1 //===- ARM64InstrFormats.td - ARM64 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 ARM64 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 // ARM64 Instruction Format
25 class ARM64Inst<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 = "ARM64";
37 bits<2> Form = F.Value;
39 let Constraints = cstr;
42 // Pseudo instructions (don't have encoding information)
43 class Pseudo<dag oops, dag iops, list<dag> pattern, string cstr = "">
44 : ARM64Inst<PseudoFrm, cstr> {
45 dag OutOperandList = oops;
46 dag InOperandList = iops;
47 let Pattern = pattern;
48 let isCodeGenOnly = 1;
51 // Real instructions (have encoding information)
52 class EncodedI<string cstr, list<dag> pattern> : ARM64Inst<NormalFrm, cstr> {
53 let Pattern = pattern;
57 // Normal instructions
58 class I<dag oops, dag iops, string asm, string operands, string cstr,
60 : EncodedI<cstr, pattern> {
61 dag OutOperandList = oops;
62 dag InOperandList = iops;
63 let AsmString = !strconcat(asm, operands);
66 class TriOpFrag<dag res> : PatFrag<(ops node:$LHS, node:$MHS, node:$RHS), res>;
67 class BinOpFrag<dag res> : PatFrag<(ops node:$LHS, node:$RHS), res>;
68 class UnOpFrag<dag res> : PatFrag<(ops node:$LHS), res>;
70 // Helper fragment for an extract of the high portion of a 128-bit vector.
71 def extract_high_v16i8 :
72 UnOpFrag<(extract_subvector (v16i8 node:$LHS), (i64 8))>;
73 def extract_high_v8i16 :
74 UnOpFrag<(extract_subvector (v8i16 node:$LHS), (i64 4))>;
75 def extract_high_v4i32 :
76 UnOpFrag<(extract_subvector (v4i32 node:$LHS), (i64 2))>;
77 def extract_high_v2i64 :
78 UnOpFrag<(extract_subvector (v2i64 node:$LHS), (i64 1))>;
80 //===----------------------------------------------------------------------===//
81 // Asm Operand Classes.
84 // Shifter operand for arithmetic shifted encodings.
85 def ShifterOperand : AsmOperandClass {
89 // Shifter operand for mov immediate encodings.
90 def MovImm32ShifterOperand : AsmOperandClass {
91 let SuperClasses = [ShifterOperand];
92 let Name = "MovImm32Shifter";
94 def MovImm64ShifterOperand : AsmOperandClass {
95 let SuperClasses = [ShifterOperand];
96 let Name = "MovImm64Shifter";
99 // Shifter operand for arithmetic register shifted encodings.
100 def ArithmeticShifterOperand : AsmOperandClass {
101 let SuperClasses = [ShifterOperand];
102 let Name = "ArithmeticShifter";
105 // Shifter operand for arithmetic shifted encodings for ADD/SUB instructions.
106 def AddSubShifterOperand : AsmOperandClass {
107 let SuperClasses = [ArithmeticShifterOperand];
108 let Name = "AddSubShifter";
111 // Shifter operand for logical vector 128/64-bit shifted encodings.
112 def LogicalVecShifterOperand : AsmOperandClass {
113 let SuperClasses = [ShifterOperand];
114 let Name = "LogicalVecShifter";
116 def LogicalVecHalfWordShifterOperand : AsmOperandClass {
117 let SuperClasses = [LogicalVecShifterOperand];
118 let Name = "LogicalVecHalfWordShifter";
121 // The "MSL" shifter on the vector MOVI instruction.
122 def MoveVecShifterOperand : AsmOperandClass {
123 let SuperClasses = [ShifterOperand];
124 let Name = "MoveVecShifter";
127 // Extend operand for arithmetic encodings.
128 def ExtendOperand : AsmOperandClass { let Name = "Extend"; }
129 def ExtendOperand64 : AsmOperandClass {
130 let SuperClasses = [ExtendOperand];
131 let Name = "Extend64";
133 // 'extend' that's a lsl of a 64-bit register.
134 def ExtendOperandLSL64 : AsmOperandClass {
135 let SuperClasses = [ExtendOperand];
136 let Name = "ExtendLSL64";
139 // 8-bit floating-point immediate encodings.
140 def FPImmOperand : AsmOperandClass {
142 let ParserMethod = "tryParseFPImm";
145 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
146 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
147 // are encoded as the eight bit value 'abcdefgh'.
148 def SIMDImmType10Operand : AsmOperandClass { let Name = "SIMDImmType10"; }
151 //===----------------------------------------------------------------------===//
152 // Operand Definitions.
155 // ADR[P] instruction labels.
156 def AdrpOperand : AsmOperandClass {
157 let Name = "AdrpLabel";
158 let ParserMethod = "tryParseAdrpLabel";
159 let DiagnosticType = "InvalidLabel";
161 def adrplabel : Operand<i64> {
162 let EncoderMethod = "getAdrLabelOpValue";
163 let PrintMethod = "printAdrpLabel";
164 let ParserMatchClass = AdrpOperand;
167 def AdrOperand : AsmOperandClass {
168 let Name = "AdrLabel";
169 let ParserMethod = "tryParseAdrLabel";
170 let DiagnosticType = "InvalidLabel";
172 def adrlabel : Operand<i64> {
173 let EncoderMethod = "getAdrLabelOpValue";
174 let ParserMatchClass = AdrOperand;
177 // simm9 predicate - True if the immediate is in the range [-256, 255].
178 def SImm9Operand : AsmOperandClass {
180 let DiagnosticType = "InvalidMemoryIndexedSImm9";
182 def simm9 : Operand<i64>, ImmLeaf<i64, [{ return Imm >= -256 && Imm < 256; }]> {
183 let ParserMatchClass = SImm9Operand;
186 // simm7s4 predicate - True if the immediate is a multiple of 4 in the range
188 def SImm7s4Operand : AsmOperandClass {
189 let Name = "SImm7s4";
190 let DiagnosticType = "InvalidMemoryIndexed32SImm7";
192 def simm7s4 : Operand<i32> {
193 let ParserMatchClass = SImm7s4Operand;
194 let PrintMethod = "printImmScale4";
197 // simm7s8 predicate - True if the immediate is a multiple of 8 in the range
199 def SImm7s8Operand : AsmOperandClass {
200 let Name = "SImm7s8";
201 let DiagnosticType = "InvalidMemoryIndexed64SImm7";
203 def simm7s8 : Operand<i32> {
204 let ParserMatchClass = SImm7s8Operand;
205 let PrintMethod = "printImmScale8";
208 // simm7s16 predicate - True if the immediate is a multiple of 16 in the range
210 def SImm7s16Operand : AsmOperandClass {
211 let Name = "SImm7s16";
212 let DiagnosticType = "InvalidMemoryIndexed64SImm7";
214 def simm7s16 : Operand<i32> {
215 let ParserMatchClass = SImm7s16Operand;
216 let PrintMethod = "printImmScale16";
219 // imm0_65535 predicate - True if the immediate is in the range [0,65535].
220 def Imm0_65535Operand : AsmOperandClass { let Name = "Imm0_65535"; }
221 def imm0_65535 : Operand<i32>, ImmLeaf<i32, [{
222 return ((uint32_t)Imm) < 65536;
224 let ParserMatchClass = Imm0_65535Operand;
227 def Imm1_8Operand : AsmOperandClass {
229 let DiagnosticType = "InvalidImm1_8";
231 def Imm1_16Operand : AsmOperandClass {
232 let Name = "Imm1_16";
233 let DiagnosticType = "InvalidImm1_16";
235 def Imm1_32Operand : AsmOperandClass {
236 let Name = "Imm1_32";
237 let DiagnosticType = "InvalidImm1_32";
239 def Imm1_64Operand : AsmOperandClass {
240 let Name = "Imm1_64";
241 let DiagnosticType = "InvalidImm1_64";
244 def MovZSymbolG3AsmOperand : AsmOperandClass {
245 let Name = "MovZSymbolG3";
246 let RenderMethod = "addImmOperands";
249 def movz_symbol_g3 : Operand<i32> {
250 let ParserMatchClass = MovZSymbolG3AsmOperand;
253 def MovZSymbolG2AsmOperand : AsmOperandClass {
254 let Name = "MovZSymbolG2";
255 let RenderMethod = "addImmOperands";
258 def movz_symbol_g2 : Operand<i32> {
259 let ParserMatchClass = MovZSymbolG2AsmOperand;
262 def MovZSymbolG1AsmOperand : AsmOperandClass {
263 let Name = "MovZSymbolG1";
264 let RenderMethod = "addImmOperands";
267 def movz_symbol_g1 : Operand<i32> {
268 let ParserMatchClass = MovZSymbolG1AsmOperand;
271 def MovZSymbolG0AsmOperand : AsmOperandClass {
272 let Name = "MovZSymbolG0";
273 let RenderMethod = "addImmOperands";
276 def movz_symbol_g0 : Operand<i32> {
277 let ParserMatchClass = MovZSymbolG0AsmOperand;
280 def MovKSymbolG3AsmOperand : AsmOperandClass {
281 let Name = "MovKSymbolG3";
282 let RenderMethod = "addImmOperands";
285 def movk_symbol_g3 : Operand<i32> {
286 let ParserMatchClass = MovKSymbolG3AsmOperand;
289 def MovKSymbolG2AsmOperand : AsmOperandClass {
290 let Name = "MovKSymbolG2";
291 let RenderMethod = "addImmOperands";
294 def movk_symbol_g2 : Operand<i32> {
295 let ParserMatchClass = MovKSymbolG2AsmOperand;
298 def MovKSymbolG1AsmOperand : AsmOperandClass {
299 let Name = "MovKSymbolG1";
300 let RenderMethod = "addImmOperands";
303 def movk_symbol_g1 : Operand<i32> {
304 let ParserMatchClass = MovKSymbolG1AsmOperand;
307 def MovKSymbolG0AsmOperand : AsmOperandClass {
308 let Name = "MovKSymbolG0";
309 let RenderMethod = "addImmOperands";
312 def movk_symbol_g0 : Operand<i32> {
313 let ParserMatchClass = MovKSymbolG0AsmOperand;
316 class fixedpoint_i32<ValueType FloatVT>
318 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<32>", [fpimm, ld]> {
319 let EncoderMethod = "getFixedPointScaleOpValue";
320 let DecoderMethod = "DecodeFixedPointScaleImm32";
321 let ParserMatchClass = Imm1_32Operand;
324 class fixedpoint_i64<ValueType FloatVT>
326 ComplexPattern<FloatVT, 1, "SelectCVTFixedPosOperand<64>", [fpimm, ld]> {
327 let EncoderMethod = "getFixedPointScaleOpValue";
328 let DecoderMethod = "DecodeFixedPointScaleImm64";
329 let ParserMatchClass = Imm1_64Operand;
332 def fixedpoint_f32_i32 : fixedpoint_i32<f32>;
333 def fixedpoint_f64_i32 : fixedpoint_i32<f64>;
335 def fixedpoint_f32_i64 : fixedpoint_i64<f32>;
336 def fixedpoint_f64_i64 : fixedpoint_i64<f64>;
338 def vecshiftR8 : Operand<i32>, ImmLeaf<i32, [{
339 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
341 let EncoderMethod = "getVecShiftR8OpValue";
342 let DecoderMethod = "DecodeVecShiftR8Imm";
343 let ParserMatchClass = Imm1_8Operand;
345 def vecshiftR16 : Operand<i32>, ImmLeaf<i32, [{
346 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
348 let EncoderMethod = "getVecShiftR16OpValue";
349 let DecoderMethod = "DecodeVecShiftR16Imm";
350 let ParserMatchClass = Imm1_16Operand;
352 def vecshiftR16Narrow : Operand<i32>, ImmLeaf<i32, [{
353 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
355 let EncoderMethod = "getVecShiftR16OpValue";
356 let DecoderMethod = "DecodeVecShiftR16ImmNarrow";
357 let ParserMatchClass = Imm1_8Operand;
359 def vecshiftR32 : Operand<i32>, ImmLeaf<i32, [{
360 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
362 let EncoderMethod = "getVecShiftR32OpValue";
363 let DecoderMethod = "DecodeVecShiftR32Imm";
364 let ParserMatchClass = Imm1_32Operand;
366 def vecshiftR32Narrow : Operand<i32>, ImmLeaf<i32, [{
367 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
369 let EncoderMethod = "getVecShiftR32OpValue";
370 let DecoderMethod = "DecodeVecShiftR32ImmNarrow";
371 let ParserMatchClass = Imm1_16Operand;
373 def vecshiftR64 : Operand<i32>, ImmLeaf<i32, [{
374 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 65);
376 let EncoderMethod = "getVecShiftR64OpValue";
377 let DecoderMethod = "DecodeVecShiftR64Imm";
378 let ParserMatchClass = Imm1_64Operand;
380 def vecshiftR64Narrow : Operand<i32>, ImmLeaf<i32, [{
381 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
383 let EncoderMethod = "getVecShiftR64OpValue";
384 let DecoderMethod = "DecodeVecShiftR64ImmNarrow";
385 let ParserMatchClass = Imm1_32Operand;
388 def Imm0_7Operand : AsmOperandClass { let Name = "Imm0_7"; }
389 def Imm0_15Operand : AsmOperandClass { let Name = "Imm0_15"; }
390 def Imm0_31Operand : AsmOperandClass { let Name = "Imm0_31"; }
391 def Imm0_63Operand : AsmOperandClass { let Name = "Imm0_63"; }
393 def vecshiftL8 : Operand<i32>, ImmLeaf<i32, [{
394 return (((uint32_t)Imm) < 8);
396 let EncoderMethod = "getVecShiftL8OpValue";
397 let DecoderMethod = "DecodeVecShiftL8Imm";
398 let ParserMatchClass = Imm0_7Operand;
400 def vecshiftL16 : Operand<i32>, ImmLeaf<i32, [{
401 return (((uint32_t)Imm) < 16);
403 let EncoderMethod = "getVecShiftL16OpValue";
404 let DecoderMethod = "DecodeVecShiftL16Imm";
405 let ParserMatchClass = Imm0_15Operand;
407 def vecshiftL32 : Operand<i32>, ImmLeaf<i32, [{
408 return (((uint32_t)Imm) < 32);
410 let EncoderMethod = "getVecShiftL32OpValue";
411 let DecoderMethod = "DecodeVecShiftL32Imm";
412 let ParserMatchClass = Imm0_31Operand;
414 def vecshiftL64 : Operand<i32>, ImmLeaf<i32, [{
415 return (((uint32_t)Imm) < 64);
417 let EncoderMethod = "getVecShiftL64OpValue";
418 let DecoderMethod = "DecodeVecShiftL64Imm";
419 let ParserMatchClass = Imm0_63Operand;
423 // Crazy immediate formats used by 32-bit and 64-bit logical immediate
424 // instructions for splatting repeating bit patterns across the immediate.
425 def logical_imm32_XFORM : SDNodeXForm<imm, [{
426 uint64_t enc = ARM64_AM::encodeLogicalImmediate(N->getZExtValue(), 32);
427 return CurDAG->getTargetConstant(enc, MVT::i32);
429 def logical_imm64_XFORM : SDNodeXForm<imm, [{
430 uint64_t enc = ARM64_AM::encodeLogicalImmediate(N->getZExtValue(), 64);
431 return CurDAG->getTargetConstant(enc, MVT::i32);
434 def LogicalImm32Operand : AsmOperandClass { let Name = "LogicalImm32"; }
435 def LogicalImm64Operand : AsmOperandClass { let Name = "LogicalImm64"; }
436 def logical_imm32 : Operand<i32>, PatLeaf<(imm), [{
437 return ARM64_AM::isLogicalImmediate(N->getZExtValue(), 32);
438 }], logical_imm32_XFORM> {
439 let PrintMethod = "printLogicalImm32";
440 let ParserMatchClass = LogicalImm32Operand;
442 def logical_imm64 : Operand<i64>, PatLeaf<(imm), [{
443 return ARM64_AM::isLogicalImmediate(N->getZExtValue(), 64);
444 }], logical_imm64_XFORM> {
445 let PrintMethod = "printLogicalImm64";
446 let ParserMatchClass = LogicalImm64Operand;
449 // imm0_255 predicate - True if the immediate is in the range [0,255].
450 def Imm0_255Operand : AsmOperandClass { let Name = "Imm0_255"; }
451 def imm0_255 : Operand<i32>, ImmLeaf<i32, [{
452 return ((uint32_t)Imm) < 256;
454 let ParserMatchClass = Imm0_255Operand;
457 // imm0_127 predicate - True if the immediate is in the range [0,127]
458 def Imm0_127Operand : AsmOperandClass { let Name = "Imm0_127"; }
459 def imm0_127 : Operand<i32>, ImmLeaf<i32, [{
460 return ((uint32_t)Imm) < 128;
462 let ParserMatchClass = Imm0_127Operand;
465 // NOTE: These imm0_N operands have to be of type i64 because i64 is the size
466 // for all shift-amounts.
468 // imm0_63 predicate - True if the immediate is in the range [0,63]
469 def imm0_63 : Operand<i64>, ImmLeaf<i64, [{
470 return ((uint64_t)Imm) < 64;
472 let ParserMatchClass = Imm0_63Operand;
475 // imm0_31 predicate - True if the immediate is in the range [0,31]
476 def imm0_31 : Operand<i64>, ImmLeaf<i64, [{
477 return ((uint64_t)Imm) < 32;
479 let ParserMatchClass = Imm0_31Operand;
482 // imm0_15 predicate - True if the immediate is in the range [0,15]
483 def imm0_15 : Operand<i64>, ImmLeaf<i64, [{
484 return ((uint64_t)Imm) < 16;
486 let ParserMatchClass = Imm0_15Operand;
489 // imm0_7 predicate - True if the immediate is in the range [0,7]
490 def imm0_7 : Operand<i64>, ImmLeaf<i64, [{
491 return ((uint64_t)Imm) < 8;
493 let ParserMatchClass = Imm0_7Operand;
496 // An arithmetic shifter operand:
497 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr
499 def arith_shift : Operand<i32> {
500 let PrintMethod = "printShifter";
501 let ParserMatchClass = ArithmeticShifterOperand;
504 class arith_shifted_reg<ValueType Ty, RegisterClass regclass>
506 ComplexPattern<Ty, 2, "SelectArithShiftedRegister", []> {
507 let PrintMethod = "printShiftedRegister";
508 let MIOperandInfo = (ops regclass, arith_shift);
511 def arith_shifted_reg32 : arith_shifted_reg<i32, GPR32>;
512 def arith_shifted_reg64 : arith_shifted_reg<i64, GPR64>;
514 // An arithmetic shifter operand:
515 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr, 11 = ror
517 def logical_shift : Operand<i32> {
518 let PrintMethod = "printShifter";
519 let ParserMatchClass = ShifterOperand;
522 class logical_shifted_reg<ValueType Ty, RegisterClass regclass>
524 ComplexPattern<Ty, 2, "SelectLogicalShiftedRegister", []> {
525 let PrintMethod = "printShiftedRegister";
526 let MIOperandInfo = (ops regclass, logical_shift);
529 def logical_shifted_reg32 : logical_shifted_reg<i32, GPR32>;
530 def logical_shifted_reg64 : logical_shifted_reg<i64, GPR64>;
532 // A logical vector shifter operand:
533 // {7-6} - shift type: 00 = lsl
534 // {5-0} - imm6: #0, #8, #16, or #24
535 def logical_vec_shift : Operand<i32> {
536 let PrintMethod = "printShifter";
537 let EncoderMethod = "getVecShifterOpValue";
538 let ParserMatchClass = LogicalVecShifterOperand;
541 // A logical vector half-word shifter operand:
542 // {7-6} - shift type: 00 = lsl
543 // {5-0} - imm6: #0 or #8
544 def logical_vec_hw_shift : Operand<i32> {
545 let PrintMethod = "printShifter";
546 let EncoderMethod = "getVecShifterOpValue";
547 let ParserMatchClass = LogicalVecHalfWordShifterOperand;
550 // A vector move shifter operand:
551 // {0} - imm1: #8 or #16
552 def move_vec_shift : Operand<i32> {
553 let PrintMethod = "printShifter";
554 let EncoderMethod = "getMoveVecShifterOpValue";
555 let ParserMatchClass = MoveVecShifterOperand;
558 // An ADD/SUB immediate shifter operand:
559 // {7-6} - shift type: 00 = lsl
560 // {5-0} - imm6: #0 or #12
561 def addsub_shift : Operand<i32> {
562 let ParserMatchClass = AddSubShifterOperand;
565 class addsub_shifted_imm<ValueType Ty>
566 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectArithImmed", [imm]> {
567 let PrintMethod = "printAddSubImm";
568 let EncoderMethod = "getAddSubImmOpValue";
569 let MIOperandInfo = (ops i32imm, addsub_shift);
572 def addsub_shifted_imm32 : addsub_shifted_imm<i32>;
573 def addsub_shifted_imm64 : addsub_shifted_imm<i64>;
575 class neg_addsub_shifted_imm<ValueType Ty>
576 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectNegArithImmed", [imm]> {
577 let PrintMethod = "printAddSubImm";
578 let EncoderMethod = "getAddSubImmOpValue";
579 let MIOperandInfo = (ops i32imm, addsub_shift);
582 def neg_addsub_shifted_imm32 : neg_addsub_shifted_imm<i32>;
583 def neg_addsub_shifted_imm64 : neg_addsub_shifted_imm<i64>;
585 // An extend operand:
586 // {5-3} - extend type
588 def arith_extend : Operand<i32> {
589 let PrintMethod = "printExtend";
590 let ParserMatchClass = ExtendOperand;
592 def arith_extend64 : Operand<i32> {
593 let PrintMethod = "printExtend";
594 let ParserMatchClass = ExtendOperand64;
597 // 'extend' that's a lsl of a 64-bit register.
598 def arith_extendlsl64 : Operand<i32> {
599 let PrintMethod = "printExtend";
600 let ParserMatchClass = ExtendOperandLSL64;
603 class arith_extended_reg32<ValueType Ty> : Operand<Ty>,
604 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
605 let PrintMethod = "printExtendedRegister";
606 let MIOperandInfo = (ops GPR32, arith_extend);
609 class arith_extended_reg32to64<ValueType Ty> : Operand<Ty>,
610 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
611 let PrintMethod = "printExtendedRegister";
612 let MIOperandInfo = (ops GPR32, arith_extend64);
615 // Floating-point immediate.
616 def fpimm32 : Operand<f32>,
617 PatLeaf<(f32 fpimm), [{
618 return ARM64_AM::getFP32Imm(N->getValueAPF()) != -1;
619 }], SDNodeXForm<fpimm, [{
620 APFloat InVal = N->getValueAPF();
621 uint32_t enc = ARM64_AM::getFP32Imm(InVal);
622 return CurDAG->getTargetConstant(enc, MVT::i32);
624 let ParserMatchClass = FPImmOperand;
625 let PrintMethod = "printFPImmOperand";
627 def fpimm64 : Operand<f64>,
628 PatLeaf<(f64 fpimm), [{
629 return ARM64_AM::getFP64Imm(N->getValueAPF()) != -1;
630 }], SDNodeXForm<fpimm, [{
631 APFloat InVal = N->getValueAPF();
632 uint32_t enc = ARM64_AM::getFP64Imm(InVal);
633 return CurDAG->getTargetConstant(enc, MVT::i32);
635 let ParserMatchClass = FPImmOperand;
636 let PrintMethod = "printFPImmOperand";
639 def fpimm8 : Operand<i32> {
640 let ParserMatchClass = FPImmOperand;
641 let PrintMethod = "printFPImmOperand";
644 def fpimm0 : PatLeaf<(fpimm), [{
645 return N->isExactlyValue(+0.0);
648 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
649 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
650 // are encoded as the eight bit value 'abcdefgh'.
651 def simdimmtype10 : Operand<i32>,
652 PatLeaf<(f64 fpimm), [{
653 return ARM64_AM::isAdvSIMDModImmType10(N->getValueAPF()
656 }], SDNodeXForm<fpimm, [{
657 APFloat InVal = N->getValueAPF();
658 uint32_t enc = ARM64_AM::encodeAdvSIMDModImmType10(N->getValueAPF()
661 return CurDAG->getTargetConstant(enc, MVT::i32);
663 let ParserMatchClass = SIMDImmType10Operand;
664 let PrintMethod = "printSIMDType10Operand";
672 // Base encoding for system instruction operands.
673 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
674 class BaseSystemI<bit L, dag oops, dag iops, string asm, string operands>
675 : I<oops, iops, asm, operands, "", []> {
676 let Inst{31-22} = 0b1101010100;
680 // System instructions which do not have an Rt register.
681 class SimpleSystemI<bit L, dag iops, string asm, string operands>
682 : BaseSystemI<L, (outs), iops, asm, operands> {
683 let Inst{4-0} = 0b11111;
686 // System instructions which have an Rt register.
687 class RtSystemI<bit L, dag oops, dag iops, string asm, string operands>
688 : BaseSystemI<L, oops, iops, asm, operands>,
694 // Hint instructions that take both a CRm and a 3-bit immediate.
695 class HintI<string mnemonic>
696 : SimpleSystemI<0, (ins imm0_127:$imm), mnemonic#" $imm", "">,
699 let Inst{20-12} = 0b000110010;
700 let Inst{11-5} = imm;
703 // System instructions taking a single literal operand which encodes into
704 // CRm. op2 differentiates the opcodes.
705 def BarrierAsmOperand : AsmOperandClass {
706 let Name = "Barrier";
707 let ParserMethod = "tryParseBarrierOperand";
709 def barrier_op : Operand<i32> {
710 let PrintMethod = "printBarrierOption";
711 let ParserMatchClass = BarrierAsmOperand;
713 class CRmSystemI<Operand crmtype, bits<3> opc, string asm>
714 : SimpleSystemI<0, (ins crmtype:$CRm), asm, "\t$CRm">,
715 Sched<[WriteBarrier]> {
717 let Inst{20-12} = 0b000110011;
718 let Inst{11-8} = CRm;
722 // MRS/MSR system instructions. These have different operand classes because
723 // a different subset of registers can be accessed through each instruction.
724 def MRSSystemRegisterOperand : AsmOperandClass {
725 let Name = "MRSSystemRegister";
726 let ParserMethod = "tryParseSysReg";
728 // concatenation of 1, op0, op1, CRn, CRm, op2. 16-bit immediate.
729 def mrs_sysreg_op : Operand<i32> {
730 let ParserMatchClass = MRSSystemRegisterOperand;
731 let DecoderMethod = "DecodeMRSSystemRegister";
732 let PrintMethod = "printMRSSystemRegister";
735 def MSRSystemRegisterOperand : AsmOperandClass {
736 let Name = "MSRSystemRegister";
737 let ParserMethod = "tryParseSysReg";
739 def msr_sysreg_op : Operand<i32> {
740 let ParserMatchClass = MSRSystemRegisterOperand;
741 let DecoderMethod = "DecodeMSRSystemRegister";
742 let PrintMethod = "printMSRSystemRegister";
745 class MRSI : RtSystemI<1, (outs GPR64:$Rt), (ins mrs_sysreg_op:$systemreg),
746 "mrs", "\t$Rt, $systemreg"> {
749 let Inst{19-5} = systemreg;
752 // FIXME: Some of these def CPSR, others don't. Best way to model that?
753 // Explicitly modeling each of the system register as a register class
754 // would do it, but feels like overkill at this point.
755 class MSRI : RtSystemI<0, (outs), (ins msr_sysreg_op:$systemreg, GPR64:$Rt),
756 "msr", "\t$systemreg, $Rt"> {
759 let Inst{19-5} = systemreg;
762 def SystemCPSRFieldOperand : AsmOperandClass {
763 let Name = "SystemCPSRField";
764 let ParserMethod = "tryParseSysReg";
766 def cpsrfield_op : Operand<i32> {
767 let ParserMatchClass = SystemCPSRFieldOperand;
768 let PrintMethod = "printSystemCPSRField";
772 class MSRcpsrI : SimpleSystemI<0, (ins cpsrfield_op:$cpsr_field, imm0_15:$imm),
773 "msr", "\t$cpsr_field, $imm">,
777 let Inst{20-19} = 0b00;
778 let Inst{18-16} = cpsrfield{5-3};
779 let Inst{15-12} = 0b0100;
780 let Inst{11-8} = imm;
781 let Inst{7-5} = cpsrfield{2-0};
783 let DecoderMethod = "DecodeSystemCPSRInstruction";
786 // SYS and SYSL generic system instructions.
787 def SysCRAsmOperand : AsmOperandClass {
789 let ParserMethod = "tryParseSysCROperand";
792 def sys_cr_op : Operand<i32> {
793 let PrintMethod = "printSysCROperand";
794 let ParserMatchClass = SysCRAsmOperand;
797 class SystemXtI<bit L, string asm>
798 : RtSystemI<L, (outs),
799 (ins imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, GPR64:$Rt),
800 asm, "\t$op1, $Cn, $Cm, $op2, $Rt"> {
805 let Inst{20-19} = 0b01;
806 let Inst{18-16} = op1;
807 let Inst{15-12} = Cn;
812 class SystemLXtI<bit L, string asm>
813 : RtSystemI<L, (outs),
814 (ins GPR64:$Rt, imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2),
815 asm, "\t$Rt, $op1, $Cn, $Cm, $op2"> {
820 let Inst{20-19} = 0b01;
821 let Inst{18-16} = op1;
822 let Inst{15-12} = Cn;
828 // Branch (register) instructions:
836 // otherwise UNDEFINED
837 class BaseBranchReg<bits<4> opc, dag oops, dag iops, string asm,
838 string operands, list<dag> pattern>
839 : I<oops, iops, asm, operands, "", pattern>, Sched<[WriteBrReg]> {
840 let Inst{31-25} = 0b1101011;
841 let Inst{24-21} = opc;
842 let Inst{20-16} = 0b11111;
843 let Inst{15-10} = 0b000000;
844 let Inst{4-0} = 0b00000;
847 class BranchReg<bits<4> opc, string asm, list<dag> pattern>
848 : BaseBranchReg<opc, (outs), (ins GPR64:$Rn), asm, "\t$Rn", pattern> {
853 let mayLoad = 0, mayStore = 0, hasSideEffects = 1, isReturn = 1 in
854 class SpecialReturn<bits<4> opc, string asm>
855 : BaseBranchReg<opc, (outs), (ins), asm, "", []> {
856 let Inst{9-5} = 0b11111;
860 // Conditional branch instruction.
862 // Branch condition code.
863 // 4-bit immediate. Pretty-printed as .<cc>
864 def dotCcode : Operand<i32> {
865 let PrintMethod = "printDotCondCode";
868 // Conditional branch target. 19-bit immediate. The low two bits of the target
869 // offset are implied zero and so are not part of the immediate.
870 def PCRelLabel19Operand : AsmOperandClass {
871 let Name = "PCRelLabel19";
873 def am_brcond : Operand<OtherVT> {
874 let EncoderMethod = "getCondBranchTargetOpValue";
875 let DecoderMethod = "DecodePCRelLabel19";
876 let PrintMethod = "printAlignedLabel";
877 let ParserMatchClass = PCRelLabel19Operand;
880 class BranchCond : I<(outs), (ins dotCcode:$cond, am_brcond:$target),
881 "b", "$cond\t$target", "",
882 [(ARM64brcond bb:$target, imm:$cond, CPSR)]>,
885 let isTerminator = 1;
890 let Inst{31-24} = 0b01010100;
891 let Inst{23-5} = target;
893 let Inst{3-0} = cond;
897 // Compare-and-branch instructions.
899 class BaseCmpBranch<RegisterClass regtype, bit op, string asm, SDNode node>
900 : I<(outs), (ins regtype:$Rt, am_brcond:$target),
901 asm, "\t$Rt, $target", "",
902 [(node regtype:$Rt, bb:$target)]>,
905 let isTerminator = 1;
909 let Inst{30-25} = 0b011010;
911 let Inst{23-5} = target;
915 multiclass CmpBranch<bit op, string asm, SDNode node> {
916 def W : BaseCmpBranch<GPR32, op, asm, node> {
919 def X : BaseCmpBranch<GPR64, op, asm, node> {
925 // Test-bit-and-branch instructions.
927 // Test-and-branch target. 14-bit sign-extended immediate. The low two bits of
928 // the target offset are implied zero and so are not part of the immediate.
929 def BranchTarget14Operand : AsmOperandClass {
930 let Name = "BranchTarget14";
932 def am_tbrcond : Operand<OtherVT> {
933 let EncoderMethod = "getTestBranchTargetOpValue";
934 let PrintMethod = "printAlignedLabel";
935 let ParserMatchClass = BranchTarget14Operand;
938 class TestBranch<bit op, string asm, SDNode node>
939 : I<(outs), (ins GPR64:$Rt, imm0_63:$bit_off, am_tbrcond:$target),
940 asm, "\t$Rt, $bit_off, $target", "",
941 [(node GPR64:$Rt, imm0_63:$bit_off, bb:$target)]>,
944 let isTerminator = 1;
950 let Inst{31} = bit_off{5};
951 let Inst{30-25} = 0b011011;
953 let Inst{23-19} = bit_off{4-0};
954 let Inst{18-5} = target;
957 let DecoderMethod = "DecodeTestAndBranch";
961 // Unconditional branch (immediate) instructions.
963 def BranchTarget26Operand : AsmOperandClass {
964 let Name = "BranchTarget26";
966 def am_b_target : Operand<OtherVT> {
967 let EncoderMethod = "getBranchTargetOpValue";
968 let PrintMethod = "printAlignedLabel";
969 let ParserMatchClass = BranchTarget26Operand;
971 def am_bl_target : Operand<i64> {
972 let EncoderMethod = "getBranchTargetOpValue";
973 let PrintMethod = "printAlignedLabel";
974 let ParserMatchClass = BranchTarget26Operand;
977 class BImm<bit op, dag iops, string asm, list<dag> pattern>
978 : I<(outs), iops, asm, "\t$addr", "", pattern>, Sched<[WriteBr]> {
981 let Inst{30-26} = 0b00101;
982 let Inst{25-0} = addr;
984 let DecoderMethod = "DecodeUnconditionalBranch";
987 class BranchImm<bit op, string asm, list<dag> pattern>
988 : BImm<op, (ins am_b_target:$addr), asm, pattern>;
989 class CallImm<bit op, string asm, list<dag> pattern>
990 : BImm<op, (ins am_bl_target:$addr), asm, pattern>;
993 // Basic one-operand data processing instructions.
996 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
997 class BaseOneOperandData<bits<3> opc, RegisterClass regtype, string asm,
998 SDPatternOperator node>
999 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
1000 [(set regtype:$Rd, (node regtype:$Rn))]>,
1005 let Inst{30-13} = 0b101101011000000000;
1006 let Inst{12-10} = opc;
1011 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1012 multiclass OneOperandData<bits<3> opc, string asm,
1013 SDPatternOperator node = null_frag> {
1014 def Wr : BaseOneOperandData<opc, GPR32, asm, node> {
1018 def Xr : BaseOneOperandData<opc, GPR64, asm, node> {
1023 class OneWRegData<bits<3> opc, string asm, SDPatternOperator node>
1024 : BaseOneOperandData<opc, GPR32, asm, node> {
1028 class OneXRegData<bits<3> opc, string asm, SDPatternOperator node>
1029 : BaseOneOperandData<opc, GPR64, asm, node> {
1034 // Basic two-operand data processing instructions.
1036 class BaseBaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1038 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1039 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
1045 let Inst{30} = isSub;
1046 let Inst{28-21} = 0b11010000;
1047 let Inst{20-16} = Rm;
1048 let Inst{15-10} = 0;
1053 class BaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1055 : BaseBaseAddSubCarry<isSub, regtype, asm,
1056 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, CPSR))]>;
1058 class BaseAddSubCarrySetFlags<bit isSub, RegisterClass regtype, string asm,
1060 : BaseBaseAddSubCarry<isSub, regtype, asm,
1061 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, CPSR)),
1066 multiclass AddSubCarry<bit isSub, string asm, string asm_setflags,
1067 SDNode OpNode, SDNode OpNode_setflags> {
1068 def Wr : BaseAddSubCarry<isSub, GPR32, asm, OpNode> {
1072 def Xr : BaseAddSubCarry<isSub, GPR64, asm, OpNode> {
1078 def SWr : BaseAddSubCarrySetFlags<isSub, GPR32, asm_setflags,
1083 def SXr : BaseAddSubCarrySetFlags<isSub, GPR64, asm_setflags,
1090 class BaseTwoOperand<bits<4> opc, RegisterClass regtype, string asm,
1091 SDPatternOperator OpNode>
1092 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1093 asm, "\t$Rd, $Rn, $Rm", "",
1094 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]> {
1098 let Inst{30-21} = 0b0011010110;
1099 let Inst{20-16} = Rm;
1100 let Inst{15-14} = 0b00;
1101 let Inst{13-10} = opc;
1106 class BaseDiv<bit isSigned, RegisterClass regtype, string asm,
1107 SDPatternOperator OpNode>
1108 : BaseTwoOperand<{0,0,1,?}, regtype, asm, OpNode> {
1109 let Inst{10} = isSigned;
1112 multiclass Div<bit isSigned, string asm, SDPatternOperator OpNode> {
1113 def Wr : BaseDiv<isSigned, GPR32, asm, OpNode>,
1114 Sched<[WriteID32]> {
1117 def Xr : BaseDiv<isSigned, GPR64, asm, OpNode>,
1118 Sched<[WriteID64]> {
1123 class BaseShift<bits<2> shift_type, RegisterClass regtype, string asm,
1124 SDPatternOperator OpNode = null_frag>
1125 : BaseTwoOperand<{1,0,?,?}, regtype, asm, OpNode>,
1127 let Inst{11-10} = shift_type;
1130 multiclass Shift<bits<2> shift_type, string asm, SDNode OpNode> {
1131 def Wr : BaseShift<shift_type, GPR32, asm> {
1135 def Xr : BaseShift<shift_type, GPR64, asm, OpNode> {
1139 def : Pat<(i32 (OpNode GPR32:$Rn, i64:$Rm)),
1140 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn,
1141 (EXTRACT_SUBREG i64:$Rm, sub_32))>;
1143 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (zext GPR32:$Rm)))),
1144 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1146 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (anyext GPR32:$Rm)))),
1147 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1149 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (sext GPR32:$Rm)))),
1150 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1153 class ShiftAlias<string asm, Instruction inst, RegisterClass regtype>
1154 : InstAlias<asm#" $dst, $src1, $src2",
1155 (inst regtype:$dst, regtype:$src1, regtype:$src2)>;
1157 class BaseMulAccum<bit isSub, bits<3> opc, RegisterClass multype,
1158 RegisterClass addtype, string asm,
1160 : I<(outs addtype:$Rd), (ins multype:$Rn, multype:$Rm, addtype:$Ra),
1161 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pattern> {
1166 let Inst{30-24} = 0b0011011;
1167 let Inst{23-21} = opc;
1168 let Inst{20-16} = Rm;
1169 let Inst{15} = isSub;
1170 let Inst{14-10} = Ra;
1175 multiclass MulAccum<bit isSub, string asm, SDNode AccNode> {
1176 def Wrrr : BaseMulAccum<isSub, 0b000, GPR32, GPR32, asm,
1177 [(set GPR32:$Rd, (AccNode GPR32:$Ra, (mul GPR32:$Rn, GPR32:$Rm)))]>,
1178 Sched<[WriteIM32]> {
1182 def Xrrr : BaseMulAccum<isSub, 0b000, GPR64, GPR64, asm,
1183 [(set GPR64:$Rd, (AccNode GPR64:$Ra, (mul GPR64:$Rn, GPR64:$Rm)))]>,
1184 Sched<[WriteIM64]> {
1189 class WideMulAccum<bit isSub, bits<3> opc, string asm,
1190 SDNode AccNode, SDNode ExtNode>
1191 : BaseMulAccum<isSub, opc, GPR32, GPR64, asm,
1192 [(set GPR64:$Rd, (AccNode GPR64:$Ra,
1193 (mul (ExtNode GPR32:$Rn), (ExtNode GPR32:$Rm))))]>,
1194 Sched<[WriteIM32]> {
1198 class MulHi<bits<3> opc, string asm, SDNode OpNode>
1199 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64:$Rm),
1200 asm, "\t$Rd, $Rn, $Rm", "",
1201 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64:$Rm))]>,
1202 Sched<[WriteIM64]> {
1206 let Inst{31-24} = 0b10011011;
1207 let Inst{23-21} = opc;
1208 let Inst{20-16} = Rm;
1213 // The Ra field of SMULH and UMULH is unused: it should be assembled as 31
1214 // (i.e. all bits 1) but is ignored by the processor.
1215 let PostEncoderMethod = "fixMulHigh";
1218 class MulAccumWAlias<string asm, Instruction inst>
1219 : InstAlias<asm#" $dst, $src1, $src2",
1220 (inst GPR32:$dst, GPR32:$src1, GPR32:$src2, WZR)>;
1221 class MulAccumXAlias<string asm, Instruction inst>
1222 : InstAlias<asm#" $dst, $src1, $src2",
1223 (inst GPR64:$dst, GPR64:$src1, GPR64:$src2, XZR)>;
1224 class WideMulAccumAlias<string asm, Instruction inst>
1225 : InstAlias<asm#" $dst, $src1, $src2",
1226 (inst GPR64:$dst, GPR32:$src1, GPR32:$src2, XZR)>;
1228 class BaseCRC32<bit sf, bits<2> sz, bit C, RegisterClass StreamReg,
1229 SDPatternOperator OpNode, string asm>
1230 : I<(outs GPR32:$Rd), (ins GPR32:$Rn, StreamReg:$Rm),
1231 asm, "\t$Rd, $Rn, $Rm", "",
1232 [(set GPR32:$Rd, (OpNode GPR32:$Rn, StreamReg:$Rm))]>,
1233 Sched<[WriteISReg]> {
1239 let Inst{30-21} = 0b0011010110;
1240 let Inst{20-16} = Rm;
1241 let Inst{15-13} = 0b010;
1243 let Inst{11-10} = sz;
1249 // Address generation.
1252 class ADRI<bit page, string asm, Operand adr, list<dag> pattern>
1253 : I<(outs GPR64:$Xd), (ins adr:$label), asm, "\t$Xd, $label", "",
1258 let Inst{31} = page;
1259 let Inst{30-29} = label{1-0};
1260 let Inst{28-24} = 0b10000;
1261 let Inst{23-5} = label{20-2};
1264 let DecoderMethod = "DecodeAdrInstruction";
1271 def movimm32_imm : Operand<i32> {
1272 let ParserMatchClass = Imm0_65535Operand;
1273 let EncoderMethod = "getMoveWideImmOpValue";
1275 def movimm32_shift : Operand<i32> {
1276 let PrintMethod = "printShifter";
1277 let ParserMatchClass = MovImm32ShifterOperand;
1279 def movimm64_shift : Operand<i32> {
1280 let PrintMethod = "printShifter";
1281 let ParserMatchClass = MovImm64ShifterOperand;
1283 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1284 class BaseMoveImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1286 : I<(outs regtype:$Rd), (ins movimm32_imm:$imm, shifter:$shift),
1287 asm, "\t$Rd, $imm$shift", "", []>,
1292 let Inst{30-29} = opc;
1293 let Inst{28-23} = 0b100101;
1294 let Inst{22-21} = shift{5-4};
1295 let Inst{20-5} = imm;
1298 let DecoderMethod = "DecodeMoveImmInstruction";
1301 multiclass MoveImmediate<bits<2> opc, string asm> {
1302 def Wi : BaseMoveImmediate<opc, GPR32, movimm32_shift, asm> {
1306 def Xi : BaseMoveImmediate<opc, GPR64, movimm64_shift, asm> {
1311 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1312 class BaseInsertImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1314 : I<(outs regtype:$Rd),
1315 (ins regtype:$src, movimm32_imm:$imm, shifter:$shift),
1316 asm, "\t$Rd, $imm$shift", "$src = $Rd", []>,
1321 let Inst{30-29} = opc;
1322 let Inst{28-23} = 0b100101;
1323 let Inst{22-21} = shift{5-4};
1324 let Inst{20-5} = imm;
1327 let DecoderMethod = "DecodeMoveImmInstruction";
1330 multiclass InsertImmediate<bits<2> opc, string asm> {
1331 def Wi : BaseInsertImmediate<opc, GPR32, movimm32_shift, asm> {
1335 def Xi : BaseInsertImmediate<opc, GPR64, movimm64_shift, asm> {
1344 class BaseAddSubImm<bit isSub, bit setFlags, RegisterClass dstRegtype,
1345 RegisterClass srcRegtype, addsub_shifted_imm immtype,
1346 string asm, SDPatternOperator OpNode>
1347 : I<(outs dstRegtype:$Rd), (ins srcRegtype:$Rn, immtype:$imm),
1348 asm, "\t$Rd, $Rn, $imm", "",
1349 [(set dstRegtype:$Rd, (OpNode srcRegtype:$Rn, immtype:$imm))]>,
1354 let Inst{30} = isSub;
1355 let Inst{29} = setFlags;
1356 let Inst{28-24} = 0b10001;
1357 let Inst{23-22} = imm{13-12}; // '00' => lsl #0, '01' => lsl #12
1358 let Inst{21-10} = imm{11-0};
1361 let DecoderMethod = "DecodeBaseAddSubImm";
1364 class BaseAddSubRegPseudo<RegisterClass regtype,
1365 SDPatternOperator OpNode>
1366 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1367 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
1370 class BaseAddSubSReg<bit isSub, bit setFlags, RegisterClass regtype,
1371 arith_shifted_reg shifted_regtype, string asm,
1372 SDPatternOperator OpNode>
1373 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
1374 asm, "\t$Rd, $Rn, $Rm", "",
1375 [(set regtype:$Rd, (OpNode regtype:$Rn, shifted_regtype:$Rm))]>,
1376 Sched<[WriteISReg]> {
1377 // The operands are in order to match the 'addr' MI operands, so we
1378 // don't need an encoder method and by-name matching. Just use the default
1379 // in-order handling. Since we're using by-order, make sure the names
1385 let Inst{30} = isSub;
1386 let Inst{29} = setFlags;
1387 let Inst{28-24} = 0b01011;
1388 let Inst{23-22} = shift{7-6};
1390 let Inst{20-16} = src2;
1391 let Inst{15-10} = shift{5-0};
1392 let Inst{9-5} = src1;
1393 let Inst{4-0} = dst;
1395 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
1398 class BaseAddSubEReg<bit isSub, bit setFlags, RegisterClass dstRegtype,
1399 RegisterClass src1Regtype, Operand src2Regtype,
1400 string asm, SDPatternOperator OpNode>
1401 : I<(outs dstRegtype:$R1),
1402 (ins src1Regtype:$R2, src2Regtype:$R3),
1403 asm, "\t$R1, $R2, $R3", "",
1404 [(set dstRegtype:$R1, (OpNode src1Regtype:$R2, src2Regtype:$R3))]>,
1405 Sched<[WriteIEReg]> {
1410 let Inst{30} = isSub;
1411 let Inst{29} = setFlags;
1412 let Inst{28-24} = 0b01011;
1413 let Inst{23-21} = 0b001;
1414 let Inst{20-16} = Rm;
1415 let Inst{15-13} = ext{5-3};
1416 let Inst{12-10} = ext{2-0};
1420 let DecoderMethod = "DecodeAddSubERegInstruction";
1423 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1424 class BaseAddSubEReg64<bit isSub, bit setFlags, RegisterClass dstRegtype,
1425 RegisterClass src1Regtype, RegisterClass src2Regtype,
1426 Operand ext_op, string asm>
1427 : I<(outs dstRegtype:$Rd),
1428 (ins src1Regtype:$Rn, src2Regtype:$Rm, ext_op:$ext),
1429 asm, "\t$Rd, $Rn, $Rm$ext", "", []>,
1430 Sched<[WriteIEReg]> {
1435 let Inst{30} = isSub;
1436 let Inst{29} = setFlags;
1437 let Inst{28-24} = 0b01011;
1438 let Inst{23-21} = 0b001;
1439 let Inst{20-16} = Rm;
1440 let Inst{15} = ext{5};
1441 let Inst{12-10} = ext{2-0};
1445 let DecoderMethod = "DecodeAddSubERegInstruction";
1448 // Aliases for register+register add/subtract.
1449 class AddSubRegAlias<string asm, Instruction inst, RegisterClass dstRegtype,
1450 RegisterClass src1Regtype, RegisterClass src2Regtype,
1452 : InstAlias<asm#" $dst, $src1, $src2",
1453 (inst dstRegtype:$dst, src1Regtype:$src1, src2Regtype:$src2,
1456 multiclass AddSub<bit isSub, string mnemonic,
1457 SDPatternOperator OpNode = null_frag> {
1458 let hasSideEffects = 0 in {
1459 // Add/Subtract immediate
1460 def Wri : BaseAddSubImm<isSub, 0, GPR32sp, GPR32sp, addsub_shifted_imm32,
1464 def Xri : BaseAddSubImm<isSub, 0, GPR64sp, GPR64sp, addsub_shifted_imm64,
1469 // Add/Subtract register - Only used for CodeGen
1470 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
1471 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
1473 // Add/Subtract shifted register
1474 def Wrs : BaseAddSubSReg<isSub, 0, GPR32, arith_shifted_reg32, mnemonic,
1478 def Xrs : BaseAddSubSReg<isSub, 0, GPR64, arith_shifted_reg64, mnemonic,
1484 // Add/Subtract extended register
1485 let AddedComplexity = 1, hasSideEffects = 0 in {
1486 def Wrx : BaseAddSubEReg<isSub, 0, GPR32sp, GPR32sp,
1487 arith_extended_reg32<i32>, mnemonic, OpNode> {
1490 def Xrx : BaseAddSubEReg<isSub, 0, GPR64sp, GPR64sp,
1491 arith_extended_reg32to64<i64>, mnemonic, OpNode> {
1496 def Xrx64 : BaseAddSubEReg64<isSub, 0, GPR64sp, GPR64sp, GPR64,
1497 arith_extendlsl64, mnemonic> {
1498 // UXTX and SXTX only.
1499 let Inst{14-13} = 0b11;
1503 // Register/register aliases with no shift when SP is not used.
1504 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
1505 GPR32, GPR32, GPR32, 0>;
1506 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
1507 GPR64, GPR64, GPR64, 0>;
1509 // Register/register aliases with no shift when either the destination or
1510 // first source register is SP. This relies on the shifted register aliases
1511 // above matching first in the case when SP is not used.
1512 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
1513 GPR32sp, GPR32sp, GPR32, 16>; // UXTW #0
1514 def : AddSubRegAlias<mnemonic,
1515 !cast<Instruction>(NAME#"Xrx64"),
1516 GPR64sp, GPR64sp, GPR64, 24>; // UXTX #0
1519 multiclass AddSubS<bit isSub, string mnemonic, SDNode OpNode> {
1520 let isCompare = 1, Defs = [CPSR] in {
1521 // Add/Subtract immediate
1522 def Wri : BaseAddSubImm<isSub, 1, GPR32, GPR32sp, addsub_shifted_imm32,
1526 def Xri : BaseAddSubImm<isSub, 1, GPR64, GPR64sp, addsub_shifted_imm64,
1531 // Add/Subtract register
1532 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
1533 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
1535 // Add/Subtract shifted register
1536 def Wrs : BaseAddSubSReg<isSub, 1, GPR32, arith_shifted_reg32, mnemonic,
1540 def Xrs : BaseAddSubSReg<isSub, 1, GPR64, arith_shifted_reg64, mnemonic,
1545 // Add/Subtract extended register
1546 let AddedComplexity = 1 in {
1547 def Wrx : BaseAddSubEReg<isSub, 1, GPR32, GPR32sp,
1548 arith_extended_reg32<i32>, mnemonic, OpNode> {
1551 def Xrx : BaseAddSubEReg<isSub, 1, GPR64, GPR64sp,
1552 arith_extended_reg32<i64>, mnemonic, OpNode> {
1557 def Xrx64 : BaseAddSubEReg64<isSub, 1, GPR64, GPR64sp, GPR64,
1558 arith_extendlsl64, mnemonic> {
1559 // UXTX and SXTX only.
1560 let Inst{14-13} = 0b11;
1565 // Register/register aliases with no shift when SP is not used.
1566 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
1567 GPR32, GPR32, GPR32, 0>;
1568 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
1569 GPR64, GPR64, GPR64, 0>;
1571 // Register/register aliases with no shift when the first source register
1572 // is SP. This relies on the shifted register aliases above matching first
1573 // in the case when SP is not used.
1574 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
1575 GPR32, GPR32sp, GPR32, 16>; // UXTW #0
1576 def : AddSubRegAlias<mnemonic,
1577 !cast<Instruction>(NAME#"Xrx64"),
1578 GPR64, GPR64sp, GPR64, 24>; // UXTX #0
1584 def SDTA64EXTR : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>,
1586 def ARM64Extr : SDNode<"ARM64ISD::EXTR", SDTA64EXTR>;
1588 class BaseExtractImm<RegisterClass regtype, Operand imm_type, string asm,
1590 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, imm_type:$imm),
1591 asm, "\t$Rd, $Rn, $Rm, $imm", "", patterns>,
1592 Sched<[WriteExtr, ReadExtrHi]> {
1598 let Inst{30-23} = 0b00100111;
1600 let Inst{20-16} = Rm;
1601 let Inst{15-10} = imm;
1606 multiclass ExtractImm<string asm> {
1607 def Wrri : BaseExtractImm<GPR32, imm0_31, asm,
1609 (ARM64Extr GPR32:$Rn, GPR32:$Rm, imm0_31:$imm))]> {
1612 // imm<5> must be zero.
1615 def Xrri : BaseExtractImm<GPR64, imm0_63, asm,
1617 (ARM64Extr GPR64:$Rn, GPR64:$Rm, imm0_63:$imm))]> {
1628 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1629 class BaseBitfieldImm<bits<2> opc,
1630 RegisterClass regtype, Operand imm_type, string asm>
1631 : I<(outs regtype:$Rd), (ins regtype:$Rn, imm_type:$immr, imm_type:$imms),
1632 asm, "\t$Rd, $Rn, $immr, $imms", "", []>,
1639 let Inst{30-29} = opc;
1640 let Inst{28-23} = 0b100110;
1641 let Inst{21-16} = immr;
1642 let Inst{15-10} = imms;
1647 multiclass BitfieldImm<bits<2> opc, string asm> {
1648 def Wri : BaseBitfieldImm<opc, GPR32, imm0_31, asm> {
1651 // imms<5> and immr<5> must be zero, else ReservedValue().
1655 def Xri : BaseBitfieldImm<opc, GPR64, imm0_63, asm> {
1661 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1662 class BaseBitfieldImmWith2RegArgs<bits<2> opc,
1663 RegisterClass regtype, Operand imm_type, string asm>
1664 : I<(outs regtype:$Rd), (ins regtype:$src, regtype:$Rn, imm_type:$immr,
1666 asm, "\t$Rd, $Rn, $immr, $imms", "$src = $Rd", []>,
1673 let Inst{30-29} = opc;
1674 let Inst{28-23} = 0b100110;
1675 let Inst{21-16} = immr;
1676 let Inst{15-10} = imms;
1681 multiclass BitfieldImmWith2RegArgs<bits<2> opc, string asm> {
1682 def Wri : BaseBitfieldImmWith2RegArgs<opc, GPR32, imm0_31, asm> {
1685 // imms<5> and immr<5> must be zero, else ReservedValue().
1689 def Xri : BaseBitfieldImmWith2RegArgs<opc, GPR64, imm0_63, asm> {
1699 // Logical (immediate)
1700 class BaseLogicalImm<bits<2> opc, RegisterClass dregtype,
1701 RegisterClass sregtype, Operand imm_type, string asm,
1703 : I<(outs dregtype:$Rd), (ins sregtype:$Rn, imm_type:$imm),
1704 asm, "\t$Rd, $Rn, $imm", "", pattern>,
1709 let Inst{30-29} = opc;
1710 let Inst{28-23} = 0b100100;
1711 let Inst{22} = imm{12};
1712 let Inst{21-16} = imm{11-6};
1713 let Inst{15-10} = imm{5-0};
1717 let DecoderMethod = "DecodeLogicalImmInstruction";
1720 // Logical (shifted register)
1721 class BaseLogicalSReg<bits<2> opc, bit N, RegisterClass regtype,
1722 logical_shifted_reg shifted_regtype, string asm,
1724 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
1725 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
1726 Sched<[WriteISReg]> {
1727 // The operands are in order to match the 'addr' MI operands, so we
1728 // don't need an encoder method and by-name matching. Just use the default
1729 // in-order handling. Since we're using by-order, make sure the names
1735 let Inst{30-29} = opc;
1736 let Inst{28-24} = 0b01010;
1737 let Inst{23-22} = shift{7-6};
1739 let Inst{20-16} = src2;
1740 let Inst{15-10} = shift{5-0};
1741 let Inst{9-5} = src1;
1742 let Inst{4-0} = dst;
1744 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
1747 // Aliases for register+register logical instructions.
1748 class LogicalRegAlias<string asm, Instruction inst, RegisterClass regtype>
1749 : InstAlias<asm#" $dst, $src1, $src2",
1750 (inst regtype:$dst, regtype:$src1, regtype:$src2, 0)>;
1752 let AddedComplexity = 6 in
1753 multiclass LogicalImm<bits<2> opc, string mnemonic, SDNode OpNode> {
1754 def Wri : BaseLogicalImm<opc, GPR32sp, GPR32, logical_imm32, mnemonic,
1755 [(set GPR32sp:$Rd, (OpNode GPR32:$Rn,
1756 logical_imm32:$imm))]> {
1758 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
1760 def Xri : BaseLogicalImm<opc, GPR64sp, GPR64, logical_imm64, mnemonic,
1761 [(set GPR64sp:$Rd, (OpNode GPR64:$Rn,
1762 logical_imm64:$imm))]> {
1767 multiclass LogicalImmS<bits<2> opc, string mnemonic, SDNode OpNode> {
1768 let isCompare = 1, Defs = [CPSR] in {
1769 def Wri : BaseLogicalImm<opc, GPR32, GPR32, logical_imm32, mnemonic,
1770 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_imm32:$imm))]> {
1772 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
1774 def Xri : BaseLogicalImm<opc, GPR64, GPR64, logical_imm64, mnemonic,
1775 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_imm64:$imm))]> {
1778 } // end Defs = [CPSR]
1781 class BaseLogicalRegPseudo<RegisterClass regtype, SDPatternOperator OpNode>
1782 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1783 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
1786 // Split from LogicalImm as not all instructions have both.
1787 multiclass LogicalReg<bits<2> opc, bit N, string mnemonic,
1788 SDPatternOperator OpNode> {
1789 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
1790 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
1792 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
1793 [(set GPR32:$Rd, (OpNode GPR32:$Rn,
1794 logical_shifted_reg32:$Rm))]> {
1797 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
1798 [(set GPR64:$Rd, (OpNode GPR64:$Rn,
1799 logical_shifted_reg64:$Rm))]> {
1803 def : LogicalRegAlias<mnemonic,
1804 !cast<Instruction>(NAME#"Wrs"), GPR32>;
1805 def : LogicalRegAlias<mnemonic,
1806 !cast<Instruction>(NAME#"Xrs"), GPR64>;
1809 // Split from LogicalReg to allow setting CPSR Defs
1810 multiclass LogicalRegS<bits<2> opc, bit N, string mnemonic,
1811 SDPatternOperator OpNode = null_frag> {
1812 let Defs = [CPSR], mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
1813 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
1814 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
1816 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
1817 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_shifted_reg32:$Rm))]> {
1820 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
1821 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_shifted_reg64:$Rm))]> {
1826 def : LogicalRegAlias<mnemonic,
1827 !cast<Instruction>(NAME#"Wrs"), GPR32>;
1828 def : LogicalRegAlias<mnemonic,
1829 !cast<Instruction>(NAME#"Xrs"), GPR64>;
1833 // Conditionally set flags
1837 // 4-bit immediate. Pretty-printed as <cc>
1838 def ccode : Operand<i32> {
1839 let PrintMethod = "printCondCode";
1842 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1843 class BaseCondSetFlagsImm<bit op, RegisterClass regtype, string asm>
1844 : I<(outs), (ins regtype:$Rn, imm0_31:$imm, imm0_15:$nzcv, ccode:$cond),
1845 asm, "\t$Rn, $imm, $nzcv, $cond", "", []>,
1856 let Inst{29-21} = 0b111010010;
1857 let Inst{20-16} = imm;
1858 let Inst{15-12} = cond;
1859 let Inst{11-10} = 0b10;
1862 let Inst{3-0} = nzcv;
1865 multiclass CondSetFlagsImm<bit op, string asm> {
1866 def Wi : BaseCondSetFlagsImm<op, GPR32, asm> {
1869 def Xi : BaseCondSetFlagsImm<op, GPR64, asm> {
1874 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1875 class BaseCondSetFlagsReg<bit op, RegisterClass regtype, string asm>
1876 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm0_15:$nzcv, ccode:$cond),
1877 asm, "\t$Rn, $Rm, $nzcv, $cond", "", []>,
1888 let Inst{29-21} = 0b111010010;
1889 let Inst{20-16} = Rm;
1890 let Inst{15-12} = cond;
1891 let Inst{11-10} = 0b00;
1894 let Inst{3-0} = nzcv;
1897 multiclass CondSetFlagsReg<bit op, string asm> {
1898 def Wr : BaseCondSetFlagsReg<op, GPR32, asm> {
1901 def Xr : BaseCondSetFlagsReg<op, GPR64, asm> {
1907 // Conditional select
1910 class BaseCondSelect<bit op, bits<2> op2, RegisterClass regtype, string asm>
1911 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
1912 asm, "\t$Rd, $Rn, $Rm, $cond", "",
1914 (ARM64csel regtype:$Rn, regtype:$Rm, (i32 imm:$cond), CPSR))]>,
1924 let Inst{29-21} = 0b011010100;
1925 let Inst{20-16} = Rm;
1926 let Inst{15-12} = cond;
1927 let Inst{11-10} = op2;
1932 multiclass CondSelect<bit op, bits<2> op2, string asm> {
1933 def Wr : BaseCondSelect<op, op2, GPR32, asm> {
1936 def Xr : BaseCondSelect<op, op2, GPR64, asm> {
1941 class BaseCondSelectOp<bit op, bits<2> op2, RegisterClass regtype, string asm,
1943 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
1944 asm, "\t$Rd, $Rn, $Rm, $cond", "",
1946 (ARM64csel regtype:$Rn, (frag regtype:$Rm),
1947 (i32 imm:$cond), CPSR))]>,
1957 let Inst{29-21} = 0b011010100;
1958 let Inst{20-16} = Rm;
1959 let Inst{15-12} = cond;
1960 let Inst{11-10} = op2;
1965 def inv_cond_XFORM : SDNodeXForm<imm, [{
1966 ARM64CC::CondCode CC = static_cast<ARM64CC::CondCode>(N->getZExtValue());
1967 return CurDAG->getTargetConstant(ARM64CC::getInvertedCondCode(CC), MVT::i32);
1970 multiclass CondSelectOp<bit op, bits<2> op2, string asm, PatFrag frag> {
1971 def Wr : BaseCondSelectOp<op, op2, GPR32, asm, frag> {
1974 def Xr : BaseCondSelectOp<op, op2, GPR64, asm, frag> {
1978 def : Pat<(ARM64csel (frag GPR32:$Rm), GPR32:$Rn, (i32 imm:$cond), CPSR),
1979 (!cast<Instruction>(NAME # Wr) GPR32:$Rn, GPR32:$Rm,
1980 (inv_cond_XFORM imm:$cond))>;
1982 def : Pat<(ARM64csel (frag GPR64:$Rm), GPR64:$Rn, (i32 imm:$cond), CPSR),
1983 (!cast<Instruction>(NAME # Xr) GPR64:$Rn, GPR64:$Rm,
1984 (inv_cond_XFORM imm:$cond))>;
1988 // Special Mask Value
1990 def maski8_or_more : Operand<i32>,
1991 ImmLeaf<i32, [{ return (Imm & 0xff) == 0xff; }]> {
1993 def maski16_or_more : Operand<i32>,
1994 ImmLeaf<i32, [{ return (Imm & 0xffff) == 0xffff; }]> {
2002 // (unsigned immediate)
2003 // Indexed for 8-bit registers. offset is in range [0,4095].
2004 def MemoryIndexed8Operand : AsmOperandClass {
2005 let Name = "MemoryIndexed8";
2006 let DiagnosticType = "InvalidMemoryIndexed8";
2008 def am_indexed8 : Operand<i64>,
2009 ComplexPattern<i64, 2, "SelectAddrModeIndexed8", []> {
2010 let PrintMethod = "printAMIndexed8";
2012 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale1>";
2013 let ParserMatchClass = MemoryIndexed8Operand;
2014 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2017 // Indexed for 16-bit registers. offset is multiple of 2 in range [0,8190],
2018 // stored as immval/2 (the 12-bit literal that encodes directly into the insn).
2019 def MemoryIndexed16Operand : AsmOperandClass {
2020 let Name = "MemoryIndexed16";
2021 let DiagnosticType = "InvalidMemoryIndexed16";
2023 def am_indexed16 : Operand<i64>,
2024 ComplexPattern<i64, 2, "SelectAddrModeIndexed16", []> {
2025 let PrintMethod = "printAMIndexed16";
2027 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale2>";
2028 let ParserMatchClass = MemoryIndexed16Operand;
2029 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2032 // Indexed for 32-bit registers. offset is multiple of 4 in range [0,16380],
2033 // stored as immval/4 (the 12-bit literal that encodes directly into the insn).
2034 def MemoryIndexed32Operand : AsmOperandClass {
2035 let Name = "MemoryIndexed32";
2036 let DiagnosticType = "InvalidMemoryIndexed32";
2038 def am_indexed32 : Operand<i64>,
2039 ComplexPattern<i64, 2, "SelectAddrModeIndexed32", []> {
2040 let PrintMethod = "printAMIndexed32";
2042 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale4>";
2043 let ParserMatchClass = MemoryIndexed32Operand;
2044 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2047 // Indexed for 64-bit registers. offset is multiple of 8 in range [0,32760],
2048 // stored as immval/8 (the 12-bit literal that encodes directly into the insn).
2049 def MemoryIndexed64Operand : AsmOperandClass {
2050 let Name = "MemoryIndexed64";
2051 let DiagnosticType = "InvalidMemoryIndexed64";
2053 def am_indexed64 : Operand<i64>,
2054 ComplexPattern<i64, 2, "SelectAddrModeIndexed64", []> {
2055 let PrintMethod = "printAMIndexed64";
2057 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale8>";
2058 let ParserMatchClass = MemoryIndexed64Operand;
2059 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2062 // Indexed for 128-bit registers. offset is multiple of 16 in range [0,65520],
2063 // stored as immval/16 (the 12-bit literal that encodes directly into the insn).
2064 def MemoryIndexed128Operand : AsmOperandClass {
2065 let Name = "MemoryIndexed128";
2066 let DiagnosticType = "InvalidMemoryIndexed128";
2068 def am_indexed128 : Operand<i64>,
2069 ComplexPattern<i64, 2, "SelectAddrModeIndexed128", []> {
2070 let PrintMethod = "printAMIndexed128";
2072 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale16>";
2073 let ParserMatchClass = MemoryIndexed128Operand;
2074 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2078 def MemoryNoIndexOperand : AsmOperandClass { let Name = "MemoryNoIndex"; }
2079 def am_noindex : Operand<i64>,
2080 ComplexPattern<i64, 1, "SelectAddrModeNoIndex", []> {
2081 let PrintMethod = "printAMNoIndex";
2082 let ParserMatchClass = MemoryNoIndexOperand;
2083 let MIOperandInfo = (ops GPR64sp:$base);
2086 class BaseLoadStoreUI<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2087 string asm, list<dag> pattern>
2088 : I<oops, iops, asm, "\t$Rt, $addr", "", pattern> {
2092 bits<5> base = addr{4-0};
2093 bits<12> offset = addr{16-5};
2095 let Inst{31-30} = sz;
2096 let Inst{29-27} = 0b111;
2098 let Inst{25-24} = 0b01;
2099 let Inst{23-22} = opc;
2100 let Inst{21-10} = offset;
2101 let Inst{9-5} = base;
2102 let Inst{4-0} = dst;
2104 let DecoderMethod = "DecodeUnsignedLdStInstruction";
2107 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2108 class LoadUI<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2109 Operand indextype, string asm, list<dag> pattern>
2110 : BaseLoadStoreUI<sz, V, opc,
2111 (outs regtype:$Rt), (ins indextype:$addr), asm, pattern>,
2114 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2115 class StoreUI<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2116 Operand indextype, string asm, list<dag> pattern>
2117 : BaseLoadStoreUI<sz, V, opc,
2118 (outs), (ins regtype:$Rt, indextype:$addr), asm, pattern>,
2121 def PrefetchOperand : AsmOperandClass {
2122 let Name = "Prefetch";
2123 let ParserMethod = "tryParsePrefetch";
2125 def prfop : Operand<i32> {
2126 let PrintMethod = "printPrefetchOp";
2127 let ParserMatchClass = PrefetchOperand;
2130 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2131 class PrefetchUI<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2132 : BaseLoadStoreUI<sz, V, opc,
2133 (outs), (ins prfop:$Rt, am_indexed64:$addr), asm, pat>,
2140 // Load literal address: 19-bit immediate. The low two bits of the target
2141 // offset are implied zero and so are not part of the immediate.
2142 def am_ldrlit : Operand<OtherVT> {
2143 let EncoderMethod = "getLoadLiteralOpValue";
2144 let DecoderMethod = "DecodePCRelLabel19";
2145 let PrintMethod = "printAlignedLabel";
2146 let ParserMatchClass = PCRelLabel19Operand;
2149 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2150 class LoadLiteral<bits<2> opc, bit V, RegisterClass regtype, string asm>
2151 : I<(outs regtype:$Rt), (ins am_ldrlit:$label),
2152 asm, "\t$Rt, $label", "", []>,
2156 let Inst{31-30} = opc;
2157 let Inst{29-27} = 0b011;
2159 let Inst{25-24} = 0b00;
2160 let Inst{23-5} = label;
2164 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2165 class PrefetchLiteral<bits<2> opc, bit V, string asm, list<dag> pat>
2166 : I<(outs), (ins prfop:$Rt, am_ldrlit:$label),
2167 asm, "\t$Rt, $label", "", pat>,
2171 let Inst{31-30} = opc;
2172 let Inst{29-27} = 0b011;
2174 let Inst{25-24} = 0b00;
2175 let Inst{23-5} = label;
2180 // Load/store register offset
2183 class MemROAsmOperand<int sz> : AsmOperandClass {
2184 let Name = "MemoryRegisterOffset"#sz;
2187 def MemROAsmOperand8 : MemROAsmOperand<8>;
2188 def MemROAsmOperand16 : MemROAsmOperand<16>;
2189 def MemROAsmOperand32 : MemROAsmOperand<32>;
2190 def MemROAsmOperand64 : MemROAsmOperand<64>;
2191 def MemROAsmOperand128 : MemROAsmOperand<128>;
2193 class ro_indexed<int sz> : Operand<i64> { // ComplexPattern<...>
2194 let PrintMethod = "printMemoryRegOffset"#sz;
2195 let MIOperandInfo = (ops GPR64sp:$base, GPR64:$offset, i32imm:$extend);
2198 def ro_indexed8 : ro_indexed<8>, ComplexPattern<i64, 3, "SelectAddrModeRO8", []> {
2199 let ParserMatchClass = MemROAsmOperand8;
2202 def ro_indexed16 : ro_indexed<16>, ComplexPattern<i64, 3, "SelectAddrModeRO16", []> {
2203 let ParserMatchClass = MemROAsmOperand16;
2206 def ro_indexed32 : ro_indexed<32>, ComplexPattern<i64, 3, "SelectAddrModeRO32", []> {
2207 let ParserMatchClass = MemROAsmOperand32;
2210 def ro_indexed64 : ro_indexed<64>, ComplexPattern<i64, 3, "SelectAddrModeRO64", []> {
2211 let ParserMatchClass = MemROAsmOperand64;
2214 def ro_indexed128 : ro_indexed<128>, ComplexPattern<i64, 3, "SelectAddrModeRO128", []> {
2215 let ParserMatchClass = MemROAsmOperand128;
2218 class LoadStore8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2219 string asm, dag ins, dag outs, list<dag> pat>
2220 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2221 // The operands are in order to match the 'addr' MI operands, so we
2222 // don't need an encoder method and by-name matching. Just use the default
2223 // in-order handling. Since we're using by-order, make sure the names
2229 let Inst{31-30} = sz;
2230 let Inst{29-27} = 0b111;
2232 let Inst{25-24} = 0b00;
2233 let Inst{23-22} = opc;
2235 let Inst{20-16} = offset;
2236 let Inst{15-13} = extend{3-1};
2238 let Inst{12} = extend{0};
2239 let Inst{11-10} = 0b10;
2240 let Inst{9-5} = base;
2241 let Inst{4-0} = dst;
2243 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2246 class Load8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2247 string asm, list<dag> pat>
2248 : LoadStore8RO<sz, V, opc, regtype, asm,
2249 (outs regtype:$Rt), (ins ro_indexed8:$addr), pat>,
2250 Sched<[WriteLDIdx, ReadAdrBase]>;
2252 class Store8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2253 string asm, list<dag> pat>
2254 : LoadStore8RO<sz, V, opc, regtype, asm,
2255 (outs), (ins regtype:$Rt, ro_indexed8:$addr), pat>,
2256 Sched<[WriteSTIdx, ReadAdrBase]>;
2258 class LoadStore16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2259 string asm, dag ins, dag outs, list<dag> pat>
2260 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2261 // The operands are in order to match the 'addr' MI operands, so we
2262 // don't need an encoder method and by-name matching. Just use the default
2263 // in-order handling. Since we're using by-order, make sure the names
2269 let Inst{31-30} = sz;
2270 let Inst{29-27} = 0b111;
2272 let Inst{25-24} = 0b00;
2273 let Inst{23-22} = opc;
2275 let Inst{20-16} = offset;
2276 let Inst{15-13} = extend{3-1};
2278 let Inst{12} = extend{0};
2279 let Inst{11-10} = 0b10;
2280 let Inst{9-5} = base;
2281 let Inst{4-0} = dst;
2283 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2286 class Load16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2287 string asm, list<dag> pat>
2288 : LoadStore16RO<sz, V, opc, regtype, asm,
2289 (outs regtype:$Rt), (ins ro_indexed16:$addr), pat>,
2290 Sched<[WriteLDIdx, ReadAdrBase]>;
2292 class Store16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2293 string asm, list<dag> pat>
2294 : LoadStore16RO<sz, V, opc, regtype, asm,
2295 (outs), (ins regtype:$Rt, ro_indexed16:$addr), pat>,
2296 Sched<[WriteSTIdx, ReadAdrBase]>;
2298 class LoadStore32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2299 string asm, dag ins, dag outs, list<dag> pat>
2300 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2301 // The operands are in order to match the 'addr' MI operands, so we
2302 // don't need an encoder method and by-name matching. Just use the default
2303 // in-order handling. Since we're using by-order, make sure the names
2309 let Inst{31-30} = sz;
2310 let Inst{29-27} = 0b111;
2312 let Inst{25-24} = 0b00;
2313 let Inst{23-22} = opc;
2315 let Inst{20-16} = offset;
2316 let Inst{15-13} = extend{3-1};
2318 let Inst{12} = extend{0};
2319 let Inst{11-10} = 0b10;
2320 let Inst{9-5} = base;
2321 let Inst{4-0} = dst;
2323 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2326 class Load32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2327 string asm, list<dag> pat>
2328 : LoadStore32RO<sz, V, opc, regtype, asm,
2329 (outs regtype:$Rt), (ins ro_indexed32:$addr), pat>,
2330 Sched<[WriteLDIdx, ReadAdrBase]>;
2332 class Store32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2333 string asm, list<dag> pat>
2334 : LoadStore32RO<sz, V, opc, regtype, asm,
2335 (outs), (ins regtype:$Rt, ro_indexed32:$addr), pat>,
2336 Sched<[WriteSTIdx, ReadAdrBase]>;
2338 class LoadStore64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2339 string asm, dag ins, dag outs, list<dag> pat>
2340 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2341 // The operands are in order to match the 'addr' MI operands, so we
2342 // don't need an encoder method and by-name matching. Just use the default
2343 // in-order handling. Since we're using by-order, make sure the names
2349 let Inst{31-30} = sz;
2350 let Inst{29-27} = 0b111;
2352 let Inst{25-24} = 0b00;
2353 let Inst{23-22} = opc;
2355 let Inst{20-16} = offset;
2356 let Inst{15-13} = extend{3-1};
2358 let Inst{12} = extend{0};
2359 let Inst{11-10} = 0b10;
2360 let Inst{9-5} = base;
2361 let Inst{4-0} = dst;
2363 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2366 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2367 class Load64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2368 string asm, list<dag> pat>
2369 : LoadStore64RO<sz, V, opc, regtype, asm,
2370 (outs regtype:$Rt), (ins ro_indexed64:$addr), pat>,
2371 Sched<[WriteLDIdx, ReadAdrBase]>;
2373 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2374 class Store64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2375 string asm, list<dag> pat>
2376 : LoadStore64RO<sz, V, opc, regtype, asm,
2377 (outs), (ins regtype:$Rt, ro_indexed64:$addr), pat>,
2378 Sched<[WriteSTIdx, ReadAdrBase]>;
2381 class LoadStore128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2382 string asm, dag ins, dag outs, list<dag> pat>
2383 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2384 // The operands are in order to match the 'addr' MI operands, so we
2385 // don't need an encoder method and by-name matching. Just use the default
2386 // in-order handling. Since we're using by-order, make sure the names
2392 let Inst{31-30} = sz;
2393 let Inst{29-27} = 0b111;
2395 let Inst{25-24} = 0b00;
2396 let Inst{23-22} = opc;
2398 let Inst{20-16} = offset;
2399 let Inst{15-13} = extend{3-1};
2401 let Inst{12} = extend{0};
2402 let Inst{11-10} = 0b10;
2403 let Inst{9-5} = base;
2404 let Inst{4-0} = dst;
2406 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2409 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2410 class Load128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2411 string asm, list<dag> pat>
2412 : LoadStore128RO<sz, V, opc, regtype, asm,
2413 (outs regtype:$Rt), (ins ro_indexed128:$addr), pat>,
2414 Sched<[WriteLDIdx, ReadAdrBase]>;
2416 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2417 class Store128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2418 string asm, list<dag> pat>
2419 : LoadStore128RO<sz, V, opc, regtype, asm,
2420 (outs), (ins regtype:$Rt, ro_indexed128:$addr), pat>,
2421 Sched<[WriteSTIdx, ReadAdrBase]>;
2423 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2424 class PrefetchRO<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2425 : I<(outs), (ins prfop:$Rt, ro_indexed64:$addr), asm,
2426 "\t$Rt, $addr", "", pat>,
2428 // The operands are in order to match the 'addr' MI operands, so we
2429 // don't need an encoder method and by-name matching. Just use the default
2430 // in-order handling. Since we're using by-order, make sure the names
2436 let Inst{31-30} = sz;
2437 let Inst{29-27} = 0b111;
2439 let Inst{25-24} = 0b00;
2440 let Inst{23-22} = opc;
2442 let Inst{20-16} = offset;
2443 let Inst{15-13} = extend{3-1};
2445 let Inst{12} = extend{0};
2446 let Inst{11-10} = 0b10;
2447 let Inst{9-5} = base;
2448 let Inst{4-0} = dst;
2450 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2454 // Load/store unscaled immediate
2457 def MemoryUnscaledOperand : AsmOperandClass {
2458 let Name = "MemoryUnscaled";
2459 let DiagnosticType = "InvalidMemoryIndexedSImm9";
2461 class am_unscaled_operand : Operand<i64> {
2462 let PrintMethod = "printAMUnscaled";
2463 let ParserMatchClass = MemoryUnscaledOperand;
2464 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2466 class am_unscaled_wb_operand : Operand<i64> {
2467 let PrintMethod = "printAMUnscaledWB";
2468 let ParserMatchClass = MemoryUnscaledOperand;
2469 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2471 def am_unscaled : am_unscaled_operand;
2472 def am_unscaled_wb: am_unscaled_wb_operand;
2473 def am_unscaled8 : am_unscaled_operand,
2474 ComplexPattern<i64, 2, "SelectAddrModeUnscaled8", []>;
2475 def am_unscaled16 : am_unscaled_operand,
2476 ComplexPattern<i64, 2, "SelectAddrModeUnscaled16", []>;
2477 def am_unscaled32 : am_unscaled_operand,
2478 ComplexPattern<i64, 2, "SelectAddrModeUnscaled32", []>;
2479 def am_unscaled64 : am_unscaled_operand,
2480 ComplexPattern<i64, 2, "SelectAddrModeUnscaled64", []>;
2481 def am_unscaled128 : am_unscaled_operand,
2482 ComplexPattern<i64, 2, "SelectAddrModeUnscaled128", []>;
2484 class BaseLoadStoreUnscale<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2485 string asm, list<dag> pattern>
2486 : I<oops, iops, asm, "\t$Rt, $addr", "", pattern> {
2487 // The operands are in order to match the 'addr' MI operands, so we
2488 // don't need an encoder method and by-name matching. Just use the default
2489 // in-order handling. Since we're using by-order, make sure the names
2494 let Inst{31-30} = sz;
2495 let Inst{29-27} = 0b111;
2497 let Inst{25-24} = 0b00;
2498 let Inst{23-22} = opc;
2500 let Inst{20-12} = offset;
2501 let Inst{11-10} = 0b00;
2502 let Inst{9-5} = base;
2503 let Inst{4-0} = dst;
2505 let DecoderMethod = "DecodeSignedLdStInstruction";
2508 let AddedComplexity = 1 in // try this before LoadUI
2509 class LoadUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2510 Operand amtype, string asm, list<dag> pattern>
2511 : BaseLoadStoreUnscale<sz, V, opc, (outs regtype:$Rt),
2512 (ins amtype:$addr), asm, pattern>,
2515 let AddedComplexity = 1 in // try this before StoreUI
2516 class StoreUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2517 Operand amtype, string asm, list<dag> pattern>
2518 : BaseLoadStoreUnscale<sz, V, opc, (outs),
2519 (ins regtype:$Rt, amtype:$addr), asm, pattern>,
2522 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2523 class PrefetchUnscaled<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2524 : BaseLoadStoreUnscale<sz, V, opc, (outs),
2525 (ins prfop:$Rt, am_unscaled:$addr), asm, pat>,
2529 // Load/store unscaled immediate, unprivileged
2532 class BaseLoadStoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
2533 dag oops, dag iops, string asm>
2534 : I<oops, iops, asm, "\t$Rt, $addr", "", []> {
2535 // The operands are in order to match the 'addr' MI operands, so we
2536 // don't need an encoder method and by-name matching. Just use the default
2537 // in-order handling. Since we're using by-order, make sure the names
2542 let Inst{31-30} = sz;
2543 let Inst{29-27} = 0b111;
2545 let Inst{25-24} = 0b00;
2546 let Inst{23-22} = opc;
2548 let Inst{20-12} = offset;
2549 let Inst{11-10} = 0b10;
2550 let Inst{9-5} = base;
2551 let Inst{4-0} = dst;
2553 let DecoderMethod = "DecodeSignedLdStInstruction";
2556 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in {
2557 class LoadUnprivileged<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2559 : BaseLoadStoreUnprivileged<sz, V, opc,
2560 (outs regtype:$Rt), (ins am_unscaled:$addr), asm>,
2564 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in {
2565 class StoreUnprivileged<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2567 : BaseLoadStoreUnprivileged<sz, V, opc,
2568 (outs), (ins regtype:$Rt, am_unscaled:$addr), asm>,
2573 // Load/store pre-indexed
2576 class BaseLoadStorePreIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2577 string asm, string cstr>
2578 : I<oops, iops, asm, "\t$Rt, $addr!", cstr, []> {
2579 // The operands are in order to match the 'addr' MI operands, so we
2580 // don't need an encoder method and by-name matching. Just use the default
2581 // in-order handling.
2585 let Inst{31-30} = sz;
2586 let Inst{29-27} = 0b111;
2588 let Inst{25-24} = 0;
2589 let Inst{23-22} = opc;
2591 let Inst{20-12} = offset;
2592 let Inst{11-10} = 0b11;
2593 let Inst{9-5} = base;
2594 let Inst{4-0} = dst;
2596 let DecoderMethod = "DecodeSignedLdStInstruction";
2599 let hasSideEffects = 0 in {
2600 let mayStore = 0, mayLoad = 1 in
2601 // FIXME: Modeling the write-back of these instructions for isel is tricky.
2602 // we need the complex addressing mode for the memory reference, but
2603 // we also need the write-back specified as a tied operand to the
2604 // base register. That combination does not play nicely with
2605 // the asm matcher and friends.
2606 class LoadPreIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2608 : BaseLoadStorePreIdx<sz, V, opc,
2609 (outs regtype:$Rt/*, GPR64sp:$wback*/),
2610 (ins am_unscaled_wb:$addr), asm, ""/*"$addr.base = $wback"*/>,
2611 Sched<[WriteLD, WriteAdr]>;
2613 let mayStore = 1, mayLoad = 0 in
2614 class StorePreIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2616 : BaseLoadStorePreIdx<sz, V, opc,
2617 (outs/* GPR64sp:$wback*/),
2618 (ins regtype:$Rt, am_unscaled_wb:$addr),
2619 asm, ""/*"$addr.base = $wback"*/>,
2620 Sched<[WriteAdr, WriteST]>;
2621 } // hasSideEffects = 0
2623 // ISel pseudo-instructions which have the tied operands. When the MC lowering
2624 // logic finally gets smart enough to strip off tied operands that are just
2625 // for isel convenience, we can get rid of these pseudos and just reference
2626 // the real instructions directly.
2628 // Ironically, also because of the writeback operands, we can't put the
2629 // matcher pattern directly on the instruction, but need to define it
2632 // Loads aren't matched with patterns here at all, but rather in C++
2634 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in {
2635 class LoadPreIdxPseudo<RegisterClass regtype>
2636 : Pseudo<(outs regtype:$Rt, GPR64sp:$wback),
2637 (ins am_noindex:$addr, simm9:$offset), [],
2638 "$addr.base = $wback,@earlyclobber $wback">,
2639 Sched<[WriteLD, WriteAdr]>;
2640 class LoadPostIdxPseudo<RegisterClass regtype>
2641 : Pseudo<(outs regtype:$Rt, GPR64sp:$wback),
2642 (ins am_noindex:$addr, simm9:$offset), [],
2643 "$addr.base = $wback,@earlyclobber $wback">,
2644 Sched<[WriteLD, WriteI]>;
2646 multiclass StorePreIdxPseudo<RegisterClass regtype, ValueType Ty,
2647 SDPatternOperator OpNode> {
2648 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
2649 def _isel: Pseudo<(outs GPR64sp:$wback),
2650 (ins regtype:$Rt, am_noindex:$addr, simm9:$offset), [],
2651 "$addr.base = $wback,@earlyclobber $wback">,
2652 Sched<[WriteAdr, WriteST]>;
2654 def : Pat<(OpNode (Ty regtype:$Rt), am_noindex:$addr, simm9:$offset),
2655 (!cast<Instruction>(NAME#_isel) regtype:$Rt, am_noindex:$addr,
2660 // Load/store post-indexed
2663 // (pre-index) load/stores.
2664 class BaseLoadStorePostIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2665 string asm, string cstr>
2666 : I<oops, iops, asm, "\t$Rt, $addr, $idx", cstr, []> {
2667 // The operands are in order to match the 'addr' MI operands, so we
2668 // don't need an encoder method and by-name matching. Just use the default
2669 // in-order handling.
2673 let Inst{31-30} = sz;
2674 let Inst{29-27} = 0b111;
2676 let Inst{25-24} = 0b00;
2677 let Inst{23-22} = opc;
2679 let Inst{20-12} = offset;
2680 let Inst{11-10} = 0b01;
2681 let Inst{9-5} = base;
2682 let Inst{4-0} = dst;
2684 let DecoderMethod = "DecodeSignedLdStInstruction";
2687 let hasSideEffects = 0 in {
2688 let mayStore = 0, mayLoad = 1 in
2689 // FIXME: Modeling the write-back of these instructions for isel is tricky.
2690 // we need the complex addressing mode for the memory reference, but
2691 // we also need the write-back specified as a tied operand to the
2692 // base register. That combination does not play nicely with
2693 // the asm matcher and friends.
2694 class LoadPostIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2696 : BaseLoadStorePostIdx<sz, V, opc,
2697 (outs regtype:$Rt/*, GPR64sp:$wback*/),
2698 (ins am_noindex:$addr, simm9:$idx),
2699 asm, ""/*"$addr.base = $wback"*/>,
2700 Sched<[WriteLD, WriteI]>;
2702 let mayStore = 1, mayLoad = 0 in
2703 class StorePostIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2705 : BaseLoadStorePostIdx<sz, V, opc,
2706 (outs/* GPR64sp:$wback*/),
2707 (ins regtype:$Rt, am_noindex:$addr, simm9:$idx),
2708 asm, ""/*"$addr.base = $wback"*/>,
2709 Sched<[WriteAdr, WriteST, ReadAdrBase]>;
2710 } // hasSideEffects = 0
2712 // ISel pseudo-instructions which have the tied operands. When the MC lowering
2713 // logic finally gets smart enough to strip off tied operands that are just
2714 // for isel convenience, we can get rid of these pseudos and just reference
2715 // the real instructions directly.
2717 // Ironically, also because of the writeback operands, we can't put the
2718 // matcher pattern directly on the instruction, but need to define it
2720 multiclass StorePostIdxPseudo<RegisterClass regtype, ValueType Ty,
2721 SDPatternOperator OpNode, Instruction Insn> {
2722 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
2723 def _isel: Pseudo<(outs GPR64sp:$wback),
2724 (ins regtype:$Rt, am_noindex:$addr, simm9:$idx), [],
2725 "$addr.base = $wback,@earlyclobber $wback">,
2726 PseudoInstExpansion<(Insn regtype:$Rt, am_noindex:$addr, simm9:$idx)>,
2727 Sched<[WriteAdr, WriteST, ReadAdrBase]>;
2729 def : Pat<(OpNode (Ty regtype:$Rt), am_noindex:$addr, simm9:$idx),
2730 (!cast<Instruction>(NAME#_isel) regtype:$Rt, am_noindex:$addr,
2738 // (indexed, offset)
2740 class BaseLoadStorePairOffset<bits<2> opc, bit V, bit L, dag oops, dag iops,
2742 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr", "", []> {
2743 // The operands are in order to match the 'addr' MI operands, so we
2744 // don't need an encoder method and by-name matching. Just use the default
2745 // in-order handling. Since we're using by-order, make sure the names
2751 let Inst{31-30} = opc;
2752 let Inst{29-27} = 0b101;
2754 let Inst{25-23} = 0b010;
2756 let Inst{21-15} = offset;
2757 let Inst{14-10} = dst2;
2758 let Inst{9-5} = base;
2759 let Inst{4-0} = dst;
2761 let DecoderMethod = "DecodePairLdStInstruction";
2764 let hasSideEffects = 0 in {
2765 let mayStore = 0, mayLoad = 1 in
2766 class LoadPairOffset<bits<2> opc, bit V, RegisterClass regtype,
2767 Operand indextype, string asm>
2768 : BaseLoadStorePairOffset<opc, V, 1,
2769 (outs regtype:$Rt, regtype:$Rt2),
2770 (ins indextype:$addr), asm>,
2771 Sched<[WriteLD, WriteLDHi]>;
2773 let mayLoad = 0, mayStore = 1 in
2774 class StorePairOffset<bits<2> opc, bit V, RegisterClass regtype,
2775 Operand indextype, string asm>
2776 : BaseLoadStorePairOffset<opc, V, 0, (outs),
2777 (ins regtype:$Rt, regtype:$Rt2, indextype:$addr),
2780 } // hasSideEffects = 0
2784 def MemoryIndexed32SImm7 : AsmOperandClass {
2785 let Name = "MemoryIndexed32SImm7";
2786 let DiagnosticType = "InvalidMemoryIndexed32SImm7";
2788 def am_indexed32simm7 : Operand<i32> { // ComplexPattern<...>
2789 let PrintMethod = "printAMIndexed32";
2790 let ParserMatchClass = MemoryIndexed32SImm7;
2791 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2793 def am_indexed32simm7_wb : Operand<i32> { // ComplexPattern<...>
2794 let PrintMethod = "printAMIndexed32WB";
2795 let ParserMatchClass = MemoryIndexed32SImm7;
2796 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2799 def MemoryIndexed64SImm7 : AsmOperandClass {
2800 let Name = "MemoryIndexed64SImm7";
2801 let DiagnosticType = "InvalidMemoryIndexed64SImm7";
2803 def am_indexed64simm7 : Operand<i32> { // ComplexPattern<...>
2804 let PrintMethod = "printAMIndexed64";
2805 let ParserMatchClass = MemoryIndexed64SImm7;
2806 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2808 def am_indexed64simm7_wb : Operand<i32> { // ComplexPattern<...>
2809 let PrintMethod = "printAMIndexed64WB";
2810 let ParserMatchClass = MemoryIndexed64SImm7;
2811 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2814 def MemoryIndexed128SImm7 : AsmOperandClass {
2815 let Name = "MemoryIndexed128SImm7";
2816 let DiagnosticType = "InvalidMemoryIndexed128SImm7";
2818 def am_indexed128simm7 : Operand<i32> { // ComplexPattern<...>
2819 let PrintMethod = "printAMIndexed128";
2820 let ParserMatchClass = MemoryIndexed128SImm7;
2821 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2823 def am_indexed128simm7_wb : Operand<i32> { // ComplexPattern<...>
2824 let PrintMethod = "printAMIndexed128WB";
2825 let ParserMatchClass = MemoryIndexed128SImm7;
2826 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2829 class BaseLoadStorePairPreIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
2831 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr!", "", []> {
2832 // The operands are in order to match the 'addr' MI operands, so we
2833 // don't need an encoder method and by-name matching. Just use the default
2834 // in-order handling. Since we're using by-order, make sure the names
2840 let Inst{31-30} = opc;
2841 let Inst{29-27} = 0b101;
2843 let Inst{25-23} = 0b011;
2845 let Inst{21-15} = offset;
2846 let Inst{14-10} = dst2;
2847 let Inst{9-5} = base;
2848 let Inst{4-0} = dst;
2850 let DecoderMethod = "DecodePairLdStInstruction";
2853 let hasSideEffects = 0 in {
2854 let mayStore = 0, mayLoad = 1 in
2855 class LoadPairPreIdx<bits<2> opc, bit V, RegisterClass regtype,
2856 Operand addrmode, string asm>
2857 : BaseLoadStorePairPreIdx<opc, V, 1,
2858 (outs regtype:$Rt, regtype:$Rt2),
2859 (ins addrmode:$addr), asm>,
2860 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
2862 let mayStore = 1, mayLoad = 0 in
2863 class StorePairPreIdx<bits<2> opc, bit V, RegisterClass regtype,
2864 Operand addrmode, string asm>
2865 : BaseLoadStorePairPreIdx<opc, V, 0, (outs),
2866 (ins regtype:$Rt, regtype:$Rt2, addrmode:$addr),
2868 Sched<[WriteAdr, WriteSTP]>;
2869 } // hasSideEffects = 0
2873 class BaseLoadStorePairPostIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
2875 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr, $idx", "", []> {
2876 // The operands are in order to match the 'addr' MI operands, so we
2877 // don't need an encoder method and by-name matching. Just use the default
2878 // in-order handling. Since we're using by-order, make sure the names
2884 let Inst{31-30} = opc;
2885 let Inst{29-27} = 0b101;
2887 let Inst{25-23} = 0b001;
2889 let Inst{21-15} = offset;
2890 let Inst{14-10} = dst2;
2891 let Inst{9-5} = base;
2892 let Inst{4-0} = dst;
2894 let DecoderMethod = "DecodePairLdStInstruction";
2897 let hasSideEffects = 0 in {
2898 let mayStore = 0, mayLoad = 1 in
2899 class LoadPairPostIdx<bits<2> opc, bit V, RegisterClass regtype,
2900 Operand idxtype, string asm>
2901 : BaseLoadStorePairPostIdx<opc, V, 1,
2902 (outs regtype:$Rt, regtype:$Rt2),
2903 (ins am_noindex:$addr, idxtype:$idx), asm>,
2904 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
2906 let mayStore = 1, mayLoad = 0 in
2907 class StorePairPostIdx<bits<2> opc, bit V, RegisterClass regtype,
2908 Operand idxtype, string asm>
2909 : BaseLoadStorePairPostIdx<opc, V, 0, (outs),
2910 (ins regtype:$Rt, regtype:$Rt2,
2911 am_noindex:$addr, idxtype:$idx),
2913 Sched<[WriteAdr, WriteSTP]>;
2914 } // hasSideEffects = 0
2918 class BaseLoadStorePairNoAlloc<bits<2> opc, bit V, bit L, dag oops, dag iops,
2920 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr", "", []> {
2921 // The operands are in order to match the 'addr' MI operands, so we
2922 // don't need an encoder method and by-name matching. Just use the default
2923 // in-order handling. Since we're using by-order, make sure the names
2929 let Inst{31-30} = opc;
2930 let Inst{29-27} = 0b101;
2932 let Inst{25-23} = 0b000;
2934 let Inst{21-15} = offset;
2935 let Inst{14-10} = dst2;
2936 let Inst{9-5} = base;
2937 let Inst{4-0} = dst;
2939 let DecoderMethod = "DecodePairLdStInstruction";
2942 let hasSideEffects = 0 in {
2943 let mayStore = 0, mayLoad = 1 in
2944 class LoadPairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
2945 Operand indextype, string asm>
2946 : BaseLoadStorePairNoAlloc<opc, V, 1,
2947 (outs regtype:$Rt, regtype:$Rt2),
2948 (ins indextype:$addr), asm>,
2949 Sched<[WriteLD, WriteLDHi]>;
2951 let mayStore = 1, mayLoad = 0 in
2952 class StorePairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
2953 Operand indextype, string asm>
2954 : BaseLoadStorePairNoAlloc<opc, V, 0, (outs),
2955 (ins regtype:$Rt, regtype:$Rt2, indextype:$addr),
2958 } // hasSideEffects = 0
2961 // Load/store exclusive
2964 // True exclusive operations write to and/or read from the system's exclusive
2965 // monitors, which as far as a compiler is concerned can be modelled as a
2966 // random shared memory address. Hence LoadExclusive mayStore.
2968 // Since these instructions have the undefined register bits set to 1 in
2969 // their canonical form, we need a post encoder method to set those bits
2970 // to 1 when encoding these instructions. We do this using the
2971 // fixLoadStoreExclusive function. This function has template parameters:
2973 // fixLoadStoreExclusive<int hasRs, int hasRt2>
2975 // hasRs indicates that the instruction uses the Rs field, so we won't set
2976 // it to 1 (and the same for Rt2). We don't need template parameters for
2977 // the other register fields since Rt and Rn are always used.
2979 let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
2980 class BaseLoadStoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2981 dag oops, dag iops, string asm, string operands>
2982 : I<oops, iops, asm, operands, "", []> {
2983 let Inst{31-30} = sz;
2984 let Inst{29-24} = 0b001000;
2990 let DecoderMethod = "DecodeExclusiveLdStInstruction";
2993 // Neither Rs nor Rt2 operands.
2994 class LoadStoreExclusiveSimple<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2995 dag oops, dag iops, string asm, string operands>
2996 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, oops, iops, asm, operands> {
2999 let Inst{9-5} = base;
3000 let Inst{4-0} = reg;
3002 let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
3005 // Simple load acquires don't set the exclusive monitor
3006 let mayLoad = 1, mayStore = 0 in
3007 class LoadAcquire<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3008 RegisterClass regtype, string asm>
3009 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
3010 (ins am_noindex:$addr), asm, "\t$Rt, $addr">,
3013 class LoadExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3014 RegisterClass regtype, string asm>
3015 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
3016 (ins am_noindex:$addr), asm, "\t$Rt, $addr">,
3019 class LoadExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3020 RegisterClass regtype, string asm>
3021 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3022 (outs regtype:$Rt, regtype:$Rt2),
3023 (ins am_noindex:$addr), asm,
3024 "\t$Rt, $Rt2, $addr">,
3025 Sched<[WriteLD, WriteLDHi]> {
3029 let Inst{14-10} = dst2;
3030 let Inst{9-5} = base;
3031 let Inst{4-0} = dst1;
3033 let PostEncoderMethod = "fixLoadStoreExclusive<0,1>";
3036 // Simple store release operations do not check the exclusive monitor.
3037 let mayLoad = 0, mayStore = 1 in
3038 class StoreRelease<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3039 RegisterClass regtype, string asm>
3040 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs),
3041 (ins regtype:$Rt, am_noindex:$addr),
3042 asm, "\t$Rt, $addr">,
3045 let mayLoad = 1, mayStore = 1 in
3046 class StoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3047 RegisterClass regtype, string asm>
3048 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, (outs GPR32:$Ws),
3049 (ins regtype:$Rt, am_noindex:$addr),
3050 asm, "\t$Ws, $Rt, $addr">,
3055 let Inst{20-16} = status;
3056 let Inst{9-5} = base;
3057 let Inst{4-0} = reg;
3059 let Constraints = "@earlyclobber $Ws";
3060 let PostEncoderMethod = "fixLoadStoreExclusive<1,0>";
3063 class StoreExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3064 RegisterClass regtype, string asm>
3065 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3067 (ins regtype:$Rt, regtype:$Rt2, am_noindex:$addr),
3068 asm, "\t$Ws, $Rt, $Rt2, $addr">,
3074 let Inst{20-16} = status;
3075 let Inst{14-10} = dst2;
3076 let Inst{9-5} = base;
3077 let Inst{4-0} = dst1;
3079 let Constraints = "@earlyclobber $Ws";
3083 // Exception generation
3086 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3087 class ExceptionGeneration<bits<3> op1, bits<2> ll, string asm>
3088 : I<(outs), (ins imm0_65535:$imm), asm, "\t$imm", "", []>,
3091 let Inst{31-24} = 0b11010100;
3092 let Inst{23-21} = op1;
3093 let Inst{20-5} = imm;
3094 let Inst{4-2} = 0b000;
3098 let Predicates = [HasFPARMv8] in {
3101 // Floating point to integer conversion
3104 class BaseFPToIntegerUnscaled<bits<2> type, bits<2> rmode, bits<3> opcode,
3105 RegisterClass srcType, RegisterClass dstType,
3106 string asm, list<dag> pattern>
3107 : I<(outs dstType:$Rd), (ins srcType:$Rn),
3108 asm, "\t$Rd, $Rn", "", pattern>,
3109 Sched<[WriteFCvt]> {
3112 let Inst{30-29} = 0b00;
3113 let Inst{28-24} = 0b11110;
3114 let Inst{23-22} = type;
3116 let Inst{20-19} = rmode;
3117 let Inst{18-16} = opcode;
3118 let Inst{15-10} = 0;
3123 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3124 class BaseFPToInteger<bits<2> type, bits<2> rmode, bits<3> opcode,
3125 RegisterClass srcType, RegisterClass dstType,
3126 Operand immType, string asm, list<dag> pattern>
3127 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
3128 asm, "\t$Rd, $Rn, $scale", "", pattern>,
3129 Sched<[WriteFCvt]> {
3133 let Inst{30-29} = 0b00;
3134 let Inst{28-24} = 0b11110;
3135 let Inst{23-22} = type;
3137 let Inst{20-19} = rmode;
3138 let Inst{18-16} = opcode;
3139 let Inst{15-10} = scale;
3144 multiclass FPToIntegerUnscaled<bits<2> rmode, bits<3> opcode, string asm,
3145 SDPatternOperator OpN> {
3146 // Unscaled single-precision to 32-bit
3147 def UWSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR32, asm,
3148 [(set GPR32:$Rd, (OpN FPR32:$Rn))]> {
3149 let Inst{31} = 0; // 32-bit GPR flag
3152 // Unscaled single-precision to 64-bit
3153 def UXSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR64, asm,
3154 [(set GPR64:$Rd, (OpN FPR32:$Rn))]> {
3155 let Inst{31} = 1; // 64-bit GPR flag
3158 // Unscaled double-precision to 32-bit
3159 def UWDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR32, asm,
3160 [(set GPR32:$Rd, (OpN (f64 FPR64:$Rn)))]> {
3161 let Inst{31} = 0; // 32-bit GPR flag
3164 // Unscaled double-precision to 64-bit
3165 def UXDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR64, asm,
3166 [(set GPR64:$Rd, (OpN (f64 FPR64:$Rn)))]> {
3167 let Inst{31} = 1; // 64-bit GPR flag
3171 multiclass FPToIntegerScaled<bits<2> rmode, bits<3> opcode, string asm,
3172 SDPatternOperator OpN> {
3173 // Scaled single-precision to 32-bit
3174 def SWSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR32,
3175 fixedpoint_f32_i32, asm,
3176 [(set GPR32:$Rd, (OpN (fmul FPR32:$Rn,
3177 fixedpoint_f32_i32:$scale)))]> {
3178 let Inst{31} = 0; // 32-bit GPR flag
3182 // Scaled single-precision to 64-bit
3183 def SXSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR64,
3184 fixedpoint_f32_i64, asm,
3185 [(set GPR64:$Rd, (OpN (fmul FPR32:$Rn,
3186 fixedpoint_f32_i64:$scale)))]> {
3187 let Inst{31} = 1; // 64-bit GPR flag
3190 // Scaled double-precision to 32-bit
3191 def SWDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR32,
3192 fixedpoint_f64_i32, asm,
3193 [(set GPR32:$Rd, (OpN (fmul FPR64:$Rn,
3194 fixedpoint_f64_i32:$scale)))]> {
3195 let Inst{31} = 0; // 32-bit GPR flag
3199 // Scaled double-precision to 64-bit
3200 def SXDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR64,
3201 fixedpoint_f64_i64, asm,
3202 [(set GPR64:$Rd, (OpN (fmul FPR64:$Rn,
3203 fixedpoint_f64_i64:$scale)))]> {
3204 let Inst{31} = 1; // 64-bit GPR flag
3209 // Integer to floating point conversion
3212 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
3213 class BaseIntegerToFP<bit isUnsigned,
3214 RegisterClass srcType, RegisterClass dstType,
3215 Operand immType, string asm, list<dag> pattern>
3216 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
3217 asm, "\t$Rd, $Rn, $scale", "", pattern>,
3218 Sched<[WriteFCvt]> {
3222 let Inst{30-23} = 0b00111100;
3223 let Inst{21-17} = 0b00001;
3224 let Inst{16} = isUnsigned;
3225 let Inst{15-10} = scale;
3230 class BaseIntegerToFPUnscaled<bit isUnsigned,
3231 RegisterClass srcType, RegisterClass dstType,
3232 ValueType dvt, string asm, SDNode node>
3233 : I<(outs dstType:$Rd), (ins srcType:$Rn),
3234 asm, "\t$Rd, $Rn", "", [(set (dvt dstType:$Rd), (node srcType:$Rn))]>,
3235 Sched<[WriteFCvt]> {
3239 let Inst{30-23} = 0b00111100;
3240 let Inst{21-17} = 0b10001;
3241 let Inst{16} = isUnsigned;
3242 let Inst{15-10} = 0b000000;
3247 multiclass IntegerToFP<bit isUnsigned, string asm, SDNode node> {
3249 def UWSri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR32, f32, asm, node> {
3250 let Inst{31} = 0; // 32-bit GPR flag
3251 let Inst{22} = 0; // 32-bit FPR flag
3254 def UWDri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR64, f64, asm, node> {
3255 let Inst{31} = 0; // 32-bit GPR flag
3256 let Inst{22} = 1; // 64-bit FPR flag
3259 def UXSri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR32, f32, asm, node> {
3260 let Inst{31} = 1; // 64-bit GPR flag
3261 let Inst{22} = 0; // 32-bit FPR flag
3264 def UXDri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR64, f64, asm, node> {
3265 let Inst{31} = 1; // 64-bit GPR flag
3266 let Inst{22} = 1; // 64-bit FPR flag
3270 def SWSri: BaseIntegerToFP<isUnsigned, GPR32, FPR32, fixedpoint_f32_i32, asm,
3272 (fdiv (node GPR32:$Rn),
3273 fixedpoint_f32_i32:$scale))]> {
3274 let Inst{31} = 0; // 32-bit GPR flag
3275 let Inst{22} = 0; // 32-bit FPR flag
3279 def SWDri: BaseIntegerToFP<isUnsigned, GPR32, FPR64, fixedpoint_f64_i32, asm,
3281 (fdiv (node GPR32:$Rn),
3282 fixedpoint_f64_i32:$scale))]> {
3283 let Inst{31} = 0; // 32-bit GPR flag
3284 let Inst{22} = 1; // 64-bit FPR flag
3288 def SXSri: BaseIntegerToFP<isUnsigned, GPR64, FPR32, fixedpoint_f32_i64, asm,
3290 (fdiv (node GPR64:$Rn),
3291 fixedpoint_f32_i64:$scale))]> {
3292 let Inst{31} = 1; // 64-bit GPR flag
3293 let Inst{22} = 0; // 32-bit FPR flag
3296 def SXDri: BaseIntegerToFP<isUnsigned, GPR64, FPR64, fixedpoint_f64_i64, asm,
3298 (fdiv (node GPR64:$Rn),
3299 fixedpoint_f64_i64:$scale))]> {
3300 let Inst{31} = 1; // 64-bit GPR flag
3301 let Inst{22} = 1; // 64-bit FPR flag
3306 // Unscaled integer <-> floating point conversion (i.e. FMOV)
3309 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3310 class BaseUnscaledConversion<bits<2> rmode, bits<3> opcode,
3311 RegisterClass srcType, RegisterClass dstType,
3313 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "",
3314 // We use COPY_TO_REGCLASS for these bitconvert operations.
3315 // copyPhysReg() expands the resultant COPY instructions after
3316 // regalloc is done. This gives greater freedom for the allocator
3317 // and related passes (coalescing, copy propagation, et. al.) to
3318 // be more effective.
3319 [/*(set (dvt dstType:$Rd), (bitconvert (svt srcType:$Rn)))*/]>,
3320 Sched<[WriteFCopy]> {
3323 let Inst{30-23} = 0b00111100;
3325 let Inst{20-19} = rmode;
3326 let Inst{18-16} = opcode;
3327 let Inst{15-10} = 0b000000;
3332 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3333 class BaseUnscaledConversionToHigh<bits<2> rmode, bits<3> opcode,
3334 RegisterClass srcType, RegisterOperand dstType, string asm,
3336 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm,
3337 "{\t$Rd"#kind#"[1], $Rn|"#kind#"\t$Rd[1], $Rn}", "", []>,
3338 Sched<[WriteFCopy]> {
3341 let Inst{30-23} = 0b00111101;
3343 let Inst{20-19} = rmode;
3344 let Inst{18-16} = opcode;
3345 let Inst{15-10} = 0b000000;
3350 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3351 class BaseUnscaledConversionFromHigh<bits<2> rmode, bits<3> opcode,
3352 RegisterOperand srcType, RegisterClass dstType, string asm,
3354 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm,
3355 "{\t$Rd, $Rn"#kind#"[1]|"#kind#"\t$Rd, $Rn[1]}", "", []>,
3356 Sched<[WriteFCopy]> {
3359 let Inst{30-23} = 0b00111101;
3361 let Inst{20-19} = rmode;
3362 let Inst{18-16} = opcode;
3363 let Inst{15-10} = 0b000000;
3370 multiclass UnscaledConversion<string asm> {
3371 def WSr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR32, asm> {
3372 let Inst{31} = 0; // 32-bit GPR flag
3373 let Inst{22} = 0; // 32-bit FPR flag
3376 def XDr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR64, asm> {
3377 let Inst{31} = 1; // 64-bit GPR flag
3378 let Inst{22} = 1; // 64-bit FPR flag
3381 def SWr : BaseUnscaledConversion<0b00, 0b110, FPR32, GPR32, asm> {
3382 let Inst{31} = 0; // 32-bit GPR flag
3383 let Inst{22} = 0; // 32-bit FPR flag
3386 def DXr : BaseUnscaledConversion<0b00, 0b110, FPR64, GPR64, asm> {
3387 let Inst{31} = 1; // 64-bit GPR flag
3388 let Inst{22} = 1; // 64-bit FPR flag
3391 def XDHighr : BaseUnscaledConversionToHigh<0b01, 0b111, GPR64, V128,
3397 def DXHighr : BaseUnscaledConversionFromHigh<0b01, 0b110, V128, GPR64,
3405 // Floating point conversion
3408 class BaseFPConversion<bits<2> type, bits<2> opcode, RegisterClass dstType,
3409 RegisterClass srcType, string asm, list<dag> pattern>
3410 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "", pattern>,
3411 Sched<[WriteFCvt]> {
3414 let Inst{31-24} = 0b00011110;
3415 let Inst{23-22} = type;
3416 let Inst{21-17} = 0b10001;
3417 let Inst{16-15} = opcode;
3418 let Inst{14-10} = 0b10000;
3423 multiclass FPConversion<string asm> {
3424 // Double-precision to Half-precision
3425 def HDr : BaseFPConversion<0b01, 0b11, FPR16, FPR64, asm,
3426 [(set FPR16:$Rd, (fround FPR64:$Rn))]>;
3428 // Double-precision to Single-precision
3429 def SDr : BaseFPConversion<0b01, 0b00, FPR32, FPR64, asm,
3430 [(set FPR32:$Rd, (fround FPR64:$Rn))]>;
3432 // Half-precision to Double-precision
3433 def DHr : BaseFPConversion<0b11, 0b01, FPR64, FPR16, asm,
3434 [(set FPR64:$Rd, (fextend FPR16:$Rn))]>;
3436 // Half-precision to Single-precision
3437 def SHr : BaseFPConversion<0b11, 0b00, FPR32, FPR16, asm,
3438 [(set FPR32:$Rd, (fextend FPR16:$Rn))]>;
3440 // Single-precision to Double-precision
3441 def DSr : BaseFPConversion<0b00, 0b01, FPR64, FPR32, asm,
3442 [(set FPR64:$Rd, (fextend FPR32:$Rn))]>;
3444 // Single-precision to Half-precision
3445 def HSr : BaseFPConversion<0b00, 0b11, FPR16, FPR32, asm,
3446 [(set FPR16:$Rd, (fround FPR32:$Rn))]>;
3450 // Single operand floating point data processing
3453 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3454 class BaseSingleOperandFPData<bits<4> opcode, RegisterClass regtype,
3455 ValueType vt, string asm, SDPatternOperator node>
3456 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
3457 [(set (vt regtype:$Rd), (node (vt regtype:$Rn)))]>,
3461 let Inst{31-23} = 0b000111100;
3462 let Inst{21-19} = 0b100;
3463 let Inst{18-15} = opcode;
3464 let Inst{14-10} = 0b10000;
3469 multiclass SingleOperandFPData<bits<4> opcode, string asm,
3470 SDPatternOperator node = null_frag> {
3471 def Sr : BaseSingleOperandFPData<opcode, FPR32, f32, asm, node> {
3472 let Inst{22} = 0; // 32-bit size flag
3475 def Dr : BaseSingleOperandFPData<opcode, FPR64, f64, asm, node> {
3476 let Inst{22} = 1; // 64-bit size flag
3481 // Two operand floating point data processing
3484 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3485 class BaseTwoOperandFPData<bits<4> opcode, RegisterClass regtype,
3486 string asm, list<dag> pat>
3487 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
3488 asm, "\t$Rd, $Rn, $Rm", "", pat>,
3493 let Inst{31-23} = 0b000111100;
3495 let Inst{20-16} = Rm;
3496 let Inst{15-12} = opcode;
3497 let Inst{11-10} = 0b10;
3502 multiclass TwoOperandFPData<bits<4> opcode, string asm,
3503 SDPatternOperator node = null_frag> {
3504 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
3505 [(set (f32 FPR32:$Rd),
3506 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]> {
3507 let Inst{22} = 0; // 32-bit size flag
3510 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
3511 [(set (f64 FPR64:$Rd),
3512 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]> {
3513 let Inst{22} = 1; // 64-bit size flag
3517 multiclass TwoOperandFPDataNeg<bits<4> opcode, string asm, SDNode node> {
3518 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
3519 [(set FPR32:$Rd, (fneg (node FPR32:$Rn, (f32 FPR32:$Rm))))]> {
3520 let Inst{22} = 0; // 32-bit size flag
3523 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
3524 [(set FPR64:$Rd, (fneg (node FPR64:$Rn, (f64 FPR64:$Rm))))]> {
3525 let Inst{22} = 1; // 64-bit size flag
3531 // Three operand floating point data processing
3534 class BaseThreeOperandFPData<bit isNegated, bit isSub,
3535 RegisterClass regtype, string asm, list<dag> pat>
3536 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, regtype: $Ra),
3537 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pat>,
3538 Sched<[WriteFMul]> {
3543 let Inst{31-23} = 0b000111110;
3544 let Inst{21} = isNegated;
3545 let Inst{20-16} = Rm;
3546 let Inst{15} = isSub;
3547 let Inst{14-10} = Ra;
3552 multiclass ThreeOperandFPData<bit isNegated, bit isSub,string asm,
3553 SDPatternOperator node> {
3554 def Srrr : BaseThreeOperandFPData<isNegated, isSub, FPR32, asm,
3556 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm), (f32 FPR32:$Ra)))]> {
3557 let Inst{22} = 0; // 32-bit size flag
3560 def Drrr : BaseThreeOperandFPData<isNegated, isSub, FPR64, asm,
3562 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm), (f64 FPR64:$Ra)))]> {
3563 let Inst{22} = 1; // 64-bit size flag
3568 // Floating point data comparisons
3571 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3572 class BaseOneOperandFPComparison<bit signalAllNans,
3573 RegisterClass regtype, string asm,
3575 : I<(outs), (ins regtype:$Rn), asm, "\t$Rn, #0.0", "", pat>,
3576 Sched<[WriteFCmp]> {
3578 let Inst{31-23} = 0b000111100;
3581 let Inst{15-10} = 0b001000;
3583 let Inst{4} = signalAllNans;
3584 let Inst{3-0} = 0b1000;
3586 // Rm should be 0b00000 canonically, but we need to accept any value.
3587 let PostEncoderMethod = "fixOneOperandFPComparison";
3590 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3591 class BaseTwoOperandFPComparison<bit signalAllNans, RegisterClass regtype,
3592 string asm, list<dag> pat>
3593 : I<(outs), (ins regtype:$Rn, regtype:$Rm), asm, "\t$Rn, $Rm", "", pat>,
3594 Sched<[WriteFCmp]> {
3597 let Inst{31-23} = 0b000111100;
3599 let Inst{20-16} = Rm;
3600 let Inst{15-10} = 0b001000;
3602 let Inst{4} = signalAllNans;
3603 let Inst{3-0} = 0b0000;
3606 multiclass FPComparison<bit signalAllNans, string asm,
3607 SDPatternOperator OpNode = null_frag> {
3608 let Defs = [CPSR] in {
3609 def Srr : BaseTwoOperandFPComparison<signalAllNans, FPR32, asm,
3610 [(OpNode FPR32:$Rn, (f32 FPR32:$Rm)), (implicit CPSR)]> {
3614 def Sri : BaseOneOperandFPComparison<signalAllNans, FPR32, asm,
3615 [(OpNode (f32 FPR32:$Rn), fpimm0), (implicit CPSR)]> {
3619 def Drr : BaseTwoOperandFPComparison<signalAllNans, FPR64, asm,
3620 [(OpNode FPR64:$Rn, (f64 FPR64:$Rm)), (implicit CPSR)]> {
3624 def Dri : BaseOneOperandFPComparison<signalAllNans, FPR64, asm,
3625 [(OpNode (f64 FPR64:$Rn), fpimm0), (implicit CPSR)]> {
3632 // Floating point conditional comparisons
3635 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3636 class BaseFPCondComparison<bit signalAllNans,
3637 RegisterClass regtype, string asm>
3638 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm0_15:$nzcv, ccode:$cond),
3639 asm, "\t$Rn, $Rm, $nzcv, $cond", "", []>,
3640 Sched<[WriteFCmp]> {
3646 let Inst{31-23} = 0b000111100;
3648 let Inst{20-16} = Rm;
3649 let Inst{15-12} = cond;
3650 let Inst{11-10} = 0b01;
3652 let Inst{4} = signalAllNans;
3653 let Inst{3-0} = nzcv;
3656 multiclass FPCondComparison<bit signalAllNans, string asm> {
3657 let Defs = [CPSR], Uses = [CPSR] in {
3658 def Srr : BaseFPCondComparison<signalAllNans, FPR32, asm> {
3662 def Drr : BaseFPCondComparison<signalAllNans, FPR64, asm> {
3665 } // Defs = [CPSR], Uses = [CPSR]
3669 // Floating point conditional select
3672 class BaseFPCondSelect<RegisterClass regtype, ValueType vt, string asm>
3673 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
3674 asm, "\t$Rd, $Rn, $Rm, $cond", "",
3676 (ARM64csel (vt regtype:$Rn), regtype:$Rm,
3677 (i32 imm:$cond), CPSR))]>,
3684 let Inst{31-23} = 0b000111100;
3686 let Inst{20-16} = Rm;
3687 let Inst{15-12} = cond;
3688 let Inst{11-10} = 0b11;
3693 multiclass FPCondSelect<string asm> {
3694 let Uses = [CPSR] in {
3695 def Srrr : BaseFPCondSelect<FPR32, f32, asm> {
3699 def Drrr : BaseFPCondSelect<FPR64, f64, asm> {
3706 // Floating move immediate
3709 class BaseFPMoveImmediate<RegisterClass regtype, Operand fpimmtype, string asm>
3710 : I<(outs regtype:$Rd), (ins fpimmtype:$imm), asm, "\t$Rd, $imm", "",
3711 [(set regtype:$Rd, fpimmtype:$imm)]>,
3712 Sched<[WriteFImm]> {
3715 let Inst{31-23} = 0b000111100;
3717 let Inst{20-13} = imm;
3718 let Inst{12-5} = 0b10000000;
3722 multiclass FPMoveImmediate<string asm> {
3723 def Si : BaseFPMoveImmediate<FPR32, fpimm32, asm> {
3727 def Di : BaseFPMoveImmediate<FPR64, fpimm64, asm> {
3731 } // end of 'let Predicates = [HasFPARMv8]'
3733 //----------------------------------------------------------------------------
3735 //----------------------------------------------------------------------------
3737 def VectorIndexBOperand : AsmOperandClass { let Name = "VectorIndexB"; }
3738 def VectorIndexHOperand : AsmOperandClass { let Name = "VectorIndexH"; }
3739 def VectorIndexSOperand : AsmOperandClass { let Name = "VectorIndexS"; }
3740 def VectorIndexDOperand : AsmOperandClass { let Name = "VectorIndexD"; }
3741 def VectorIndexB : Operand<i64>, ImmLeaf<i64, [{
3742 return ((uint64_t)Imm) < 16;
3744 let ParserMatchClass = VectorIndexBOperand;
3745 let PrintMethod = "printVectorIndex";
3746 let MIOperandInfo = (ops i64imm);
3748 def VectorIndexH : Operand<i64>, ImmLeaf<i64, [{
3749 return ((uint64_t)Imm) < 8;
3751 let ParserMatchClass = VectorIndexHOperand;
3752 let PrintMethod = "printVectorIndex";
3753 let MIOperandInfo = (ops i64imm);
3755 def VectorIndexS : Operand<i64>, ImmLeaf<i64, [{
3756 return ((uint64_t)Imm) < 4;
3758 let ParserMatchClass = VectorIndexSOperand;
3759 let PrintMethod = "printVectorIndex";
3760 let MIOperandInfo = (ops i64imm);
3762 def VectorIndexD : Operand<i64>, ImmLeaf<i64, [{
3763 return ((uint64_t)Imm) < 2;
3765 let ParserMatchClass = VectorIndexDOperand;
3766 let PrintMethod = "printVectorIndex";
3767 let MIOperandInfo = (ops i64imm);
3770 def MemorySIMDNoIndexOperand : AsmOperandClass {
3771 let Name = "MemorySIMDNoIndex";
3772 let ParserMethod = "tryParseNoIndexMemory";
3774 def am_simdnoindex : Operand<i64>,
3775 ComplexPattern<i64, 1, "SelectAddrModeNoIndex", []> {
3776 let PrintMethod = "printAMNoIndex";
3777 let ParserMatchClass = MemorySIMDNoIndexOperand;
3778 let MIOperandInfo = (ops GPR64sp:$base);
3779 let DecoderMethod = "DecodeGPR64spRegisterClass";
3782 let Predicates = [HasNEON] in {
3784 //----------------------------------------------------------------------------
3785 // AdvSIMD three register vector instructions
3786 //----------------------------------------------------------------------------
3788 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3789 class BaseSIMDThreeSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
3790 RegisterOperand regtype, string asm, string kind,
3792 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
3793 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
3794 "|" # kind # "\t$Rd, $Rn, $Rm|}", "", pattern>,
3802 let Inst{28-24} = 0b01110;
3803 let Inst{23-22} = size;
3805 let Inst{20-16} = Rm;
3806 let Inst{15-11} = opcode;
3812 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3813 class BaseSIMDThreeSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
3814 RegisterOperand regtype, string asm, string kind,
3816 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn, regtype:$Rm), asm,
3817 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
3818 "|" # kind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
3826 let Inst{28-24} = 0b01110;
3827 let Inst{23-22} = size;
3829 let Inst{20-16} = Rm;
3830 let Inst{15-11} = opcode;
3836 // All operand sizes distinguished in the encoding.
3837 multiclass SIMDThreeSameVector<bit U, bits<5> opc, string asm,
3838 SDPatternOperator OpNode> {
3839 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3841 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3842 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3844 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3845 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3847 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3848 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3850 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3851 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3853 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3854 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3856 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3857 def v2i64 : BaseSIMDThreeSameVector<1, U, 0b11, opc, V128,
3859 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
3862 // As above, but D sized elements unsupported.
3863 multiclass SIMDThreeSameVectorBHS<bit U, bits<5> opc, string asm,
3864 SDPatternOperator OpNode> {
3865 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3867 [(set V64:$Rd, (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))]>;
3868 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3870 [(set V128:$Rd, (v16i8 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm))))]>;
3871 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3873 [(set V64:$Rd, (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))]>;
3874 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3876 [(set V128:$Rd, (v8i16 (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm))))]>;
3877 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3879 [(set V64:$Rd, (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))]>;
3880 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3882 [(set V128:$Rd, (v4i32 (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm))))]>;
3885 multiclass SIMDThreeSameVectorBHSTied<bit U, bits<5> opc, string asm,
3886 SDPatternOperator OpNode> {
3887 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, 0b00, opc, V64,
3889 [(set (v8i8 V64:$dst),
3890 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3891 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, 0b00, opc, V128,
3893 [(set (v16i8 V128:$dst),
3894 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3895 def v4i16 : BaseSIMDThreeSameVectorTied<0, U, 0b01, opc, V64,
3897 [(set (v4i16 V64:$dst),
3898 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3899 def v8i16 : BaseSIMDThreeSameVectorTied<1, U, 0b01, opc, V128,
3901 [(set (v8i16 V128:$dst),
3902 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3903 def v2i32 : BaseSIMDThreeSameVectorTied<0, U, 0b10, opc, V64,
3905 [(set (v2i32 V64:$dst),
3906 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3907 def v4i32 : BaseSIMDThreeSameVectorTied<1, U, 0b10, opc, V128,
3909 [(set (v4i32 V128:$dst),
3910 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3913 // As above, but only B sized elements supported.
3914 multiclass SIMDThreeSameVectorB<bit U, bits<5> opc, string asm,
3915 SDPatternOperator OpNode> {
3916 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3918 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3919 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3921 [(set (v16i8 V128:$Rd),
3922 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3925 // As above, but only S and D sized floating point elements supported.
3926 multiclass SIMDThreeSameVectorFP<bit U, bit S, bits<5> opc,
3927 string asm, SDPatternOperator OpNode> {
3928 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0}, opc, V64,
3930 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3931 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0}, opc, V128,
3933 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3934 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,1}, opc, V128,
3936 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3939 multiclass SIMDThreeSameVectorFPCmp<bit U, bit S, bits<5> opc,
3941 SDPatternOperator OpNode> {
3942 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0}, opc, V64,
3944 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3945 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0}, opc, V128,
3947 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3948 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,1}, opc, V128,
3950 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3953 multiclass SIMDThreeSameVectorFPTied<bit U, bit S, bits<5> opc,
3954 string asm, SDPatternOperator OpNode> {
3955 def v2f32 : BaseSIMDThreeSameVectorTied<0, U, {S,0}, opc, V64,
3957 [(set (v2f32 V64:$dst),
3958 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3959 def v4f32 : BaseSIMDThreeSameVectorTied<1, U, {S,0}, opc, V128,
3961 [(set (v4f32 V128:$dst),
3962 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3963 def v2f64 : BaseSIMDThreeSameVectorTied<1, U, {S,1}, opc, V128,
3965 [(set (v2f64 V128:$dst),
3966 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3969 // As above, but D and B sized elements unsupported.
3970 multiclass SIMDThreeSameVectorHS<bit U, bits<5> opc, string asm,
3971 SDPatternOperator OpNode> {
3972 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3974 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3975 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3977 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3978 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3980 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3981 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3983 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3986 // Logical three vector ops share opcode bits, and only use B sized elements.
3987 multiclass SIMDLogicalThreeVector<bit U, bits<2> size, string asm,
3988 SDPatternOperator OpNode = null_frag> {
3989 def v8i8 : BaseSIMDThreeSameVector<0, U, size, 0b00011, V64,
3991 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn, V64:$Rm))]>;
3992 def v16i8 : BaseSIMDThreeSameVector<1, U, size, 0b00011, V128,
3994 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn, V128:$Rm))]>;
3996 def : Pat<(v4i16 (OpNode V64:$LHS, V64:$RHS)),
3997 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
3998 def : Pat<(v2i32 (OpNode V64:$LHS, V64:$RHS)),
3999 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
4000 def : Pat<(v1i64 (OpNode V64:$LHS, V64:$RHS)),
4001 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
4003 def : Pat<(v8i16 (OpNode V128:$LHS, V128:$RHS)),
4004 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
4005 def : Pat<(v4i32 (OpNode V128:$LHS, V128:$RHS)),
4006 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
4007 def : Pat<(v2i64 (OpNode V128:$LHS, V128:$RHS)),
4008 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
4011 multiclass SIMDLogicalThreeVectorTied<bit U, bits<2> size,
4012 string asm, SDPatternOperator OpNode> {
4013 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, size, 0b00011, V64,
4015 [(set (v8i8 V64:$dst),
4016 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4017 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, size, 0b00011, V128,
4019 [(set (v16i8 V128:$dst),
4020 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
4021 (v16i8 V128:$Rm)))]>;
4023 def : Pat<(v4i16 (OpNode (v4i16 V64:$LHS), (v4i16 V64:$MHS),
4025 (!cast<Instruction>(NAME#"v8i8")
4026 V64:$LHS, V64:$MHS, V64:$RHS)>;
4027 def : Pat<(v2i32 (OpNode (v2i32 V64:$LHS), (v2i32 V64:$MHS),
4029 (!cast<Instruction>(NAME#"v8i8")
4030 V64:$LHS, V64:$MHS, V64:$RHS)>;
4031 def : Pat<(v1i64 (OpNode (v1i64 V64:$LHS), (v1i64 V64:$MHS),
4033 (!cast<Instruction>(NAME#"v8i8")
4034 V64:$LHS, V64:$MHS, V64:$RHS)>;
4036 def : Pat<(v8i16 (OpNode (v8i16 V128:$LHS), (v8i16 V128:$MHS),
4037 (v8i16 V128:$RHS))),
4038 (!cast<Instruction>(NAME#"v16i8")
4039 V128:$LHS, V128:$MHS, V128:$RHS)>;
4040 def : Pat<(v4i32 (OpNode (v4i32 V128:$LHS), (v4i32 V128:$MHS),
4041 (v4i32 V128:$RHS))),
4042 (!cast<Instruction>(NAME#"v16i8")
4043 V128:$LHS, V128:$MHS, V128:$RHS)>;
4044 def : Pat<(v2i64 (OpNode (v2i64 V128:$LHS), (v2i64 V128:$MHS),
4045 (v2i64 V128:$RHS))),
4046 (!cast<Instruction>(NAME#"v16i8")
4047 V128:$LHS, V128:$MHS, V128:$RHS)>;
4051 //----------------------------------------------------------------------------
4052 // AdvSIMD two register vector instructions.
4053 //----------------------------------------------------------------------------
4055 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4056 class BaseSIMDTwoSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4057 RegisterOperand regtype, string asm, string dstkind,
4058 string srckind, list<dag> pattern>
4059 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
4060 "{\t$Rd" # dstkind # ", $Rn" # srckind #
4061 "|" # dstkind # "\t$Rd, $Rn}", "", pattern>,
4068 let Inst{28-24} = 0b01110;
4069 let Inst{23-22} = size;
4070 let Inst{21-17} = 0b10000;
4071 let Inst{16-12} = opcode;
4072 let Inst{11-10} = 0b10;
4077 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4078 class BaseSIMDTwoSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4079 RegisterOperand regtype, string asm, string dstkind,
4080 string srckind, list<dag> pattern>
4081 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn), asm,
4082 "{\t$Rd" # dstkind # ", $Rn" # srckind #
4083 "|" # dstkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
4090 let Inst{28-24} = 0b01110;
4091 let Inst{23-22} = size;
4092 let Inst{21-17} = 0b10000;
4093 let Inst{16-12} = opcode;
4094 let Inst{11-10} = 0b10;
4099 // Supports B, H, and S element sizes.
4100 multiclass SIMDTwoVectorBHS<bit U, bits<5> opc, string asm,
4101 SDPatternOperator OpNode> {
4102 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4104 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4105 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4106 asm, ".16b", ".16b",
4107 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4108 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4110 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4111 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4113 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4114 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4116 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4117 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4119 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4122 class BaseSIMDVectorLShiftLongBySize<bit Q, bits<2> size,
4123 RegisterOperand regtype, string asm, string dstkind,
4124 string srckind, string amount>
4125 : I<(outs V128:$Rd), (ins regtype:$Rn), asm,
4126 "{\t$Rd" # dstkind # ", $Rn" # srckind # ", #" # amount #
4127 "|" # dstkind # "\t$Rd, $Rn, #" # amount # "}", "", []>,
4133 let Inst{29-24} = 0b101110;
4134 let Inst{23-22} = size;
4135 let Inst{21-10} = 0b100001001110;
4140 multiclass SIMDVectorLShiftLongBySizeBHS {
4141 let neverHasSideEffects = 1 in {
4142 def v8i8 : BaseSIMDVectorLShiftLongBySize<0, 0b00, V64,
4143 "shll", ".8h", ".8b", "8">;
4144 def v16i8 : BaseSIMDVectorLShiftLongBySize<1, 0b00, V128,
4145 "shll2", ".8h", ".16b", "8">;
4146 def v4i16 : BaseSIMDVectorLShiftLongBySize<0, 0b01, V64,
4147 "shll", ".4s", ".4h", "16">;
4148 def v8i16 : BaseSIMDVectorLShiftLongBySize<1, 0b01, V128,
4149 "shll2", ".4s", ".8h", "16">;
4150 def v2i32 : BaseSIMDVectorLShiftLongBySize<0, 0b10, V64,
4151 "shll", ".2d", ".2s", "32">;
4152 def v4i32 : BaseSIMDVectorLShiftLongBySize<1, 0b10, V128,
4153 "shll2", ".2d", ".4s", "32">;
4157 // Supports all element sizes.
4158 multiclass SIMDLongTwoVector<bit U, bits<5> opc, string asm,
4159 SDPatternOperator OpNode> {
4160 def v8i8_v4i16 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4162 [(set (v4i16 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4163 def v16i8_v8i16 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4165 [(set (v8i16 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4166 def v4i16_v2i32 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4168 [(set (v2i32 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4169 def v8i16_v4i32 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4171 [(set (v4i32 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4172 def v2i32_v1i64 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4174 [(set (v1i64 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4175 def v4i32_v2i64 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4177 [(set (v2i64 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4180 multiclass SIMDLongTwoVectorTied<bit U, bits<5> opc, string asm,
4181 SDPatternOperator OpNode> {
4182 def v8i8_v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, V64,
4184 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd),
4186 def v16i8_v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, V128,
4188 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd),
4189 (v16i8 V128:$Rn)))]>;
4190 def v4i16_v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, V64,
4192 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd),
4193 (v4i16 V64:$Rn)))]>;
4194 def v8i16_v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, V128,
4196 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd),
4197 (v8i16 V128:$Rn)))]>;
4198 def v2i32_v1i64 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, V64,
4200 [(set (v1i64 V64:$dst), (OpNode (v1i64 V64:$Rd),
4201 (v2i32 V64:$Rn)))]>;
4202 def v4i32_v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, V128,
4204 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd),
4205 (v4i32 V128:$Rn)))]>;
4208 // Supports all element sizes, except 1xD.
4209 multiclass SIMDTwoVectorBHSDTied<bit U, bits<5> opc, string asm,
4210 SDPatternOperator OpNode> {
4211 def v8i8 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, V64,
4213 [(set (v8i8 V64:$dst), (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn)))]>;
4214 def v16i8 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, V128,
4215 asm, ".16b", ".16b",
4216 [(set (v16i8 V128:$dst), (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
4217 def v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, V64,
4219 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn)))]>;
4220 def v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, V128,
4222 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn)))]>;
4223 def v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, V64,
4225 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn)))]>;
4226 def v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, V128,
4228 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
4229 def v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b11, opc, V128,
4231 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn)))]>;
4234 multiclass SIMDTwoVectorBHSD<bit U, bits<5> opc, string asm,
4235 SDPatternOperator OpNode = null_frag> {
4236 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4238 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4239 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4240 asm, ".16b", ".16b",
4241 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4242 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4244 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4245 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4247 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4248 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4250 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4251 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4253 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4254 def v2i64 : BaseSIMDTwoSameVector<1, U, 0b11, opc, V128,
4256 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4260 // Supports only B element sizes.
4261 multiclass SIMDTwoVectorB<bit U, bits<2> size, bits<5> opc, string asm,
4262 SDPatternOperator OpNode> {
4263 def v8i8 : BaseSIMDTwoSameVector<0, U, size, opc, V64,
4265 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4266 def v16i8 : BaseSIMDTwoSameVector<1, U, size, opc, V128,
4267 asm, ".16b", ".16b",
4268 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4272 // Supports only B and H element sizes.
4273 multiclass SIMDTwoVectorBH<bit U, bits<5> opc, string asm,
4274 SDPatternOperator OpNode> {
4275 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4277 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn))]>;
4278 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4279 asm, ".16b", ".16b",
4280 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn))]>;
4281 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4283 [(set (v4i16 V64:$Rd), (OpNode V64:$Rn))]>;
4284 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4286 [(set (v8i16 V128:$Rd), (OpNode V128:$Rn))]>;
4289 // Supports only S and D element sizes, uses high bit of the size field
4290 // as an extra opcode bit.
4291 multiclass SIMDTwoVectorFP<bit U, bit S, bits<5> opc, string asm,
4292 SDPatternOperator OpNode> {
4293 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4295 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
4296 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4298 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
4299 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4301 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4304 // Supports only S element size.
4305 multiclass SIMDTwoVectorS<bit U, bit S, bits<5> opc, string asm,
4306 SDPatternOperator OpNode> {
4307 def v2i32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4309 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4310 def v4i32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4312 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4316 multiclass SIMDTwoVectorFPToInt<bit U, bit S, bits<5> opc, string asm,
4317 SDPatternOperator OpNode> {
4318 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4320 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
4321 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4323 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
4324 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4326 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4329 multiclass SIMDTwoVectorIntToFP<bit U, bit S, bits<5> opc, string asm,
4330 SDPatternOperator OpNode> {
4331 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4333 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4334 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4336 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4337 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4339 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4343 class BaseSIMDMixedTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4344 RegisterOperand inreg, RegisterOperand outreg,
4345 string asm, string outkind, string inkind,
4347 : I<(outs outreg:$Rd), (ins inreg:$Rn), asm,
4348 "{\t$Rd" # outkind # ", $Rn" # inkind #
4349 "|" # outkind # "\t$Rd, $Rn}", "", pattern>,
4356 let Inst{28-24} = 0b01110;
4357 let Inst{23-22} = size;
4358 let Inst{21-17} = 0b10000;
4359 let Inst{16-12} = opcode;
4360 let Inst{11-10} = 0b10;
4365 class BaseSIMDMixedTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4366 RegisterOperand inreg, RegisterOperand outreg,
4367 string asm, string outkind, string inkind,
4369 : I<(outs outreg:$dst), (ins outreg:$Rd, inreg:$Rn), asm,
4370 "{\t$Rd" # outkind # ", $Rn" # inkind #
4371 "|" # outkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
4378 let Inst{28-24} = 0b01110;
4379 let Inst{23-22} = size;
4380 let Inst{21-17} = 0b10000;
4381 let Inst{16-12} = opcode;
4382 let Inst{11-10} = 0b10;
4387 multiclass SIMDMixedTwoVector<bit U, bits<5> opc, string asm,
4388 SDPatternOperator OpNode> {
4389 def v8i8 : BaseSIMDMixedTwoVector<0, U, 0b00, opc, V128, V64,
4391 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4392 def v16i8 : BaseSIMDMixedTwoVectorTied<1, U, 0b00, opc, V128, V128,
4393 asm#"2", ".16b", ".8h", []>;
4394 def v4i16 : BaseSIMDMixedTwoVector<0, U, 0b01, opc, V128, V64,
4396 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4397 def v8i16 : BaseSIMDMixedTwoVectorTied<1, U, 0b01, opc, V128, V128,
4398 asm#"2", ".8h", ".4s", []>;
4399 def v2i32 : BaseSIMDMixedTwoVector<0, U, 0b10, opc, V128, V64,
4401 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4402 def v4i32 : BaseSIMDMixedTwoVectorTied<1, U, 0b10, opc, V128, V128,
4403 asm#"2", ".4s", ".2d", []>;
4405 def : Pat<(concat_vectors (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn))),
4406 (!cast<Instruction>(NAME # "v16i8")
4407 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4408 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn))),
4409 (!cast<Instruction>(NAME # "v8i16")
4410 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4411 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn))),
4412 (!cast<Instruction>(NAME # "v4i32")
4413 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4416 class BaseSIMDCmpTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4417 RegisterOperand regtype, string asm, string kind,
4418 ValueType dty, ValueType sty, SDNode OpNode>
4419 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
4420 "{\t$Rd" # kind # ", $Rn" # kind # ", #0" #
4421 "|" # kind # "\t$Rd, $Rn, #0}", "",
4422 [(set (dty regtype:$Rd), (OpNode (sty regtype:$Rn)))]>,
4429 let Inst{28-24} = 0b01110;
4430 let Inst{23-22} = size;
4431 let Inst{21-17} = 0b10000;
4432 let Inst{16-12} = opcode;
4433 let Inst{11-10} = 0b10;
4438 // Comparisons support all element sizes, except 1xD.
4439 multiclass SIMDCmpTwoVector<bit U, bits<5> opc, string asm,
4441 def v8i8rz : BaseSIMDCmpTwoVector<0, U, 0b00, opc, V64,
4443 v8i8, v8i8, OpNode>;
4444 def v16i8rz : BaseSIMDCmpTwoVector<1, U, 0b00, opc, V128,
4446 v16i8, v16i8, OpNode>;
4447 def v4i16rz : BaseSIMDCmpTwoVector<0, U, 0b01, opc, V64,
4449 v4i16, v4i16, OpNode>;
4450 def v8i16rz : BaseSIMDCmpTwoVector<1, U, 0b01, opc, V128,
4452 v8i16, v8i16, OpNode>;
4453 def v2i32rz : BaseSIMDCmpTwoVector<0, U, 0b10, opc, V64,
4455 v2i32, v2i32, OpNode>;
4456 def v4i32rz : BaseSIMDCmpTwoVector<1, U, 0b10, opc, V128,
4458 v4i32, v4i32, OpNode>;
4459 def v2i64rz : BaseSIMDCmpTwoVector<1, U, 0b11, opc, V128,
4461 v2i64, v2i64, OpNode>;
4464 // FP Comparisons support only S and D element sizes.
4465 multiclass SIMDFPCmpTwoVector<bit U, bit S, bits<5> opc,
4466 string asm, SDNode OpNode> {
4467 def v2i32rz : BaseSIMDCmpTwoVector<0, U, {S,0}, opc, V64,
4469 v2i32, v2f32, OpNode>;
4470 def v4i32rz : BaseSIMDCmpTwoVector<1, U, {S,0}, opc, V128,
4472 v4i32, v4f32, OpNode>;
4473 def v2i64rz : BaseSIMDCmpTwoVector<1, U, {S,1}, opc, V128,
4475 v2i64, v2f64, OpNode>;
4478 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4479 class BaseSIMDFPCvtTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4480 RegisterOperand outtype, RegisterOperand intype,
4481 string asm, string VdTy, string VnTy,
4483 : I<(outs outtype:$Rd), (ins intype:$Rn), asm,
4484 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "", pattern>,
4491 let Inst{28-24} = 0b01110;
4492 let Inst{23-22} = size;
4493 let Inst{21-17} = 0b10000;
4494 let Inst{16-12} = opcode;
4495 let Inst{11-10} = 0b10;
4500 class BaseSIMDFPCvtTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4501 RegisterOperand outtype, RegisterOperand intype,
4502 string asm, string VdTy, string VnTy,
4504 : I<(outs outtype:$dst), (ins outtype:$Rd, intype:$Rn), asm,
4505 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "$Rd = $dst", pattern>,
4512 let Inst{28-24} = 0b01110;
4513 let Inst{23-22} = size;
4514 let Inst{21-17} = 0b10000;
4515 let Inst{16-12} = opcode;
4516 let Inst{11-10} = 0b10;
4521 multiclass SIMDFPWidenTwoVector<bit U, bit S, bits<5> opc, string asm> {
4522 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V128, V64,
4523 asm, ".4s", ".4h", []>;
4524 def v8i16 : BaseSIMDFPCvtTwoVector<1, U, {S,0}, opc, V128, V128,
4525 asm#"2", ".4s", ".8h", []>;
4526 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V128, V64,
4527 asm, ".2d", ".2s", []>;
4528 def v4i32 : BaseSIMDFPCvtTwoVector<1, U, {S,1}, opc, V128, V128,
4529 asm#"2", ".2d", ".4s", []>;
4532 multiclass SIMDFPNarrowTwoVector<bit U, bit S, bits<5> opc, string asm> {
4533 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V64, V128,
4534 asm, ".4h", ".4s", []>;
4535 def v8i16 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,0}, opc, V128, V128,
4536 asm#"2", ".8h", ".4s", []>;
4537 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
4538 asm, ".2s", ".2d", []>;
4539 def v4i32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
4540 asm#"2", ".4s", ".2d", []>;
4543 multiclass SIMDFPInexactCvtTwoVector<bit U, bit S, bits<5> opc, string asm,
4545 def v2f32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
4547 [(set (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4548 def v4f32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
4549 asm#"2", ".4s", ".2d", []>;
4551 def : Pat<(concat_vectors (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn))),
4552 (!cast<Instruction>(NAME # "v4f32")
4553 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4556 //----------------------------------------------------------------------------
4557 // AdvSIMD three register different-size vector instructions.
4558 //----------------------------------------------------------------------------
4560 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4561 class BaseSIMDDifferentThreeVector<bit U, bits<3> size, bits<4> opcode,
4562 RegisterOperand outtype, RegisterOperand intype1,
4563 RegisterOperand intype2, string asm,
4564 string outkind, string inkind1, string inkind2,
4566 : I<(outs outtype:$Rd), (ins intype1:$Rn, intype2:$Rm), asm,
4567 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
4568 "|" # outkind # "\t$Rd, $Rn, $Rm}", "", pattern>,
4574 let Inst{30} = size{0};
4576 let Inst{28-24} = 0b01110;
4577 let Inst{23-22} = size{2-1};
4579 let Inst{20-16} = Rm;
4580 let Inst{15-12} = opcode;
4581 let Inst{11-10} = 0b00;
4586 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4587 class BaseSIMDDifferentThreeVectorTied<bit U, bits<3> size, bits<4> opcode,
4588 RegisterOperand outtype, RegisterOperand intype1,
4589 RegisterOperand intype2, string asm,
4590 string outkind, string inkind1, string inkind2,
4592 : I<(outs outtype:$dst), (ins outtype:$Rd, intype1:$Rn, intype2:$Rm), asm,
4593 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
4594 "|" # outkind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
4600 let Inst{30} = size{0};
4602 let Inst{28-24} = 0b01110;
4603 let Inst{23-22} = size{2-1};
4605 let Inst{20-16} = Rm;
4606 let Inst{15-12} = opcode;
4607 let Inst{11-10} = 0b00;
4612 // FIXME: TableGen doesn't know how to deal with expanded types that also
4613 // change the element count (in this case, placing the results in
4614 // the high elements of the result register rather than the low
4615 // elements). Until that's fixed, we can't code-gen those.
4616 multiclass SIMDNarrowThreeVectorBHS<bit U, bits<4> opc, string asm,
4618 def v8i16_v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4620 asm, ".8b", ".8h", ".8h",
4621 [(set (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
4622 def v8i16_v16i8 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4624 asm#"2", ".16b", ".8h", ".8h",
4626 def v4i32_v4i16 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4628 asm, ".4h", ".4s", ".4s",
4629 [(set (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
4630 def v4i32_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4632 asm#"2", ".8h", ".4s", ".4s",
4634 def v2i64_v2i32 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4636 asm, ".2s", ".2d", ".2d",
4637 [(set (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
4638 def v2i64_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4640 asm#"2", ".4s", ".2d", ".2d",
4644 // Patterns for the '2' variants involve INSERT_SUBREG, which you can't put in
4645 // a version attached to an instruction.
4646 def : Pat<(concat_vectors (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn),
4648 (!cast<Instruction>(NAME # "v8i16_v16i8")
4649 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4650 V128:$Rn, V128:$Rm)>;
4651 def : Pat<(concat_vectors (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn),
4653 (!cast<Instruction>(NAME # "v4i32_v8i16")
4654 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4655 V128:$Rn, V128:$Rm)>;
4656 def : Pat<(concat_vectors (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn),
4658 (!cast<Instruction>(NAME # "v2i64_v4i32")
4659 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4660 V128:$Rn, V128:$Rm)>;
4663 multiclass SIMDDifferentThreeVectorBD<bit U, bits<4> opc, string asm,
4665 def v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4667 asm, ".8h", ".8b", ".8b",
4668 [(set (v8i16 V128:$Rd), (IntOp (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4669 def v16i8 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4671 asm#"2", ".8h", ".16b", ".16b", []>;
4672 let Predicates = [HasCrypto] in {
4673 def v1i64 : BaseSIMDDifferentThreeVector<U, 0b110, opc,
4675 asm, ".1q", ".1d", ".1d", []>;
4676 def v2i64 : BaseSIMDDifferentThreeVector<U, 0b111, opc,
4678 asm#"2", ".1q", ".2d", ".2d", []>;
4681 def : Pat<(v8i16 (IntOp (v8i8 (extract_high_v16i8 V128:$Rn)),
4682 (v8i8 (extract_high_v16i8 V128:$Rm)))),
4683 (!cast<Instruction>(NAME#"v16i8") V128:$Rn, V128:$Rm)>;
4686 multiclass SIMDLongThreeVectorHS<bit U, bits<4> opc, string asm,
4687 SDPatternOperator OpNode> {
4688 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4690 asm, ".4s", ".4h", ".4h",
4691 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4692 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4694 asm#"2", ".4s", ".8h", ".8h",
4695 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
4696 (extract_high_v8i16 V128:$Rm)))]>;
4697 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4699 asm, ".2d", ".2s", ".2s",
4700 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4701 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4703 asm#"2", ".2d", ".4s", ".4s",
4704 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
4705 (extract_high_v4i32 V128:$Rm)))]>;
4708 multiclass SIMDLongThreeVectorBHSabdl<bit U, bits<4> opc, string asm,
4709 SDPatternOperator OpNode = null_frag> {
4710 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4712 asm, ".8h", ".8b", ".8b",
4713 [(set (v8i16 V128:$Rd),
4714 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))))]>;
4715 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4717 asm#"2", ".8h", ".16b", ".16b",
4718 [(set (v8i16 V128:$Rd),
4719 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
4720 (extract_high_v16i8 V128:$Rm)))))]>;
4721 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4723 asm, ".4s", ".4h", ".4h",
4724 [(set (v4i32 V128:$Rd),
4725 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))))]>;
4726 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4728 asm#"2", ".4s", ".8h", ".8h",
4729 [(set (v4i32 V128:$Rd),
4730 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
4731 (extract_high_v8i16 V128:$Rm)))))]>;
4732 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4734 asm, ".2d", ".2s", ".2s",
4735 [(set (v2i64 V128:$Rd),
4736 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))))]>;
4737 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4739 asm#"2", ".2d", ".4s", ".4s",
4740 [(set (v2i64 V128:$Rd),
4741 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
4742 (extract_high_v4i32 V128:$Rm)))))]>;
4745 multiclass SIMDLongThreeVectorTiedBHSabal<bit U, bits<4> opc,
4747 SDPatternOperator OpNode> {
4748 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
4750 asm, ".8h", ".8b", ".8b",
4751 [(set (v8i16 V128:$dst),
4752 (add (v8i16 V128:$Rd),
4753 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))))]>;
4754 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4756 asm#"2", ".8h", ".16b", ".16b",
4757 [(set (v8i16 V128:$dst),
4758 (add (v8i16 V128:$Rd),
4759 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
4760 (extract_high_v16i8 V128:$Rm))))))]>;
4761 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4763 asm, ".4s", ".4h", ".4h",
4764 [(set (v4i32 V128:$dst),
4765 (add (v4i32 V128:$Rd),
4766 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))))]>;
4767 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4769 asm#"2", ".4s", ".8h", ".8h",
4770 [(set (v4i32 V128:$dst),
4771 (add (v4i32 V128:$Rd),
4772 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
4773 (extract_high_v8i16 V128:$Rm))))))]>;
4774 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4776 asm, ".2d", ".2s", ".2s",
4777 [(set (v2i64 V128:$dst),
4778 (add (v2i64 V128:$Rd),
4779 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))))]>;
4780 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4782 asm#"2", ".2d", ".4s", ".4s",
4783 [(set (v2i64 V128:$dst),
4784 (add (v2i64 V128:$Rd),
4785 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
4786 (extract_high_v4i32 V128:$Rm))))))]>;
4789 multiclass SIMDLongThreeVectorBHS<bit U, bits<4> opc, string asm,
4790 SDPatternOperator OpNode = null_frag> {
4791 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4793 asm, ".8h", ".8b", ".8b",
4794 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4795 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4797 asm#"2", ".8h", ".16b", ".16b",
4798 [(set (v8i16 V128:$Rd), (OpNode (extract_high_v16i8 V128:$Rn),
4799 (extract_high_v16i8 V128:$Rm)))]>;
4800 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4802 asm, ".4s", ".4h", ".4h",
4803 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4804 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4806 asm#"2", ".4s", ".8h", ".8h",
4807 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
4808 (extract_high_v8i16 V128:$Rm)))]>;
4809 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4811 asm, ".2d", ".2s", ".2s",
4812 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4813 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4815 asm#"2", ".2d", ".4s", ".4s",
4816 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
4817 (extract_high_v4i32 V128:$Rm)))]>;
4820 multiclass SIMDLongThreeVectorTiedBHS<bit U, bits<4> opc,
4822 SDPatternOperator OpNode> {
4823 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
4825 asm, ".8h", ".8b", ".8b",
4826 [(set (v8i16 V128:$dst),
4827 (OpNode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4828 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4830 asm#"2", ".8h", ".16b", ".16b",
4831 [(set (v8i16 V128:$dst),
4832 (OpNode (v8i16 V128:$Rd),
4833 (extract_high_v16i8 V128:$Rn),
4834 (extract_high_v16i8 V128:$Rm)))]>;
4835 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4837 asm, ".4s", ".4h", ".4h",
4838 [(set (v4i32 V128:$dst),
4839 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4840 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4842 asm#"2", ".4s", ".8h", ".8h",
4843 [(set (v4i32 V128:$dst),
4844 (OpNode (v4i32 V128:$Rd),
4845 (extract_high_v8i16 V128:$Rn),
4846 (extract_high_v8i16 V128:$Rm)))]>;
4847 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4849 asm, ".2d", ".2s", ".2s",
4850 [(set (v2i64 V128:$dst),
4851 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4852 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4854 asm#"2", ".2d", ".4s", ".4s",
4855 [(set (v2i64 V128:$dst),
4856 (OpNode (v2i64 V128:$Rd),
4857 (extract_high_v4i32 V128:$Rn),
4858 (extract_high_v4i32 V128:$Rm)))]>;
4861 multiclass SIMDLongThreeVectorSQDMLXTiedHS<bit U, bits<4> opc, string asm,
4862 SDPatternOperator Accum> {
4863 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4865 asm, ".4s", ".4h", ".4h",
4866 [(set (v4i32 V128:$dst),
4867 (Accum (v4i32 V128:$Rd),
4868 (v4i32 (int_arm64_neon_sqdmull (v4i16 V64:$Rn),
4869 (v4i16 V64:$Rm)))))]>;
4870 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4872 asm#"2", ".4s", ".8h", ".8h",
4873 [(set (v4i32 V128:$dst),
4874 (Accum (v4i32 V128:$Rd),
4875 (v4i32 (int_arm64_neon_sqdmull (extract_high_v8i16 V128:$Rn),
4876 (extract_high_v8i16 V128:$Rm)))))]>;
4877 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4879 asm, ".2d", ".2s", ".2s",
4880 [(set (v2i64 V128:$dst),
4881 (Accum (v2i64 V128:$Rd),
4882 (v2i64 (int_arm64_neon_sqdmull (v2i32 V64:$Rn),
4883 (v2i32 V64:$Rm)))))]>;
4884 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4886 asm#"2", ".2d", ".4s", ".4s",
4887 [(set (v2i64 V128:$dst),
4888 (Accum (v2i64 V128:$Rd),
4889 (v2i64 (int_arm64_neon_sqdmull (extract_high_v4i32 V128:$Rn),
4890 (extract_high_v4i32 V128:$Rm)))))]>;
4893 multiclass SIMDWideThreeVectorBHS<bit U, bits<4> opc, string asm,
4894 SDPatternOperator OpNode> {
4895 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4897 asm, ".8h", ".8h", ".8b",
4898 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i8 V64:$Rm)))]>;
4899 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4901 asm#"2", ".8h", ".8h", ".16b",
4902 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
4903 (extract_high_v16i8 V128:$Rm)))]>;
4904 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4906 asm, ".4s", ".4s", ".4h",
4907 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i16 V64:$Rm)))]>;
4908 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4910 asm#"2", ".4s", ".4s", ".8h",
4911 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
4912 (extract_high_v8i16 V128:$Rm)))]>;
4913 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4915 asm, ".2d", ".2d", ".2s",
4916 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i32 V64:$Rm)))]>;
4917 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4919 asm#"2", ".2d", ".2d", ".4s",
4920 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
4921 (extract_high_v4i32 V128:$Rm)))]>;
4924 //----------------------------------------------------------------------------
4925 // AdvSIMD bitwise extract from vector
4926 //----------------------------------------------------------------------------
4928 class BaseSIMDBitwiseExtract<bit size, RegisterOperand regtype, ValueType vty,
4929 string asm, string kind>
4930 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, i32imm:$imm), asm,
4931 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $imm" #
4932 "|" # kind # "\t$Rd, $Rn, $Rm, $imm}", "",
4933 [(set (vty regtype:$Rd),
4934 (ARM64ext regtype:$Rn, regtype:$Rm, (i32 imm:$imm)))]>,
4941 let Inst{30} = size;
4942 let Inst{29-21} = 0b101110000;
4943 let Inst{20-16} = Rm;
4945 let Inst{14-11} = imm;
4952 multiclass SIMDBitwiseExtract<string asm> {
4953 def v8i8 : BaseSIMDBitwiseExtract<0, V64, v8i8, asm, ".8b"> {
4956 def v16i8 : BaseSIMDBitwiseExtract<1, V128, v16i8, asm, ".16b">;
4959 //----------------------------------------------------------------------------
4960 // AdvSIMD zip vector
4961 //----------------------------------------------------------------------------
4963 class BaseSIMDZipVector<bits<3> size, bits<3> opc, RegisterOperand regtype,
4964 string asm, string kind, SDNode OpNode, ValueType valty>
4965 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
4966 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
4967 "|" # kind # "\t$Rd, $Rn, $Rm}", "",
4968 [(set (valty regtype:$Rd), (OpNode regtype:$Rn, regtype:$Rm))]>,
4974 let Inst{30} = size{0};
4975 let Inst{29-24} = 0b001110;
4976 let Inst{23-22} = size{2-1};
4978 let Inst{20-16} = Rm;
4980 let Inst{14-12} = opc;
4981 let Inst{11-10} = 0b10;
4986 multiclass SIMDZipVector<bits<3>opc, string asm,
4988 def v8i8 : BaseSIMDZipVector<0b000, opc, V64,
4989 asm, ".8b", OpNode, v8i8>;
4990 def v16i8 : BaseSIMDZipVector<0b001, opc, V128,
4991 asm, ".16b", OpNode, v16i8>;
4992 def v4i16 : BaseSIMDZipVector<0b010, opc, V64,
4993 asm, ".4h", OpNode, v4i16>;
4994 def v8i16 : BaseSIMDZipVector<0b011, opc, V128,
4995 asm, ".8h", OpNode, v8i16>;
4996 def v2i32 : BaseSIMDZipVector<0b100, opc, V64,
4997 asm, ".2s", OpNode, v2i32>;
4998 def v4i32 : BaseSIMDZipVector<0b101, opc, V128,
4999 asm, ".4s", OpNode, v4i32>;
5000 def v2i64 : BaseSIMDZipVector<0b111, opc, V128,
5001 asm, ".2d", OpNode, v2i64>;
5003 def : Pat<(v2f32 (OpNode V64:$Rn, V64:$Rm)),
5004 (!cast<Instruction>(NAME#"v2i32") V64:$Rn, V64:$Rm)>;
5005 def : Pat<(v4f32 (OpNode V128:$Rn, V128:$Rm)),
5006 (!cast<Instruction>(NAME#"v4i32") V128:$Rn, V128:$Rm)>;
5007 def : Pat<(v2f64 (OpNode V128:$Rn, V128:$Rm)),
5008 (!cast<Instruction>(NAME#"v2i64") V128:$Rn, V128:$Rm)>;
5011 //----------------------------------------------------------------------------
5012 // AdvSIMD three register scalar instructions
5013 //----------------------------------------------------------------------------
5015 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
5016 class BaseSIMDThreeScalar<bit U, bits<2> size, bits<5> opcode,
5017 RegisterClass regtype, string asm,
5019 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
5020 "\t$Rd, $Rn, $Rm", "", pattern>,
5025 let Inst{31-30} = 0b01;
5027 let Inst{28-24} = 0b11110;
5028 let Inst{23-22} = size;
5030 let Inst{20-16} = Rm;
5031 let Inst{15-11} = opcode;
5037 multiclass SIMDThreeScalarD<bit U, bits<5> opc, string asm,
5038 SDPatternOperator OpNode> {
5039 def v1i64 : BaseSIMDThreeScalar<U, 0b11, opc, FPR64, asm,
5040 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
5043 multiclass SIMDThreeScalarBHSD<bit U, bits<5> opc, string asm,
5044 SDPatternOperator OpNode> {
5045 def v1i64 : BaseSIMDThreeScalar<U, 0b11, opc, FPR64, asm,
5046 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
5047 def v1i32 : BaseSIMDThreeScalar<U, 0b10, opc, FPR32, asm, []>;
5048 def v1i16 : BaseSIMDThreeScalar<U, 0b01, opc, FPR16, asm, []>;
5049 def v1i8 : BaseSIMDThreeScalar<U, 0b00, opc, FPR8 , asm, []>;
5051 def : Pat<(i64 (OpNode (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
5052 (!cast<Instruction>(NAME#"v1i64") FPR64:$Rn, FPR64:$Rm)>;
5053 def : Pat<(i32 (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm))),
5054 (!cast<Instruction>(NAME#"v1i32") FPR32:$Rn, FPR32:$Rm)>;
5057 multiclass SIMDThreeScalarHS<bit U, bits<5> opc, string asm,
5058 SDPatternOperator OpNode> {
5059 def v1i32 : BaseSIMDThreeScalar<U, 0b10, opc, FPR32, asm,
5060 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
5061 def v1i16 : BaseSIMDThreeScalar<U, 0b01, opc, FPR16, asm, []>;
5064 multiclass SIMDThreeScalarSD<bit U, bit S, bits<5> opc, string asm,
5065 SDPatternOperator OpNode = null_frag> {
5066 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5067 def #NAME#64 : BaseSIMDThreeScalar<U, {S,1}, opc, FPR64, asm,
5068 [(set (f64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
5069 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0}, opc, FPR32, asm,
5070 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
5073 def : Pat<(v1f64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
5074 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
5077 multiclass SIMDThreeScalarFPCmp<bit U, bit S, bits<5> opc, string asm,
5078 SDPatternOperator OpNode = null_frag> {
5079 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5080 def #NAME#64 : BaseSIMDThreeScalar<U, {S,1}, opc, FPR64, asm,
5081 [(set (i64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
5082 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0}, opc, FPR32, asm,
5083 [(set (i32 FPR32:$Rd), (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]>;
5086 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
5087 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
5090 class BaseSIMDThreeScalarMixed<bit U, bits<2> size, bits<5> opcode,
5091 dag oops, dag iops, string asm, string cstr, list<dag> pat>
5092 : I<oops, iops, asm,
5093 "\t$Rd, $Rn, $Rm", cstr, pat>,
5098 let Inst{31-30} = 0b01;
5100 let Inst{28-24} = 0b11110;
5101 let Inst{23-22} = size;
5103 let Inst{20-16} = Rm;
5104 let Inst{15-11} = opcode;
5110 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5111 multiclass SIMDThreeScalarMixedHS<bit U, bits<5> opc, string asm,
5112 SDPatternOperator OpNode = null_frag> {
5113 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
5115 (ins FPR16:$Rn, FPR16:$Rm), asm, "", []>;
5116 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
5118 (ins FPR32:$Rn, FPR32:$Rm), asm, "",
5119 [(set (i64 FPR64:$Rd), (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
5122 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5123 multiclass SIMDThreeScalarMixedTiedHS<bit U, bits<5> opc, string asm,
5124 SDPatternOperator OpNode = null_frag> {
5125 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
5127 (ins FPR32:$Rd, FPR16:$Rn, FPR16:$Rm),
5128 asm, "$Rd = $dst", []>;
5129 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
5131 (ins FPR64:$Rd, FPR32:$Rn, FPR32:$Rm),
5133 [(set (i64 FPR64:$dst),
5134 (OpNode (i64 FPR64:$Rd), (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
5137 //----------------------------------------------------------------------------
5138 // AdvSIMD two register scalar instructions
5139 //----------------------------------------------------------------------------
5141 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5142 class BaseSIMDTwoScalar<bit U, bits<2> size, bits<5> opcode,
5143 RegisterClass regtype, RegisterClass regtype2,
5144 string asm, list<dag> pat>
5145 : I<(outs regtype:$Rd), (ins regtype2:$Rn), asm,
5146 "\t$Rd, $Rn", "", pat>,
5150 let Inst{31-30} = 0b01;
5152 let Inst{28-24} = 0b11110;
5153 let Inst{23-22} = size;
5154 let Inst{21-17} = 0b10000;
5155 let Inst{16-12} = opcode;
5156 let Inst{11-10} = 0b10;
5161 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5162 class BaseSIMDTwoScalarTied<bit U, bits<2> size, bits<5> opcode,
5163 RegisterClass regtype, RegisterClass regtype2,
5164 string asm, list<dag> pat>
5165 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype2:$Rn), asm,
5166 "\t$Rd, $Rn", "$Rd = $dst", pat>,
5170 let Inst{31-30} = 0b01;
5172 let Inst{28-24} = 0b11110;
5173 let Inst{23-22} = size;
5174 let Inst{21-17} = 0b10000;
5175 let Inst{16-12} = opcode;
5176 let Inst{11-10} = 0b10;
5182 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5183 class BaseSIMDCmpTwoScalar<bit U, bits<2> size, bits<5> opcode,
5184 RegisterClass regtype, string asm>
5185 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5186 "\t$Rd, $Rn, #0", "", []>,
5190 let Inst{31-30} = 0b01;
5192 let Inst{28-24} = 0b11110;
5193 let Inst{23-22} = size;
5194 let Inst{21-17} = 0b10000;
5195 let Inst{16-12} = opcode;
5196 let Inst{11-10} = 0b10;
5201 class SIMDInexactCvtTwoScalar<bits<5> opcode, string asm>
5202 : I<(outs FPR32:$Rd), (ins FPR64:$Rn), asm, "\t$Rd, $Rn", "",
5203 [(set (f32 FPR32:$Rd), (int_arm64_sisd_fcvtxn (f64 FPR64:$Rn)))]>,
5207 let Inst{31-17} = 0b011111100110000;
5208 let Inst{16-12} = opcode;
5209 let Inst{11-10} = 0b10;
5214 multiclass SIMDCmpTwoScalarD<bit U, bits<5> opc, string asm,
5215 SDPatternOperator OpNode> {
5216 def v1i64rz : BaseSIMDCmpTwoScalar<U, 0b11, opc, FPR64, asm>;
5218 def : Pat<(v1i64 (OpNode FPR64:$Rn)),
5219 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
5222 multiclass SIMDCmpTwoScalarSD<bit U, bit S, bits<5> opc, string asm,
5223 SDPatternOperator OpNode> {
5224 def v1i64rz : BaseSIMDCmpTwoScalar<U, {S,1}, opc, FPR64, asm>;
5225 def v1i32rz : BaseSIMDCmpTwoScalar<U, {S,0}, opc, FPR32, asm>;
5227 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn))),
5228 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
5231 multiclass SIMDTwoScalarD<bit U, bits<5> opc, string asm,
5232 SDPatternOperator OpNode = null_frag> {
5233 def v1i64 : BaseSIMDTwoScalar<U, 0b11, opc, FPR64, FPR64, asm,
5234 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn)))]>;
5236 def : Pat<(i64 (OpNode (i64 FPR64:$Rn))),
5237 (!cast<Instruction>(NAME # "v1i64") FPR64:$Rn)>;
5240 multiclass SIMDTwoScalarSD<bit U, bit S, bits<5> opc, string asm> {
5241 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, opc, FPR64, FPR64, asm,[]>;
5242 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, opc, FPR32, FPR32, asm,[]>;
5245 multiclass SIMDTwoScalarCVTSD<bit U, bit S, bits<5> opc, string asm,
5246 SDPatternOperator OpNode> {
5247 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, opc, FPR64, FPR64, asm,
5248 [(set FPR64:$Rd, (OpNode (f64 FPR64:$Rn)))]>;
5249 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, opc, FPR32, FPR32, asm,
5250 [(set FPR32:$Rd, (OpNode (f32 FPR32:$Rn)))]>;
5253 multiclass SIMDTwoScalarBHSD<bit U, bits<5> opc, string asm,
5254 SDPatternOperator OpNode = null_frag> {
5255 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5256 def v1i64 : BaseSIMDTwoScalar<U, 0b11, opc, FPR64, FPR64, asm,
5257 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
5258 def v1i32 : BaseSIMDTwoScalar<U, 0b10, opc, FPR32, FPR32, asm,
5259 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
5260 def v1i16 : BaseSIMDTwoScalar<U, 0b01, opc, FPR16, FPR16, asm, []>;
5261 def v1i8 : BaseSIMDTwoScalar<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
5264 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn))),
5265 (!cast<Instruction>(NAME # v1i64) FPR64:$Rn)>;
5268 multiclass SIMDTwoScalarBHSDTied<bit U, bits<5> opc, string asm,
5270 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5271 def v1i64 : BaseSIMDTwoScalarTied<U, 0b11, opc, FPR64, FPR64, asm,
5272 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn)))]>;
5273 def v1i32 : BaseSIMDTwoScalarTied<U, 0b10, opc, FPR32, FPR32, asm,
5274 [(set (i32 FPR32:$dst), (OpNode (i32 FPR32:$Rd), (i32 FPR32:$Rn)))]>;
5275 def v1i16 : BaseSIMDTwoScalarTied<U, 0b01, opc, FPR16, FPR16, asm, []>;
5276 def v1i8 : BaseSIMDTwoScalarTied<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
5279 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn))),
5280 (!cast<Instruction>(NAME # v1i64) FPR64:$Rd, FPR64:$Rn)>;
5285 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5286 multiclass SIMDTwoScalarMixedBHS<bit U, bits<5> opc, string asm,
5287 SDPatternOperator OpNode = null_frag> {
5288 def v1i32 : BaseSIMDTwoScalar<U, 0b10, opc, FPR32, FPR64, asm,
5289 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
5290 def v1i16 : BaseSIMDTwoScalar<U, 0b01, opc, FPR16, FPR32, asm, []>;
5291 def v1i8 : BaseSIMDTwoScalar<U, 0b00, opc, FPR8 , FPR16, asm, []>;
5294 //----------------------------------------------------------------------------
5295 // AdvSIMD scalar pairwise instructions
5296 //----------------------------------------------------------------------------
5298 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5299 class BaseSIMDPairwiseScalar<bit U, bits<2> size, bits<5> opcode,
5300 RegisterOperand regtype, RegisterOperand vectype,
5301 string asm, string kind>
5302 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
5303 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", []>,
5307 let Inst{31-30} = 0b01;
5309 let Inst{28-24} = 0b11110;
5310 let Inst{23-22} = size;
5311 let Inst{21-17} = 0b11000;
5312 let Inst{16-12} = opcode;
5313 let Inst{11-10} = 0b10;
5318 multiclass SIMDPairwiseScalarD<bit U, bits<5> opc, string asm> {
5319 def v2i64p : BaseSIMDPairwiseScalar<U, 0b11, opc, FPR64Op, V128,
5323 multiclass SIMDPairwiseScalarSD<bit U, bit S, bits<5> opc, string asm> {
5324 def v2i32p : BaseSIMDPairwiseScalar<U, {S,0}, opc, FPR32Op, V64,
5326 def v2i64p : BaseSIMDPairwiseScalar<U, {S,1}, opc, FPR64Op, V128,
5330 //----------------------------------------------------------------------------
5331 // AdvSIMD across lanes instructions
5332 //----------------------------------------------------------------------------
5334 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5335 class BaseSIMDAcrossLanes<bit Q, bit U, bits<2> size, bits<5> opcode,
5336 RegisterClass regtype, RegisterOperand vectype,
5337 string asm, string kind, list<dag> pattern>
5338 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
5339 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", pattern>,
5346 let Inst{28-24} = 0b01110;
5347 let Inst{23-22} = size;
5348 let Inst{21-17} = 0b11000;
5349 let Inst{16-12} = opcode;
5350 let Inst{11-10} = 0b10;
5355 multiclass SIMDAcrossLanesBHS<bit U, bits<5> opcode,
5357 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR8, V64,
5359 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR8, V128,
5361 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR16, V64,
5363 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR16, V128,
5365 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR32, V128,
5369 multiclass SIMDAcrossLanesHSD<bit U, bits<5> opcode, string asm> {
5370 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR16, V64,
5372 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR16, V128,
5374 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR32, V64,
5376 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR32, V128,
5378 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR64, V128,
5382 multiclass SIMDAcrossLanesS<bits<5> opcode, bit sz1, string asm,
5384 def v4i32v : BaseSIMDAcrossLanes<1, 1, {sz1, 0}, opcode, FPR32, V128,
5386 [(set FPR32:$Rd, (intOp (v4f32 V128:$Rn)))]>;
5389 //----------------------------------------------------------------------------
5390 // AdvSIMD INS/DUP instructions
5391 //----------------------------------------------------------------------------
5393 // FIXME: There has got to be a better way to factor these. ugh.
5395 class BaseSIMDInsDup<bit Q, bit op, dag outs, dag ins, string asm,
5396 string operands, string constraints, list<dag> pattern>
5397 : I<outs, ins, asm, operands, constraints, pattern>,
5404 let Inst{28-21} = 0b01110000;
5411 class SIMDDupFromMain<bit Q, bits<5> imm5, string size, ValueType vectype,
5412 RegisterOperand vecreg, RegisterClass regtype>
5413 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins regtype:$Rn), "dup",
5414 "{\t$Rd" # size # ", $Rn" #
5415 "|" # size # "\t$Rd, $Rn}", "",
5416 [(set (vectype vecreg:$Rd), (ARM64dup regtype:$Rn))]> {
5417 let Inst{20-16} = imm5;
5418 let Inst{14-11} = 0b0001;
5421 class SIMDDupFromElement<bit Q, string dstkind, string srckind,
5422 ValueType vectype, ValueType insreg,
5423 RegisterOperand vecreg, Operand idxtype,
5424 ValueType elttype, SDNode OpNode>
5425 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins V128:$Rn, idxtype:$idx), "dup",
5426 "{\t$Rd" # dstkind # ", $Rn" # srckind # "$idx" #
5427 "|" # dstkind # "\t$Rd, $Rn$idx}", "",
5428 [(set (vectype vecreg:$Rd),
5429 (OpNode (insreg V128:$Rn), idxtype:$idx))]> {
5430 let Inst{14-11} = 0b0000;
5433 class SIMDDup64FromElement
5434 : SIMDDupFromElement<1, ".2d", ".d", v2i64, v2i64, V128,
5435 VectorIndexD, i64, ARM64duplane64> {
5438 let Inst{19-16} = 0b1000;
5441 class SIMDDup32FromElement<bit Q, string size, ValueType vectype,
5442 RegisterOperand vecreg>
5443 : SIMDDupFromElement<Q, size, ".s", vectype, v4i32, vecreg,
5444 VectorIndexS, i64, ARM64duplane32> {
5446 let Inst{20-19} = idx;
5447 let Inst{18-16} = 0b100;
5450 class SIMDDup16FromElement<bit Q, string size, ValueType vectype,
5451 RegisterOperand vecreg>
5452 : SIMDDupFromElement<Q, size, ".h", vectype, v8i16, vecreg,
5453 VectorIndexH, i64, ARM64duplane16> {
5455 let Inst{20-18} = idx;
5456 let Inst{17-16} = 0b10;
5459 class SIMDDup8FromElement<bit Q, string size, ValueType vectype,
5460 RegisterOperand vecreg>
5461 : SIMDDupFromElement<Q, size, ".b", vectype, v16i8, vecreg,
5462 VectorIndexB, i64, ARM64duplane8> {
5464 let Inst{20-17} = idx;
5468 class BaseSIMDMov<bit Q, string size, bits<4> imm4, RegisterClass regtype,
5469 Operand idxtype, string asm, list<dag> pattern>
5470 : BaseSIMDInsDup<Q, 0, (outs regtype:$Rd), (ins V128:$Rn, idxtype:$idx), asm,
5471 "{\t$Rd, $Rn" # size # "$idx" #
5472 "|" # size # "\t$Rd, $Rn$idx}", "", pattern> {
5473 let Inst{14-11} = imm4;
5476 class SIMDSMov<bit Q, string size, RegisterClass regtype,
5478 : BaseSIMDMov<Q, size, 0b0101, regtype, idxtype, "smov", []>;
5479 class SIMDUMov<bit Q, string size, ValueType vectype, RegisterClass regtype,
5481 : BaseSIMDMov<Q, size, 0b0111, regtype, idxtype, "umov",
5482 [(set regtype:$Rd, (vector_extract (vectype V128:$Rn), idxtype:$idx))]>;
5484 class SIMDMovAlias<string asm, string size, Instruction inst,
5485 RegisterClass regtype, Operand idxtype>
5486 : InstAlias<asm#"{\t$dst, $src"#size#"$idx" #
5487 "|" # size # "\t$dst, $src$idx}",
5488 (inst regtype:$dst, V128:$src, idxtype:$idx)>;
5491 def vi8to32 : SIMDSMov<0, ".b", GPR32, VectorIndexB> {
5493 let Inst{20-17} = idx;
5496 def vi8to64 : SIMDSMov<1, ".b", GPR64, VectorIndexB> {
5498 let Inst{20-17} = idx;
5501 def vi16to32 : SIMDSMov<0, ".h", GPR32, VectorIndexH> {
5503 let Inst{20-18} = idx;
5504 let Inst{17-16} = 0b10;
5506 def vi16to64 : SIMDSMov<1, ".h", GPR64, VectorIndexH> {
5508 let Inst{20-18} = idx;
5509 let Inst{17-16} = 0b10;
5511 def vi32to64 : SIMDSMov<1, ".s", GPR64, VectorIndexS> {
5513 let Inst{20-19} = idx;
5514 let Inst{18-16} = 0b100;
5519 def vi8 : SIMDUMov<0, ".b", v16i8, GPR32, VectorIndexB> {
5521 let Inst{20-17} = idx;
5524 def vi16 : SIMDUMov<0, ".h", v8i16, GPR32, VectorIndexH> {
5526 let Inst{20-18} = idx;
5527 let Inst{17-16} = 0b10;
5529 def vi32 : SIMDUMov<0, ".s", v4i32, GPR32, VectorIndexS> {
5531 let Inst{20-19} = idx;
5532 let Inst{18-16} = 0b100;
5534 def vi64 : SIMDUMov<1, ".d", v2i64, GPR64, VectorIndexD> {
5537 let Inst{19-16} = 0b1000;
5539 def : SIMDMovAlias<"mov", ".s",
5540 !cast<Instruction>(NAME#"vi32"),
5541 GPR32, VectorIndexS>;
5542 def : SIMDMovAlias<"mov", ".d",
5543 !cast<Instruction>(NAME#"vi64"),
5544 GPR64, VectorIndexD>;
5547 class SIMDInsFromMain<string size, ValueType vectype,
5548 RegisterClass regtype, Operand idxtype>
5549 : BaseSIMDInsDup<1, 0, (outs V128:$dst),
5550 (ins V128:$Rd, idxtype:$idx, regtype:$Rn), "ins",
5551 "{\t$Rd" # size # "$idx, $Rn" #
5552 "|" # size # "\t$Rd$idx, $Rn}",
5555 (vector_insert (vectype V128:$Rd), regtype:$Rn, idxtype:$idx))]> {
5556 let Inst{14-11} = 0b0011;
5559 class SIMDInsFromElement<string size, ValueType vectype,
5560 ValueType elttype, Operand idxtype>
5561 : BaseSIMDInsDup<1, 1, (outs V128:$dst),
5562 (ins V128:$Rd, idxtype:$idx, V128:$Rn, idxtype:$idx2), "ins",
5563 "{\t$Rd" # size # "$idx, $Rn" # size # "$idx2" #
5564 "|" # size # "\t$Rd$idx, $Rn$idx2}",
5569 (elttype (vector_extract (vectype V128:$Rn), idxtype:$idx2)),
5572 class SIMDInsMainMovAlias<string size, Instruction inst,
5573 RegisterClass regtype, Operand idxtype>
5574 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" #
5575 "|" # size #"\t$dst$idx, $src}",
5576 (inst V128:$dst, idxtype:$idx, regtype:$src)>;
5577 class SIMDInsElementMovAlias<string size, Instruction inst,
5579 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" # size # "$idx2" #
5580 # "|" # size #" $dst$idx, $src$idx2}",
5581 (inst V128:$dst, idxtype:$idx, V128:$src, idxtype:$idx2)>;
5584 multiclass SIMDIns {
5585 def vi8gpr : SIMDInsFromMain<".b", v16i8, GPR32, VectorIndexB> {
5587 let Inst{20-17} = idx;
5590 def vi16gpr : SIMDInsFromMain<".h", v8i16, GPR32, VectorIndexH> {
5592 let Inst{20-18} = idx;
5593 let Inst{17-16} = 0b10;
5595 def vi32gpr : SIMDInsFromMain<".s", v4i32, GPR32, VectorIndexS> {
5597 let Inst{20-19} = idx;
5598 let Inst{18-16} = 0b100;
5600 def vi64gpr : SIMDInsFromMain<".d", v2i64, GPR64, VectorIndexD> {
5603 let Inst{19-16} = 0b1000;
5606 def vi8lane : SIMDInsFromElement<".b", v16i8, i32, VectorIndexB> {
5609 let Inst{20-17} = idx;
5611 let Inst{14-11} = idx2;
5613 def vi16lane : SIMDInsFromElement<".h", v8i16, i32, VectorIndexH> {
5616 let Inst{20-18} = idx;
5617 let Inst{17-16} = 0b10;
5618 let Inst{14-12} = idx2;
5621 def vi32lane : SIMDInsFromElement<".s", v4i32, i32, VectorIndexS> {
5624 let Inst{20-19} = idx;
5625 let Inst{18-16} = 0b100;
5626 let Inst{14-13} = idx2;
5627 let Inst{12-11} = 0;
5629 def vi64lane : SIMDInsFromElement<".d", v2i64, i64, VectorIndexD> {
5633 let Inst{19-16} = 0b1000;
5634 let Inst{14} = idx2;
5635 let Inst{13-11} = 0;
5638 // For all forms of the INS instruction, the "mov" mnemonic is the
5639 // preferred alias. Why they didn't just call the instruction "mov" in
5640 // the first place is a very good question indeed...
5641 def : SIMDInsMainMovAlias<".b", !cast<Instruction>(NAME#"vi8gpr"),
5642 GPR32, VectorIndexB>;
5643 def : SIMDInsMainMovAlias<".h", !cast<Instruction>(NAME#"vi16gpr"),
5644 GPR32, VectorIndexH>;
5645 def : SIMDInsMainMovAlias<".s", !cast<Instruction>(NAME#"vi32gpr"),
5646 GPR32, VectorIndexS>;
5647 def : SIMDInsMainMovAlias<".d", !cast<Instruction>(NAME#"vi64gpr"),
5648 GPR64, VectorIndexD>;
5650 def : SIMDInsElementMovAlias<".b", !cast<Instruction>(NAME#"vi8lane"),
5652 def : SIMDInsElementMovAlias<".h", !cast<Instruction>(NAME#"vi16lane"),
5654 def : SIMDInsElementMovAlias<".s", !cast<Instruction>(NAME#"vi32lane"),
5656 def : SIMDInsElementMovAlias<".d", !cast<Instruction>(NAME#"vi64lane"),
5660 //----------------------------------------------------------------------------
5662 //----------------------------------------------------------------------------
5664 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
5665 class BaseSIMDTableLookup<bit Q, bits<2> len, bit op, RegisterOperand vectype,
5666 RegisterOperand listtype, string asm, string kind>
5667 : I<(outs vectype:$Vd), (ins listtype:$Vn, vectype:$Vm), asm,
5668 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "", []>,
5675 let Inst{29-21} = 0b001110000;
5676 let Inst{20-16} = Vm;
5678 let Inst{14-13} = len;
5680 let Inst{11-10} = 0b00;
5685 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
5686 class BaseSIMDTableLookupTied<bit Q, bits<2> len, bit op, RegisterOperand vectype,
5687 RegisterOperand listtype, string asm, string kind>
5688 : I<(outs vectype:$dst), (ins vectype:$Vd, listtype:$Vn, vectype:$Vm), asm,
5689 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "$Vd = $dst", []>,
5696 let Inst{29-21} = 0b001110000;
5697 let Inst{20-16} = Vm;
5699 let Inst{14-13} = len;
5701 let Inst{11-10} = 0b00;
5706 class SIMDTableLookupAlias<string asm, Instruction inst,
5707 RegisterOperand vectype, RegisterOperand listtype>
5708 : InstAlias<!strconcat(asm, "\t$dst, $lst, $index"),
5709 (inst vectype:$dst, listtype:$lst, vectype:$index), 0>;
5711 multiclass SIMDTableLookup<bit op, string asm> {
5712 def v8i8One : BaseSIMDTableLookup<0, 0b00, op, V64, VecListOne16b,
5714 def v8i8Two : BaseSIMDTableLookup<0, 0b01, op, V64, VecListTwo16b,
5716 def v8i8Three : BaseSIMDTableLookup<0, 0b10, op, V64, VecListThree16b,
5718 def v8i8Four : BaseSIMDTableLookup<0, 0b11, op, V64, VecListFour16b,
5720 def v16i8One : BaseSIMDTableLookup<1, 0b00, op, V128, VecListOne16b,
5722 def v16i8Two : BaseSIMDTableLookup<1, 0b01, op, V128, VecListTwo16b,
5724 def v16i8Three: BaseSIMDTableLookup<1, 0b10, op, V128, VecListThree16b,
5726 def v16i8Four : BaseSIMDTableLookup<1, 0b11, op, V128, VecListFour16b,
5729 def : SIMDTableLookupAlias<asm # ".8b",
5730 !cast<Instruction>(NAME#"v8i8One"),
5731 V64, VecListOne128>;
5732 def : SIMDTableLookupAlias<asm # ".8b",
5733 !cast<Instruction>(NAME#"v8i8Two"),
5734 V64, VecListTwo128>;
5735 def : SIMDTableLookupAlias<asm # ".8b",
5736 !cast<Instruction>(NAME#"v8i8Three"),
5737 V64, VecListThree128>;
5738 def : SIMDTableLookupAlias<asm # ".8b",
5739 !cast<Instruction>(NAME#"v8i8Four"),
5740 V64, VecListFour128>;
5741 def : SIMDTableLookupAlias<asm # ".16b",
5742 !cast<Instruction>(NAME#"v16i8One"),
5743 V128, VecListOne128>;
5744 def : SIMDTableLookupAlias<asm # ".16b",
5745 !cast<Instruction>(NAME#"v16i8Two"),
5746 V128, VecListTwo128>;
5747 def : SIMDTableLookupAlias<asm # ".16b",
5748 !cast<Instruction>(NAME#"v16i8Three"),
5749 V128, VecListThree128>;
5750 def : SIMDTableLookupAlias<asm # ".16b",
5751 !cast<Instruction>(NAME#"v16i8Four"),
5752 V128, VecListFour128>;
5755 multiclass SIMDTableLookupTied<bit op, string asm> {
5756 def v8i8One : BaseSIMDTableLookupTied<0, 0b00, op, V64, VecListOne16b,
5758 def v8i8Two : BaseSIMDTableLookupTied<0, 0b01, op, V64, VecListTwo16b,
5760 def v8i8Three : BaseSIMDTableLookupTied<0, 0b10, op, V64, VecListThree16b,
5762 def v8i8Four : BaseSIMDTableLookupTied<0, 0b11, op, V64, VecListFour16b,
5764 def v16i8One : BaseSIMDTableLookupTied<1, 0b00, op, V128, VecListOne16b,
5766 def v16i8Two : BaseSIMDTableLookupTied<1, 0b01, op, V128, VecListTwo16b,
5768 def v16i8Three: BaseSIMDTableLookupTied<1, 0b10, op, V128, VecListThree16b,
5770 def v16i8Four : BaseSIMDTableLookupTied<1, 0b11, op, V128, VecListFour16b,
5773 def : SIMDTableLookupAlias<asm # ".8b",
5774 !cast<Instruction>(NAME#"v8i8One"),
5775 V64, VecListOne128>;
5776 def : SIMDTableLookupAlias<asm # ".8b",
5777 !cast<Instruction>(NAME#"v8i8Two"),
5778 V64, VecListTwo128>;
5779 def : SIMDTableLookupAlias<asm # ".8b",
5780 !cast<Instruction>(NAME#"v8i8Three"),
5781 V64, VecListThree128>;
5782 def : SIMDTableLookupAlias<asm # ".8b",
5783 !cast<Instruction>(NAME#"v8i8Four"),
5784 V64, VecListFour128>;
5785 def : SIMDTableLookupAlias<asm # ".16b",
5786 !cast<Instruction>(NAME#"v16i8One"),
5787 V128, VecListOne128>;
5788 def : SIMDTableLookupAlias<asm # ".16b",
5789 !cast<Instruction>(NAME#"v16i8Two"),
5790 V128, VecListTwo128>;
5791 def : SIMDTableLookupAlias<asm # ".16b",
5792 !cast<Instruction>(NAME#"v16i8Three"),
5793 V128, VecListThree128>;
5794 def : SIMDTableLookupAlias<asm # ".16b",
5795 !cast<Instruction>(NAME#"v16i8Four"),
5796 V128, VecListFour128>;
5800 //----------------------------------------------------------------------------
5801 // AdvSIMD scalar CPY
5802 //----------------------------------------------------------------------------
5803 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5804 class BaseSIMDScalarCPY<RegisterClass regtype, RegisterOperand vectype,
5805 string kind, Operand idxtype>
5806 : I<(outs regtype:$dst), (ins vectype:$src, idxtype:$idx), "mov",
5807 "{\t$dst, $src" # kind # "$idx" #
5808 "|\t$dst, $src$idx}", "", []>,
5812 let Inst{31-21} = 0b01011110000;
5813 let Inst{15-10} = 0b000001;
5814 let Inst{9-5} = src;
5815 let Inst{4-0} = dst;
5818 class SIMDScalarCPYAlias<string asm, string size, Instruction inst,
5819 RegisterClass regtype, RegisterOperand vectype, Operand idxtype>
5820 : InstAlias<asm # "{\t$dst, $src" # size # "$index" #
5821 # "|\t$dst, $src$index}",
5822 (inst regtype:$dst, vectype:$src, idxtype:$index)>;
5825 multiclass SIMDScalarCPY<string asm> {
5826 def i8 : BaseSIMDScalarCPY<FPR8, V128, ".b", VectorIndexB> {
5828 let Inst{20-17} = idx;
5831 def i16 : BaseSIMDScalarCPY<FPR16, V128, ".h", VectorIndexH> {
5833 let Inst{20-18} = idx;
5834 let Inst{17-16} = 0b10;
5836 def i32 : BaseSIMDScalarCPY<FPR32, V128, ".s", VectorIndexS> {
5838 let Inst{20-19} = idx;
5839 let Inst{18-16} = 0b100;
5841 def i64 : BaseSIMDScalarCPY<FPR64, V128, ".d", VectorIndexD> {
5844 let Inst{19-16} = 0b1000;
5847 def : Pat<(v1i64 (scalar_to_vector (i64 (vector_extract (v2i64 V128:$src),
5848 VectorIndexD:$idx)))),
5849 (!cast<Instruction>(NAME # i64) V128:$src, VectorIndexD:$idx)>;
5851 // 'DUP' mnemonic aliases.
5852 def : SIMDScalarCPYAlias<"dup", ".b",
5853 !cast<Instruction>(NAME#"i8"),
5854 FPR8, V128, VectorIndexB>;
5855 def : SIMDScalarCPYAlias<"dup", ".h",
5856 !cast<Instruction>(NAME#"i16"),
5857 FPR16, V128, VectorIndexH>;
5858 def : SIMDScalarCPYAlias<"dup", ".s",
5859 !cast<Instruction>(NAME#"i32"),
5860 FPR32, V128, VectorIndexS>;
5861 def : SIMDScalarCPYAlias<"dup", ".d",
5862 !cast<Instruction>(NAME#"i64"),
5863 FPR64, V128, VectorIndexD>;
5866 //----------------------------------------------------------------------------
5867 // AdvSIMD modified immediate instructions
5868 //----------------------------------------------------------------------------
5870 class BaseSIMDModifiedImm<bit Q, bit op, dag oops, dag iops,
5871 string asm, string op_string,
5872 string cstr, list<dag> pattern>
5873 : I<oops, iops, asm, op_string, cstr, pattern>,
5880 let Inst{28-19} = 0b0111100000;
5881 let Inst{18-16} = imm8{7-5};
5882 let Inst{11-10} = 0b01;
5883 let Inst{9-5} = imm8{4-0};
5887 class BaseSIMDModifiedImmVector<bit Q, bit op, RegisterOperand vectype,
5888 Operand immtype, dag opt_shift_iop,
5889 string opt_shift, string asm, string kind,
5891 : BaseSIMDModifiedImm<Q, op, (outs vectype:$Rd),
5892 !con((ins immtype:$imm8), opt_shift_iop), asm,
5893 "{\t$Rd" # kind # ", $imm8" # opt_shift #
5894 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
5896 let DecoderMethod = "DecodeModImmInstruction";
5899 class BaseSIMDModifiedImmVectorTied<bit Q, bit op, RegisterOperand vectype,
5900 Operand immtype, dag opt_shift_iop,
5901 string opt_shift, string asm, string kind,
5903 : BaseSIMDModifiedImm<Q, op, (outs vectype:$dst),
5904 !con((ins vectype:$Rd, immtype:$imm8), opt_shift_iop),
5905 asm, "{\t$Rd" # kind # ", $imm8" # opt_shift #
5906 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
5907 "$Rd = $dst", pattern> {
5908 let DecoderMethod = "DecodeModImmTiedInstruction";
5911 class BaseSIMDModifiedImmVectorShift<bit Q, bit op, bits<2> b15_b12,
5912 RegisterOperand vectype, string asm,
5913 string kind, list<dag> pattern>
5914 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5915 (ins logical_vec_shift:$shift),
5916 "$shift", asm, kind, pattern> {
5918 let Inst{15} = b15_b12{1};
5919 let Inst{14-13} = shift;
5920 let Inst{12} = b15_b12{0};
5923 class BaseSIMDModifiedImmVectorShiftTied<bit Q, bit op, bits<2> b15_b12,
5924 RegisterOperand vectype, string asm,
5925 string kind, list<dag> pattern>
5926 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
5927 (ins logical_vec_shift:$shift),
5928 "$shift", asm, kind, pattern> {
5930 let Inst{15} = b15_b12{1};
5931 let Inst{14-13} = shift;
5932 let Inst{12} = b15_b12{0};
5936 class BaseSIMDModifiedImmVectorShiftHalf<bit Q, bit op, bits<2> b15_b12,
5937 RegisterOperand vectype, string asm,
5938 string kind, list<dag> pattern>
5939 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5940 (ins logical_vec_hw_shift:$shift),
5941 "$shift", asm, kind, pattern> {
5943 let Inst{15} = b15_b12{1};
5945 let Inst{13} = shift{0};
5946 let Inst{12} = b15_b12{0};
5949 class BaseSIMDModifiedImmVectorShiftHalfTied<bit Q, bit op, bits<2> b15_b12,
5950 RegisterOperand vectype, string asm,
5951 string kind, list<dag> pattern>
5952 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
5953 (ins logical_vec_hw_shift:$shift),
5954 "$shift", asm, kind, pattern> {
5956 let Inst{15} = b15_b12{1};
5958 let Inst{13} = shift{0};
5959 let Inst{12} = b15_b12{0};
5962 multiclass SIMDModifiedImmVectorShift<bit op, bits<2> hw_cmode, bits<2> w_cmode,
5964 def v4i16 : BaseSIMDModifiedImmVectorShiftHalf<0, op, hw_cmode, V64,
5966 def v8i16 : BaseSIMDModifiedImmVectorShiftHalf<1, op, hw_cmode, V128,
5969 def v2i32 : BaseSIMDModifiedImmVectorShift<0, op, w_cmode, V64,
5971 def v4i32 : BaseSIMDModifiedImmVectorShift<1, op, w_cmode, V128,
5975 multiclass SIMDModifiedImmVectorShiftTied<bit op, bits<2> hw_cmode,
5976 bits<2> w_cmode, string asm,
5978 def v4i16 : BaseSIMDModifiedImmVectorShiftHalfTied<0, op, hw_cmode, V64,
5980 [(set (v4i16 V64:$dst), (OpNode V64:$Rd,
5982 (i32 imm:$shift)))]>;
5983 def v8i16 : BaseSIMDModifiedImmVectorShiftHalfTied<1, op, hw_cmode, V128,
5985 [(set (v8i16 V128:$dst), (OpNode V128:$Rd,
5987 (i32 imm:$shift)))]>;
5989 def v2i32 : BaseSIMDModifiedImmVectorShiftTied<0, op, w_cmode, V64,
5991 [(set (v2i32 V64:$dst), (OpNode V64:$Rd,
5993 (i32 imm:$shift)))]>;
5994 def v4i32 : BaseSIMDModifiedImmVectorShiftTied<1, op, w_cmode, V128,
5996 [(set (v4i32 V128:$dst), (OpNode V128:$Rd,
5998 (i32 imm:$shift)))]>;
6001 class SIMDModifiedImmMoveMSL<bit Q, bit op, bits<4> cmode,
6002 RegisterOperand vectype, string asm,
6003 string kind, list<dag> pattern>
6004 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
6005 (ins move_vec_shift:$shift),
6006 "$shift", asm, kind, pattern> {
6008 let Inst{15-13} = cmode{3-1};
6009 let Inst{12} = shift;
6012 class SIMDModifiedImmVectorNoShift<bit Q, bit op, bits<4> cmode,
6013 RegisterOperand vectype,
6014 Operand imm_type, string asm,
6015 string kind, list<dag> pattern>
6016 : BaseSIMDModifiedImmVector<Q, op, vectype, imm_type, (ins), "",
6017 asm, kind, pattern> {
6018 let Inst{15-12} = cmode;
6021 class SIMDModifiedImmScalarNoShift<bit Q, bit op, bits<4> cmode, string asm,
6023 : BaseSIMDModifiedImm<Q, op, (outs FPR64:$Rd), (ins simdimmtype10:$imm8), asm,
6024 "\t$Rd, $imm8", "", pattern> {
6025 let Inst{15-12} = cmode;
6026 let DecoderMethod = "DecodeModImmInstruction";
6029 //----------------------------------------------------------------------------
6030 // AdvSIMD indexed element
6031 //----------------------------------------------------------------------------
6033 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6034 class BaseSIMDIndexed<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
6035 RegisterOperand dst_reg, RegisterOperand lhs_reg,
6036 RegisterOperand rhs_reg, Operand vec_idx, string asm,
6037 string apple_kind, string dst_kind, string lhs_kind,
6038 string rhs_kind, list<dag> pattern>
6039 : I<(outs dst_reg:$Rd), (ins lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx),
6041 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
6042 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "", pattern>,
6051 let Inst{28} = Scalar;
6052 let Inst{27-24} = 0b1111;
6053 let Inst{23-22} = size;
6054 // Bit 21 must be set by the derived class.
6055 let Inst{20-16} = Rm;
6056 let Inst{15-12} = opc;
6057 // Bit 11 must be set by the derived class.
6063 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
6064 class BaseSIMDIndexedTied<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
6065 RegisterOperand dst_reg, RegisterOperand lhs_reg,
6066 RegisterOperand rhs_reg, Operand vec_idx, string asm,
6067 string apple_kind, string dst_kind, string lhs_kind,
6068 string rhs_kind, list<dag> pattern>
6069 : I<(outs dst_reg:$dst),
6070 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx), asm,
6071 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
6072 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "$Rd = $dst", pattern>,
6081 let Inst{28} = Scalar;
6082 let Inst{27-24} = 0b1111;
6083 let Inst{23-22} = size;
6084 // Bit 21 must be set by the derived class.
6085 let Inst{20-16} = Rm;
6086 let Inst{15-12} = opc;
6087 // Bit 11 must be set by the derived class.
6093 multiclass SIMDFPIndexedSD<bit U, bits<4> opc, string asm,
6094 SDPatternOperator OpNode> {
6095 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6098 asm, ".2s", ".2s", ".2s", ".s",
6099 [(set (v2f32 V64:$Rd),
6100 (OpNode (v2f32 V64:$Rn),
6101 (v2f32 (ARM64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
6103 let Inst{11} = idx{1};
6104 let Inst{21} = idx{0};
6107 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6110 asm, ".4s", ".4s", ".4s", ".s",
6111 [(set (v4f32 V128:$Rd),
6112 (OpNode (v4f32 V128:$Rn),
6113 (v4f32 (ARM64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
6115 let Inst{11} = idx{1};
6116 let Inst{21} = idx{0};
6119 def v2i64_indexed : BaseSIMDIndexed<1, U, 0, 0b11, opc,
6122 asm, ".2d", ".2d", ".2d", ".d",
6123 [(set (v2f64 V128:$Rd),
6124 (OpNode (v2f64 V128:$Rn),
6125 (v2f64 (ARM64duplane64 (v2f64 V128:$Rm), VectorIndexD:$idx))))]> {
6127 let Inst{11} = idx{0};
6131 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6132 FPR32Op, FPR32Op, V128, VectorIndexS,
6133 asm, ".s", "", "", ".s",
6134 [(set (f32 FPR32Op:$Rd),
6135 (OpNode (f32 FPR32Op:$Rn),
6136 (f32 (vector_extract (v4f32 V128:$Rm),
6137 VectorIndexS:$idx))))]> {
6139 let Inst{11} = idx{1};
6140 let Inst{21} = idx{0};
6143 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b11, opc,
6144 FPR64Op, FPR64Op, V128, VectorIndexD,
6145 asm, ".d", "", "", ".d",
6146 [(set (f64 FPR64Op:$Rd),
6147 (OpNode (f64 FPR64Op:$Rn),
6148 (f64 (vector_extract (v2f64 V128:$Rm),
6149 VectorIndexD:$idx))))]> {
6151 let Inst{11} = idx{0};
6156 multiclass SIMDFPIndexedSDTiedPatterns<string INST, SDPatternOperator OpNode> {
6157 // 2 variants for the .2s version: DUPLANE from 128-bit and DUP scalar.
6158 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6159 (ARM64duplane32 (v4f32 V128:$Rm),
6160 VectorIndexS:$idx))),
6161 (!cast<Instruction>(INST # v2i32_indexed)
6162 V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6163 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6164 (ARM64dup (f32 FPR32Op:$Rm)))),
6165 (!cast<Instruction>(INST # "v2i32_indexed") V64:$Rd, V64:$Rn,
6166 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6169 // 2 variants for the .4s version: DUPLANE from 128-bit and DUP scalar.
6170 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6171 (ARM64duplane32 (v4f32 V128:$Rm),
6172 VectorIndexS:$idx))),
6173 (!cast<Instruction>(INST # "v4i32_indexed")
6174 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6175 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6176 (ARM64dup (f32 FPR32Op:$Rm)))),
6177 (!cast<Instruction>(INST # "v4i32_indexed") V128:$Rd, V128:$Rn,
6178 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6180 // 2 variants for the .2d version: DUPLANE from 128-bit and DUP scalar.
6181 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6182 (ARM64duplane64 (v2f64 V128:$Rm),
6183 VectorIndexD:$idx))),
6184 (!cast<Instruction>(INST # "v2i64_indexed")
6185 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6186 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6187 (ARM64dup (f64 FPR64Op:$Rm)))),
6188 (!cast<Instruction>(INST # "v2i64_indexed") V128:$Rd, V128:$Rn,
6189 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
6191 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
6192 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6193 (vector_extract (v4f32 V128:$Rm), VectorIndexS:$idx))),
6194 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
6195 V128:$Rm, VectorIndexS:$idx)>;
6196 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6197 (vector_extract (v2f32 V64:$Rm), VectorIndexS:$idx))),
6198 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
6199 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
6201 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
6202 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
6203 (vector_extract (v2f64 V128:$Rm), VectorIndexD:$idx))),
6204 (!cast<Instruction>(INST # "v1i64_indexed") FPR64:$Rd, FPR64:$Rn,
6205 V128:$Rm, VectorIndexD:$idx)>;
6208 multiclass SIMDFPIndexedSDTied<bit U, bits<4> opc, string asm> {
6209 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc, V64, V64,
6211 asm, ".2s", ".2s", ".2s", ".s", []> {
6213 let Inst{11} = idx{1};
6214 let Inst{21} = idx{0};
6217 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6220 asm, ".4s", ".4s", ".4s", ".s", []> {
6222 let Inst{11} = idx{1};
6223 let Inst{21} = idx{0};
6226 def v2i64_indexed : BaseSIMDIndexedTied<1, U, 0, 0b11, opc,
6229 asm, ".2d", ".2d", ".2d", ".d", []> {
6231 let Inst{11} = idx{0};
6236 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
6237 FPR32Op, FPR32Op, V128, VectorIndexS,
6238 asm, ".s", "", "", ".s", []> {
6240 let Inst{11} = idx{1};
6241 let Inst{21} = idx{0};
6244 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b11, opc,
6245 FPR64Op, FPR64Op, V128, VectorIndexD,
6246 asm, ".d", "", "", ".d", []> {
6248 let Inst{11} = idx{0};
6253 multiclass SIMDIndexedHS<bit U, bits<4> opc, string asm,
6254 SDPatternOperator OpNode> {
6255 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc, V64, V64,
6256 V128_lo, VectorIndexH,
6257 asm, ".4h", ".4h", ".4h", ".h",
6258 [(set (v4i16 V64:$Rd),
6259 (OpNode (v4i16 V64:$Rn),
6260 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6262 let Inst{11} = idx{2};
6263 let Inst{21} = idx{1};
6264 let Inst{20} = idx{0};
6267 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6269 V128_lo, VectorIndexH,
6270 asm, ".8h", ".8h", ".8h", ".h",
6271 [(set (v8i16 V128:$Rd),
6272 (OpNode (v8i16 V128:$Rn),
6273 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6275 let Inst{11} = idx{2};
6276 let Inst{21} = idx{1};
6277 let Inst{20} = idx{0};
6280 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6283 asm, ".2s", ".2s", ".2s", ".s",
6284 [(set (v2i32 V64:$Rd),
6285 (OpNode (v2i32 V64:$Rn),
6286 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6288 let Inst{11} = idx{1};
6289 let Inst{21} = idx{0};
6292 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6295 asm, ".4s", ".4s", ".4s", ".s",
6296 [(set (v4i32 V128:$Rd),
6297 (OpNode (v4i32 V128:$Rn),
6298 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6300 let Inst{11} = idx{1};
6301 let Inst{21} = idx{0};
6304 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
6305 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
6306 asm, ".h", "", "", ".h", []> {
6308 let Inst{11} = idx{2};
6309 let Inst{21} = idx{1};
6310 let Inst{20} = idx{0};
6313 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6314 FPR32Op, FPR32Op, V128, VectorIndexS,
6315 asm, ".s", "", "", ".s",
6316 [(set (i32 FPR32Op:$Rd),
6317 (OpNode FPR32Op:$Rn,
6318 (i32 (vector_extract (v4i32 V128:$Rm),
6319 VectorIndexS:$idx))))]> {
6321 let Inst{11} = idx{1};
6322 let Inst{21} = idx{0};
6326 multiclass SIMDVectorIndexedHS<bit U, bits<4> opc, string asm,
6327 SDPatternOperator OpNode> {
6328 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6330 V128_lo, VectorIndexH,
6331 asm, ".4h", ".4h", ".4h", ".h",
6332 [(set (v4i16 V64:$Rd),
6333 (OpNode (v4i16 V64:$Rn),
6334 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6336 let Inst{11} = idx{2};
6337 let Inst{21} = idx{1};
6338 let Inst{20} = idx{0};
6341 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6343 V128_lo, VectorIndexH,
6344 asm, ".8h", ".8h", ".8h", ".h",
6345 [(set (v8i16 V128:$Rd),
6346 (OpNode (v8i16 V128:$Rn),
6347 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6349 let Inst{11} = idx{2};
6350 let Inst{21} = idx{1};
6351 let Inst{20} = idx{0};
6354 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6357 asm, ".2s", ".2s", ".2s", ".s",
6358 [(set (v2i32 V64:$Rd),
6359 (OpNode (v2i32 V64:$Rn),
6360 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6362 let Inst{11} = idx{1};
6363 let Inst{21} = idx{0};
6366 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6369 asm, ".4s", ".4s", ".4s", ".s",
6370 [(set (v4i32 V128:$Rd),
6371 (OpNode (v4i32 V128:$Rn),
6372 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6374 let Inst{11} = idx{1};
6375 let Inst{21} = idx{0};
6379 multiclass SIMDVectorIndexedHSTied<bit U, bits<4> opc, string asm,
6380 SDPatternOperator OpNode> {
6381 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc, V64, V64,
6382 V128_lo, VectorIndexH,
6383 asm, ".4h", ".4h", ".4h", ".h",
6384 [(set (v4i16 V64:$dst),
6385 (OpNode (v4i16 V64:$Rd),(v4i16 V64:$Rn),
6386 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6388 let Inst{11} = idx{2};
6389 let Inst{21} = idx{1};
6390 let Inst{20} = idx{0};
6393 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6395 V128_lo, VectorIndexH,
6396 asm, ".8h", ".8h", ".8h", ".h",
6397 [(set (v8i16 V128:$dst),
6398 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
6399 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6401 let Inst{11} = idx{2};
6402 let Inst{21} = idx{1};
6403 let Inst{20} = idx{0};
6406 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6409 asm, ".2s", ".2s", ".2s", ".s",
6410 [(set (v2i32 V64:$dst),
6411 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
6412 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6414 let Inst{11} = idx{1};
6415 let Inst{21} = idx{0};
6418 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6421 asm, ".4s", ".4s", ".4s", ".s",
6422 [(set (v4i32 V128:$dst),
6423 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
6424 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6426 let Inst{11} = idx{1};
6427 let Inst{21} = idx{0};
6431 multiclass SIMDIndexedLongSD<bit U, bits<4> opc, string asm,
6432 SDPatternOperator OpNode> {
6433 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6435 V128_lo, VectorIndexH,
6436 asm, ".4s", ".4s", ".4h", ".h",
6437 [(set (v4i32 V128:$Rd),
6438 (OpNode (v4i16 V64:$Rn),
6439 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6441 let Inst{11} = idx{2};
6442 let Inst{21} = idx{1};
6443 let Inst{20} = idx{0};
6446 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6448 V128_lo, VectorIndexH,
6449 asm#"2", ".4s", ".4s", ".8h", ".h",
6450 [(set (v4i32 V128:$Rd),
6451 (OpNode (extract_high_v8i16 V128:$Rn),
6452 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6453 VectorIndexH:$idx))))]> {
6456 let Inst{11} = idx{2};
6457 let Inst{21} = idx{1};
6458 let Inst{20} = idx{0};
6461 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6464 asm, ".2d", ".2d", ".2s", ".s",
6465 [(set (v2i64 V128:$Rd),
6466 (OpNode (v2i32 V64:$Rn),
6467 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6469 let Inst{11} = idx{1};
6470 let Inst{21} = idx{0};
6473 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6476 asm#"2", ".2d", ".2d", ".4s", ".s",
6477 [(set (v2i64 V128:$Rd),
6478 (OpNode (extract_high_v4i32 V128:$Rn),
6479 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6480 VectorIndexS:$idx))))]> {
6482 let Inst{11} = idx{1};
6483 let Inst{21} = idx{0};
6486 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
6487 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
6488 asm, ".h", "", "", ".h", []> {
6490 let Inst{11} = idx{2};
6491 let Inst{21} = idx{1};
6492 let Inst{20} = idx{0};
6495 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6496 FPR64Op, FPR32Op, V128, VectorIndexS,
6497 asm, ".s", "", "", ".s", []> {
6499 let Inst{11} = idx{1};
6500 let Inst{21} = idx{0};
6504 multiclass SIMDIndexedLongSQDMLXSDTied<bit U, bits<4> opc, string asm,
6505 SDPatternOperator Accum> {
6506 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
6508 V128_lo, VectorIndexH,
6509 asm, ".4s", ".4s", ".4h", ".h",
6510 [(set (v4i32 V128:$dst),
6511 (Accum (v4i32 V128:$Rd),
6512 (v4i32 (int_arm64_neon_sqdmull
6514 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6515 VectorIndexH:$idx))))))]> {
6517 let Inst{11} = idx{2};
6518 let Inst{21} = idx{1};
6519 let Inst{20} = idx{0};
6522 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but an
6523 // intermediate EXTRACT_SUBREG would be untyped.
6524 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
6525 (i32 (vector_extract (v4i32
6526 (int_arm64_neon_sqdmull (v4i16 V64:$Rn),
6527 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6528 VectorIndexH:$idx)))),
6531 (!cast<Instruction>(NAME # v4i16_indexed)
6532 (SUBREG_TO_REG (i32 0), FPR32Op:$Rd, ssub), V64:$Rn,
6533 V128_lo:$Rm, VectorIndexH:$idx),
6536 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6538 V128_lo, VectorIndexH,
6539 asm#"2", ".4s", ".4s", ".8h", ".h",
6540 [(set (v4i32 V128:$dst),
6541 (Accum (v4i32 V128:$Rd),
6542 (v4i32 (int_arm64_neon_sqdmull
6543 (extract_high_v8i16 V128:$Rn),
6545 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6546 VectorIndexH:$idx))))))]> {
6548 let Inst{11} = idx{2};
6549 let Inst{21} = idx{1};
6550 let Inst{20} = idx{0};
6553 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6556 asm, ".2d", ".2d", ".2s", ".s",
6557 [(set (v2i64 V128:$dst),
6558 (Accum (v2i64 V128:$Rd),
6559 (v2i64 (int_arm64_neon_sqdmull
6561 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm),
6562 VectorIndexS:$idx))))))]> {
6564 let Inst{11} = idx{1};
6565 let Inst{21} = idx{0};
6568 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6571 asm#"2", ".2d", ".2d", ".4s", ".s",
6572 [(set (v2i64 V128:$dst),
6573 (Accum (v2i64 V128:$Rd),
6574 (v2i64 (int_arm64_neon_sqdmull
6575 (extract_high_v4i32 V128:$Rn),
6577 (ARM64duplane32 (v4i32 V128:$Rm),
6578 VectorIndexS:$idx))))))]> {
6580 let Inst{11} = idx{1};
6581 let Inst{21} = idx{0};
6584 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
6585 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
6586 asm, ".h", "", "", ".h", []> {
6588 let Inst{11} = idx{2};
6589 let Inst{21} = idx{1};
6590 let Inst{20} = idx{0};
6594 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
6595 FPR64Op, FPR32Op, V128, VectorIndexS,
6596 asm, ".s", "", "", ".s",
6597 [(set (i64 FPR64Op:$dst),
6598 (Accum (i64 FPR64Op:$Rd),
6599 (i64 (int_arm64_neon_sqdmulls_scalar
6601 (i32 (vector_extract (v4i32 V128:$Rm),
6602 VectorIndexS:$idx))))))]> {
6605 let Inst{11} = idx{1};
6606 let Inst{21} = idx{0};
6610 multiclass SIMDVectorIndexedLongSD<bit U, bits<4> opc, string asm,
6611 SDPatternOperator OpNode> {
6612 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6613 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6615 V128_lo, VectorIndexH,
6616 asm, ".4s", ".4s", ".4h", ".h",
6617 [(set (v4i32 V128:$Rd),
6618 (OpNode (v4i16 V64:$Rn),
6619 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6621 let Inst{11} = idx{2};
6622 let Inst{21} = idx{1};
6623 let Inst{20} = idx{0};
6626 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6628 V128_lo, VectorIndexH,
6629 asm#"2", ".4s", ".4s", ".8h", ".h",
6630 [(set (v4i32 V128:$Rd),
6631 (OpNode (extract_high_v8i16 V128:$Rn),
6632 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6633 VectorIndexH:$idx))))]> {
6636 let Inst{11} = idx{2};
6637 let Inst{21} = idx{1};
6638 let Inst{20} = idx{0};
6641 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6644 asm, ".2d", ".2d", ".2s", ".s",
6645 [(set (v2i64 V128:$Rd),
6646 (OpNode (v2i32 V64:$Rn),
6647 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6649 let Inst{11} = idx{1};
6650 let Inst{21} = idx{0};
6653 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6656 asm#"2", ".2d", ".2d", ".4s", ".s",
6657 [(set (v2i64 V128:$Rd),
6658 (OpNode (extract_high_v4i32 V128:$Rn),
6659 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6660 VectorIndexS:$idx))))]> {
6662 let Inst{11} = idx{1};
6663 let Inst{21} = idx{0};
6668 multiclass SIMDVectorIndexedLongSDTied<bit U, bits<4> opc, string asm,
6669 SDPatternOperator OpNode> {
6670 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6671 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
6673 V128_lo, VectorIndexH,
6674 asm, ".4s", ".4s", ".4h", ".h",
6675 [(set (v4i32 V128:$dst),
6676 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn),
6677 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6679 let Inst{11} = idx{2};
6680 let Inst{21} = idx{1};
6681 let Inst{20} = idx{0};
6684 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6686 V128_lo, VectorIndexH,
6687 asm#"2", ".4s", ".4s", ".8h", ".h",
6688 [(set (v4i32 V128:$dst),
6689 (OpNode (v4i32 V128:$Rd),
6690 (extract_high_v8i16 V128:$Rn),
6691 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6692 VectorIndexH:$idx))))]> {
6694 let Inst{11} = idx{2};
6695 let Inst{21} = idx{1};
6696 let Inst{20} = idx{0};
6699 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6702 asm, ".2d", ".2d", ".2s", ".s",
6703 [(set (v2i64 V128:$dst),
6704 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn),
6705 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6707 let Inst{11} = idx{1};
6708 let Inst{21} = idx{0};
6711 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6714 asm#"2", ".2d", ".2d", ".4s", ".s",
6715 [(set (v2i64 V128:$dst),
6716 (OpNode (v2i64 V128:$Rd),
6717 (extract_high_v4i32 V128:$Rn),
6718 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6719 VectorIndexS:$idx))))]> {
6721 let Inst{11} = idx{1};
6722 let Inst{21} = idx{0};
6727 //----------------------------------------------------------------------------
6728 // AdvSIMD scalar shift by immediate
6729 //----------------------------------------------------------------------------
6731 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6732 class BaseSIMDScalarShift<bit U, bits<5> opc, bits<7> fixed_imm,
6733 RegisterClass regtype1, RegisterClass regtype2,
6734 Operand immtype, string asm, list<dag> pattern>
6735 : I<(outs regtype1:$Rd), (ins regtype2:$Rn, immtype:$imm),
6736 asm, "\t$Rd, $Rn, $imm", "", pattern>,
6741 let Inst{31-30} = 0b01;
6743 let Inst{28-23} = 0b111110;
6744 let Inst{22-16} = fixed_imm;
6745 let Inst{15-11} = opc;
6751 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6752 class BaseSIMDScalarShiftTied<bit U, bits<5> opc, bits<7> fixed_imm,
6753 RegisterClass regtype1, RegisterClass regtype2,
6754 Operand immtype, string asm, list<dag> pattern>
6755 : I<(outs regtype1:$dst), (ins regtype1:$Rd, regtype2:$Rn, immtype:$imm),
6756 asm, "\t$Rd, $Rn, $imm", "$Rd = $dst", pattern>,
6761 let Inst{31-30} = 0b01;
6763 let Inst{28-23} = 0b111110;
6764 let Inst{22-16} = fixed_imm;
6765 let Inst{15-11} = opc;
6772 multiclass SIMDScalarRShiftSD<bit U, bits<5> opc, string asm> {
6773 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6774 FPR32, FPR32, vecshiftR32, asm, []> {
6775 let Inst{20-16} = imm{4-0};
6778 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6779 FPR64, FPR64, vecshiftR64, asm, []> {
6780 let Inst{21-16} = imm{5-0};
6784 multiclass SIMDScalarRShiftD<bit U, bits<5> opc, string asm,
6785 SDPatternOperator OpNode> {
6786 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6787 FPR64, FPR64, vecshiftR64, asm,
6788 [(set (i64 FPR64:$Rd),
6789 (OpNode (i64 FPR64:$Rn), (i32 vecshiftR64:$imm)))]> {
6790 let Inst{21-16} = imm{5-0};
6793 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftR64:$imm))),
6794 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftR64:$imm)>;
6797 multiclass SIMDScalarRShiftDTied<bit U, bits<5> opc, string asm,
6798 SDPatternOperator OpNode = null_frag> {
6799 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
6800 FPR64, FPR64, vecshiftR64, asm,
6801 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn),
6802 (i32 vecshiftR64:$imm)))]> {
6803 let Inst{21-16} = imm{5-0};
6806 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
6807 (i32 vecshiftR64:$imm))),
6808 (!cast<Instruction>(NAME # "d") FPR64:$Rd, FPR64:$Rn,
6812 multiclass SIMDScalarLShiftD<bit U, bits<5> opc, string asm,
6813 SDPatternOperator OpNode> {
6814 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6815 FPR64, FPR64, vecshiftL64, asm,
6816 [(set (v1i64 FPR64:$Rd),
6817 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
6818 let Inst{21-16} = imm{5-0};
6822 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6823 multiclass SIMDScalarLShiftDTied<bit U, bits<5> opc, string asm> {
6824 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
6825 FPR64, FPR64, vecshiftL64, asm, []> {
6826 let Inst{21-16} = imm{5-0};
6830 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6831 multiclass SIMDScalarRShiftBHS<bit U, bits<5> opc, string asm,
6832 SDPatternOperator OpNode = null_frag> {
6833 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6834 FPR8, FPR16, vecshiftR8, asm, []> {
6835 let Inst{18-16} = imm{2-0};
6838 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6839 FPR16, FPR32, vecshiftR16, asm, []> {
6840 let Inst{19-16} = imm{3-0};
6843 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6844 FPR32, FPR64, vecshiftR32, asm,
6845 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn), vecshiftR32:$imm))]> {
6846 let Inst{20-16} = imm{4-0};
6850 multiclass SIMDScalarLShiftBHSD<bit U, bits<5> opc, string asm,
6851 SDPatternOperator OpNode> {
6852 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6853 FPR8, FPR8, vecshiftL8, asm, []> {
6854 let Inst{18-16} = imm{2-0};
6857 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6858 FPR16, FPR16, vecshiftL16, asm, []> {
6859 let Inst{19-16} = imm{3-0};
6862 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6863 FPR32, FPR32, vecshiftL32, asm,
6864 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn), (i32 vecshiftL32:$imm)))]> {
6865 let Inst{20-16} = imm{4-0};
6868 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6869 FPR64, FPR64, vecshiftL64, asm,
6870 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn),
6871 (i32 vecshiftL64:$imm)))]> {
6872 let Inst{21-16} = imm{5-0};
6876 multiclass SIMDScalarRShiftBHSD<bit U, bits<5> opc, string asm> {
6877 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6878 FPR8, FPR8, vecshiftR8, asm, []> {
6879 let Inst{18-16} = imm{2-0};
6882 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6883 FPR16, FPR16, vecshiftR16, asm, []> {
6884 let Inst{19-16} = imm{3-0};
6887 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6888 FPR32, FPR32, vecshiftR32, asm, []> {
6889 let Inst{20-16} = imm{4-0};
6892 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6893 FPR64, FPR64, vecshiftR64, asm, []> {
6894 let Inst{21-16} = imm{5-0};
6898 //----------------------------------------------------------------------------
6899 // AdvSIMD vector x indexed element
6900 //----------------------------------------------------------------------------
6902 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6903 class BaseSIMDVectorShift<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
6904 RegisterOperand dst_reg, RegisterOperand src_reg,
6906 string asm, string dst_kind, string src_kind,
6908 : I<(outs dst_reg:$Rd), (ins src_reg:$Rn, immtype:$imm),
6909 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
6910 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "", pattern>,
6917 let Inst{28-23} = 0b011110;
6918 let Inst{22-16} = fixed_imm;
6919 let Inst{15-11} = opc;
6925 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6926 class BaseSIMDVectorShiftTied<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
6927 RegisterOperand vectype1, RegisterOperand vectype2,
6929 string asm, string dst_kind, string src_kind,
6931 : I<(outs vectype1:$dst), (ins vectype1:$Rd, vectype2:$Rn, immtype:$imm),
6932 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
6933 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "$Rd = $dst", pattern>,
6940 let Inst{28-23} = 0b011110;
6941 let Inst{22-16} = fixed_imm;
6942 let Inst{15-11} = opc;
6948 multiclass SIMDVectorRShiftSD<bit U, bits<5> opc, string asm,
6950 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6951 V64, V64, vecshiftR32,
6953 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (i32 imm:$imm)))]> {
6955 let Inst{20-16} = imm;
6958 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
6959 V128, V128, vecshiftR32,
6961 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (i32 imm:$imm)))]> {
6963 let Inst{20-16} = imm;
6966 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
6967 V128, V128, vecshiftR64,
6969 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (i32 imm:$imm)))]> {
6971 let Inst{21-16} = imm;
6975 multiclass SIMDVectorRShiftSDToFP<bit U, bits<5> opc, string asm,
6977 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6978 V64, V64, vecshiftR32,
6980 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (i32 imm:$imm)))]> {
6982 let Inst{20-16} = imm;
6985 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
6986 V128, V128, vecshiftR32,
6988 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (i32 imm:$imm)))]> {
6990 let Inst{20-16} = imm;
6993 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
6994 V128, V128, vecshiftR64,
6996 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (i32 imm:$imm)))]> {
6998 let Inst{21-16} = imm;
7002 multiclass SIMDVectorRShiftNarrowBHS<bit U, bits<5> opc, string asm,
7003 SDPatternOperator OpNode> {
7004 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7005 V64, V128, vecshiftR16Narrow,
7007 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))]> {
7009 let Inst{18-16} = imm;
7012 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
7013 V128, V128, vecshiftR16Narrow,
7014 asm#"2", ".16b", ".8h", []> {
7016 let Inst{18-16} = imm;
7017 let hasSideEffects = 0;
7020 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7021 V64, V128, vecshiftR32Narrow,
7023 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))]> {
7025 let Inst{19-16} = imm;
7028 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
7029 V128, V128, vecshiftR32Narrow,
7030 asm#"2", ".8h", ".4s", []> {
7032 let Inst{19-16} = imm;
7033 let hasSideEffects = 0;
7036 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7037 V64, V128, vecshiftR64Narrow,
7039 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))]> {
7041 let Inst{20-16} = imm;
7044 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
7045 V128, V128, vecshiftR64Narrow,
7046 asm#"2", ".4s", ".2d", []> {
7048 let Inst{20-16} = imm;
7049 let hasSideEffects = 0;
7052 // TableGen doesn't like patters w/ INSERT_SUBREG on the instructions
7053 // themselves, so put them here instead.
7055 // Patterns involving what's effectively an insert high and a normal
7056 // intrinsic, represented by CONCAT_VECTORS.
7057 def : Pat<(concat_vectors (v8i8 V64:$Rd),(OpNode (v8i16 V128:$Rn),
7058 vecshiftR16Narrow:$imm)),
7059 (!cast<Instruction>(NAME # "v16i8_shift")
7060 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
7061 V128:$Rn, vecshiftR16Narrow:$imm)>;
7062 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn),
7063 vecshiftR32Narrow:$imm)),
7064 (!cast<Instruction>(NAME # "v8i16_shift")
7065 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
7066 V128:$Rn, vecshiftR32Narrow:$imm)>;
7067 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn),
7068 vecshiftR64Narrow:$imm)),
7069 (!cast<Instruction>(NAME # "v4i32_shift")
7070 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
7071 V128:$Rn, vecshiftR64Narrow:$imm)>;
7074 multiclass SIMDVectorLShiftBHSD<bit U, bits<5> opc, string asm,
7075 SDPatternOperator OpNode> {
7076 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7077 V64, V64, vecshiftL8,
7079 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
7080 (i32 vecshiftL8:$imm)))]> {
7082 let Inst{18-16} = imm;
7085 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7086 V128, V128, vecshiftL8,
7087 asm, ".16b", ".16b",
7088 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
7089 (i32 vecshiftL8:$imm)))]> {
7091 let Inst{18-16} = imm;
7094 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7095 V64, V64, vecshiftL16,
7097 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
7098 (i32 vecshiftL16:$imm)))]> {
7100 let Inst{19-16} = imm;
7103 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7104 V128, V128, vecshiftL16,
7106 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
7107 (i32 vecshiftL16:$imm)))]> {
7109 let Inst{19-16} = imm;
7112 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7113 V64, V64, vecshiftL32,
7115 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
7116 (i32 vecshiftL32:$imm)))]> {
7118 let Inst{20-16} = imm;
7121 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7122 V128, V128, vecshiftL32,
7124 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
7125 (i32 vecshiftL32:$imm)))]> {
7127 let Inst{20-16} = imm;
7130 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
7131 V128, V128, vecshiftL64,
7133 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
7134 (i32 vecshiftL64:$imm)))]> {
7136 let Inst{21-16} = imm;
7140 multiclass SIMDVectorRShiftBHSD<bit U, bits<5> opc, string asm,
7141 SDPatternOperator OpNode> {
7142 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7143 V64, V64, vecshiftR8,
7145 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
7146 (i32 vecshiftR8:$imm)))]> {
7148 let Inst{18-16} = imm;
7151 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7152 V128, V128, vecshiftR8,
7153 asm, ".16b", ".16b",
7154 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
7155 (i32 vecshiftR8:$imm)))]> {
7157 let Inst{18-16} = imm;
7160 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7161 V64, V64, vecshiftR16,
7163 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
7164 (i32 vecshiftR16:$imm)))]> {
7166 let Inst{19-16} = imm;
7169 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7170 V128, V128, vecshiftR16,
7172 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
7173 (i32 vecshiftR16:$imm)))]> {
7175 let Inst{19-16} = imm;
7178 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7179 V64, V64, vecshiftR32,
7181 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
7182 (i32 vecshiftR32:$imm)))]> {
7184 let Inst{20-16} = imm;
7187 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7188 V128, V128, vecshiftR32,
7190 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
7191 (i32 vecshiftR32:$imm)))]> {
7193 let Inst{20-16} = imm;
7196 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
7197 V128, V128, vecshiftR64,
7199 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
7200 (i32 vecshiftR64:$imm)))]> {
7202 let Inst{21-16} = imm;
7206 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7207 multiclass SIMDVectorRShiftBHSDTied<bit U, bits<5> opc, string asm,
7208 SDPatternOperator OpNode = null_frag> {
7209 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
7210 V64, V64, vecshiftR8, asm, ".8b", ".8b",
7211 [(set (v8i8 V64:$dst),
7212 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
7213 (i32 vecshiftR8:$imm)))]> {
7215 let Inst{18-16} = imm;
7218 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
7219 V128, V128, vecshiftR8, asm, ".16b", ".16b",
7220 [(set (v16i8 V128:$dst),
7221 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
7222 (i32 vecshiftR8:$imm)))]> {
7224 let Inst{18-16} = imm;
7227 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
7228 V64, V64, vecshiftR16, asm, ".4h", ".4h",
7229 [(set (v4i16 V64:$dst),
7230 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
7231 (i32 vecshiftR16:$imm)))]> {
7233 let Inst{19-16} = imm;
7236 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
7237 V128, V128, vecshiftR16, asm, ".8h", ".8h",
7238 [(set (v8i16 V128:$dst),
7239 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7240 (i32 vecshiftR16:$imm)))]> {
7242 let Inst{19-16} = imm;
7245 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
7246 V64, V64, vecshiftR32, asm, ".2s", ".2s",
7247 [(set (v2i32 V64:$dst),
7248 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7249 (i32 vecshiftR32:$imm)))]> {
7251 let Inst{20-16} = imm;
7254 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
7255 V128, V128, vecshiftR32, asm, ".4s", ".4s",
7256 [(set (v4i32 V128:$dst),
7257 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7258 (i32 vecshiftR32:$imm)))]> {
7260 let Inst{20-16} = imm;
7263 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
7264 V128, V128, vecshiftR64,
7265 asm, ".2d", ".2d", [(set (v2i64 V128:$dst),
7266 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
7267 (i32 vecshiftR64:$imm)))]> {
7269 let Inst{21-16} = imm;
7273 multiclass SIMDVectorLShiftBHSDTied<bit U, bits<5> opc, string asm,
7274 SDPatternOperator OpNode = null_frag> {
7275 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
7276 V64, V64, vecshiftL8,
7278 [(set (v8i8 V64:$dst),
7279 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
7280 (i32 vecshiftL8:$imm)))]> {
7282 let Inst{18-16} = imm;
7285 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
7286 V128, V128, vecshiftL8,
7287 asm, ".16b", ".16b",
7288 [(set (v16i8 V128:$dst),
7289 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
7290 (i32 vecshiftL8:$imm)))]> {
7292 let Inst{18-16} = imm;
7295 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
7296 V64, V64, vecshiftL16,
7298 [(set (v4i16 V64:$dst),
7299 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
7300 (i32 vecshiftL16:$imm)))]> {
7302 let Inst{19-16} = imm;
7305 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
7306 V128, V128, vecshiftL16,
7308 [(set (v8i16 V128:$dst),
7309 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7310 (i32 vecshiftL16:$imm)))]> {
7312 let Inst{19-16} = imm;
7315 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
7316 V64, V64, vecshiftL32,
7318 [(set (v2i32 V64:$dst),
7319 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7320 (i32 vecshiftL32:$imm)))]> {
7322 let Inst{20-16} = imm;
7325 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
7326 V128, V128, vecshiftL32,
7328 [(set (v4i32 V128:$dst),
7329 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7330 (i32 vecshiftL32:$imm)))]> {
7332 let Inst{20-16} = imm;
7335 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
7336 V128, V128, vecshiftL64,
7338 [(set (v2i64 V128:$dst),
7339 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
7340 (i32 vecshiftL64:$imm)))]> {
7342 let Inst{21-16} = imm;
7346 multiclass SIMDVectorLShiftLongBHSD<bit U, bits<5> opc, string asm,
7347 SDPatternOperator OpNode> {
7348 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7349 V128, V64, vecshiftL8, asm, ".8h", ".8b",
7350 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), vecshiftL8:$imm))]> {
7352 let Inst{18-16} = imm;
7355 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7356 V128, V128, vecshiftL8,
7357 asm#"2", ".8h", ".16b",
7358 [(set (v8i16 V128:$Rd),
7359 (OpNode (extract_high_v16i8 V128:$Rn), vecshiftL8:$imm))]> {
7361 let Inst{18-16} = imm;
7364 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7365 V128, V64, vecshiftL16, asm, ".4s", ".4h",
7366 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), vecshiftL16:$imm))]> {
7368 let Inst{19-16} = imm;
7371 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7372 V128, V128, vecshiftL16,
7373 asm#"2", ".4s", ".8h",
7374 [(set (v4i32 V128:$Rd),
7375 (OpNode (extract_high_v8i16 V128:$Rn), vecshiftL16:$imm))]> {
7378 let Inst{19-16} = imm;
7381 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7382 V128, V64, vecshiftL32, asm, ".2d", ".2s",
7383 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), vecshiftL32:$imm))]> {
7385 let Inst{20-16} = imm;
7388 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7389 V128, V128, vecshiftL32,
7390 asm#"2", ".2d", ".4s",
7391 [(set (v2i64 V128:$Rd),
7392 (OpNode (extract_high_v4i32 V128:$Rn), vecshiftL32:$imm))]> {
7394 let Inst{20-16} = imm;
7400 // Vector load/store
7402 // SIMD ldX/stX no-index memory references don't allow the optional
7403 // ", #0" constant and handle post-indexing explicitly, so we use
7404 // a more specialized parse method for them. Otherwise, it's the same as
7405 // the general am_noindex handling.
7407 class BaseSIMDLdSt<bit Q, bit L, bits<4> opcode, bits<2> size,
7408 string asm, dag oops, dag iops, list<dag> pattern>
7409 : I<oops, iops, asm, "\t$Vt, $vaddr", "", pattern> {
7414 let Inst{29-23} = 0b0011000;
7416 let Inst{21-16} = 0b000000;
7417 let Inst{15-12} = opcode;
7418 let Inst{11-10} = size;
7419 let Inst{9-5} = vaddr;
7423 class BaseSIMDLdStPost<bit Q, bit L, bits<4> opcode, bits<2> size,
7424 string asm, dag oops, dag iops>
7425 : I<oops, iops, asm, "\t$Vt, $vaddr, $Xm", "", []> {
7431 let Inst{29-23} = 0b0011001;
7434 let Inst{20-16} = Xm;
7435 let Inst{15-12} = opcode;
7436 let Inst{11-10} = size;
7437 let Inst{9-5} = vaddr;
7439 let DecoderMethod = "DecodeSIMDLdStPost";
7442 // The immediate form of AdvSIMD post-indexed addressing is encoded with
7443 // register post-index addressing from the zero register.
7444 multiclass SIMDLdStAliases<string asm, string layout, string Count,
7445 int Offset, int Size> {
7446 // E.g. "ld1 { v0.8b, v1.8b }, [x1], #16"
7447 // "ld1\t$Vt, $vaddr, #16"
7448 // may get mapped to
7449 // (LD1Twov8b_POST VecListTwo8b:$Vt, am_simdnoindex:$vaddr, XZR)
7450 def : InstAlias<asm # "\t$Vt, $vaddr, #" # Offset,
7451 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7452 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
7453 am_simdnoindex:$vaddr, XZR), 1>;
7455 // E.g. "ld1.8b { v0, v1 }, [x1], #16"
7456 // "ld1.8b\t$Vt, $vaddr, #16"
7457 // may get mapped to
7458 // (LD1Twov8b_POST VecListTwo64:$Vt, am_simdnoindex:$vaddr, XZR)
7459 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, #" # Offset,
7460 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7461 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7462 am_simdnoindex:$vaddr, XZR), 0>;
7464 // E.g. "ld1.8b { v0, v1 }, [x1]"
7465 // "ld1\t$Vt, $vaddr"
7466 // may get mapped to
7467 // (LD1Twov8b VecListTwo64:$Vt, am_simdnoindex:$vaddr)
7468 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr",
7469 (!cast<Instruction>(NAME # Count # "v" # layout)
7470 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7471 am_simdnoindex:$vaddr), 0>;
7473 // E.g. "ld1.8b { v0, v1 }, [x1], x2"
7474 // "ld1\t$Vt, $vaddr, $Xm"
7475 // may get mapped to
7476 // (LD1Twov8b_POST VecListTwo64:$Vt, am_simdnoindex:$vaddr, GPR64pi8:$Xm)
7477 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, $Xm",
7478 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7479 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7480 am_simdnoindex:$vaddr,
7481 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
7484 multiclass BaseSIMDLdN<string Count, string asm, string veclist, int Offset128,
7485 int Offset64, bits<4> opcode> {
7486 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
7487 def v16b: BaseSIMDLdSt<1, 1, opcode, 0b00, asm,
7488 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
7489 (ins am_simdnoindex:$vaddr), []>;
7490 def v8h : BaseSIMDLdSt<1, 1, opcode, 0b01, asm,
7491 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
7492 (ins am_simdnoindex:$vaddr), []>;
7493 def v4s : BaseSIMDLdSt<1, 1, opcode, 0b10, asm,
7494 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
7495 (ins am_simdnoindex:$vaddr), []>;
7496 def v2d : BaseSIMDLdSt<1, 1, opcode, 0b11, asm,
7497 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
7498 (ins am_simdnoindex:$vaddr), []>;
7499 def v8b : BaseSIMDLdSt<0, 1, opcode, 0b00, asm,
7500 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
7501 (ins am_simdnoindex:$vaddr), []>;
7502 def v4h : BaseSIMDLdSt<0, 1, opcode, 0b01, asm,
7503 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
7504 (ins am_simdnoindex:$vaddr), []>;
7505 def v2s : BaseSIMDLdSt<0, 1, opcode, 0b10, asm,
7506 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
7507 (ins am_simdnoindex:$vaddr), []>;
7510 def v16b_POST: BaseSIMDLdStPost<1, 1, opcode, 0b00, asm,
7511 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
7512 (ins am_simdnoindex:$vaddr,
7513 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7514 def v8h_POST : BaseSIMDLdStPost<1, 1, opcode, 0b01, asm,
7515 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
7516 (ins am_simdnoindex:$vaddr,
7517 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7518 def v4s_POST : BaseSIMDLdStPost<1, 1, opcode, 0b10, asm,
7519 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
7520 (ins am_simdnoindex:$vaddr,
7521 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7522 def v2d_POST : BaseSIMDLdStPost<1, 1, opcode, 0b11, asm,
7523 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
7524 (ins am_simdnoindex:$vaddr,
7525 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7526 def v8b_POST : BaseSIMDLdStPost<0, 1, opcode, 0b00, asm,
7527 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
7528 (ins am_simdnoindex:$vaddr,
7529 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7530 def v4h_POST : BaseSIMDLdStPost<0, 1, opcode, 0b01, asm,
7531 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
7532 (ins am_simdnoindex:$vaddr,
7533 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7534 def v2s_POST : BaseSIMDLdStPost<0, 1, opcode, 0b10, asm,
7535 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
7536 (ins am_simdnoindex:$vaddr,
7537 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7540 defm : SIMDLdStAliases<asm, "16b", Count, Offset128, 128>;
7541 defm : SIMDLdStAliases<asm, "8h", Count, Offset128, 128>;
7542 defm : SIMDLdStAliases<asm, "4s", Count, Offset128, 128>;
7543 defm : SIMDLdStAliases<asm, "2d", Count, Offset128, 128>;
7544 defm : SIMDLdStAliases<asm, "8b", Count, Offset64, 64>;
7545 defm : SIMDLdStAliases<asm, "4h", Count, Offset64, 64>;
7546 defm : SIMDLdStAliases<asm, "2s", Count, Offset64, 64>;
7549 // Only ld1/st1 has a v1d version.
7550 multiclass BaseSIMDStN<string Count, string asm, string veclist, int Offset128,
7551 int Offset64, bits<4> opcode> {
7552 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in {
7553 def v16b : BaseSIMDLdSt<1, 0, opcode, 0b00, asm, (outs),
7554 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
7555 am_simdnoindex:$vaddr), []>;
7556 def v8h : BaseSIMDLdSt<1, 0, opcode, 0b01, asm, (outs),
7557 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
7558 am_simdnoindex:$vaddr), []>;
7559 def v4s : BaseSIMDLdSt<1, 0, opcode, 0b10, asm, (outs),
7560 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
7561 am_simdnoindex:$vaddr), []>;
7562 def v2d : BaseSIMDLdSt<1, 0, opcode, 0b11, asm, (outs),
7563 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
7564 am_simdnoindex:$vaddr), []>;
7565 def v8b : BaseSIMDLdSt<0, 0, opcode, 0b00, asm, (outs),
7566 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
7567 am_simdnoindex:$vaddr), []>;
7568 def v4h : BaseSIMDLdSt<0, 0, opcode, 0b01, asm, (outs),
7569 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
7570 am_simdnoindex:$vaddr), []>;
7571 def v2s : BaseSIMDLdSt<0, 0, opcode, 0b10, asm, (outs),
7572 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
7573 am_simdnoindex:$vaddr), []>;
7575 def v16b_POST : BaseSIMDLdStPost<1, 0, opcode, 0b00, asm, (outs),
7576 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
7577 am_simdnoindex:$vaddr,
7578 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7579 def v8h_POST : BaseSIMDLdStPost<1, 0, opcode, 0b01, asm, (outs),
7580 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
7581 am_simdnoindex:$vaddr,
7582 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7583 def v4s_POST : BaseSIMDLdStPost<1, 0, opcode, 0b10, asm, (outs),
7584 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
7585 am_simdnoindex:$vaddr,
7586 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7587 def v2d_POST : BaseSIMDLdStPost<1, 0, opcode, 0b11, asm, (outs),
7588 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
7589 am_simdnoindex:$vaddr,
7590 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7591 def v8b_POST : BaseSIMDLdStPost<0, 0, opcode, 0b00, asm, (outs),
7592 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
7593 am_simdnoindex:$vaddr,
7594 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7595 def v4h_POST : BaseSIMDLdStPost<0, 0, opcode, 0b01, asm, (outs),
7596 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
7597 am_simdnoindex:$vaddr,
7598 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7599 def v2s_POST : BaseSIMDLdStPost<0, 0, opcode, 0b10, asm, (outs),
7600 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
7601 am_simdnoindex:$vaddr,
7602 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7605 defm : SIMDLdStAliases<asm, "16b", Count, Offset128, 128>;
7606 defm : SIMDLdStAliases<asm, "8h", Count, Offset128, 128>;
7607 defm : SIMDLdStAliases<asm, "4s", Count, Offset128, 128>;
7608 defm : SIMDLdStAliases<asm, "2d", Count, Offset128, 128>;
7609 defm : SIMDLdStAliases<asm, "8b", Count, Offset64, 64>;
7610 defm : SIMDLdStAliases<asm, "4h", Count, Offset64, 64>;
7611 defm : SIMDLdStAliases<asm, "2s", Count, Offset64, 64>;
7614 multiclass BaseSIMDLd1<string Count, string asm, string veclist,
7615 int Offset128, int Offset64, bits<4> opcode>
7616 : BaseSIMDLdN<Count, asm, veclist, Offset128, Offset64, opcode> {
7618 // LD1 instructions have extra "1d" variants.
7619 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
7620 def v1d : BaseSIMDLdSt<0, 1, opcode, 0b11, asm,
7621 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
7622 (ins am_simdnoindex:$vaddr), []>;
7624 def v1d_POST : BaseSIMDLdStPost<0, 1, opcode, 0b11, asm,
7625 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
7626 (ins am_simdnoindex:$vaddr,
7627 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7630 defm : SIMDLdStAliases<asm, "1d", Count, Offset64, 64>;
7633 multiclass BaseSIMDSt1<string Count, string asm, string veclist,
7634 int Offset128, int Offset64, bits<4> opcode>
7635 : BaseSIMDStN<Count, asm, veclist, Offset128, Offset64, opcode> {
7637 // ST1 instructions have extra "1d" variants.
7638 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in {
7639 def v1d : BaseSIMDLdSt<0, 0, opcode, 0b11, asm, (outs),
7640 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
7641 am_simdnoindex:$vaddr), []>;
7643 def v1d_POST : BaseSIMDLdStPost<0, 0, opcode, 0b11, asm, (outs),
7644 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
7645 am_simdnoindex:$vaddr,
7646 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7649 defm : SIMDLdStAliases<asm, "1d", Count, Offset64, 64>;
7652 multiclass SIMDLd1Multiple<string asm> {
7653 defm One : BaseSIMDLd1<"One", asm, "VecListOne", 16, 8, 0b0111>;
7654 defm Two : BaseSIMDLd1<"Two", asm, "VecListTwo", 32, 16, 0b1010>;
7655 defm Three : BaseSIMDLd1<"Three", asm, "VecListThree", 48, 24, 0b0110>;
7656 defm Four : BaseSIMDLd1<"Four", asm, "VecListFour", 64, 32, 0b0010>;
7659 multiclass SIMDSt1Multiple<string asm> {
7660 defm One : BaseSIMDSt1<"One", asm, "VecListOne", 16, 8, 0b0111>;
7661 defm Two : BaseSIMDSt1<"Two", asm, "VecListTwo", 32, 16, 0b1010>;
7662 defm Three : BaseSIMDSt1<"Three", asm, "VecListThree", 48, 24, 0b0110>;
7663 defm Four : BaseSIMDSt1<"Four", asm, "VecListFour", 64, 32, 0b0010>;
7666 multiclass SIMDLd2Multiple<string asm> {
7667 defm Two : BaseSIMDLdN<"Two", asm, "VecListTwo", 32, 16, 0b1000>;
7670 multiclass SIMDSt2Multiple<string asm> {
7671 defm Two : BaseSIMDStN<"Two", asm, "VecListTwo", 32, 16, 0b1000>;
7674 multiclass SIMDLd3Multiple<string asm> {
7675 defm Three : BaseSIMDLdN<"Three", asm, "VecListThree", 48, 24, 0b0100>;
7678 multiclass SIMDSt3Multiple<string asm> {
7679 defm Three : BaseSIMDStN<"Three", asm, "VecListThree", 48, 24, 0b0100>;
7682 multiclass SIMDLd4Multiple<string asm> {
7683 defm Four : BaseSIMDLdN<"Four", asm, "VecListFour", 64, 32, 0b0000>;
7686 multiclass SIMDSt4Multiple<string asm> {
7687 defm Four : BaseSIMDStN<"Four", asm, "VecListFour", 64, 32, 0b0000>;
7691 // AdvSIMD Load/store single-element
7694 class BaseSIMDLdStSingle<bit L, bit R, bits<3> opcode,
7695 string asm, string operands, dag oops, dag iops,
7697 : I<oops, iops, asm, operands, "", pattern> {
7701 let Inst{29-24} = 0b001101;
7704 let Inst{15-13} = opcode;
7705 let Inst{9-5} = vaddr;
7707 let DecoderMethod = "DecodeSIMDLdStSingle";
7710 class BaseSIMDLdStSingleTied<bit L, bit R, bits<3> opcode,
7711 string asm, string operands, dag oops, dag iops,
7713 : I<oops, iops, asm, operands, "$Vt = $dst", pattern> {
7717 let Inst{29-24} = 0b001101;
7720 let Inst{15-13} = opcode;
7721 let Inst{9-5} = vaddr;
7723 let DecoderMethod = "DecodeSIMDLdStSingleTied";
7727 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7728 class BaseSIMDLdR<bit Q, bit R, bits<3> opcode, bit S, bits<2> size, string asm,
7730 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, $vaddr",
7731 (outs listtype:$Vt), (ins am_simdnoindex:$vaddr), []> {
7734 let Inst{20-16} = 0b00000;
7736 let Inst{11-10} = size;
7738 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7739 class BaseSIMDLdRPost<bit Q, bit R, bits<3> opcode, bit S, bits<2> size,
7740 string asm, Operand listtype, Operand GPR64pi>
7741 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, $vaddr, $Xm",
7742 (outs listtype:$Vt),
7743 (ins am_simdnoindex:$vaddr, GPR64pi:$Xm), []> {
7747 let Inst{20-16} = Xm;
7749 let Inst{11-10} = size;
7752 multiclass SIMDLdrAliases<string asm, string layout, string Count,
7753 int Offset, int Size> {
7754 // E.g. "ld1r { v0.8b }, [x1], #1"
7755 // "ld1r.8b\t$Vt, $vaddr, #1"
7756 // may get mapped to
7757 // (LD1Rv8b_POST VecListOne8b:$Vt, am_simdnoindex:$vaddr, XZR)
7758 def : InstAlias<asm # "\t$Vt, $vaddr, #" # Offset,
7759 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7760 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
7761 am_simdnoindex:$vaddr, XZR), 1>;
7763 // E.g. "ld1r.8b { v0 }, [x1], #1"
7764 // "ld1r.8b\t$Vt, $vaddr, #1"
7765 // may get mapped to
7766 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, XZR)
7767 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, #" # Offset,
7768 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7769 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7770 am_simdnoindex:$vaddr, XZR), 0>;
7772 // E.g. "ld1r.8b { v0 }, [x1]"
7773 // "ld1r.8b\t$Vt, $vaddr"
7774 // may get mapped to
7775 // (LD1Rv8b VecListOne64:$Vt, am_simdnoindex:$vaddr)
7776 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr",
7777 (!cast<Instruction>(NAME # "v" # layout)
7778 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7779 am_simdnoindex:$vaddr), 0>;
7781 // E.g. "ld1r.8b { v0 }, [x1], x2"
7782 // "ld1r.8b\t$Vt, $vaddr, $Xm"
7783 // may get mapped to
7784 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, GPR64pi1:$Xm)
7785 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, $Xm",
7786 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7787 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7788 am_simdnoindex:$vaddr,
7789 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
7792 multiclass SIMDLdR<bit R, bits<3> opcode, bit S, string asm, string Count,
7793 int Offset1, int Offset2, int Offset4, int Offset8> {
7794 def v8b : BaseSIMDLdR<0, R, opcode, S, 0b00, asm,
7795 !cast<Operand>("VecList" # Count # "8b")>;
7796 def v16b: BaseSIMDLdR<1, R, opcode, S, 0b00, asm,
7797 !cast<Operand>("VecList" # Count #"16b")>;
7798 def v4h : BaseSIMDLdR<0, R, opcode, S, 0b01, asm,
7799 !cast<Operand>("VecList" # Count #"4h")>;
7800 def v8h : BaseSIMDLdR<1, R, opcode, S, 0b01, asm,
7801 !cast<Operand>("VecList" # Count #"8h")>;
7802 def v2s : BaseSIMDLdR<0, R, opcode, S, 0b10, asm,
7803 !cast<Operand>("VecList" # Count #"2s")>;
7804 def v4s : BaseSIMDLdR<1, R, opcode, S, 0b10, asm,
7805 !cast<Operand>("VecList" # Count #"4s")>;
7806 def v1d : BaseSIMDLdR<0, R, opcode, S, 0b11, asm,
7807 !cast<Operand>("VecList" # Count #"1d")>;
7808 def v2d : BaseSIMDLdR<1, R, opcode, S, 0b11, asm,
7809 !cast<Operand>("VecList" # Count #"2d")>;
7811 def v8b_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b00, asm,
7812 !cast<Operand>("VecList" # Count # "8b"),
7813 !cast<Operand>("GPR64pi" # Offset1)>;
7814 def v16b_POST: BaseSIMDLdRPost<1, R, opcode, S, 0b00, asm,
7815 !cast<Operand>("VecList" # Count # "16b"),
7816 !cast<Operand>("GPR64pi" # Offset1)>;
7817 def v4h_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b01, asm,
7818 !cast<Operand>("VecList" # Count # "4h"),
7819 !cast<Operand>("GPR64pi" # Offset2)>;
7820 def v8h_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b01, asm,
7821 !cast<Operand>("VecList" # Count # "8h"),
7822 !cast<Operand>("GPR64pi" # Offset2)>;
7823 def v2s_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b10, asm,
7824 !cast<Operand>("VecList" # Count # "2s"),
7825 !cast<Operand>("GPR64pi" # Offset4)>;
7826 def v4s_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b10, asm,
7827 !cast<Operand>("VecList" # Count # "4s"),
7828 !cast<Operand>("GPR64pi" # Offset4)>;
7829 def v1d_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b11, asm,
7830 !cast<Operand>("VecList" # Count # "1d"),
7831 !cast<Operand>("GPR64pi" # Offset8)>;
7832 def v2d_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b11, asm,
7833 !cast<Operand>("VecList" # Count # "2d"),
7834 !cast<Operand>("GPR64pi" # Offset8)>;
7836 defm : SIMDLdrAliases<asm, "8b", Count, Offset1, 64>;
7837 defm : SIMDLdrAliases<asm, "16b", Count, Offset1, 128>;
7838 defm : SIMDLdrAliases<asm, "4h", Count, Offset2, 64>;
7839 defm : SIMDLdrAliases<asm, "8h", Count, Offset2, 128>;
7840 defm : SIMDLdrAliases<asm, "2s", Count, Offset4, 64>;
7841 defm : SIMDLdrAliases<asm, "4s", Count, Offset4, 128>;
7842 defm : SIMDLdrAliases<asm, "1d", Count, Offset8, 64>;
7843 defm : SIMDLdrAliases<asm, "2d", Count, Offset8, 128>;
7846 class SIMDLdStSingleB<bit L, bit R, bits<3> opcode, string asm,
7847 dag oops, dag iops, list<dag> pattern>
7848 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7850 // idx encoded in Q:S:size fields.
7852 let Inst{30} = idx{3};
7854 let Inst{20-16} = 0b00000;
7855 let Inst{12} = idx{2};
7856 let Inst{11-10} = idx{1-0};
7858 class SIMDLdStSingleBTied<bit L, bit R, bits<3> opcode, string asm,
7859 dag oops, dag iops, list<dag> pattern>
7860 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7862 // idx encoded in Q:S:size fields.
7864 let Inst{30} = idx{3};
7866 let Inst{20-16} = 0b00000;
7867 let Inst{12} = idx{2};
7868 let Inst{11-10} = idx{1-0};
7870 class SIMDLdStSingleBPost<bit L, bit R, bits<3> opcode, string asm,
7872 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7874 // idx encoded in Q:S:size fields.
7877 let Inst{30} = idx{3};
7879 let Inst{20-16} = Xm;
7880 let Inst{12} = idx{2};
7881 let Inst{11-10} = idx{1-0};
7883 class SIMDLdStSingleBTiedPost<bit L, bit R, bits<3> opcode, string asm,
7885 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7887 // idx encoded in Q:S:size fields.
7890 let Inst{30} = idx{3};
7892 let Inst{20-16} = Xm;
7893 let Inst{12} = idx{2};
7894 let Inst{11-10} = idx{1-0};
7897 class SIMDLdStSingleH<bit L, bit R, bits<3> opcode, bit size, string asm,
7898 dag oops, dag iops, list<dag> pattern>
7899 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7901 // idx encoded in Q:S:size<1> fields.
7903 let Inst{30} = idx{2};
7905 let Inst{20-16} = 0b00000;
7906 let Inst{12} = idx{1};
7907 let Inst{11} = idx{0};
7908 let Inst{10} = size;
7910 class SIMDLdStSingleHTied<bit L, bit R, bits<3> opcode, bit size, string asm,
7911 dag oops, dag iops, list<dag> pattern>
7912 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7914 // idx encoded in Q:S:size<1> fields.
7916 let Inst{30} = idx{2};
7918 let Inst{20-16} = 0b00000;
7919 let Inst{12} = idx{1};
7920 let Inst{11} = idx{0};
7921 let Inst{10} = size;
7924 class SIMDLdStSingleHPost<bit L, bit R, bits<3> opcode, bit size, string asm,
7926 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7928 // idx encoded in Q:S:size<1> fields.
7931 let Inst{30} = idx{2};
7933 let Inst{20-16} = Xm;
7934 let Inst{12} = idx{1};
7935 let Inst{11} = idx{0};
7936 let Inst{10} = size;
7938 class SIMDLdStSingleHTiedPost<bit L, bit R, bits<3> opcode, bit size, string asm,
7940 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7942 // idx encoded in Q:S:size<1> fields.
7945 let Inst{30} = idx{2};
7947 let Inst{20-16} = Xm;
7948 let Inst{12} = idx{1};
7949 let Inst{11} = idx{0};
7950 let Inst{10} = size;
7952 class SIMDLdStSingleS<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7953 dag oops, dag iops, list<dag> pattern>
7954 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7956 // idx encoded in Q:S fields.
7958 let Inst{30} = idx{1};
7960 let Inst{20-16} = 0b00000;
7961 let Inst{12} = idx{0};
7962 let Inst{11-10} = size;
7964 class SIMDLdStSingleSTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7965 dag oops, dag iops, list<dag> pattern>
7966 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7968 // idx encoded in Q:S fields.
7970 let Inst{30} = idx{1};
7972 let Inst{20-16} = 0b00000;
7973 let Inst{12} = idx{0};
7974 let Inst{11-10} = size;
7976 class SIMDLdStSingleSPost<bit L, bit R, bits<3> opcode, bits<2> size,
7977 string asm, dag oops, dag iops>
7978 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7980 // idx encoded in Q:S fields.
7983 let Inst{30} = idx{1};
7985 let Inst{20-16} = Xm;
7986 let Inst{12} = idx{0};
7987 let Inst{11-10} = size;
7989 class SIMDLdStSingleSTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
7990 string asm, dag oops, dag iops>
7991 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7993 // idx encoded in Q:S fields.
7996 let Inst{30} = idx{1};
7998 let Inst{20-16} = Xm;
7999 let Inst{12} = idx{0};
8000 let Inst{11-10} = size;
8002 class SIMDLdStSingleD<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
8003 dag oops, dag iops, list<dag> pattern>
8004 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
8006 // idx encoded in Q field.
8010 let Inst{20-16} = 0b00000;
8012 let Inst{11-10} = size;
8014 class SIMDLdStSingleDTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
8015 dag oops, dag iops, list<dag> pattern>
8016 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
8018 // idx encoded in Q field.
8022 let Inst{20-16} = 0b00000;
8024 let Inst{11-10} = size;
8026 class SIMDLdStSingleDPost<bit L, bit R, bits<3> opcode, bits<2> size,
8027 string asm, dag oops, dag iops>
8028 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
8030 // idx encoded in Q field.
8035 let Inst{20-16} = Xm;
8037 let Inst{11-10} = size;
8039 class SIMDLdStSingleDTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
8040 string asm, dag oops, dag iops>
8041 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
8043 // idx encoded in Q field.
8048 let Inst{20-16} = Xm;
8050 let Inst{11-10} = size;
8053 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8054 multiclass SIMDLdSingleBTied<bit R, bits<3> opcode, string asm,
8055 RegisterOperand listtype,
8056 RegisterOperand GPR64pi> {
8057 def i8 : SIMDLdStSingleBTied<1, R, opcode, asm,
8058 (outs listtype:$dst),
8059 (ins listtype:$Vt, VectorIndexB:$idx,
8060 am_simdnoindex:$vaddr), []>;
8062 def i8_POST : SIMDLdStSingleBTiedPost<1, R, opcode, asm,
8063 (outs listtype:$dst),
8064 (ins listtype:$Vt, VectorIndexB:$idx,
8065 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8067 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8068 multiclass SIMDLdSingleHTied<bit R, bits<3> opcode, bit size, string asm,
8069 RegisterOperand listtype,
8070 RegisterOperand GPR64pi> {
8071 def i16 : SIMDLdStSingleHTied<1, R, opcode, size, asm,
8072 (outs listtype:$dst),
8073 (ins listtype:$Vt, VectorIndexH:$idx,
8074 am_simdnoindex:$vaddr), []>;
8076 def i16_POST : SIMDLdStSingleHTiedPost<1, R, opcode, size, asm,
8077 (outs listtype:$dst),
8078 (ins listtype:$Vt, VectorIndexH:$idx,
8079 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8081 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8082 multiclass SIMDLdSingleSTied<bit R, bits<3> opcode, bits<2> size,string asm,
8083 RegisterOperand listtype,
8084 RegisterOperand GPR64pi> {
8085 def i32 : SIMDLdStSingleSTied<1, R, opcode, size, asm,
8086 (outs listtype:$dst),
8087 (ins listtype:$Vt, VectorIndexS:$idx,
8088 am_simdnoindex:$vaddr), []>;
8090 def i32_POST : SIMDLdStSingleSTiedPost<1, R, opcode, size, asm,
8091 (outs listtype:$dst),
8092 (ins listtype:$Vt, VectorIndexS:$idx,
8093 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8095 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8096 multiclass SIMDLdSingleDTied<bit R, bits<3> opcode, bits<2> size, string asm,
8097 RegisterOperand listtype, RegisterOperand GPR64pi> {
8098 def i64 : SIMDLdStSingleDTied<1, R, opcode, size, asm,
8099 (outs listtype:$dst),
8100 (ins listtype:$Vt, VectorIndexD:$idx,
8101 am_simdnoindex:$vaddr), []>;
8103 def i64_POST : SIMDLdStSingleDTiedPost<1, R, opcode, size, asm,
8104 (outs listtype:$dst),
8105 (ins listtype:$Vt, VectorIndexD:$idx,
8106 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8108 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8109 multiclass SIMDStSingleB<bit R, bits<3> opcode, string asm,
8110 RegisterOperand listtype, RegisterOperand GPR64pi> {
8111 def i8 : SIMDLdStSingleB<0, R, opcode, asm,
8112 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
8113 am_simdnoindex:$vaddr), []>;
8115 def i8_POST : SIMDLdStSingleBPost<0, R, opcode, asm,
8116 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
8117 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8119 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8120 multiclass SIMDStSingleH<bit R, bits<3> opcode, bit size, string asm,
8121 RegisterOperand listtype, RegisterOperand GPR64pi> {
8122 def i16 : SIMDLdStSingleH<0, R, opcode, size, asm,
8123 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
8124 am_simdnoindex:$vaddr), []>;
8126 def i16_POST : SIMDLdStSingleHPost<0, R, opcode, size, asm,
8127 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
8128 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8130 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8131 multiclass SIMDStSingleS<bit R, bits<3> opcode, bits<2> size,string asm,
8132 RegisterOperand listtype, RegisterOperand GPR64pi> {
8133 def i32 : SIMDLdStSingleS<0, R, opcode, size, asm,
8134 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
8135 am_simdnoindex:$vaddr), []>;
8137 def i32_POST : SIMDLdStSingleSPost<0, R, opcode, size, asm,
8138 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
8139 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8141 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8142 multiclass SIMDStSingleD<bit R, bits<3> opcode, bits<2> size, string asm,
8143 RegisterOperand listtype, RegisterOperand GPR64pi> {
8144 def i64 : SIMDLdStSingleD<0, R, opcode, size, asm,
8145 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
8146 am_simdnoindex:$vaddr), []>;
8148 def i64_POST : SIMDLdStSingleDPost<0, R, opcode, size, asm,
8149 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
8150 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8153 multiclass SIMDLdStSingleAliases<string asm, string layout, string Type,
8154 string Count, int Offset, Operand idxtype> {
8155 // E.g. "ld1 { v0.8b }[0], [x1], #1"
8156 // "ld1\t$Vt, $vaddr, #1"
8157 // may get mapped to
8158 // (LD1Rv8b_POST VecListOne8b:$Vt, am_simdnoindex:$vaddr, XZR)
8159 def : InstAlias<asm # "\t$Vt$idx, $vaddr, #" # Offset,
8160 (!cast<Instruction>(NAME # Type # "_POST")
8161 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
8162 idxtype:$idx, am_simdnoindex:$vaddr, XZR), 1>;
8164 // E.g. "ld1.8b { v0 }[0], [x1], #1"
8165 // "ld1.8b\t$Vt, $vaddr, #1"
8166 // may get mapped to
8167 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, XZR)
8168 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr, #" # Offset,
8169 (!cast<Instruction>(NAME # Type # "_POST")
8170 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8171 idxtype:$idx, am_simdnoindex:$vaddr, XZR), 0>;
8173 // E.g. "ld1.8b { v0 }[0], [x1]"
8174 // "ld1.8b\t$Vt, $vaddr"
8175 // may get mapped to
8176 // (LD1Rv8b VecListOne64:$Vt, am_simdnoindex:$vaddr)
8177 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr",
8178 (!cast<Instruction>(NAME # Type)
8179 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8180 idxtype:$idx, am_simdnoindex:$vaddr), 0>;
8182 // E.g. "ld1.8b { v0 }[0], [x1], x2"
8183 // "ld1.8b\t$Vt, $vaddr, $Xm"
8184 // may get mapped to
8185 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, GPR64pi1:$Xm)
8186 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr, $Xm",
8187 (!cast<Instruction>(NAME # Type # "_POST")
8188 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8189 idxtype:$idx, am_simdnoindex:$vaddr,
8190 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
8193 multiclass SIMDLdSt1SingleAliases<string asm> {
8194 defm : SIMDLdStSingleAliases<asm, "b", "i8", "One", 1, VectorIndexB>;
8195 defm : SIMDLdStSingleAliases<asm, "h", "i16", "One", 2, VectorIndexH>;
8196 defm : SIMDLdStSingleAliases<asm, "s", "i32", "One", 4, VectorIndexS>;
8197 defm : SIMDLdStSingleAliases<asm, "d", "i64", "One", 8, VectorIndexD>;
8200 multiclass SIMDLdSt2SingleAliases<string asm> {
8201 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Two", 2, VectorIndexB>;
8202 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Two", 4, VectorIndexH>;
8203 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Two", 8, VectorIndexS>;
8204 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Two", 16, VectorIndexD>;
8207 multiclass SIMDLdSt3SingleAliases<string asm> {
8208 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Three", 3, VectorIndexB>;
8209 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Three", 6, VectorIndexH>;
8210 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Three", 12, VectorIndexS>;
8211 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Three", 24, VectorIndexD>;
8214 multiclass SIMDLdSt4SingleAliases<string asm> {
8215 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Four", 4, VectorIndexB>;
8216 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Four", 8, VectorIndexH>;
8217 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Four", 16, VectorIndexS>;
8218 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Four", 32, VectorIndexD>;
8220 } // end of 'let Predicates = [HasNEON]'
8222 //----------------------------------------------------------------------------
8223 // Crypto extensions
8224 //----------------------------------------------------------------------------
8226 let Predicates = [HasCrypto] in {
8227 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8228 class AESBase<bits<4> opc, string asm, dag outs, dag ins, string cstr,
8230 : I<outs, ins, asm, "{\t$Rd.16b, $Rn.16b|.16b\t$Rd, $Rn}", cstr, pat>,
8234 let Inst{31-16} = 0b0100111000101000;
8235 let Inst{15-12} = opc;
8236 let Inst{11-10} = 0b10;
8241 class AESInst<bits<4> opc, string asm, Intrinsic OpNode>
8242 : AESBase<opc, asm, (outs V128:$Rd), (ins V128:$Rn), "",
8243 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
8245 class AESTiedInst<bits<4> opc, string asm, Intrinsic OpNode>
8246 : AESBase<opc, asm, (outs V128:$dst), (ins V128:$Rd, V128:$Rn),
8248 [(set (v16i8 V128:$dst),
8249 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
8251 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8252 class SHA3OpTiedInst<bits<3> opc, string asm, string dst_lhs_kind,
8253 dag oops, dag iops, list<dag> pat>
8254 : I<oops, iops, asm,
8255 "{\t$Rd" # dst_lhs_kind # ", $Rn" # dst_lhs_kind # ", $Rm.4s" #
8256 "|.4s\t$Rd, $Rn, $Rm}", "$Rd = $dst", pat>,
8261 let Inst{31-21} = 0b01011110000;
8262 let Inst{20-16} = Rm;
8264 let Inst{14-12} = opc;
8265 let Inst{11-10} = 0b00;
8270 class SHATiedInstQSV<bits<3> opc, string asm, Intrinsic OpNode>
8271 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
8272 (ins FPR128:$Rd, FPR32:$Rn, V128:$Rm),
8273 [(set (v4i32 FPR128:$dst),
8274 (OpNode (v4i32 FPR128:$Rd), (i32 FPR32:$Rn),
8275 (v4i32 V128:$Rm)))]>;
8277 class SHATiedInstVVV<bits<3> opc, string asm, Intrinsic OpNode>
8278 : SHA3OpTiedInst<opc, asm, ".4s", (outs V128:$dst),
8279 (ins V128:$Rd, V128:$Rn, V128:$Rm),
8280 [(set (v4i32 V128:$dst),
8281 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8282 (v4i32 V128:$Rm)))]>;
8284 class SHATiedInstQQV<bits<3> opc, string asm, Intrinsic OpNode>
8285 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
8286 (ins FPR128:$Rd, FPR128:$Rn, V128:$Rm),
8287 [(set (v4i32 FPR128:$dst),
8288 (OpNode (v4i32 FPR128:$Rd), (v4i32 FPR128:$Rn),
8289 (v4i32 V128:$Rm)))]>;
8291 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8292 class SHA2OpInst<bits<4> opc, string asm, string kind,
8293 string cstr, dag oops, dag iops,
8295 : I<oops, iops, asm, "{\t$Rd" # kind # ", $Rn" # kind #
8296 "|" # kind # "\t$Rd, $Rn}", cstr, pat>,
8300 let Inst{31-16} = 0b0101111000101000;
8301 let Inst{15-12} = opc;
8302 let Inst{11-10} = 0b10;
8307 class SHATiedInstVV<bits<4> opc, string asm, Intrinsic OpNode>
8308 : SHA2OpInst<opc, asm, ".4s", "$Rd = $dst", (outs V128:$dst),
8309 (ins V128:$Rd, V128:$Rn),
8310 [(set (v4i32 V128:$dst),
8311 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
8313 class SHAInstSS<bits<4> opc, string asm, Intrinsic OpNode>
8314 : SHA2OpInst<opc, asm, "", "", (outs FPR32:$Rd), (ins FPR32:$Rn),
8315 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
8316 } // end of 'let Predicates = [HasCrypto]'
8318 // Allow the size specifier tokens to be upper case, not just lower.
8319 def : TokenAlias<".8B", ".8b">;
8320 def : TokenAlias<".4H", ".4h">;
8321 def : TokenAlias<".2S", ".2s">;
8322 def : TokenAlias<".1D", ".1d">;
8323 def : TokenAlias<".16B", ".16b">;
8324 def : TokenAlias<".8H", ".8h">;
8325 def : TokenAlias<".4S", ".4s">;
8326 def : TokenAlias<".2D", ".2d">;
8327 def : TokenAlias<".1Q", ".1q">;
8328 def : TokenAlias<".B", ".b">;
8329 def : TokenAlias<".H", ".h">;
8330 def : TokenAlias<".S", ".s">;
8331 def : TokenAlias<".D", ".d">;
8332 def : TokenAlias<".Q", ".q">;