$dst), (OpNode (v4f32 V128:$Rd),
9765 (rottype i32:$rot)))]>;
9767 def v2f64 : BaseSIMDThreeSameVectorComplex<1, U, 0b11, opcode, V128, rottype,
9769 [(set (v2f64 V128:$dst), (OpNode (v2f64 V128:$Rd),
9772 (rottype i32:$rot)))]>;
9776 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
9777 class BaseSIMDThreeSameVectorTiedComplex<bit Q, bit U, bits<2> size,
9779 RegisterOperand regtype,
9780 Operand rottype, string asm,
9781 string kind, list<dag> pattern>
9782 : I<(outs regtype:$dst),
9783 (ins regtype:$Rd, regtype:$Rn, regtype:$Rm, rottype:$rot), asm,
9784 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $rot"
9785 "|" # kind # "\t$Rd, $Rn, $Rm, $rot}", "$Rd = $dst", pattern>,
9794 let Inst{28-24} = 0b01110;
9795 let Inst{23-22} = size;
9797 let Inst{20-16} = Rm;
9798 let Inst{15-13} = opcode;
9799 let Inst{12-11} = rot;
9805 multiclass SIMDThreeSameVectorTiedComplexHSD<bit U, bits<3> opcode,
9806 Operand rottype, string asm,
9807 SDPatternOperator OpNode> {
9808 let Predicates = [HasV8_3a, HasNEON, HasFullFP16] in {
9809 def v4f16 : BaseSIMDThreeSameVectorTiedComplex<0, U, 0b01, opcode, V64,
9810 rottype, asm, ".4h",
9811 [(set (v4f16 V64:$dst), (OpNode (v4f16 V64:$Rd),
9814 (rottype i32:$rot)))]>;
9816 def v8f16 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b01, opcode, V128,
9817 rottype, asm, ".8h",
9818 [(set (v8f16 V128:$dst), (OpNode (v8f16 V128:$Rd),
9821 (rottype i32:$rot)))]>;
9824 let Predicates = [HasV8_3a, HasNEON] in {
9825 def v2f32 : BaseSIMDThreeSameVectorTiedComplex<0, U, 0b10, opcode, V64,
9826 rottype, asm, ".2s",
9827 [(set (v2f32 V64:$dst), (OpNode (v2f32 V64:$Rd),
9830 (rottype i32:$rot)))]>;
9832 def v4f32 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b10, opcode, V128,
9833 rottype, asm, ".4s",
9834 [(set (v4f32 V128:$dst), (OpNode (v4f32 V128:$Rd),
9837 (rottype i32:$rot)))]>;
9839 def v2f64 : BaseSIMDThreeSameVectorTiedComplex<1, U, 0b11, opcode, V128,
9840 rottype, asm, ".2d",
9841 [(set (v2f64 V128:$dst), (OpNode (v2f64 V128:$Rd),
9844 (rottype i32:$rot)))]>;
9848 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
9849 class BaseSIMDIndexedTiedComplex<bit Q, bit U, bit Scalar, bits<2> size,
9850 bit opc1, bit opc2, RegisterOperand dst_reg,
9851 RegisterOperand lhs_reg,
9852 RegisterOperand rhs_reg, Operand vec_idx,
9853 Operand rottype, string asm, string apple_kind,
9854 string dst_kind, string lhs_kind,
9855 string rhs_kind, list<dag> pattern>
9856 : I<(outs dst_reg:$dst),
9857 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx, rottype:$rot),
9859 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind #
9860 "$idx, $rot" # "|" # apple_kind #
9861 "\t$Rd, $Rn, $Rm$idx, $rot}", "$Rd = $dst", pattern>,
9871 let Inst{28} = Scalar;
9872 let Inst{27-24} = 0b1111;
9873 let Inst{23-22} = size;
9874 // Bit 21 must be set by the derived class.
9875 let Inst{20-16} = Rm;
9876 let Inst{15} = opc1;
9877 let Inst{14-13} = rot;
9878 let Inst{12} = opc2;
9879 // Bit 11 must be set by the derived class.
9885 // The complex instructions index by pairs of elements, so the VectorIndexes
9886 // don't match the lane types, and the index bits are different to the other
9888 multiclass SIMDIndexedTiedComplexHSD<bit U, bit opc1, bit opc2, Operand rottype,
9889 string asm, SDPatternOperator OpNode> {
9890 let Predicates = [HasV8_3a,HasNEON,HasFullFP16] in {
9891 def v4f16_indexed : BaseSIMDIndexedTiedComplex<0, 1, 0, 0b01, opc1, opc2, V64,
9892 V64, V128, VectorIndexD, rottype, asm, ".4h", ".4h",
9896 let Inst{21} = idx{0};
9899 def v8f16_indexed : BaseSIMDIndexedTiedComplex<1, 1, 0, 0b01, opc1, opc2,
9900 V128, V128, V128, VectorIndexS, rottype, asm, ".8h",
9901 ".8h", ".8h", ".h", []> {
9903 let Inst{11} = idx{1};
9904 let Inst{21} = idx{0};
9906 } // Predicates = [HasV8_3a,HasNEON,HasFullFP16]
9908 let Predicates = [HasV8_3a,HasNEON] in {
9909 def v4f32_indexed : BaseSIMDIndexedTiedComplex<1, 1, 0, 0b10, opc1, opc2,
9910 V128, V128, V128, VectorIndexD, rottype, asm, ".4s",
9911 ".4s", ".4s", ".s", []> {
9913 let Inst{11} = idx{0};
9916 } // Predicates = [HasV8_3a,HasNEON]
9919 //----------------------------------------------------------------------------
9920 // Crypto extensions
9921 //----------------------------------------------------------------------------
9923 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
9924 class AESBase<bits<4> opc, string asm, dag outs, dag ins, string cstr,
9926 : I<outs, ins, asm, "{\t$Rd.16b, $Rn.16b|.16b\t$Rd, $Rn}", cstr, pat>,
9930 let Inst{31-16} = 0b0100111000101000;
9931 let Inst{15-12} = opc;
9932 let Inst{11-10} = 0b10;
9937 class AESInst<bits<4> opc, string asm, Intrinsic OpNode>
9938 : AESBase<opc, asm, (outs V128:$Rd), (ins V128:$Rn), "",
9939 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
9941 class AESTiedInst<bits<4> opc, string asm, Intrinsic OpNode>
9942 : AESBase<opc, asm, (outs V128:$dst), (ins V128:$Rd, V128:$Rn),
9944 [(set (v16i8 V128:$dst),
9945 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
9947 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
9948 class SHA3OpTiedInst<bits<3> opc, string asm, string dst_lhs_kind,
9949 dag oops, dag iops, list<dag> pat>
9950 : I<oops, iops, asm,
9951 "{\t$Rd" # dst_lhs_kind # ", $Rn" # dst_lhs_kind # ", $Rm.4s" #
9952 "|.4s\t$Rd, $Rn, $Rm}", "$Rd = $dst", pat>,
9957 let Inst{31-21} = 0b01011110000;
9958 let Inst{20-16} = Rm;
9960 let Inst{14-12} = opc;
9961 let Inst{11-10} = 0b00;
9966 class SHATiedInstQSV<bits<3> opc, string asm, Intrinsic OpNode>
9967 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
9968 (ins FPR128:$Rd, FPR32:$Rn, V128:$Rm),
9969 [(set (v4i32 FPR128:$dst),
9970 (OpNode (v4i32 FPR128:$Rd), (i32 FPR32:$Rn),
9971 (v4i32 V128:$Rm)))]>;
9973 class SHATiedInstVVV<bits<3> opc, string asm, Intrinsic OpNode>
9974 : SHA3OpTiedInst<opc, asm, ".4s", (outs V128:$dst),
9975 (ins V128:$Rd, V128:$Rn, V128:$Rm),
9976 [(set (v4i32 V128:$dst),
9977 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
9978 (v4i32 V128:$Rm)))]>;
9980 class SHATiedInstQQV<bits<3> opc, string asm, Intrinsic OpNode>
9981 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
9982 (ins FPR128:$Rd, FPR128:$Rn, V128:$Rm),
9983 [(set (v4i32 FPR128:$dst),
9984 (OpNode (v4i32 FPR128:$Rd), (v4i32 FPR128:$Rn),
9985 (v4i32 V128:$Rm)))]>;
9987 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
9988 class SHA2OpInst<bits<4> opc, string asm, string kind,
9989 string cstr, dag oops, dag iops,
9991 : I<oops, iops, asm, "{\t$Rd" # kind # ", $Rn" # kind #
9992 "|" # kind # "\t$Rd, $Rn}", cstr, pat>,
9996 let Inst{31-16} = 0b0101111000101000;
9997 let Inst{15-12} = opc;
9998 let Inst{11-10} = 0b10;
10000 let Inst{4-0} = Rd;
10003 class SHATiedInstVV<bits<4> opc, string asm, Intrinsic OpNode>
10004 : SHA2OpInst<opc, asm, ".4s", "$Rd = $dst", (outs V128:$dst),
10005 (ins V128:$Rd, V128:$Rn),
10006 [(set (v4i32 V128:$dst),
10007 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
10009 class SHAInstSS<bits<4> opc, string asm, Intrinsic OpNode>
10010 : SHA2OpInst<opc, asm, "", "", (outs FPR32:$Rd), (ins FPR32:$Rn),
10011 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
10013 // Armv8.2-A Crypto extensions
10014 class BaseCryptoV82<dag oops, dag iops, string asm, string asmops, string cst,
10016 : I <oops, iops, asm, asmops, cst, pattern>, Sched<[WriteV]> {
10019 let Inst{31-25} = 0b1100111;
10020 let Inst{9-5} = Vn;
10021 let Inst{4-0} = Vd;
10024 class CryptoRRTied<bits<1>op0, bits<2>op1, string asm, string asmops>
10025 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm, asmops,
10027 let Inst{31-25} = 0b1100111;
10028 let Inst{24-21} = 0b0110;
10029 let Inst{20-15} = 0b000001;
10030 let Inst{14} = op0;
10031 let Inst{13-12} = 0b00;
10032 let Inst{11-10} = op1;
10034 class CryptoRRTied_2D<bits<1>op0, bits<2>op1, string asm>
10035 : CryptoRRTied<op0, op1, asm, "{\t$Vd.2d, $Vn.2d}">;
10036 class CryptoRRTied_4S<bits<1>op0, bits<2>op1, string asm>
10037 : CryptoRRTied<op0, op1, asm, "{\t$Vd.4s, $Vn.4s}">;
10039 class CryptoRRR<bits<1> op0, bits<2>op1, dag oops, dag iops, string asm,
10040 string asmops, string cst>
10041 : BaseCryptoV82<oops, iops, asm , asmops, cst, []> {
10043 let Inst{24-21} = 0b0011;
10044 let Inst{20-16} = Vm;
10045 let Inst{15} = 0b1;
10046 let Inst{14} = op0;
10047 let Inst{13-12} = 0b00;
10048 let Inst{11-10} = op1;
10050 class CryptoRRR_2D<bits<1> op0, bits<2>op1, string asm>
10051 : CryptoRRR<op0, op1, (outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm,
10052 "{\t$Vd.2d, $Vn.2d, $Vm.2d}", "">;
10053 class CryptoRRRTied_2D<bits<1> op0, bits<2>op1, string asm>
10054 : CryptoRRR<op0, op1, (outs V128:$Vdst), (ins V128:$Vd, V128:$Vn, V128:$Vm), asm,
10055 "{\t$Vd.2d, $Vn.2d, $Vm.2d}", "$Vd = $Vdst">;
10056 class CryptoRRR_4S<bits<1> op0, bits<2>op1, string asm>
10057 : CryptoRRR<op0, op1, (outs V128:$Vd), (ins V128:$Vn, V128:$Vm), asm,
10058 "{\t$Vd.4s, $Vn.4s, $Vm.4s}", "">;
10059 class CryptoRRRTied_4S<bits<1> op0, bits<2>op1, string asm>
10060 : CryptoRRR<op0, op1, (outs V128:$Vdst), (ins V128:$Vd, V128:$Vn, V128:$Vm), asm,
10061 "{\t$Vd.4s, $Vn.4s, $Vm.4s}", "$Vd = $Vdst">;
10062 class CryptoRRRTied<bits<1> op0, bits<2>op1, string asm>
10063 : CryptoRRR<op0, op1, (outs FPR128:$Vdst), (ins FPR128:$Vd, FPR128:$Vn, V128:$Vm),
10064 asm, "{\t$Vd, $Vn, $Vm.2d}", "$Vd = $Vdst">;
10066 class CryptoRRRR<bits<2>op0, string asm, string asmops>
10067 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm, V128:$Va), asm,
10071 let Inst{24-23} = 0b00;
10072 let Inst{22-21} = op0;
10073 let Inst{20-16} = Vm;
10074 let Inst{15} = 0b0;
10075 let Inst{14-10} = Va;
10077 class CryptoRRRR_16B<bits<2>op0, string asm>
10078 : CryptoRRRR<op0, asm, "{\t$Vd.16b, $Vn.16b, $Vm.16b, $Va.16b}"> {
10080 class CryptoRRRR_4S<bits<2>op0, string asm>
10081 : CryptoRRRR<op0, asm, "{\t$Vd.4s, $Vn.4s, $Vm.4s, $Va.4s}"> {
10084 class CryptoRRRi6<string asm>
10085 : BaseCryptoV82<(outs V128:$Vd), (ins V128:$Vn, V128:$Vm, uimm6:$imm), asm,
10086 "{\t$Vd.2d, $Vn.2d, $Vm.2d, $imm}", "", []> {
10089 let Inst{24-21} = 0b0100;
10090 let Inst{20-16} = Vm;
10091 let Inst{15-10} = imm;
10092 let Inst{9-5} = Vn;
10093 let Inst{4-0} = Vd;
10096 class CryptoRRRi2Tied<bits<1>op0, bits<2>op1, string asm>
10097 : BaseCryptoV82<(outs V128:$Vdst),
10098 (ins V128:$Vd, V128:$Vn, V128:$Vm, VectorIndexS:$imm),
10099 asm, "{\t$Vd.4s, $Vn.4s, $Vm.s$imm}", "$Vd = $Vdst", []> {
10102 let Inst{24-21} = 0b0010;
10103 let Inst{20-16} = Vm;
10104 let Inst{15} = 0b1;
10105 let Inst{14} = op0;
10106 let Inst{13-12} = imm;
10107 let Inst{11-10} = op1;
10110 //----------------------------------------------------------------------------
10111 // v8.1 atomic instructions extension:
10115 // * LDOPregister<OP>, and aliases STOPregister<OP>
10117 // Instruction encodings:
10119 // 31 30|29 24|23|22|21|20 16|15|14 10|9 5|4 0
10120 // CAS SZ |001000|1 |A |1 |Rs |R |11111 |Rn |Rt
10121 // CASP 0|SZ|001000|0 |A |1 |Rs |R |11111 |Rn |Rt
10122 // SWP SZ |111000|A |R |1 |Rs |1 |OPC|00|Rn |Rt
10123 // LD SZ |111000|A |R |1 |Rs |0 |OPC|00|Rn |Rt
10124 // ST SZ |111000|A |R |1 |Rs |0 |OPC|00|Rn |11111
10126 // Instruction syntax:
10128 // CAS{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10129 // CAS{<order>} <Xs>, <Xt>, [<Xn|SP>]
10130 // CASP{<order>} <Ws>, <W(s+1)>, <Wt>, <W(t+1)>, [<Xn|SP>]
10131 // CASP{<order>} <Xs>, <X(s+1)>, <Xt>, <X(t+1)>, [<Xn|SP>]
10132 // SWP{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10133 // SWP{<order>} <Xs>, <Xt>, [<Xn|SP>]
10134 // LD<OP>{<order>}[<size>] <Ws>, <Wt>, [<Xn|SP>]
10135 // LD<OP>{<order>} <Xs>, <Xt>, [<Xn|SP>]
10136 // ST<OP>{<order>}[<size>] <Ws>, [<Xn|SP>]
10137 // ST<OP>{<order>} <Xs>, [<Xn|SP>]
10139 let Predicates = [HasLSE], mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
10140 class BaseCASEncoding<dag oops, dag iops, string asm, string operands,
10141 string cstr, list<dag> pattern>
10142 : I<oops, iops, asm, operands, cstr, pattern> {
10150 let Inst{31-30} = Sz;
10151 let Inst{29-24} = 0b001000;
10153 let Inst{22} = Acq;
10154 let Inst{21} = 0b1;
10155 let Inst{20-16} = Rs;
10156 let Inst{15} = Rel;
10157 let Inst{14-10} = 0b11111;
10158 let Inst{9-5} = Rn;
10159 let Inst{4-0} = Rt;
10160 let Predicates = [HasLSE];
10163 class BaseCAS<string order, string size, RegisterClass RC>
10164 : BaseCASEncoding<(outs RC:$out),(ins RC:$Rs, RC:$Rt, GPR64sp:$Rn),
10165 "cas" # order # size, "\t$Rs, $Rt, [$Rn]",
10167 Sched<[WriteAtomic]> {
10171 multiclass CompareAndSwap<bits<1> Acq, bits<1> Rel, string order> {
10172 let Sz = 0b00, Acq = Acq, Rel = Rel in def B : BaseCAS<order, "b", GPR32>;
10173 let Sz = 0b01, Acq = Acq, Rel = Rel in def H : BaseCAS<order, "h", GPR32>;
10174 let Sz = 0b10, Acq = Acq, Rel = Rel in def W : BaseCAS<order, "", GPR32>;
10175 let Sz = 0b11, Acq = Acq, Rel = Rel in def X : BaseCAS<order, "", GPR64>;
10178 class BaseCASP<string order, string size, RegisterOperand RC>
10179 : BaseCASEncoding<(outs RC:$out),(ins RC:$Rs, RC:$Rt, GPR64sp:$Rn),
10180 "casp" # order # size, "\t$Rs, $Rt, [$Rn]",
10182 Sched<[WriteAtomic]> {
10186 multiclass CompareAndSwapPair<bits<1> Acq, bits<1> Rel, string order> {
10187 let Sz = 0b00, Acq = Acq, Rel = Rel in
10188 def W : BaseCASP<order, "", WSeqPairClassOperand>;
10189 let Sz = 0b01, Acq = Acq, Rel = Rel in
10190 def X : BaseCASP<order, "", XSeqPairClassOperand>;
10193 let Predicates = [HasLSE] in
10194 class BaseSWP<string order, string size, RegisterClass RC>
10195 : I<(outs RC:$Rt),(ins RC:$Rs, GPR64sp:$Rn), "swp" # order # size,
10196 "\t$Rs, $Rt, [$Rn]","",[]>,
10197 Sched<[WriteAtomic]> {
10202 bits<3> opc = 0b000;
10205 let Inst{31-30} = Sz;
10206 let Inst{29-24} = 0b111000;
10207 let Inst{23} = Acq;
10208 let Inst{22} = Rel;
10209 let Inst{21} = 0b1;
10210 let Inst{20-16} = Rs;
10211 let Inst{15} = 0b1;
10212 let Inst{14-12} = opc;
10213 let Inst{11-10} = 0b00;
10214 let Inst{9-5} = Rn;
10215 let Inst{4-0} = Rt;
10216 let Predicates = [HasLSE];
10219 multiclass Swap<bits<1> Acq, bits<1> Rel, string order> {
10220 let Sz = 0b00, Acq = Acq, Rel = Rel in def B : BaseSWP<order, "b", GPR32>;
10221 let Sz = 0b01, Acq = Acq, Rel = Rel in def H : BaseSWP<order, "h", GPR32>;
10222 let Sz = 0b10, Acq = Acq, Rel = Rel in def W : BaseSWP<order, "", GPR32>;
10223 let Sz = 0b11, Acq = Acq, Rel = Rel in def X : BaseSWP<order, "", GPR64>;
10226 let Predicates = [HasLSE], mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
10227 class BaseLDOPregister<string op, string order, string size, RegisterClass RC>
10228 : I<(outs RC:$Rt),(ins RC:$Rs, GPR64sp:$Rn), "ld" # op # order # size,
10229 "\t$Rs, $Rt, [$Rn]","",[]>,
10230 Sched<[WriteAtomic]> {
10238 let Inst{31-30} = Sz;
10239 let Inst{29-24} = 0b111000;
10240 let Inst{23} = Acq;
10241 let Inst{22} = Rel;
10242 let Inst{21} = 0b1;
10243 let Inst{20-16} = Rs;
10244 let Inst{15} = 0b0;
10245 let Inst{14-12} = opc;
10246 let Inst{11-10} = 0b00;
10247 let Inst{9-5} = Rn;
10248 let Inst{4-0} = Rt;
10249 let Predicates = [HasLSE];
10252 multiclass LDOPregister<bits<3> opc, string op, bits<1> Acq, bits<1> Rel,
10254 let Sz = 0b00, Acq = Acq, Rel = Rel, opc = opc in
10255 def B : BaseLDOPregister<op, order, "b", GPR32>;
10256 let Sz = 0b01, Acq = Acq, Rel = Rel, opc = opc in
10257 def H : BaseLDOPregister<op, order, "h", GPR32>;
10258 let Sz = 0b10, Acq = Acq, Rel = Rel, opc = opc in
10259 def W : BaseLDOPregister<op, order, "", GPR32>;
10260 let Sz = 0b11, Acq = Acq, Rel = Rel, opc = opc in
10261 def X : BaseLDOPregister<op, order, "", GPR64>;
10264 // Differing SrcRHS and DstRHS allow you to cover CLR & SUB by giving a more
10265 // complex DAG for DstRHS.
10266 let Predicates = [HasLSE] in
10267 multiclass LDOPregister_patterns_ord_dag<string inst, string suffix, string op,
10268 string size, dag SrcRHS, dag DstRHS> {
10269 def : Pat<(!cast<PatFrag>(op#"_"#size#"_monotonic") GPR64sp:$Rn, SrcRHS),
10270 (!cast<Instruction>(inst # suffix) DstRHS, GPR64sp:$Rn)>;
10271 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acquire") GPR64sp:$Rn, SrcRHS),
10272 (!cast<Instruction>(inst # "A" # suffix) DstRHS, GPR64sp:$Rn)>;
10273 def : Pat<(!cast<PatFrag>(op#"_"#size#"_release") GPR64sp:$Rn, SrcRHS),
10274 (!cast<Instruction>(inst # "L" # suffix) DstRHS, GPR64sp:$Rn)>;
10275 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acq_rel") GPR64sp:$Rn, SrcRHS),
10276 (!cast<Instruction>(inst # "AL" # suffix) DstRHS, GPR64sp:$Rn)>;
10277 def : Pat<(!cast<PatFrag>(op#"_"#size#"_seq_cst") GPR64sp:$Rn, SrcRHS),
10278 (!cast<Instruction>(inst # "AL" # suffix) DstRHS, GPR64sp:$Rn)>;
10281 multiclass LDOPregister_patterns_ord<string inst, string suffix, string op,
10282 string size, dag RHS> {
10283 defm : LDOPregister_patterns_ord_dag<inst, suffix, op, size, RHS, RHS>;
10286 multiclass LDOPregister_patterns_ord_mod<string inst, string suffix, string op,
10287 string size, dag LHS, dag RHS> {
10288 defm : LDOPregister_patterns_ord_dag<inst, suffix, op, size, LHS, RHS>;
10291 multiclass LDOPregister_patterns<string inst, string op> {
10292 defm : LDOPregister_patterns_ord<inst, "X", op, "64", (i64 GPR64:$Rm)>;
10293 defm : LDOPregister_patterns_ord<inst, "W", op, "32", (i32 GPR32:$Rm)>;
10294 defm : LDOPregister_patterns_ord<inst, "H", op, "16", (i32 GPR32:$Rm)>;
10295 defm : LDOPregister_patterns_ord<inst, "B", op, "8", (i32 GPR32:$Rm)>;
10298 multiclass LDOPregister_patterns_mod<string inst, string op, string mod> {
10299 defm : LDOPregister_patterns_ord_mod<inst, "X", op, "64",
10301 (i64 (!cast<Instruction>(mod#Xrr) XZR, GPR64:$Rm))>;
10302 defm : LDOPregister_patterns_ord_mod<inst, "W", op, "32",
10304 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10305 defm : LDOPregister_patterns_ord_mod<inst, "H", op, "16",
10307 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10308 defm : LDOPregister_patterns_ord_mod<inst, "B", op, "8",
10310 (i32 (!cast<Instruction>(mod#Wrr) WZR, GPR32:$Rm))>;
10313 let Predicates = [HasLSE] in
10314 multiclass CASregister_patterns_ord_dag<string inst, string suffix, string op,
10315 string size, dag OLD, dag NEW> {
10316 def : Pat<(!cast<PatFrag>(op#"_"#size#"_monotonic") GPR64sp:$Rn, OLD, NEW),
10317 (!cast<Instruction>(inst # suffix) OLD, NEW, GPR64sp:$Rn)>;
10318 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acquire") GPR64sp:$Rn, OLD, NEW),
10319 (!cast<Instruction>(inst # "A" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10320 def : Pat<(!cast<PatFrag>(op#"_"#size#"_release") GPR64sp:$Rn, OLD, NEW),
10321 (!cast<Instruction>(inst # "L" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10322 def : Pat<(!cast<PatFrag>(op#"_"#size#"_acq_rel") GPR64sp:$Rn, OLD, NEW),
10323 (!cast<Instruction>(inst # "AL" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10324 def : Pat<(!cast<PatFrag>(op#"_"#size#"_seq_cst") GPR64sp:$Rn, OLD, NEW),
10325 (!cast<Instruction>(inst # "AL" # suffix) OLD, NEW, GPR64sp:$Rn)>;
10328 multiclass CASregister_patterns_ord<string inst, string suffix, string op,
10329 string size, dag OLD, dag NEW> {
10330 defm : CASregister_patterns_ord_dag<inst, suffix, op, size, OLD, NEW>;
10333 multiclass CASregister_patterns<string inst, string op> {
10334 defm : CASregister_patterns_ord<inst, "X", op, "64",
10335 (i64 GPR64:$Rold), (i64 GPR64:$Rnew)>;
10336 defm : CASregister_patterns_ord<inst, "W", op, "32",
10337 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10338 defm : CASregister_patterns_ord<inst, "H", op, "16",
10339 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10340 defm : CASregister_patterns_ord<inst, "B", op, "8",
10341 (i32 GPR32:$Rold), (i32 GPR32:$Rnew)>;
10344 let Predicates = [HasLSE] in
10345 class BaseSTOPregister<string asm, RegisterClass OP, Register Reg,
10346 Instruction inst> :
10347 InstAlias<asm # "\t$Rs, [$Rn]", (inst Reg, OP:$Rs, GPR64sp:$Rn)>;
10349 multiclass STOPregister<string asm, string instr> {
10350 def : BaseSTOPregister<asm # "lb", GPR32, WZR,
10351 !cast<Instruction>(instr # "LB")>;
10352 def : BaseSTOPregister<asm # "lh", GPR32, WZR,
10353 !cast<Instruction>(instr # "LH")>;
10354 def : BaseSTOPregister<asm # "l", GPR32, WZR,
10355 !cast<Instruction>(instr # "LW")>;
10356 def : BaseSTOPregister<asm # "l", GPR64, XZR,
10357 !cast<Instruction>(instr # "LX")>;
10358 def : BaseSTOPregister<asm # "b", GPR32, WZR,
10359 !cast<Instruction>(instr # "B")>;
10360 def : BaseSTOPregister<asm # "h", GPR32, WZR,
10361 !cast<Instruction>(instr # "H")>;
10362 def : BaseSTOPregister<asm, GPR32, WZR,
10363 !cast<Instruction>(instr # "W")>;
10364 def : BaseSTOPregister<asm, GPR64, XZR,
10365 !cast<Instruction>(instr # "X")>;
10368 //----------------------------------------------------------------------------
10369 // Allow the size specifier tokens to be upper case, not just lower.
10370 def : TokenAlias<".4B", ".4b">; // Add dot product
10371 def : TokenAlias<".8B", ".8b">;
10372 def : TokenAlias<".4H", ".4h">;
10373 def : TokenAlias<".2S", ".2s">;
10374 def : TokenAlias<".1D", ".1d">;
10375 def : TokenAlias<".16B", ".16b">;
10376 def : TokenAlias<".8H", ".8h">;
10377 def : TokenAlias<".4S", ".4s">;
10378 def : TokenAlias<".2D", ".2d">;
10379 def : TokenAlias<".1Q", ".1q">;
10380 def : TokenAlias<".2H", ".2h">;
10381 def : TokenAlias<".B", ".b">;
10382 def : TokenAlias<".H", ".h">;
10383 def : TokenAlias<".S", ".s">;
10384 def : TokenAlias<".D", ".d">;
10385 def : TokenAlias<".Q", ".q">;