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 MovKSymbolG2AsmOperand : AsmOperandClass {
281 let Name = "MovKSymbolG2";
282 let RenderMethod = "addImmOperands";
285 def movk_symbol_g2 : Operand<i32> {
286 let ParserMatchClass = MovKSymbolG2AsmOperand;
289 def MovKSymbolG1AsmOperand : AsmOperandClass {
290 let Name = "MovKSymbolG1";
291 let RenderMethod = "addImmOperands";
294 def movk_symbol_g1 : Operand<i32> {
295 let ParserMatchClass = MovKSymbolG1AsmOperand;
298 def MovKSymbolG0AsmOperand : AsmOperandClass {
299 let Name = "MovKSymbolG0";
300 let RenderMethod = "addImmOperands";
303 def movk_symbol_g0 : Operand<i32> {
304 let ParserMatchClass = MovKSymbolG0AsmOperand;
307 def fixedpoint32 : Operand<i32> {
308 let EncoderMethod = "getFixedPointScaleOpValue";
309 let DecoderMethod = "DecodeFixedPointScaleImm32";
310 let ParserMatchClass = Imm1_32Operand;
312 def fixedpoint64 : Operand<i64> {
313 let EncoderMethod = "getFixedPointScaleOpValue";
314 let DecoderMethod = "DecodeFixedPointScaleImm64";
315 let ParserMatchClass = Imm1_64Operand;
318 def vecshiftR8 : Operand<i32>, ImmLeaf<i32, [{
319 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
321 let EncoderMethod = "getVecShiftR8OpValue";
322 let DecoderMethod = "DecodeVecShiftR8Imm";
323 let ParserMatchClass = Imm1_8Operand;
325 def vecshiftR16 : Operand<i32>, ImmLeaf<i32, [{
326 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
328 let EncoderMethod = "getVecShiftR16OpValue";
329 let DecoderMethod = "DecodeVecShiftR16Imm";
330 let ParserMatchClass = Imm1_16Operand;
332 def vecshiftR16Narrow : Operand<i32>, ImmLeaf<i32, [{
333 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 9);
335 let EncoderMethod = "getVecShiftR16OpValue";
336 let DecoderMethod = "DecodeVecShiftR16ImmNarrow";
337 let ParserMatchClass = Imm1_8Operand;
339 def vecshiftR32 : Operand<i32>, ImmLeaf<i32, [{
340 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
342 let EncoderMethod = "getVecShiftR32OpValue";
343 let DecoderMethod = "DecodeVecShiftR32Imm";
344 let ParserMatchClass = Imm1_32Operand;
346 def vecshiftR32Narrow : Operand<i32>, ImmLeaf<i32, [{
347 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 17);
349 let EncoderMethod = "getVecShiftR32OpValue";
350 let DecoderMethod = "DecodeVecShiftR32ImmNarrow";
351 let ParserMatchClass = Imm1_16Operand;
353 def vecshiftR64 : Operand<i32>, ImmLeaf<i32, [{
354 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 65);
356 let EncoderMethod = "getVecShiftR64OpValue";
357 let DecoderMethod = "DecodeVecShiftR64Imm";
358 let ParserMatchClass = Imm1_64Operand;
360 def vecshiftR64Narrow : Operand<i32>, ImmLeaf<i32, [{
361 return (((uint32_t)Imm) > 0) && (((uint32_t)Imm) < 33);
363 let EncoderMethod = "getVecShiftR64OpValue";
364 let DecoderMethod = "DecodeVecShiftR64ImmNarrow";
365 let ParserMatchClass = Imm1_32Operand;
368 def Imm0_7Operand : AsmOperandClass { let Name = "Imm0_7"; }
369 def Imm0_15Operand : AsmOperandClass { let Name = "Imm0_15"; }
370 def Imm0_31Operand : AsmOperandClass { let Name = "Imm0_31"; }
371 def Imm0_63Operand : AsmOperandClass { let Name = "Imm0_63"; }
373 def vecshiftL8 : Operand<i32>, ImmLeaf<i32, [{
374 return (((uint32_t)Imm) < 8);
376 let EncoderMethod = "getVecShiftL8OpValue";
377 let DecoderMethod = "DecodeVecShiftL8Imm";
378 let ParserMatchClass = Imm0_7Operand;
380 def vecshiftL16 : Operand<i32>, ImmLeaf<i32, [{
381 return (((uint32_t)Imm) < 16);
383 let EncoderMethod = "getVecShiftL16OpValue";
384 let DecoderMethod = "DecodeVecShiftL16Imm";
385 let ParserMatchClass = Imm0_15Operand;
387 def vecshiftL32 : Operand<i32>, ImmLeaf<i32, [{
388 return (((uint32_t)Imm) < 32);
390 let EncoderMethod = "getVecShiftL32OpValue";
391 let DecoderMethod = "DecodeVecShiftL32Imm";
392 let ParserMatchClass = Imm0_31Operand;
394 def vecshiftL64 : Operand<i32>, ImmLeaf<i32, [{
395 return (((uint32_t)Imm) < 64);
397 let EncoderMethod = "getVecShiftL64OpValue";
398 let DecoderMethod = "DecodeVecShiftL64Imm";
399 let ParserMatchClass = Imm0_63Operand;
403 // Crazy immediate formats used by 32-bit and 64-bit logical immediate
404 // instructions for splatting repeating bit patterns across the immediate.
405 def logical_imm32_XFORM : SDNodeXForm<imm, [{
406 uint64_t enc = ARM64_AM::encodeLogicalImmediate(N->getZExtValue(), 32);
407 return CurDAG->getTargetConstant(enc, MVT::i32);
409 def logical_imm64_XFORM : SDNodeXForm<imm, [{
410 uint64_t enc = ARM64_AM::encodeLogicalImmediate(N->getZExtValue(), 64);
411 return CurDAG->getTargetConstant(enc, MVT::i32);
414 def LogicalImm32Operand : AsmOperandClass { let Name = "LogicalImm32"; }
415 def LogicalImm64Operand : AsmOperandClass { let Name = "LogicalImm64"; }
416 def logical_imm32 : Operand<i32>, PatLeaf<(imm), [{
417 return ARM64_AM::isLogicalImmediate(N->getZExtValue(), 32);
418 }], logical_imm32_XFORM> {
419 let PrintMethod = "printLogicalImm32";
420 let ParserMatchClass = LogicalImm32Operand;
422 def logical_imm64 : Operand<i64>, PatLeaf<(imm), [{
423 return ARM64_AM::isLogicalImmediate(N->getZExtValue(), 64);
424 }], logical_imm64_XFORM> {
425 let PrintMethod = "printLogicalImm64";
426 let ParserMatchClass = LogicalImm64Operand;
429 // imm0_255 predicate - True if the immediate is in the range [0,255].
430 def Imm0_255Operand : AsmOperandClass { let Name = "Imm0_255"; }
431 def imm0_255 : Operand<i32>, ImmLeaf<i32, [{
432 return ((uint32_t)Imm) < 256;
434 let ParserMatchClass = Imm0_255Operand;
437 // imm0_127 predicate - True if the immediate is in the range [0,127]
438 def Imm0_127Operand : AsmOperandClass { let Name = "Imm0_127"; }
439 def imm0_127 : Operand<i32>, ImmLeaf<i32, [{
440 return ((uint32_t)Imm) < 128;
442 let ParserMatchClass = Imm0_127Operand;
445 // NOTE: These imm0_N operands have to be of type i64 because i64 is the size
446 // for all shift-amounts.
448 // imm0_63 predicate - True if the immediate is in the range [0,63]
449 def imm0_63 : Operand<i64>, ImmLeaf<i64, [{
450 return ((uint64_t)Imm) < 64;
452 let ParserMatchClass = Imm0_63Operand;
455 // imm0_31 predicate - True if the immediate is in the range [0,31]
456 def imm0_31 : Operand<i64>, ImmLeaf<i64, [{
457 return ((uint64_t)Imm) < 32;
459 let ParserMatchClass = Imm0_31Operand;
462 // imm0_15 predicate - True if the immediate is in the range [0,15]
463 def imm0_15 : Operand<i64>, ImmLeaf<i64, [{
464 return ((uint64_t)Imm) < 16;
466 let ParserMatchClass = Imm0_15Operand;
469 // imm0_7 predicate - True if the immediate is in the range [0,7]
470 def imm0_7 : Operand<i64>, ImmLeaf<i64, [{
471 return ((uint64_t)Imm) < 8;
473 let ParserMatchClass = Imm0_7Operand;
476 // An arithmetic shifter operand:
477 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr
479 def arith_shift : Operand<i32> {
480 let PrintMethod = "printShifter";
481 let ParserMatchClass = ArithmeticShifterOperand;
484 class arith_shifted_reg<ValueType Ty, RegisterClass regclass>
486 ComplexPattern<Ty, 2, "SelectArithShiftedRegister", []> {
487 let PrintMethod = "printShiftedRegister";
488 let MIOperandInfo = (ops regclass, arith_shift);
491 def arith_shifted_reg32 : arith_shifted_reg<i32, GPR32>;
492 def arith_shifted_reg64 : arith_shifted_reg<i64, GPR64>;
494 // An arithmetic shifter operand:
495 // {7-6} - shift type: 00 = lsl, 01 = lsr, 10 = asr, 11 = ror
497 def logical_shift : Operand<i32> {
498 let PrintMethod = "printShifter";
499 let ParserMatchClass = ShifterOperand;
502 class logical_shifted_reg<ValueType Ty, RegisterClass regclass>
504 ComplexPattern<Ty, 2, "SelectLogicalShiftedRegister", []> {
505 let PrintMethod = "printShiftedRegister";
506 let MIOperandInfo = (ops regclass, logical_shift);
509 def logical_shifted_reg32 : logical_shifted_reg<i32, GPR32>;
510 def logical_shifted_reg64 : logical_shifted_reg<i64, GPR64>;
512 // A logical vector shifter operand:
513 // {7-6} - shift type: 00 = lsl
514 // {5-0} - imm6: #0, #8, #16, or #24
515 def logical_vec_shift : Operand<i32> {
516 let PrintMethod = "printShifter";
517 let EncoderMethod = "getVecShifterOpValue";
518 let ParserMatchClass = LogicalVecShifterOperand;
521 // A logical vector half-word shifter operand:
522 // {7-6} - shift type: 00 = lsl
523 // {5-0} - imm6: #0 or #8
524 def logical_vec_hw_shift : Operand<i32> {
525 let PrintMethod = "printShifter";
526 let EncoderMethod = "getVecShifterOpValue";
527 let ParserMatchClass = LogicalVecHalfWordShifterOperand;
530 // A vector move shifter operand:
531 // {0} - imm1: #8 or #16
532 def move_vec_shift : Operand<i32> {
533 let PrintMethod = "printShifter";
534 let EncoderMethod = "getMoveVecShifterOpValue";
535 let ParserMatchClass = MoveVecShifterOperand;
538 // An ADD/SUB immediate shifter operand:
539 // {7-6} - shift type: 00 = lsl
540 // {5-0} - imm6: #0 or #12
541 def addsub_shift : Operand<i32> {
542 let ParserMatchClass = AddSubShifterOperand;
545 class addsub_shifted_imm<ValueType Ty>
546 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectArithImmed", [imm]> {
547 let PrintMethod = "printAddSubImm";
548 let EncoderMethod = "getAddSubImmOpValue";
549 let MIOperandInfo = (ops i32imm, addsub_shift);
552 def addsub_shifted_imm32 : addsub_shifted_imm<i32>;
553 def addsub_shifted_imm64 : addsub_shifted_imm<i64>;
555 class neg_addsub_shifted_imm<ValueType Ty>
556 : Operand<Ty>, ComplexPattern<Ty, 2, "SelectNegArithImmed", [imm]> {
557 let PrintMethod = "printAddSubImm";
558 let EncoderMethod = "getAddSubImmOpValue";
559 let MIOperandInfo = (ops i32imm, addsub_shift);
562 def neg_addsub_shifted_imm32 : neg_addsub_shifted_imm<i32>;
563 def neg_addsub_shifted_imm64 : neg_addsub_shifted_imm<i64>;
565 // An extend operand:
566 // {5-3} - extend type
568 def arith_extend : Operand<i32> {
569 let PrintMethod = "printExtend";
570 let ParserMatchClass = ExtendOperand;
572 def arith_extend64 : Operand<i32> {
573 let PrintMethod = "printExtend";
574 let ParserMatchClass = ExtendOperand64;
577 // 'extend' that's a lsl of a 64-bit register.
578 def arith_extendlsl64 : Operand<i32> {
579 let PrintMethod = "printExtend";
580 let ParserMatchClass = ExtendOperandLSL64;
583 class arith_extended_reg32<ValueType Ty> : Operand<Ty>,
584 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
585 let PrintMethod = "printExtendedRegister";
586 let MIOperandInfo = (ops GPR32, arith_extend);
589 class arith_extended_reg32to64<ValueType Ty> : Operand<Ty>,
590 ComplexPattern<Ty, 2, "SelectArithExtendedRegister", []> {
591 let PrintMethod = "printExtendedRegister";
592 let MIOperandInfo = (ops GPR32, arith_extend64);
595 // Floating-point immediate.
596 def fpimm32 : Operand<f32>,
597 PatLeaf<(f32 fpimm), [{
598 return ARM64_AM::getFP32Imm(N->getValueAPF()) != -1;
599 }], SDNodeXForm<fpimm, [{
600 APFloat InVal = N->getValueAPF();
601 uint32_t enc = ARM64_AM::getFP32Imm(InVal);
602 return CurDAG->getTargetConstant(enc, MVT::i32);
604 let ParserMatchClass = FPImmOperand;
605 let PrintMethod = "printFPImmOperand";
607 def fpimm64 : Operand<f64>,
608 PatLeaf<(f64 fpimm), [{
609 return ARM64_AM::getFP64Imm(N->getValueAPF()) != -1;
610 }], SDNodeXForm<fpimm, [{
611 APFloat InVal = N->getValueAPF();
612 uint32_t enc = ARM64_AM::getFP64Imm(InVal);
613 return CurDAG->getTargetConstant(enc, MVT::i32);
615 let ParserMatchClass = FPImmOperand;
616 let PrintMethod = "printFPImmOperand";
619 def fpimm8 : Operand<i32> {
620 let ParserMatchClass = FPImmOperand;
621 let PrintMethod = "printFPImmOperand";
624 def fpimm0 : PatLeaf<(fpimm), [{
625 return N->isExactlyValue(+0.0);
628 // 8-bit immediate for AdvSIMD where 64-bit values of the form:
629 // aaaaaaaa bbbbbbbb cccccccc dddddddd eeeeeeee ffffffff gggggggg hhhhhhhh
630 // are encoded as the eight bit value 'abcdefgh'.
631 def simdimmtype10 : Operand<i32>,
632 PatLeaf<(f64 fpimm), [{
633 return ARM64_AM::isAdvSIMDModImmType10(N->getValueAPF()
636 }], SDNodeXForm<fpimm, [{
637 APFloat InVal = N->getValueAPF();
638 uint32_t enc = ARM64_AM::encodeAdvSIMDModImmType10(N->getValueAPF()
641 return CurDAG->getTargetConstant(enc, MVT::i32);
643 let ParserMatchClass = SIMDImmType10Operand;
644 let PrintMethod = "printSIMDType10Operand";
652 // Base encoding for system instruction operands.
653 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
654 class BaseSystemI<bit L, dag oops, dag iops, string asm, string operands>
655 : I<oops, iops, asm, operands, "", []> {
656 let Inst{31-22} = 0b1101010100;
660 // System instructions which do not have an Rt register.
661 class SimpleSystemI<bit L, dag iops, string asm, string operands>
662 : BaseSystemI<L, (outs), iops, asm, operands> {
663 let Inst{4-0} = 0b11111;
666 // System instructions which have an Rt register.
667 class RtSystemI<bit L, dag oops, dag iops, string asm, string operands>
668 : BaseSystemI<L, oops, iops, asm, operands>,
674 // Hint instructions that take both a CRm and a 3-bit immediate.
675 class HintI<string mnemonic>
676 : SimpleSystemI<0, (ins imm0_127:$imm), mnemonic#" $imm", "">,
679 let Inst{20-12} = 0b000110010;
680 let Inst{11-5} = imm;
683 // System instructions taking a single literal operand which encodes into
684 // CRm. op2 differentiates the opcodes.
685 def BarrierAsmOperand : AsmOperandClass {
686 let Name = "Barrier";
687 let ParserMethod = "tryParseBarrierOperand";
689 def barrier_op : Operand<i32> {
690 let PrintMethod = "printBarrierOption";
691 let ParserMatchClass = BarrierAsmOperand;
693 class CRmSystemI<Operand crmtype, bits<3> opc, string asm>
694 : SimpleSystemI<0, (ins crmtype:$CRm), asm, "\t$CRm">,
695 Sched<[WriteBarrier]> {
697 let Inst{20-12} = 0b000110011;
698 let Inst{11-8} = CRm;
702 // MRS/MSR system instructions. These have different operand classes because
703 // a different subset of registers can be accessed through each instruction.
704 def MRSSystemRegisterOperand : AsmOperandClass {
705 let Name = "MRSSystemRegister";
706 let ParserMethod = "tryParseSysReg";
708 // concatenation of 1, op0, op1, CRn, CRm, op2. 16-bit immediate.
709 def mrs_sysreg_op : Operand<i32> {
710 let ParserMatchClass = MRSSystemRegisterOperand;
711 let DecoderMethod = "DecodeMRSSystemRegister";
712 let PrintMethod = "printMRSSystemRegister";
715 def MSRSystemRegisterOperand : AsmOperandClass {
716 let Name = "MSRSystemRegister";
717 let ParserMethod = "tryParseSysReg";
719 def msr_sysreg_op : Operand<i32> {
720 let ParserMatchClass = MSRSystemRegisterOperand;
721 let DecoderMethod = "DecodeMSRSystemRegister";
722 let PrintMethod = "printMSRSystemRegister";
725 class MRSI : RtSystemI<1, (outs GPR64:$Rt), (ins mrs_sysreg_op:$systemreg),
726 "mrs", "\t$Rt, $systemreg"> {
729 let Inst{19-5} = systemreg;
732 // FIXME: Some of these def CPSR, others don't. Best way to model that?
733 // Explicitly modeling each of the system register as a register class
734 // would do it, but feels like overkill at this point.
735 class MSRI : RtSystemI<0, (outs), (ins msr_sysreg_op:$systemreg, GPR64:$Rt),
736 "msr", "\t$systemreg, $Rt"> {
739 let Inst{19-5} = systemreg;
742 def SystemCPSRFieldOperand : AsmOperandClass {
743 let Name = "SystemCPSRField";
744 let ParserMethod = "tryParseSysReg";
746 def cpsrfield_op : Operand<i32> {
747 let ParserMatchClass = SystemCPSRFieldOperand;
748 let PrintMethod = "printSystemCPSRField";
752 class MSRcpsrI : SimpleSystemI<0, (ins cpsrfield_op:$cpsr_field, imm0_15:$imm),
753 "msr", "\t$cpsr_field, $imm">,
757 let Inst{20-19} = 0b00;
758 let Inst{18-16} = cpsrfield{5-3};
759 let Inst{15-12} = 0b0100;
760 let Inst{11-8} = imm;
761 let Inst{7-5} = cpsrfield{2-0};
763 let DecoderMethod = "DecodeSystemCPSRInstruction";
766 // SYS and SYSL generic system instructions.
767 def SysCRAsmOperand : AsmOperandClass {
769 let ParserMethod = "tryParseSysCROperand";
772 def sys_cr_op : Operand<i32> {
773 let PrintMethod = "printSysCROperand";
774 let ParserMatchClass = SysCRAsmOperand;
777 class SystemXtI<bit L, string asm>
778 : RtSystemI<L, (outs),
779 (ins imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2, GPR64:$Rt),
780 asm, "\t$op1, $Cn, $Cm, $op2, $Rt"> {
785 let Inst{20-19} = 0b01;
786 let Inst{18-16} = op1;
787 let Inst{15-12} = Cn;
792 class SystemLXtI<bit L, string asm>
793 : RtSystemI<L, (outs),
794 (ins GPR64:$Rt, imm0_7:$op1, sys_cr_op:$Cn, sys_cr_op:$Cm, imm0_7:$op2),
795 asm, "\t$Rt, $op1, $Cn, $Cm, $op2"> {
800 let Inst{20-19} = 0b01;
801 let Inst{18-16} = op1;
802 let Inst{15-12} = Cn;
808 // Branch (register) instructions:
816 // otherwise UNDEFINED
817 class BaseBranchReg<bits<4> opc, dag oops, dag iops, string asm,
818 string operands, list<dag> pattern>
819 : I<oops, iops, asm, operands, "", pattern>, Sched<[WriteBrReg]> {
820 let Inst{31-25} = 0b1101011;
821 let Inst{24-21} = opc;
822 let Inst{20-16} = 0b11111;
823 let Inst{15-10} = 0b000000;
824 let Inst{4-0} = 0b00000;
827 class BranchReg<bits<4> opc, string asm, list<dag> pattern>
828 : BaseBranchReg<opc, (outs), (ins GPR64:$Rn), asm, "\t$Rn", pattern> {
833 let mayLoad = 0, mayStore = 0, hasSideEffects = 1, isReturn = 1 in
834 class SpecialReturn<bits<4> opc, string asm>
835 : BaseBranchReg<opc, (outs), (ins), asm, "", []> {
836 let Inst{9-5} = 0b11111;
840 // Conditional branch instruction.
842 // Branch condition code.
843 // 4-bit immediate. Pretty-printed as .<cc>
844 def dotCcode : Operand<i32> {
845 let PrintMethod = "printDotCondCode";
848 // Conditional branch target. 19-bit immediate. The low two bits of the target
849 // offset are implied zero and so are not part of the immediate.
850 def BranchTarget19Operand : AsmOperandClass {
851 let Name = "BranchTarget19";
853 def am_brcond : Operand<OtherVT> {
854 let EncoderMethod = "getCondBranchTargetOpValue";
855 let DecoderMethod = "DecodeCondBranchTarget";
856 let PrintMethod = "printAlignedBranchTarget";
857 let ParserMatchClass = BranchTarget19Operand;
860 class BranchCond : I<(outs), (ins dotCcode:$cond, am_brcond:$target),
861 "b", "$cond\t$target", "",
862 [(ARM64brcond bb:$target, imm:$cond, CPSR)]>,
865 let isTerminator = 1;
870 let Inst{31-24} = 0b01010100;
871 let Inst{23-5} = target;
873 let Inst{3-0} = cond;
877 // Compare-and-branch instructions.
879 class BaseCmpBranch<RegisterClass regtype, bit op, string asm, SDNode node>
880 : I<(outs), (ins regtype:$Rt, am_brcond:$target),
881 asm, "\t$Rt, $target", "",
882 [(node regtype:$Rt, bb:$target)]>,
885 let isTerminator = 1;
889 let Inst{30-25} = 0b011010;
891 let Inst{23-5} = target;
895 multiclass CmpBranch<bit op, string asm, SDNode node> {
896 def W : BaseCmpBranch<GPR32, op, asm, node> {
899 def X : BaseCmpBranch<GPR64, op, asm, node> {
905 // Test-bit-and-branch instructions.
907 // Test-and-branch target. 14-bit sign-extended immediate. The low two bits of
908 // the target offset are implied zero and so are not part of the immediate.
909 def BranchTarget14Operand : AsmOperandClass {
910 let Name = "BranchTarget14";
912 def am_tbrcond : Operand<OtherVT> {
913 let EncoderMethod = "getTestBranchTargetOpValue";
914 let PrintMethod = "printAlignedBranchTarget";
915 let ParserMatchClass = BranchTarget14Operand;
918 class TestBranch<bit op, string asm, SDNode node>
919 : I<(outs), (ins GPR64:$Rt, imm0_63:$bit_off, am_tbrcond:$target),
920 asm, "\t$Rt, $bit_off, $target", "",
921 [(node GPR64:$Rt, imm0_63:$bit_off, bb:$target)]>,
924 let isTerminator = 1;
930 let Inst{31} = bit_off{5};
931 let Inst{30-25} = 0b011011;
933 let Inst{23-19} = bit_off{4-0};
934 let Inst{18-5} = target;
937 let DecoderMethod = "DecodeTestAndBranch";
941 // Unconditional branch (immediate) instructions.
943 def BranchTarget26Operand : AsmOperandClass {
944 let Name = "BranchTarget26";
946 def am_b_target : Operand<OtherVT> {
947 let EncoderMethod = "getBranchTargetOpValue";
948 let PrintMethod = "printAlignedBranchTarget";
949 let ParserMatchClass = BranchTarget26Operand;
951 def am_bl_target : Operand<i64> {
952 let EncoderMethod = "getBranchTargetOpValue";
953 let PrintMethod = "printAlignedBranchTarget";
954 let ParserMatchClass = BranchTarget26Operand;
957 class BImm<bit op, dag iops, string asm, list<dag> pattern>
958 : I<(outs), iops, asm, "\t$addr", "", pattern>, Sched<[WriteBr]> {
961 let Inst{30-26} = 0b00101;
962 let Inst{25-0} = addr;
964 let DecoderMethod = "DecodeUnconditionalBranch";
967 class BranchImm<bit op, string asm, list<dag> pattern>
968 : BImm<op, (ins am_b_target:$addr), asm, pattern>;
969 class CallImm<bit op, string asm, list<dag> pattern>
970 : BImm<op, (ins am_bl_target:$addr), asm, pattern>;
973 // Basic one-operand data processing instructions.
976 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
977 class BaseOneOperandData<bits<3> opc, RegisterClass regtype, string asm,
978 SDPatternOperator node>
979 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
980 [(set regtype:$Rd, (node regtype:$Rn))]>,
985 let Inst{30-13} = 0b101101011000000000;
986 let Inst{12-10} = opc;
991 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
992 multiclass OneOperandData<bits<3> opc, string asm,
993 SDPatternOperator node = null_frag> {
994 def Wr : BaseOneOperandData<opc, GPR32, asm, node> {
998 def Xr : BaseOneOperandData<opc, GPR64, asm, node> {
1003 class OneWRegData<bits<3> opc, string asm, SDPatternOperator node>
1004 : BaseOneOperandData<opc, GPR32, asm, node> {
1008 class OneXRegData<bits<3> opc, string asm, SDPatternOperator node>
1009 : BaseOneOperandData<opc, GPR64, asm, node> {
1014 // Basic two-operand data processing instructions.
1016 class BaseBaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1018 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1019 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
1025 let Inst{30} = isSub;
1026 let Inst{28-21} = 0b11010000;
1027 let Inst{20-16} = Rm;
1028 let Inst{15-10} = 0;
1033 class BaseAddSubCarry<bit isSub, RegisterClass regtype, string asm,
1035 : BaseBaseAddSubCarry<isSub, regtype, asm,
1036 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, CPSR))]>;
1038 class BaseAddSubCarrySetFlags<bit isSub, RegisterClass regtype, string asm,
1040 : BaseBaseAddSubCarry<isSub, regtype, asm,
1041 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm, CPSR)),
1046 multiclass AddSubCarry<bit isSub, string asm, string asm_setflags,
1047 SDNode OpNode, SDNode OpNode_setflags> {
1048 def Wr : BaseAddSubCarry<isSub, GPR32, asm, OpNode> {
1052 def Xr : BaseAddSubCarry<isSub, GPR64, asm, OpNode> {
1058 def SWr : BaseAddSubCarrySetFlags<isSub, GPR32, asm_setflags,
1063 def SXr : BaseAddSubCarrySetFlags<isSub, GPR64, asm_setflags,
1070 class BaseTwoOperand<bits<4> opc, RegisterClass regtype, string asm,
1071 SDPatternOperator OpNode>
1072 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1073 asm, "\t$Rd, $Rn, $Rm", "",
1074 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]> {
1078 let Inst{30-21} = 0b0011010110;
1079 let Inst{20-16} = Rm;
1080 let Inst{15-14} = 0b00;
1081 let Inst{13-10} = opc;
1086 class BaseDiv<bit isSigned, RegisterClass regtype, string asm,
1087 SDPatternOperator OpNode>
1088 : BaseTwoOperand<{0,0,1,?}, regtype, asm, OpNode> {
1089 let Inst{10} = isSigned;
1092 multiclass Div<bit isSigned, string asm, SDPatternOperator OpNode> {
1093 def Wr : BaseDiv<isSigned, GPR32, asm, OpNode>,
1094 Sched<[WriteID32]> {
1097 def Xr : BaseDiv<isSigned, GPR64, asm, OpNode>,
1098 Sched<[WriteID64]> {
1103 class BaseShift<bits<2> shift_type, RegisterClass regtype, string asm,
1104 SDPatternOperator OpNode = null_frag>
1105 : BaseTwoOperand<{1,0,?,?}, regtype, asm, OpNode>,
1107 let Inst{11-10} = shift_type;
1110 multiclass Shift<bits<2> shift_type, string asm, SDNode OpNode> {
1111 def Wr : BaseShift<shift_type, GPR32, asm> {
1115 def Xr : BaseShift<shift_type, GPR64, asm, OpNode> {
1119 def : Pat<(i32 (OpNode GPR32:$Rn, i64:$Rm)),
1120 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn,
1121 (EXTRACT_SUBREG i64:$Rm, sub_32))>;
1123 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (zext GPR32:$Rm)))),
1124 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1126 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (anyext GPR32:$Rm)))),
1127 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1129 def : Pat<(i32 (OpNode GPR32:$Rn, (i64 (sext GPR32:$Rm)))),
1130 (!cast<Instruction>(NAME # "Wr") GPR32:$Rn, GPR32:$Rm)>;
1133 class ShiftAlias<string asm, Instruction inst, RegisterClass regtype>
1134 : InstAlias<asm#" $dst, $src1, $src2",
1135 (inst regtype:$dst, regtype:$src1, regtype:$src2)>;
1137 class BaseMulAccum<bit isSub, bits<3> opc, RegisterClass multype,
1138 RegisterClass addtype, string asm,
1140 : I<(outs addtype:$Rd), (ins multype:$Rn, multype:$Rm, addtype:$Ra),
1141 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pattern> {
1146 let Inst{30-24} = 0b0011011;
1147 let Inst{23-21} = opc;
1148 let Inst{20-16} = Rm;
1149 let Inst{15} = isSub;
1150 let Inst{14-10} = Ra;
1155 multiclass MulAccum<bit isSub, string asm, SDNode AccNode> {
1156 def Wrrr : BaseMulAccum<isSub, 0b000, GPR32, GPR32, asm,
1157 [(set GPR32:$Rd, (AccNode GPR32:$Ra, (mul GPR32:$Rn, GPR32:$Rm)))]>,
1158 Sched<[WriteIM32]> {
1162 def Xrrr : BaseMulAccum<isSub, 0b000, GPR64, GPR64, asm,
1163 [(set GPR64:$Rd, (AccNode GPR64:$Ra, (mul GPR64:$Rn, GPR64:$Rm)))]>,
1164 Sched<[WriteIM64]> {
1169 class WideMulAccum<bit isSub, bits<3> opc, string asm,
1170 SDNode AccNode, SDNode ExtNode>
1171 : BaseMulAccum<isSub, opc, GPR32, GPR64, asm,
1172 [(set GPR64:$Rd, (AccNode GPR64:$Ra,
1173 (mul (ExtNode GPR32:$Rn), (ExtNode GPR32:$Rm))))]>,
1174 Sched<[WriteIM32]> {
1178 class MulHi<bits<3> opc, string asm, SDNode OpNode>
1179 : I<(outs GPR64:$Rd), (ins GPR64:$Rn, GPR64:$Rm),
1180 asm, "\t$Rd, $Rn, $Rm", "",
1181 [(set GPR64:$Rd, (OpNode GPR64:$Rn, GPR64:$Rm))]>,
1182 Sched<[WriteIM64]> {
1186 let Inst{31-24} = 0b10011011;
1187 let Inst{23-21} = opc;
1188 let Inst{20-16} = Rm;
1193 // The Ra field of SMULH and UMULH is unused: it should be assembled as 31
1194 // (i.e. all bits 1) but is ignored by the processor.
1195 let PostEncoderMethod = "fixMulHigh";
1198 class MulAccumWAlias<string asm, Instruction inst>
1199 : InstAlias<asm#" $dst, $src1, $src2",
1200 (inst GPR32:$dst, GPR32:$src1, GPR32:$src2, WZR)>;
1201 class MulAccumXAlias<string asm, Instruction inst>
1202 : InstAlias<asm#" $dst, $src1, $src2",
1203 (inst GPR64:$dst, GPR64:$src1, GPR64:$src2, XZR)>;
1204 class WideMulAccumAlias<string asm, Instruction inst>
1205 : InstAlias<asm#" $dst, $src1, $src2",
1206 (inst GPR64:$dst, GPR32:$src1, GPR32:$src2, XZR)>;
1208 class BaseCRC32<bit sf, bits<2> sz, bit C, RegisterClass StreamReg,
1209 SDPatternOperator OpNode, string asm>
1210 : I<(outs GPR32:$Rd), (ins GPR32:$Rn, StreamReg:$Rm),
1211 asm, "\t$Rd, $Rn, $Rm", "",
1212 [(set GPR32:$Rd, (OpNode GPR32:$Rn, StreamReg:$Rm))]>,
1213 Sched<[WriteISReg]> {
1219 let Inst{30-21} = 0b0011010110;
1220 let Inst{20-16} = Rm;
1221 let Inst{15-13} = 0b010;
1223 let Inst{11-10} = sz;
1229 // Address generation.
1232 class ADRI<bit page, string asm, Operand adr, list<dag> pattern>
1233 : I<(outs GPR64:$Xd), (ins adr:$label), asm, "\t$Xd, $label", "",
1238 let Inst{31} = page;
1239 let Inst{30-29} = label{1-0};
1240 let Inst{28-24} = 0b10000;
1241 let Inst{23-5} = label{20-2};
1244 let DecoderMethod = "DecodeAdrInstruction";
1251 def movimm32_imm : Operand<i32> {
1252 let ParserMatchClass = Imm0_65535Operand;
1253 let EncoderMethod = "getMoveWideImmOpValue";
1255 def movimm32_shift : Operand<i32> {
1256 let PrintMethod = "printShifter";
1257 let ParserMatchClass = MovImm32ShifterOperand;
1259 def movimm64_shift : Operand<i32> {
1260 let PrintMethod = "printShifter";
1261 let ParserMatchClass = MovImm64ShifterOperand;
1263 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1264 class BaseMoveImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1266 : I<(outs regtype:$Rd), (ins movimm32_imm:$imm, shifter:$shift),
1267 asm, "\t$Rd, $imm$shift", "", []>,
1272 let Inst{30-29} = opc;
1273 let Inst{28-23} = 0b100101;
1274 let Inst{22-21} = shift{5-4};
1275 let Inst{20-5} = imm;
1278 let DecoderMethod = "DecodeMoveImmInstruction";
1281 multiclass MoveImmediate<bits<2> opc, string asm> {
1282 def Wi : BaseMoveImmediate<opc, GPR32, movimm32_shift, asm> {
1286 def Xi : BaseMoveImmediate<opc, GPR64, movimm64_shift, asm> {
1291 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1292 class BaseInsertImmediate<bits<2> opc, RegisterClass regtype, Operand shifter,
1294 : I<(outs regtype:$Rd),
1295 (ins regtype:$src, movimm32_imm:$imm, shifter:$shift),
1296 asm, "\t$Rd, $imm$shift", "$src = $Rd", []>,
1301 let Inst{30-29} = opc;
1302 let Inst{28-23} = 0b100101;
1303 let Inst{22-21} = shift{5-4};
1304 let Inst{20-5} = imm;
1307 let DecoderMethod = "DecodeMoveImmInstruction";
1310 multiclass InsertImmediate<bits<2> opc, string asm> {
1311 def Wi : BaseInsertImmediate<opc, GPR32, movimm32_shift, asm> {
1315 def Xi : BaseInsertImmediate<opc, GPR64, movimm64_shift, asm> {
1324 class BaseAddSubImm<bit isSub, bit setFlags, RegisterClass dstRegtype,
1325 RegisterClass srcRegtype, addsub_shifted_imm immtype,
1326 string asm, SDPatternOperator OpNode>
1327 : I<(outs dstRegtype:$Rd), (ins srcRegtype:$Rn, immtype:$imm),
1328 asm, "\t$Rd, $Rn, $imm", "",
1329 [(set dstRegtype:$Rd, (OpNode srcRegtype:$Rn, immtype:$imm))]>,
1334 let Inst{30} = isSub;
1335 let Inst{29} = setFlags;
1336 let Inst{28-24} = 0b10001;
1337 let Inst{23-22} = imm{13-12}; // '00' => lsl #0, '01' => lsl #12
1338 let Inst{21-10} = imm{11-0};
1341 let DecoderMethod = "DecodeBaseAddSubImm";
1344 class BaseAddSubRegPseudo<RegisterClass regtype,
1345 SDPatternOperator OpNode>
1346 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1347 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
1350 class BaseAddSubSReg<bit isSub, bit setFlags, RegisterClass regtype,
1351 arith_shifted_reg shifted_regtype, string asm,
1352 SDPatternOperator OpNode>
1353 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
1354 asm, "\t$Rd, $Rn, $Rm", "",
1355 [(set regtype:$Rd, (OpNode regtype:$Rn, shifted_regtype:$Rm))]>,
1356 Sched<[WriteISReg]> {
1357 // The operands are in order to match the 'addr' MI operands, so we
1358 // don't need an encoder method and by-name matching. Just use the default
1359 // in-order handling. Since we're using by-order, make sure the names
1365 let Inst{30} = isSub;
1366 let Inst{29} = setFlags;
1367 let Inst{28-24} = 0b01011;
1368 let Inst{23-22} = shift{7-6};
1370 let Inst{20-16} = src2;
1371 let Inst{15-10} = shift{5-0};
1372 let Inst{9-5} = src1;
1373 let Inst{4-0} = dst;
1375 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
1378 class BaseAddSubEReg<bit isSub, bit setFlags, RegisterClass dstRegtype,
1379 RegisterClass src1Regtype, Operand src2Regtype,
1380 string asm, SDPatternOperator OpNode>
1381 : I<(outs dstRegtype:$R1),
1382 (ins src1Regtype:$R2, src2Regtype:$R3),
1383 asm, "\t$R1, $R2, $R3", "",
1384 [(set dstRegtype:$R1, (OpNode src1Regtype:$R2, src2Regtype:$R3))]>,
1385 Sched<[WriteIEReg]> {
1390 let Inst{30} = isSub;
1391 let Inst{29} = setFlags;
1392 let Inst{28-24} = 0b01011;
1393 let Inst{23-21} = 0b001;
1394 let Inst{20-16} = Rm;
1395 let Inst{15-13} = ext{5-3};
1396 let Inst{12-10} = ext{2-0};
1400 let DecoderMethod = "DecodeAddSubERegInstruction";
1403 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1404 class BaseAddSubEReg64<bit isSub, bit setFlags, RegisterClass dstRegtype,
1405 RegisterClass src1Regtype, RegisterClass src2Regtype,
1406 Operand ext_op, string asm>
1407 : I<(outs dstRegtype:$Rd),
1408 (ins src1Regtype:$Rn, src2Regtype:$Rm, ext_op:$ext),
1409 asm, "\t$Rd, $Rn, $Rm$ext", "", []>,
1410 Sched<[WriteIEReg]> {
1415 let Inst{30} = isSub;
1416 let Inst{29} = setFlags;
1417 let Inst{28-24} = 0b01011;
1418 let Inst{23-21} = 0b001;
1419 let Inst{20-16} = Rm;
1420 let Inst{15} = ext{5};
1421 let Inst{12-10} = ext{2-0};
1425 let DecoderMethod = "DecodeAddSubERegInstruction";
1428 // Aliases for register+register add/subtract.
1429 class AddSubRegAlias<string asm, Instruction inst, RegisterClass dstRegtype,
1430 RegisterClass src1Regtype, RegisterClass src2Regtype,
1432 : InstAlias<asm#" $dst, $src1, $src2",
1433 (inst dstRegtype:$dst, src1Regtype:$src1, src2Regtype:$src2,
1436 multiclass AddSub<bit isSub, string mnemonic,
1437 SDPatternOperator OpNode = null_frag> {
1438 let hasSideEffects = 0 in {
1439 // Add/Subtract immediate
1440 def Wri : BaseAddSubImm<isSub, 0, GPR32sp, GPR32sp, addsub_shifted_imm32,
1444 def Xri : BaseAddSubImm<isSub, 0, GPR64sp, GPR64sp, addsub_shifted_imm64,
1449 // Add/Subtract register - Only used for CodeGen
1450 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
1451 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
1453 // Add/Subtract shifted register
1454 def Wrs : BaseAddSubSReg<isSub, 0, GPR32, arith_shifted_reg32, mnemonic,
1458 def Xrs : BaseAddSubSReg<isSub, 0, GPR64, arith_shifted_reg64, mnemonic,
1464 // Add/Subtract extended register
1465 let AddedComplexity = 1, hasSideEffects = 0 in {
1466 def Wrx : BaseAddSubEReg<isSub, 0, GPR32sp, GPR32sp,
1467 arith_extended_reg32<i32>, mnemonic, OpNode> {
1470 def Xrx : BaseAddSubEReg<isSub, 0, GPR64sp, GPR64sp,
1471 arith_extended_reg32to64<i64>, mnemonic, OpNode> {
1476 def Xrx64 : BaseAddSubEReg64<isSub, 0, GPR64sp, GPR64sp, GPR64,
1477 arith_extendlsl64, mnemonic> {
1478 // UXTX and SXTX only.
1479 let Inst{14-13} = 0b11;
1483 // Register/register aliases with no shift when SP is not used.
1484 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
1485 GPR32, GPR32, GPR32, 0>;
1486 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
1487 GPR64, GPR64, GPR64, 0>;
1489 // Register/register aliases with no shift when either the destination or
1490 // first source register is SP. This relies on the shifted register aliases
1491 // above matching first in the case when SP is not used.
1492 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
1493 GPR32sp, GPR32sp, GPR32, 16>; // UXTW #0
1494 def : AddSubRegAlias<mnemonic,
1495 !cast<Instruction>(NAME#"Xrx64"),
1496 GPR64sp, GPR64sp, GPR64, 24>; // UXTX #0
1499 multiclass AddSubS<bit isSub, string mnemonic, SDNode OpNode> {
1500 let isCompare = 1, Defs = [CPSR] in {
1501 // Add/Subtract immediate
1502 def Wri : BaseAddSubImm<isSub, 1, GPR32, GPR32sp, addsub_shifted_imm32,
1506 def Xri : BaseAddSubImm<isSub, 1, GPR64, GPR64sp, addsub_shifted_imm64,
1511 // Add/Subtract register
1512 def Wrr : BaseAddSubRegPseudo<GPR32, OpNode>;
1513 def Xrr : BaseAddSubRegPseudo<GPR64, OpNode>;
1515 // Add/Subtract shifted register
1516 def Wrs : BaseAddSubSReg<isSub, 1, GPR32, arith_shifted_reg32, mnemonic,
1520 def Xrs : BaseAddSubSReg<isSub, 1, GPR64, arith_shifted_reg64, mnemonic,
1525 // Add/Subtract extended register
1526 let AddedComplexity = 1 in {
1527 def Wrx : BaseAddSubEReg<isSub, 1, GPR32, GPR32sp,
1528 arith_extended_reg32<i32>, mnemonic, OpNode> {
1531 def Xrx : BaseAddSubEReg<isSub, 1, GPR64, GPR64sp,
1532 arith_extended_reg32<i64>, mnemonic, OpNode> {
1537 def Xrx64 : BaseAddSubEReg64<isSub, 1, GPR64, GPR64sp, GPR64,
1538 arith_extendlsl64, mnemonic> {
1539 // UXTX and SXTX only.
1540 let Inst{14-13} = 0b11;
1545 // Register/register aliases with no shift when SP is not used.
1546 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrs"),
1547 GPR32, GPR32, GPR32, 0>;
1548 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Xrs"),
1549 GPR64, GPR64, GPR64, 0>;
1551 // Register/register aliases with no shift when the first source register
1552 // is SP. This relies on the shifted register aliases above matching first
1553 // in the case when SP is not used.
1554 def : AddSubRegAlias<mnemonic, !cast<Instruction>(NAME#"Wrx"),
1555 GPR32, GPR32sp, GPR32, 16>; // UXTW #0
1556 def : AddSubRegAlias<mnemonic,
1557 !cast<Instruction>(NAME#"Xrx64"),
1558 GPR64, GPR64sp, GPR64, 24>; // UXTX #0
1564 def SDTA64EXTR : SDTypeProfile<1, 3, [SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>,
1566 def ARM64Extr : SDNode<"ARM64ISD::EXTR", SDTA64EXTR>;
1568 class BaseExtractImm<RegisterClass regtype, Operand imm_type, string asm,
1570 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, imm_type:$imm),
1571 asm, "\t$Rd, $Rn, $Rm, $imm", "", patterns>,
1572 Sched<[WriteExtr, ReadExtrHi]> {
1578 let Inst{30-23} = 0b00100111;
1580 let Inst{20-16} = Rm;
1581 let Inst{15-10} = imm;
1586 multiclass ExtractImm<string asm> {
1587 def Wrri : BaseExtractImm<GPR32, imm0_31, asm,
1589 (ARM64Extr GPR32:$Rn, GPR32:$Rm, imm0_31:$imm))]> {
1592 // imm<5> must be zero.
1595 def Xrri : BaseExtractImm<GPR64, imm0_63, asm,
1597 (ARM64Extr GPR64:$Rn, GPR64:$Rm, imm0_63:$imm))]> {
1608 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1609 class BaseBitfieldImm<bits<2> opc,
1610 RegisterClass regtype, Operand imm_type, string asm>
1611 : I<(outs regtype:$Rd), (ins regtype:$Rn, imm_type:$immr, imm_type:$imms),
1612 asm, "\t$Rd, $Rn, $immr, $imms", "", []>,
1619 let Inst{30-29} = opc;
1620 let Inst{28-23} = 0b100110;
1621 let Inst{21-16} = immr;
1622 let Inst{15-10} = imms;
1627 multiclass BitfieldImm<bits<2> opc, string asm> {
1628 def Wri : BaseBitfieldImm<opc, GPR32, imm0_31, asm> {
1631 // imms<5> and immr<5> must be zero, else ReservedValue().
1635 def Xri : BaseBitfieldImm<opc, GPR64, imm0_63, asm> {
1641 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1642 class BaseBitfieldImmWith2RegArgs<bits<2> opc,
1643 RegisterClass regtype, Operand imm_type, string asm>
1644 : I<(outs regtype:$Rd), (ins regtype:$src, regtype:$Rn, imm_type:$immr,
1646 asm, "\t$Rd, $Rn, $immr, $imms", "$src = $Rd", []>,
1653 let Inst{30-29} = opc;
1654 let Inst{28-23} = 0b100110;
1655 let Inst{21-16} = immr;
1656 let Inst{15-10} = imms;
1661 multiclass BitfieldImmWith2RegArgs<bits<2> opc, string asm> {
1662 def Wri : BaseBitfieldImmWith2RegArgs<opc, GPR32, imm0_31, asm> {
1665 // imms<5> and immr<5> must be zero, else ReservedValue().
1669 def Xri : BaseBitfieldImmWith2RegArgs<opc, GPR64, imm0_63, asm> {
1679 // Logical (immediate)
1680 class BaseLogicalImm<bits<2> opc, RegisterClass dregtype,
1681 RegisterClass sregtype, Operand imm_type, string asm,
1683 : I<(outs dregtype:$Rd), (ins sregtype:$Rn, imm_type:$imm),
1684 asm, "\t$Rd, $Rn, $imm", "", pattern>,
1689 let Inst{30-29} = opc;
1690 let Inst{28-23} = 0b100100;
1691 let Inst{22} = imm{12};
1692 let Inst{21-16} = imm{11-6};
1693 let Inst{15-10} = imm{5-0};
1697 let DecoderMethod = "DecodeLogicalImmInstruction";
1700 // Logical (shifted register)
1701 class BaseLogicalSReg<bits<2> opc, bit N, RegisterClass regtype,
1702 logical_shifted_reg shifted_regtype, string asm,
1704 : I<(outs regtype:$Rd), (ins regtype:$Rn, shifted_regtype:$Rm),
1705 asm, "\t$Rd, $Rn, $Rm", "", pattern>,
1706 Sched<[WriteISReg]> {
1707 // The operands are in order to match the 'addr' MI operands, so we
1708 // don't need an encoder method and by-name matching. Just use the default
1709 // in-order handling. Since we're using by-order, make sure the names
1715 let Inst{30-29} = opc;
1716 let Inst{28-24} = 0b01010;
1717 let Inst{23-22} = shift{7-6};
1719 let Inst{20-16} = src2;
1720 let Inst{15-10} = shift{5-0};
1721 let Inst{9-5} = src1;
1722 let Inst{4-0} = dst;
1724 let DecoderMethod = "DecodeThreeAddrSRegInstruction";
1727 // Aliases for register+register logical instructions.
1728 class LogicalRegAlias<string asm, Instruction inst, RegisterClass regtype>
1729 : InstAlias<asm#" $dst, $src1, $src2",
1730 (inst regtype:$dst, regtype:$src1, regtype:$src2, 0)>;
1732 let AddedComplexity = 6 in
1733 multiclass LogicalImm<bits<2> opc, string mnemonic, SDNode OpNode> {
1734 def Wri : BaseLogicalImm<opc, GPR32sp, GPR32, logical_imm32, mnemonic,
1735 [(set GPR32sp:$Rd, (OpNode GPR32:$Rn,
1736 logical_imm32:$imm))]> {
1738 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
1740 def Xri : BaseLogicalImm<opc, GPR64sp, GPR64, logical_imm64, mnemonic,
1741 [(set GPR64sp:$Rd, (OpNode GPR64:$Rn,
1742 logical_imm64:$imm))]> {
1747 multiclass LogicalImmS<bits<2> opc, string mnemonic, SDNode OpNode> {
1748 let isCompare = 1, Defs = [CPSR] in {
1749 def Wri : BaseLogicalImm<opc, GPR32, GPR32, logical_imm32, mnemonic,
1750 [(set GPR32:$Rd, (OpNode GPR32:$Rn, logical_imm32:$imm))]> {
1752 let Inst{22} = 0; // 64-bit version has an additional bit of immediate.
1754 def Xri : BaseLogicalImm<opc, GPR64, GPR64, logical_imm64, mnemonic,
1755 [(set GPR64:$Rd, (OpNode GPR64:$Rn, logical_imm64:$imm))]> {
1758 } // end Defs = [CPSR]
1761 class BaseLogicalRegPseudo<RegisterClass regtype, SDPatternOperator OpNode>
1762 : Pseudo<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
1763 [(set regtype:$Rd, (OpNode regtype:$Rn, regtype:$Rm))]>,
1766 // Split from LogicalImm as not all instructions have both.
1767 multiclass LogicalReg<bits<2> opc, bit N, string mnemonic,
1768 SDPatternOperator OpNode> {
1769 def Wrr : BaseLogicalRegPseudo<GPR32, OpNode>;
1770 def Xrr : BaseLogicalRegPseudo<GPR64, OpNode>;
1772 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic,
1773 [(set GPR32:$Rd, (OpNode GPR32:$Rn,
1774 logical_shifted_reg32:$Rm))]> {
1777 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic,
1778 [(set GPR64:$Rd, (OpNode GPR64:$Rn,
1779 logical_shifted_reg64:$Rm))]> {
1783 def : LogicalRegAlias<mnemonic,
1784 !cast<Instruction>(NAME#"Wrs"), GPR32>;
1785 def : LogicalRegAlias<mnemonic,
1786 !cast<Instruction>(NAME#"Xrs"), GPR64>;
1789 // Split from LogicalReg to allow setting CPSR Defs
1790 multiclass LogicalRegS<bits<2> opc, bit N, string mnemonic> {
1791 let Defs = [CPSR], mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
1792 def Wrs : BaseLogicalSReg<opc, N, GPR32, logical_shifted_reg32, mnemonic, []>{
1795 def Xrs : BaseLogicalSReg<opc, N, GPR64, logical_shifted_reg64, mnemonic, []>{
1800 def : LogicalRegAlias<mnemonic,
1801 !cast<Instruction>(NAME#"Wrs"), GPR32>;
1802 def : LogicalRegAlias<mnemonic,
1803 !cast<Instruction>(NAME#"Xrs"), GPR64>;
1807 // Conditionally set flags
1811 // 4-bit immediate. Pretty-printed as <cc>
1812 def ccode : Operand<i32> {
1813 let PrintMethod = "printCondCode";
1816 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1817 class BaseCondSetFlagsImm<bit op, RegisterClass regtype, string asm>
1818 : I<(outs), (ins regtype:$Rn, imm0_31:$imm, imm0_15:$nzcv, ccode:$cond),
1819 asm, "\t$Rn, $imm, $nzcv, $cond", "", []>,
1830 let Inst{29-21} = 0b111010010;
1831 let Inst{20-16} = imm;
1832 let Inst{15-12} = cond;
1833 let Inst{11-10} = 0b10;
1836 let Inst{3-0} = nzcv;
1839 multiclass CondSetFlagsImm<bit op, string asm> {
1840 def Wi : BaseCondSetFlagsImm<op, GPR32, asm> {
1843 def Xi : BaseCondSetFlagsImm<op, GPR64, asm> {
1848 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
1849 class BaseCondSetFlagsReg<bit op, RegisterClass regtype, string asm>
1850 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm0_15:$nzcv, ccode:$cond),
1851 asm, "\t$Rn, $Rm, $nzcv, $cond", "", []>,
1862 let Inst{29-21} = 0b111010010;
1863 let Inst{20-16} = Rm;
1864 let Inst{15-12} = cond;
1865 let Inst{11-10} = 0b00;
1868 let Inst{3-0} = nzcv;
1871 multiclass CondSetFlagsReg<bit op, string asm> {
1872 def Wr : BaseCondSetFlagsReg<op, GPR32, asm> {
1875 def Xr : BaseCondSetFlagsReg<op, GPR64, asm> {
1881 // Conditional select
1884 class BaseCondSelect<bit op, bits<2> op2, RegisterClass regtype, string asm>
1885 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
1886 asm, "\t$Rd, $Rn, $Rm, $cond", "",
1888 (ARM64csel regtype:$Rn, regtype:$Rm, (i32 imm:$cond), CPSR))]>,
1898 let Inst{29-21} = 0b011010100;
1899 let Inst{20-16} = Rm;
1900 let Inst{15-12} = cond;
1901 let Inst{11-10} = op2;
1906 multiclass CondSelect<bit op, bits<2> op2, string asm> {
1907 def Wr : BaseCondSelect<op, op2, GPR32, asm> {
1910 def Xr : BaseCondSelect<op, op2, GPR64, asm> {
1915 class BaseCondSelectOp<bit op, bits<2> op2, RegisterClass regtype, string asm,
1917 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
1918 asm, "\t$Rd, $Rn, $Rm, $cond", "",
1920 (ARM64csel regtype:$Rn, (frag regtype:$Rm),
1921 (i32 imm:$cond), CPSR))]>,
1931 let Inst{29-21} = 0b011010100;
1932 let Inst{20-16} = Rm;
1933 let Inst{15-12} = cond;
1934 let Inst{11-10} = op2;
1939 def inv_cond_XFORM : SDNodeXForm<imm, [{
1940 ARM64CC::CondCode CC = static_cast<ARM64CC::CondCode>(N->getZExtValue());
1941 return CurDAG->getTargetConstant(ARM64CC::getInvertedCondCode(CC), MVT::i32);
1944 multiclass CondSelectOp<bit op, bits<2> op2, string asm, PatFrag frag> {
1945 def Wr : BaseCondSelectOp<op, op2, GPR32, asm, frag> {
1948 def Xr : BaseCondSelectOp<op, op2, GPR64, asm, frag> {
1952 def : Pat<(ARM64csel (frag GPR32:$Rm), GPR32:$Rn, (i32 imm:$cond), CPSR),
1953 (!cast<Instruction>(NAME # Wr) GPR32:$Rn, GPR32:$Rm,
1954 (inv_cond_XFORM imm:$cond))>;
1956 def : Pat<(ARM64csel (frag GPR64:$Rm), GPR64:$Rn, (i32 imm:$cond), CPSR),
1957 (!cast<Instruction>(NAME # Xr) GPR64:$Rn, GPR64:$Rm,
1958 (inv_cond_XFORM imm:$cond))>;
1962 // Special Mask Value
1964 def maski8_or_more : Operand<i32>,
1965 ImmLeaf<i32, [{ return (Imm & 0xff) == 0xff; }]> {
1967 def maski16_or_more : Operand<i32>,
1968 ImmLeaf<i32, [{ return (Imm & 0xffff) == 0xffff; }]> {
1976 // (unsigned immediate)
1977 // Indexed for 8-bit registers. offset is in range [0,4095].
1978 def MemoryIndexed8Operand : AsmOperandClass {
1979 let Name = "MemoryIndexed8";
1980 let DiagnosticType = "InvalidMemoryIndexed8";
1982 def am_indexed8 : Operand<i64>,
1983 ComplexPattern<i64, 2, "SelectAddrModeIndexed8", []> {
1984 let PrintMethod = "printAMIndexed8";
1986 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale1>";
1987 let ParserMatchClass = MemoryIndexed8Operand;
1988 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1991 // Indexed for 16-bit registers. offset is multiple of 2 in range [0,8190],
1992 // stored as immval/2 (the 12-bit literal that encodes directly into the insn).
1993 def MemoryIndexed16Operand : AsmOperandClass {
1994 let Name = "MemoryIndexed16";
1995 let DiagnosticType = "InvalidMemoryIndexed16";
1997 def am_indexed16 : Operand<i64>,
1998 ComplexPattern<i64, 2, "SelectAddrModeIndexed16", []> {
1999 let PrintMethod = "printAMIndexed16";
2001 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale2>";
2002 let ParserMatchClass = MemoryIndexed16Operand;
2003 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2006 // Indexed for 32-bit registers. offset is multiple of 4 in range [0,16380],
2007 // stored as immval/4 (the 12-bit literal that encodes directly into the insn).
2008 def MemoryIndexed32Operand : AsmOperandClass {
2009 let Name = "MemoryIndexed32";
2010 let DiagnosticType = "InvalidMemoryIndexed32";
2012 def am_indexed32 : Operand<i64>,
2013 ComplexPattern<i64, 2, "SelectAddrModeIndexed32", []> {
2014 let PrintMethod = "printAMIndexed32";
2016 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale4>";
2017 let ParserMatchClass = MemoryIndexed32Operand;
2018 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2021 // Indexed for 64-bit registers. offset is multiple of 8 in range [0,32760],
2022 // stored as immval/8 (the 12-bit literal that encodes directly into the insn).
2023 def MemoryIndexed64Operand : AsmOperandClass {
2024 let Name = "MemoryIndexed64";
2025 let DiagnosticType = "InvalidMemoryIndexed64";
2027 def am_indexed64 : Operand<i64>,
2028 ComplexPattern<i64, 2, "SelectAddrModeIndexed64", []> {
2029 let PrintMethod = "printAMIndexed64";
2031 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale8>";
2032 let ParserMatchClass = MemoryIndexed64Operand;
2033 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2036 // Indexed for 128-bit registers. offset is multiple of 16 in range [0,65520],
2037 // stored as immval/16 (the 12-bit literal that encodes directly into the insn).
2038 def MemoryIndexed128Operand : AsmOperandClass {
2039 let Name = "MemoryIndexed128";
2040 let DiagnosticType = "InvalidMemoryIndexed128";
2042 def am_indexed128 : Operand<i64>,
2043 ComplexPattern<i64, 2, "SelectAddrModeIndexed128", []> {
2044 let PrintMethod = "printAMIndexed128";
2046 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale16>";
2047 let ParserMatchClass = MemoryIndexed128Operand;
2048 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2052 def MemoryNoIndexOperand : AsmOperandClass { let Name = "MemoryNoIndex"; }
2053 def am_noindex : Operand<i64>,
2054 ComplexPattern<i64, 1, "SelectAddrModeNoIndex", []> {
2055 let PrintMethod = "printAMNoIndex";
2056 let ParserMatchClass = MemoryNoIndexOperand;
2057 let MIOperandInfo = (ops GPR64sp:$base);
2060 class BaseLoadStoreUI<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2061 string asm, list<dag> pattern>
2062 : I<oops, iops, asm, "\t$Rt, $addr", "", pattern> {
2066 bits<5> base = addr{4-0};
2067 bits<12> offset = addr{16-5};
2069 let Inst{31-30} = sz;
2070 let Inst{29-27} = 0b111;
2072 let Inst{25-24} = 0b01;
2073 let Inst{23-22} = opc;
2074 let Inst{21-10} = offset;
2075 let Inst{9-5} = base;
2076 let Inst{4-0} = dst;
2078 let DecoderMethod = "DecodeUnsignedLdStInstruction";
2081 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2082 class LoadUI<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2083 Operand indextype, string asm, list<dag> pattern>
2084 : BaseLoadStoreUI<sz, V, opc,
2085 (outs regtype:$Rt), (ins indextype:$addr), asm, pattern>,
2088 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2089 class StoreUI<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2090 Operand indextype, string asm, list<dag> pattern>
2091 : BaseLoadStoreUI<sz, V, opc,
2092 (outs), (ins regtype:$Rt, indextype:$addr), asm, pattern>,
2095 def PrefetchOperand : AsmOperandClass {
2096 let Name = "Prefetch";
2097 let ParserMethod = "tryParsePrefetch";
2099 def prfop : Operand<i32> {
2100 let PrintMethod = "printPrefetchOp";
2101 let ParserMatchClass = PrefetchOperand;
2104 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2105 class PrefetchUI<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2106 : BaseLoadStoreUI<sz, V, opc,
2107 (outs), (ins prfop:$Rt, am_indexed64:$addr), asm, pat>,
2114 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2115 class LoadLiteral<bits<2> opc, bit V, RegisterClass regtype, string asm>
2116 : I<(outs regtype:$Rt), (ins am_brcond:$label),
2117 asm, "\t$Rt, $label", "", []>,
2121 let Inst{31-30} = opc;
2122 let Inst{29-27} = 0b011;
2124 let Inst{25-24} = 0b00;
2125 let Inst{23-5} = label;
2129 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2130 class PrefetchLiteral<bits<2> opc, bit V, string asm, list<dag> pat>
2131 : I<(outs), (ins prfop:$Rt, am_brcond:$label),
2132 asm, "\t$Rt, $label", "", pat>,
2136 let Inst{31-30} = opc;
2137 let Inst{29-27} = 0b011;
2139 let Inst{25-24} = 0b00;
2140 let Inst{23-5} = label;
2145 // Load/store register offset
2148 class MemROAsmOperand<int sz> : AsmOperandClass {
2149 let Name = "MemoryRegisterOffset"#sz;
2152 def MemROAsmOperand8 : MemROAsmOperand<8>;
2153 def MemROAsmOperand16 : MemROAsmOperand<16>;
2154 def MemROAsmOperand32 : MemROAsmOperand<32>;
2155 def MemROAsmOperand64 : MemROAsmOperand<64>;
2156 def MemROAsmOperand128 : MemROAsmOperand<128>;
2158 class ro_indexed<int sz> : Operand<i64> { // ComplexPattern<...>
2159 let PrintMethod = "printMemoryRegOffset"#sz;
2160 let MIOperandInfo = (ops GPR64sp:$base, GPR64:$offset, i32imm:$extend);
2163 def ro_indexed8 : ro_indexed<8>, ComplexPattern<i64, 3, "SelectAddrModeRO8", []> {
2164 let ParserMatchClass = MemROAsmOperand8;
2167 def ro_indexed16 : ro_indexed<16>, ComplexPattern<i64, 3, "SelectAddrModeRO16", []> {
2168 let ParserMatchClass = MemROAsmOperand16;
2171 def ro_indexed32 : ro_indexed<32>, ComplexPattern<i64, 3, "SelectAddrModeRO32", []> {
2172 let ParserMatchClass = MemROAsmOperand32;
2175 def ro_indexed64 : ro_indexed<64>, ComplexPattern<i64, 3, "SelectAddrModeRO64", []> {
2176 let ParserMatchClass = MemROAsmOperand64;
2179 def ro_indexed128 : ro_indexed<128>, ComplexPattern<i64, 3, "SelectAddrModeRO128", []> {
2180 let ParserMatchClass = MemROAsmOperand128;
2183 class LoadStore8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2184 string asm, dag ins, dag outs, list<dag> pat>
2185 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2186 // The operands are in order to match the 'addr' MI operands, so we
2187 // don't need an encoder method and by-name matching. Just use the default
2188 // in-order handling. Since we're using by-order, make sure the names
2194 let Inst{31-30} = sz;
2195 let Inst{29-27} = 0b111;
2197 let Inst{25-24} = 0b00;
2198 let Inst{23-22} = opc;
2200 let Inst{20-16} = offset;
2201 let Inst{15-13} = extend{3-1};
2203 let Inst{12} = extend{0};
2204 let Inst{11-10} = 0b10;
2205 let Inst{9-5} = base;
2206 let Inst{4-0} = dst;
2208 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2211 class Load8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2212 string asm, list<dag> pat>
2213 : LoadStore8RO<sz, V, opc, regtype, asm,
2214 (outs regtype:$Rt), (ins ro_indexed8:$addr), pat>,
2215 Sched<[WriteLDIdx, ReadAdrBase]>;
2217 class Store8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2218 string asm, list<dag> pat>
2219 : LoadStore8RO<sz, V, opc, regtype, asm,
2220 (outs), (ins regtype:$Rt, ro_indexed8:$addr), pat>,
2221 Sched<[WriteSTIdx, ReadAdrBase]>;
2223 class LoadStore16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2224 string asm, dag ins, dag outs, list<dag> pat>
2225 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2226 // The operands are in order to match the 'addr' MI operands, so we
2227 // don't need an encoder method and by-name matching. Just use the default
2228 // in-order handling. Since we're using by-order, make sure the names
2234 let Inst{31-30} = sz;
2235 let Inst{29-27} = 0b111;
2237 let Inst{25-24} = 0b00;
2238 let Inst{23-22} = opc;
2240 let Inst{20-16} = offset;
2241 let Inst{15-13} = extend{3-1};
2243 let Inst{12} = extend{0};
2244 let Inst{11-10} = 0b10;
2245 let Inst{9-5} = base;
2246 let Inst{4-0} = dst;
2248 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2251 class Load16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2252 string asm, list<dag> pat>
2253 : LoadStore16RO<sz, V, opc, regtype, asm,
2254 (outs regtype:$Rt), (ins ro_indexed16:$addr), pat>,
2255 Sched<[WriteLDIdx, ReadAdrBase]>;
2257 class Store16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2258 string asm, list<dag> pat>
2259 : LoadStore16RO<sz, V, opc, regtype, asm,
2260 (outs), (ins regtype:$Rt, ro_indexed16:$addr), pat>,
2261 Sched<[WriteSTIdx, ReadAdrBase]>;
2263 class LoadStore32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2264 string asm, dag ins, dag outs, list<dag> pat>
2265 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2266 // The operands are in order to match the 'addr' MI operands, so we
2267 // don't need an encoder method and by-name matching. Just use the default
2268 // in-order handling. Since we're using by-order, make sure the names
2274 let Inst{31-30} = sz;
2275 let Inst{29-27} = 0b111;
2277 let Inst{25-24} = 0b00;
2278 let Inst{23-22} = opc;
2280 let Inst{20-16} = offset;
2281 let Inst{15-13} = extend{3-1};
2283 let Inst{12} = extend{0};
2284 let Inst{11-10} = 0b10;
2285 let Inst{9-5} = base;
2286 let Inst{4-0} = dst;
2288 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2291 class Load32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2292 string asm, list<dag> pat>
2293 : LoadStore32RO<sz, V, opc, regtype, asm,
2294 (outs regtype:$Rt), (ins ro_indexed32:$addr), pat>,
2295 Sched<[WriteLDIdx, ReadAdrBase]>;
2297 class Store32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2298 string asm, list<dag> pat>
2299 : LoadStore32RO<sz, V, opc, regtype, asm,
2300 (outs), (ins regtype:$Rt, ro_indexed32:$addr), pat>,
2301 Sched<[WriteSTIdx, ReadAdrBase]>;
2303 class LoadStore64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2304 string asm, dag ins, dag outs, list<dag> pat>
2305 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2306 // The operands are in order to match the 'addr' MI operands, so we
2307 // don't need an encoder method and by-name matching. Just use the default
2308 // in-order handling. Since we're using by-order, make sure the names
2314 let Inst{31-30} = sz;
2315 let Inst{29-27} = 0b111;
2317 let Inst{25-24} = 0b00;
2318 let Inst{23-22} = opc;
2320 let Inst{20-16} = offset;
2321 let Inst{15-13} = extend{3-1};
2323 let Inst{12} = extend{0};
2324 let Inst{11-10} = 0b10;
2325 let Inst{9-5} = base;
2326 let Inst{4-0} = dst;
2328 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2331 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2332 class Load64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2333 string asm, list<dag> pat>
2334 : LoadStore64RO<sz, V, opc, regtype, asm,
2335 (outs regtype:$Rt), (ins ro_indexed64:$addr), pat>,
2336 Sched<[WriteLDIdx, ReadAdrBase]>;
2338 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2339 class Store64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2340 string asm, list<dag> pat>
2341 : LoadStore64RO<sz, V, opc, regtype, asm,
2342 (outs), (ins regtype:$Rt, ro_indexed64:$addr), pat>,
2343 Sched<[WriteSTIdx, ReadAdrBase]>;
2346 class LoadStore128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2347 string asm, dag ins, dag outs, list<dag> pat>
2348 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2349 // The operands are in order to match the 'addr' MI operands, so we
2350 // don't need an encoder method and by-name matching. Just use the default
2351 // in-order handling. Since we're using by-order, make sure the names
2357 let Inst{31-30} = sz;
2358 let Inst{29-27} = 0b111;
2360 let Inst{25-24} = 0b00;
2361 let Inst{23-22} = opc;
2363 let Inst{20-16} = offset;
2364 let Inst{15-13} = extend{3-1};
2366 let Inst{12} = extend{0};
2367 let Inst{11-10} = 0b10;
2368 let Inst{9-5} = base;
2369 let Inst{4-0} = dst;
2371 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2374 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2375 class Load128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2376 string asm, list<dag> pat>
2377 : LoadStore128RO<sz, V, opc, regtype, asm,
2378 (outs regtype:$Rt), (ins ro_indexed128:$addr), pat>,
2379 Sched<[WriteLDIdx, ReadAdrBase]>;
2381 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2382 class Store128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2383 string asm, list<dag> pat>
2384 : LoadStore128RO<sz, V, opc, regtype, asm,
2385 (outs), (ins regtype:$Rt, ro_indexed128:$addr), pat>,
2386 Sched<[WriteSTIdx, ReadAdrBase]>;
2388 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2389 class PrefetchRO<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2390 : I<(outs), (ins prfop:$Rt, ro_indexed64:$addr), asm,
2391 "\t$Rt, $addr", "", pat>,
2393 // The operands are in order to match the 'addr' MI operands, so we
2394 // don't need an encoder method and by-name matching. Just use the default
2395 // in-order handling. Since we're using by-order, make sure the names
2401 let Inst{31-30} = sz;
2402 let Inst{29-27} = 0b111;
2404 let Inst{25-24} = 0b00;
2405 let Inst{23-22} = opc;
2407 let Inst{20-16} = offset;
2408 let Inst{15-13} = extend{3-1};
2410 let Inst{12} = extend{0};
2411 let Inst{11-10} = 0b10;
2412 let Inst{9-5} = base;
2413 let Inst{4-0} = dst;
2415 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2419 // Load/store unscaled immediate
2422 def MemoryUnscaledOperand : AsmOperandClass {
2423 let Name = "MemoryUnscaled";
2424 let DiagnosticType = "InvalidMemoryIndexedSImm9";
2426 class am_unscaled_operand : Operand<i64> {
2427 let PrintMethod = "printAMUnscaled";
2428 let ParserMatchClass = MemoryUnscaledOperand;
2429 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2431 class am_unscaled_wb_operand : Operand<i64> {
2432 let PrintMethod = "printAMUnscaledWB";
2433 let ParserMatchClass = MemoryUnscaledOperand;
2434 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2436 def am_unscaled : am_unscaled_operand;
2437 def am_unscaled_wb: am_unscaled_wb_operand;
2438 def am_unscaled8 : am_unscaled_operand,
2439 ComplexPattern<i64, 2, "SelectAddrModeUnscaled8", []>;
2440 def am_unscaled16 : am_unscaled_operand,
2441 ComplexPattern<i64, 2, "SelectAddrModeUnscaled16", []>;
2442 def am_unscaled32 : am_unscaled_operand,
2443 ComplexPattern<i64, 2, "SelectAddrModeUnscaled32", []>;
2444 def am_unscaled64 : am_unscaled_operand,
2445 ComplexPattern<i64, 2, "SelectAddrModeUnscaled64", []>;
2446 def am_unscaled128 : am_unscaled_operand,
2447 ComplexPattern<i64, 2, "SelectAddrModeUnscaled128", []>;
2449 class BaseLoadStoreUnscale<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2450 string asm, list<dag> pattern>
2451 : I<oops, iops, asm, "\t$Rt, $addr", "", pattern> {
2452 // The operands are in order to match the 'addr' MI operands, so we
2453 // don't need an encoder method and by-name matching. Just use the default
2454 // in-order handling. Since we're using by-order, make sure the names
2459 let Inst{31-30} = sz;
2460 let Inst{29-27} = 0b111;
2462 let Inst{25-24} = 0b00;
2463 let Inst{23-22} = opc;
2465 let Inst{20-12} = offset;
2466 let Inst{11-10} = 0b00;
2467 let Inst{9-5} = base;
2468 let Inst{4-0} = dst;
2470 let DecoderMethod = "DecodeSignedLdStInstruction";
2473 let AddedComplexity = 1 in // try this before LoadUI
2474 class LoadUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2475 Operand amtype, string asm, list<dag> pattern>
2476 : BaseLoadStoreUnscale<sz, V, opc, (outs regtype:$Rt),
2477 (ins amtype:$addr), asm, pattern>,
2480 let AddedComplexity = 1 in // try this before StoreUI
2481 class StoreUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2482 Operand amtype, string asm, list<dag> pattern>
2483 : BaseLoadStoreUnscale<sz, V, opc, (outs),
2484 (ins regtype:$Rt, amtype:$addr), asm, pattern>,
2487 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2488 class PrefetchUnscaled<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2489 : BaseLoadStoreUnscale<sz, V, opc, (outs),
2490 (ins prfop:$Rt, am_unscaled:$addr), asm, pat>,
2494 // Load/store unscaled immediate, unprivileged
2497 class BaseLoadStoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
2498 dag oops, dag iops, string asm>
2499 : I<oops, iops, asm, "\t$Rt, $addr", "", []> {
2500 // The operands are in order to match the 'addr' MI operands, so we
2501 // don't need an encoder method and by-name matching. Just use the default
2502 // in-order handling. Since we're using by-order, make sure the names
2507 let Inst{31-30} = sz;
2508 let Inst{29-27} = 0b111;
2510 let Inst{25-24} = 0b00;
2511 let Inst{23-22} = opc;
2513 let Inst{20-12} = offset;
2514 let Inst{11-10} = 0b10;
2515 let Inst{9-5} = base;
2516 let Inst{4-0} = dst;
2518 let DecoderMethod = "DecodeSignedLdStInstruction";
2521 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in {
2522 class LoadUnprivileged<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2524 : BaseLoadStoreUnprivileged<sz, V, opc,
2525 (outs regtype:$Rt), (ins am_unscaled:$addr), asm>,
2529 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in {
2530 class StoreUnprivileged<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2532 : BaseLoadStoreUnprivileged<sz, V, opc,
2533 (outs), (ins regtype:$Rt, am_unscaled:$addr), asm>,
2538 // Load/store pre-indexed
2541 class BaseLoadStorePreIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2542 string asm, string cstr>
2543 : I<oops, iops, asm, "\t$Rt, $addr!", cstr, []> {
2544 // The operands are in order to match the 'addr' MI operands, so we
2545 // don't need an encoder method and by-name matching. Just use the default
2546 // in-order handling.
2550 let Inst{31-30} = sz;
2551 let Inst{29-27} = 0b111;
2553 let Inst{25-24} = 0;
2554 let Inst{23-22} = opc;
2556 let Inst{20-12} = offset;
2557 let Inst{11-10} = 0b11;
2558 let Inst{9-5} = base;
2559 let Inst{4-0} = dst;
2561 let DecoderMethod = "DecodeSignedLdStInstruction";
2564 let hasSideEffects = 0 in {
2565 let mayStore = 0, mayLoad = 1 in
2566 // FIXME: Modeling the write-back of these instructions for isel is tricky.
2567 // we need the complex addressing mode for the memory reference, but
2568 // we also need the write-back specified as a tied operand to the
2569 // base register. That combination does not play nicely with
2570 // the asm matcher and friends.
2571 class LoadPreIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2573 : BaseLoadStorePreIdx<sz, V, opc,
2574 (outs regtype:$Rt/*, GPR64sp:$wback*/),
2575 (ins am_unscaled_wb:$addr), asm, ""/*"$addr.base = $wback"*/>,
2576 Sched<[WriteLD, WriteAdr]>;
2578 let mayStore = 1, mayLoad = 0 in
2579 class StorePreIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2581 : BaseLoadStorePreIdx<sz, V, opc,
2582 (outs/* GPR64sp:$wback*/),
2583 (ins regtype:$Rt, am_unscaled_wb:$addr),
2584 asm, ""/*"$addr.base = $wback"*/>,
2585 Sched<[WriteAdr, WriteST]>;
2586 } // hasSideEffects = 0
2588 // ISel pseudo-instructions which have the tied operands. When the MC lowering
2589 // logic finally gets smart enough to strip off tied operands that are just
2590 // for isel convenience, we can get rid of these pseudos and just reference
2591 // the real instructions directly.
2593 // Ironically, also because of the writeback operands, we can't put the
2594 // matcher pattern directly on the instruction, but need to define it
2597 // Loads aren't matched with patterns here at all, but rather in C++
2599 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in {
2600 class LoadPreIdxPseudo<RegisterClass regtype>
2601 : Pseudo<(outs regtype:$Rt, GPR64sp:$wback),
2602 (ins am_noindex:$addr, simm9:$offset), [],
2603 "$addr.base = $wback,@earlyclobber $wback">,
2604 Sched<[WriteLD, WriteAdr]>;
2605 class LoadPostIdxPseudo<RegisterClass regtype>
2606 : Pseudo<(outs regtype:$Rt, GPR64sp:$wback),
2607 (ins am_noindex:$addr, simm9:$offset), [],
2608 "$addr.base = $wback,@earlyclobber $wback">,
2609 Sched<[WriteLD, WriteI]>;
2611 multiclass StorePreIdxPseudo<RegisterClass regtype, ValueType Ty,
2612 SDPatternOperator OpNode> {
2613 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
2614 def _isel: Pseudo<(outs GPR64sp:$wback),
2615 (ins regtype:$Rt, am_noindex:$addr, simm9:$offset), [],
2616 "$addr.base = $wback,@earlyclobber $wback">,
2617 Sched<[WriteAdr, WriteST]>;
2619 def : Pat<(OpNode (Ty regtype:$Rt), am_noindex:$addr, simm9:$offset),
2620 (!cast<Instruction>(NAME#_isel) regtype:$Rt, am_noindex:$addr,
2625 // Load/store post-indexed
2628 // (pre-index) load/stores.
2629 class BaseLoadStorePostIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2630 string asm, string cstr>
2631 : I<oops, iops, asm, "\t$Rt, $addr, $idx", cstr, []> {
2632 // The operands are in order to match the 'addr' MI operands, so we
2633 // don't need an encoder method and by-name matching. Just use the default
2634 // in-order handling.
2638 let Inst{31-30} = sz;
2639 let Inst{29-27} = 0b111;
2641 let Inst{25-24} = 0b00;
2642 let Inst{23-22} = opc;
2644 let Inst{20-12} = offset;
2645 let Inst{11-10} = 0b01;
2646 let Inst{9-5} = base;
2647 let Inst{4-0} = dst;
2649 let DecoderMethod = "DecodeSignedLdStInstruction";
2652 let hasSideEffects = 0 in {
2653 let mayStore = 0, mayLoad = 1 in
2654 // FIXME: Modeling the write-back of these instructions for isel is tricky.
2655 // we need the complex addressing mode for the memory reference, but
2656 // we also need the write-back specified as a tied operand to the
2657 // base register. That combination does not play nicely with
2658 // the asm matcher and friends.
2659 class LoadPostIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2661 : BaseLoadStorePostIdx<sz, V, opc,
2662 (outs regtype:$Rt/*, GPR64sp:$wback*/),
2663 (ins am_noindex:$addr, simm9:$idx),
2664 asm, ""/*"$addr.base = $wback"*/>,
2665 Sched<[WriteLD, WriteI]>;
2667 let mayStore = 1, mayLoad = 0 in
2668 class StorePostIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2670 : BaseLoadStorePostIdx<sz, V, opc,
2671 (outs/* GPR64sp:$wback*/),
2672 (ins regtype:$Rt, am_noindex:$addr, simm9:$idx),
2673 asm, ""/*"$addr.base = $wback"*/>,
2674 Sched<[WriteAdr, WriteST, ReadAdrBase]>;
2675 } // hasSideEffects = 0
2677 // ISel pseudo-instructions which have the tied operands. When the MC lowering
2678 // logic finally gets smart enough to strip off tied operands that are just
2679 // for isel convenience, we can get rid of these pseudos and just reference
2680 // the real instructions directly.
2682 // Ironically, also because of the writeback operands, we can't put the
2683 // matcher pattern directly on the instruction, but need to define it
2685 multiclass StorePostIdxPseudo<RegisterClass regtype, ValueType Ty,
2686 SDPatternOperator OpNode, Instruction Insn> {
2687 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
2688 def _isel: Pseudo<(outs GPR64sp:$wback),
2689 (ins regtype:$Rt, am_noindex:$addr, simm9:$idx), [],
2690 "$addr.base = $wback,@earlyclobber $wback">,
2691 PseudoInstExpansion<(Insn regtype:$Rt, am_noindex:$addr, simm9:$idx)>,
2692 Sched<[WriteAdr, WriteST, ReadAdrBase]>;
2694 def : Pat<(OpNode (Ty regtype:$Rt), am_noindex:$addr, simm9:$idx),
2695 (!cast<Instruction>(NAME#_isel) regtype:$Rt, am_noindex:$addr,
2703 // (indexed, offset)
2705 class BaseLoadStorePairOffset<bits<2> opc, bit V, bit L, dag oops, dag iops,
2707 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr", "", []> {
2708 // The operands are in order to match the 'addr' MI operands, so we
2709 // don't need an encoder method and by-name matching. Just use the default
2710 // in-order handling. Since we're using by-order, make sure the names
2716 let Inst{31-30} = opc;
2717 let Inst{29-27} = 0b101;
2719 let Inst{25-23} = 0b010;
2721 let Inst{21-15} = offset;
2722 let Inst{14-10} = dst2;
2723 let Inst{9-5} = base;
2724 let Inst{4-0} = dst;
2726 let DecoderMethod = "DecodePairLdStInstruction";
2729 let hasSideEffects = 0 in {
2730 let mayStore = 0, mayLoad = 1 in
2731 class LoadPairOffset<bits<2> opc, bit V, RegisterClass regtype,
2732 Operand indextype, string asm>
2733 : BaseLoadStorePairOffset<opc, V, 1,
2734 (outs regtype:$Rt, regtype:$Rt2),
2735 (ins indextype:$addr), asm>,
2736 Sched<[WriteLD, WriteLDHi]>;
2738 let mayLoad = 0, mayStore = 1 in
2739 class StorePairOffset<bits<2> opc, bit V, RegisterClass regtype,
2740 Operand indextype, string asm>
2741 : BaseLoadStorePairOffset<opc, V, 0, (outs),
2742 (ins regtype:$Rt, regtype:$Rt2, indextype:$addr),
2745 } // hasSideEffects = 0
2749 def MemoryIndexed32SImm7 : AsmOperandClass {
2750 let Name = "MemoryIndexed32SImm7";
2751 let DiagnosticType = "InvalidMemoryIndexed32SImm7";
2753 def am_indexed32simm7 : Operand<i32> { // ComplexPattern<...>
2754 let PrintMethod = "printAMIndexed32";
2755 let ParserMatchClass = MemoryIndexed32SImm7;
2756 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2758 def am_indexed32simm7_wb : Operand<i32> { // ComplexPattern<...>
2759 let PrintMethod = "printAMIndexed32WB";
2760 let ParserMatchClass = MemoryIndexed32SImm7;
2761 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2764 def MemoryIndexed64SImm7 : AsmOperandClass {
2765 let Name = "MemoryIndexed64SImm7";
2766 let DiagnosticType = "InvalidMemoryIndexed64SImm7";
2768 def am_indexed64simm7 : Operand<i32> { // ComplexPattern<...>
2769 let PrintMethod = "printAMIndexed64";
2770 let ParserMatchClass = MemoryIndexed64SImm7;
2771 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2773 def am_indexed64simm7_wb : Operand<i32> { // ComplexPattern<...>
2774 let PrintMethod = "printAMIndexed64WB";
2775 let ParserMatchClass = MemoryIndexed64SImm7;
2776 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2779 def MemoryIndexed128SImm7 : AsmOperandClass {
2780 let Name = "MemoryIndexed128SImm7";
2781 let DiagnosticType = "InvalidMemoryIndexed128SImm7";
2783 def am_indexed128simm7 : Operand<i32> { // ComplexPattern<...>
2784 let PrintMethod = "printAMIndexed128";
2785 let ParserMatchClass = MemoryIndexed128SImm7;
2786 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2788 def am_indexed128simm7_wb : Operand<i32> { // ComplexPattern<...>
2789 let PrintMethod = "printAMIndexed128WB";
2790 let ParserMatchClass = MemoryIndexed128SImm7;
2791 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2794 class BaseLoadStorePairPreIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
2796 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr!", "", []> {
2797 // The operands are in order to match the 'addr' MI operands, so we
2798 // don't need an encoder method and by-name matching. Just use the default
2799 // in-order handling. Since we're using by-order, make sure the names
2805 let Inst{31-30} = opc;
2806 let Inst{29-27} = 0b101;
2808 let Inst{25-23} = 0b011;
2810 let Inst{21-15} = offset;
2811 let Inst{14-10} = dst2;
2812 let Inst{9-5} = base;
2813 let Inst{4-0} = dst;
2815 let DecoderMethod = "DecodePairLdStInstruction";
2818 let hasSideEffects = 0 in {
2819 let mayStore = 0, mayLoad = 1 in
2820 class LoadPairPreIdx<bits<2> opc, bit V, RegisterClass regtype,
2821 Operand addrmode, string asm>
2822 : BaseLoadStorePairPreIdx<opc, V, 1,
2823 (outs regtype:$Rt, regtype:$Rt2),
2824 (ins addrmode:$addr), asm>,
2825 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
2827 let mayStore = 1, mayLoad = 0 in
2828 class StorePairPreIdx<bits<2> opc, bit V, RegisterClass regtype,
2829 Operand addrmode, string asm>
2830 : BaseLoadStorePairPreIdx<opc, V, 0, (outs),
2831 (ins regtype:$Rt, regtype:$Rt2, addrmode:$addr),
2833 Sched<[WriteAdr, WriteSTP]>;
2834 } // hasSideEffects = 0
2838 class BaseLoadStorePairPostIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
2840 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr, $idx", "", []> {
2841 // The operands are in order to match the 'addr' MI operands, so we
2842 // don't need an encoder method and by-name matching. Just use the default
2843 // in-order handling. Since we're using by-order, make sure the names
2849 let Inst{31-30} = opc;
2850 let Inst{29-27} = 0b101;
2852 let Inst{25-23} = 0b001;
2854 let Inst{21-15} = offset;
2855 let Inst{14-10} = dst2;
2856 let Inst{9-5} = base;
2857 let Inst{4-0} = dst;
2859 let DecoderMethod = "DecodePairLdStInstruction";
2862 let hasSideEffects = 0 in {
2863 let mayStore = 0, mayLoad = 1 in
2864 class LoadPairPostIdx<bits<2> opc, bit V, RegisterClass regtype,
2865 Operand idxtype, string asm>
2866 : BaseLoadStorePairPostIdx<opc, V, 1,
2867 (outs regtype:$Rt, regtype:$Rt2),
2868 (ins am_noindex:$addr, idxtype:$idx), asm>,
2869 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
2871 let mayStore = 1, mayLoad = 0 in
2872 class StorePairPostIdx<bits<2> opc, bit V, RegisterClass regtype,
2873 Operand idxtype, string asm>
2874 : BaseLoadStorePairPostIdx<opc, V, 0, (outs),
2875 (ins regtype:$Rt, regtype:$Rt2,
2876 am_noindex:$addr, idxtype:$idx),
2878 Sched<[WriteAdr, WriteSTP]>;
2879 } // hasSideEffects = 0
2883 class BaseLoadStorePairNoAlloc<bits<2> opc, bit V, bit L, dag oops, dag iops,
2885 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr", "", []> {
2886 // The operands are in order to match the 'addr' MI operands, so we
2887 // don't need an encoder method and by-name matching. Just use the default
2888 // in-order handling. Since we're using by-order, make sure the names
2894 let Inst{31-30} = opc;
2895 let Inst{29-27} = 0b101;
2897 let Inst{25-23} = 0b000;
2899 let Inst{21-15} = offset;
2900 let Inst{14-10} = dst2;
2901 let Inst{9-5} = base;
2902 let Inst{4-0} = dst;
2904 let DecoderMethod = "DecodePairLdStInstruction";
2907 let hasSideEffects = 0 in {
2908 let mayStore = 0, mayLoad = 1 in
2909 class LoadPairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
2910 Operand indextype, string asm>
2911 : BaseLoadStorePairNoAlloc<opc, V, 1,
2912 (outs regtype:$Rt, regtype:$Rt2),
2913 (ins indextype:$addr), asm>,
2914 Sched<[WriteLD, WriteLDHi]>;
2916 let mayStore = 1, mayLoad = 0 in
2917 class StorePairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
2918 Operand indextype, string asm>
2919 : BaseLoadStorePairNoAlloc<opc, V, 0, (outs),
2920 (ins regtype:$Rt, regtype:$Rt2, indextype:$addr),
2923 } // hasSideEffects = 0
2926 // Load/store exclusive
2929 // True exclusive operations write to and/or read from the system's exclusive
2930 // monitors, which as far as a compiler is concerned can be modelled as a
2931 // random shared memory address. Hence LoadExclusive mayStore.
2933 // Since these instructions have the undefined register bits set to 1 in
2934 // their canonical form, we need a post encoder method to set those bits
2935 // to 1 when encoding these instructions. We do this using the
2936 // fixLoadStoreExclusive function. This function has template parameters:
2938 // fixLoadStoreExclusive<int hasRs, int hasRt2>
2940 // hasRs indicates that the instruction uses the Rs field, so we won't set
2941 // it to 1 (and the same for Rt2). We don't need template parameters for
2942 // the other register fields since Rt and Rn are always used.
2944 let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
2945 class BaseLoadStoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2946 dag oops, dag iops, string asm, string operands>
2947 : I<oops, iops, asm, operands, "", []> {
2948 let Inst{31-30} = sz;
2949 let Inst{29-24} = 0b001000;
2955 let DecoderMethod = "DecodeExclusiveLdStInstruction";
2958 // Neither Rs nor Rt2 operands.
2959 class LoadStoreExclusiveSimple<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2960 dag oops, dag iops, string asm, string operands>
2961 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, oops, iops, asm, operands> {
2964 let Inst{9-5} = base;
2965 let Inst{4-0} = reg;
2967 let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
2970 // Simple load acquires don't set the exclusive monitor
2971 let mayLoad = 1, mayStore = 0 in
2972 class LoadAcquire<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2973 RegisterClass regtype, string asm>
2974 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
2975 (ins am_noindex:$addr), asm, "\t$Rt, $addr">,
2978 class LoadExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2979 RegisterClass regtype, string asm>
2980 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
2981 (ins am_noindex:$addr), asm, "\t$Rt, $addr">,
2984 class LoadExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2985 RegisterClass regtype, string asm>
2986 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
2987 (outs regtype:$Rt, regtype:$Rt2),
2988 (ins am_noindex:$addr), asm,
2989 "\t$Rt, $Rt2, $addr">,
2990 Sched<[WriteLD, WriteLDHi]> {
2994 let Inst{14-10} = dst2;
2995 let Inst{9-5} = base;
2996 let Inst{4-0} = dst1;
2998 let PostEncoderMethod = "fixLoadStoreExclusive<0,1>";
3001 // Simple store release operations do not check the exclusive monitor.
3002 let mayLoad = 0, mayStore = 1 in
3003 class StoreRelease<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3004 RegisterClass regtype, string asm>
3005 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs),
3006 (ins regtype:$Rt, am_noindex:$addr),
3007 asm, "\t$Rt, $addr">,
3010 let mayLoad = 1, mayStore = 1 in
3011 class StoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3012 RegisterClass regtype, string asm>
3013 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, (outs GPR32:$Ws),
3014 (ins regtype:$Rt, am_noindex:$addr),
3015 asm, "\t$Ws, $Rt, $addr">,
3020 let Inst{20-16} = status;
3021 let Inst{9-5} = base;
3022 let Inst{4-0} = reg;
3024 let Constraints = "@earlyclobber $Ws";
3025 let PostEncoderMethod = "fixLoadStoreExclusive<1,0>";
3028 class StoreExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3029 RegisterClass regtype, string asm>
3030 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3032 (ins regtype:$Rt, regtype:$Rt2, am_noindex:$addr),
3033 asm, "\t$Ws, $Rt, $Rt2, $addr">,
3039 let Inst{20-16} = status;
3040 let Inst{14-10} = dst2;
3041 let Inst{9-5} = base;
3042 let Inst{4-0} = dst1;
3044 let Constraints = "@earlyclobber $Ws";
3048 // Exception generation
3051 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3052 class ExceptionGeneration<bits<3> op1, bits<2> ll, string asm>
3053 : I<(outs), (ins imm0_65535:$imm), asm, "\t$imm", "", []>,
3056 let Inst{31-24} = 0b11010100;
3057 let Inst{23-21} = op1;
3058 let Inst{20-5} = imm;
3059 let Inst{4-2} = 0b000;
3064 // Floating point to integer conversion
3067 class BaseFPToIntegerUnscaled<bits<2> type, bits<2> rmode, bits<3> opcode,
3068 RegisterClass srcType, RegisterClass dstType,
3069 string asm, list<dag> pattern>
3070 : I<(outs dstType:$Rd), (ins srcType:$Rn),
3071 asm, "\t$Rd, $Rn", "", pattern>,
3072 Sched<[WriteFCvt]> {
3075 let Inst{30-29} = 0b00;
3076 let Inst{28-24} = 0b11110;
3077 let Inst{23-22} = type;
3079 let Inst{20-19} = rmode;
3080 let Inst{18-16} = opcode;
3081 let Inst{15-10} = 0;
3086 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3087 class BaseFPToInteger<bits<2> type, bits<2> rmode, bits<3> opcode,
3088 RegisterClass srcType, RegisterClass dstType,
3089 Operand immType, string asm>
3090 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
3091 asm, "\t$Rd, $Rn, $scale", "", []>,
3092 Sched<[WriteFCvt]> {
3096 let Inst{30-29} = 0b00;
3097 let Inst{28-24} = 0b11110;
3098 let Inst{23-22} = type;
3100 let Inst{20-19} = rmode;
3101 let Inst{18-16} = opcode;
3102 let Inst{15-10} = scale;
3107 multiclass FPToIntegerUnscaled<bits<2> rmode, bits<3> opcode, string asm,
3108 SDPatternOperator OpN> {
3109 // Unscaled single-precision to 32-bit
3110 def UWSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR32, asm,
3111 [(set GPR32:$Rd, (OpN FPR32:$Rn))]> {
3112 let Inst{31} = 0; // 32-bit GPR flag
3115 // Unscaled single-precision to 64-bit
3116 def UXSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR64, asm,
3117 [(set GPR64:$Rd, (OpN FPR32:$Rn))]> {
3118 let Inst{31} = 1; // 64-bit GPR flag
3121 // Unscaled double-precision to 32-bit
3122 def UWDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR32, asm,
3123 [(set GPR32:$Rd, (OpN (f64 FPR64:$Rn)))]> {
3124 let Inst{31} = 0; // 32-bit GPR flag
3127 // Unscaled double-precision to 64-bit
3128 def UXDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR64, asm,
3129 [(set GPR64:$Rd, (OpN (f64 FPR64:$Rn)))]> {
3130 let Inst{31} = 1; // 64-bit GPR flag
3134 multiclass FPToIntegerScaled<bits<2> rmode, bits<3> opcode, string asm,
3135 SDPatternOperator OpN> {
3136 // Scaled single-precision to 32-bit
3137 def SWSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR32,
3138 fixedpoint32, asm> {
3139 let Inst{31} = 0; // 32-bit GPR flag
3143 // Scaled single-precision to 64-bit
3144 def SXSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR64,
3145 fixedpoint64, asm> {
3146 let Inst{31} = 1; // 64-bit GPR flag
3149 // Scaled double-precision to 32-bit
3150 def SWDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR32,
3151 fixedpoint32, asm> {
3152 let Inst{31} = 0; // 32-bit GPR flag
3156 // Scaled double-precision to 64-bit
3157 def SXDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR64,
3158 fixedpoint64, asm> {
3159 let Inst{31} = 1; // 64-bit GPR flag
3164 // Integer to floating point conversion
3167 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
3168 class BaseIntegerToFP<bit isUnsigned,
3169 RegisterClass srcType, RegisterClass dstType,
3170 Operand immType, string asm>
3171 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
3172 asm, "\t$Rd, $Rn, $scale", "", []>,
3173 Sched<[WriteFCvt]> {
3177 let Inst{30-23} = 0b00111100;
3178 let Inst{21-17} = 0b00001;
3179 let Inst{16} = isUnsigned;
3180 let Inst{15-10} = scale;
3185 class BaseIntegerToFPUnscaled<bit isUnsigned,
3186 RegisterClass srcType, RegisterClass dstType,
3187 ValueType dvt, string asm, SDNode node>
3188 : I<(outs dstType:$Rd), (ins srcType:$Rn),
3189 asm, "\t$Rd, $Rn", "", [(set (dvt dstType:$Rd), (node srcType:$Rn))]>,
3190 Sched<[WriteFCvt]> {
3194 let Inst{30-23} = 0b00111100;
3195 let Inst{21-17} = 0b10001;
3196 let Inst{16} = isUnsigned;
3197 let Inst{15-10} = 0b000000;
3202 multiclass IntegerToFP<bit isUnsigned, string asm, SDNode node> {
3204 def UWSri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR32, f32, asm, node> {
3205 let Inst{31} = 0; // 32-bit GPR flag
3206 let Inst{22} = 0; // 32-bit FPR flag
3209 def UWDri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR64, f64, asm, node> {
3210 let Inst{31} = 0; // 32-bit GPR flag
3211 let Inst{22} = 1; // 64-bit FPR flag
3214 def UXSri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR32, f32, asm, node> {
3215 let Inst{31} = 1; // 64-bit GPR flag
3216 let Inst{22} = 0; // 32-bit FPR flag
3219 def UXDri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR64, f64, asm, node> {
3220 let Inst{31} = 1; // 64-bit GPR flag
3221 let Inst{22} = 1; // 64-bit FPR flag
3225 def SWSri: BaseIntegerToFP<isUnsigned, GPR32, FPR32, fixedpoint32, asm> {
3226 let Inst{31} = 0; // 32-bit GPR flag
3227 let Inst{22} = 0; // 32-bit FPR flag
3231 def SWDri: BaseIntegerToFP<isUnsigned, GPR32, FPR64, fixedpoint32, asm> {
3232 let Inst{31} = 0; // 32-bit GPR flag
3233 let Inst{22} = 1; // 64-bit FPR flag
3237 def SXSri: BaseIntegerToFP<isUnsigned, GPR64, FPR32, fixedpoint64, asm> {
3238 let Inst{31} = 1; // 64-bit GPR flag
3239 let Inst{22} = 0; // 32-bit FPR flag
3242 def SXDri: BaseIntegerToFP<isUnsigned, GPR64, FPR64, fixedpoint64, asm> {
3243 let Inst{31} = 1; // 64-bit GPR flag
3244 let Inst{22} = 1; // 64-bit FPR flag
3249 // Unscaled integer <-> floating point conversion (i.e. FMOV)
3252 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3253 class BaseUnscaledConversion<bits<2> rmode, bits<3> opcode,
3254 RegisterClass srcType, RegisterClass dstType,
3256 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "",
3257 // We use COPY_TO_REGCLASS for these bitconvert operations.
3258 // copyPhysReg() expands the resultant COPY instructions after
3259 // regalloc is done. This gives greater freedom for the allocator
3260 // and related passes (coalescing, copy propagation, et. al.) to
3261 // be more effective.
3262 [/*(set (dvt dstType:$Rd), (bitconvert (svt srcType:$Rn)))*/]>,
3263 Sched<[WriteFCopy]> {
3266 let Inst{30-23} = 0b00111100;
3268 let Inst{20-19} = rmode;
3269 let Inst{18-16} = opcode;
3270 let Inst{15-10} = 0b000000;
3275 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3276 class BaseUnscaledConversionToHigh<bits<2> rmode, bits<3> opcode,
3277 RegisterClass srcType, RegisterOperand dstType, string asm,
3279 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm,
3280 "{\t$Rd"#kind#"[1], $Rn|"#kind#"\t$Rd[1], $Rn}", "", []>,
3281 Sched<[WriteFCopy]> {
3284 let Inst{30-23} = 0b00111101;
3286 let Inst{20-19} = rmode;
3287 let Inst{18-16} = opcode;
3288 let Inst{15-10} = 0b000000;
3293 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3294 class BaseUnscaledConversionFromHigh<bits<2> rmode, bits<3> opcode,
3295 RegisterOperand srcType, RegisterClass dstType, string asm,
3297 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm,
3298 "{\t$Rd, $Rn"#kind#"[1]|"#kind#"\t$Rd, $Rn[1]}", "", []>,
3299 Sched<[WriteFCopy]> {
3302 let Inst{30-23} = 0b00111101;
3304 let Inst{20-19} = rmode;
3305 let Inst{18-16} = opcode;
3306 let Inst{15-10} = 0b000000;
3313 multiclass UnscaledConversion<string asm> {
3314 def WSr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR32, asm> {
3315 let Inst{31} = 0; // 32-bit GPR flag
3316 let Inst{22} = 0; // 32-bit FPR flag
3319 def XDr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR64, asm> {
3320 let Inst{31} = 1; // 64-bit GPR flag
3321 let Inst{22} = 1; // 64-bit FPR flag
3324 def SWr : BaseUnscaledConversion<0b00, 0b110, FPR32, GPR32, asm> {
3325 let Inst{31} = 0; // 32-bit GPR flag
3326 let Inst{22} = 0; // 32-bit FPR flag
3329 def DXr : BaseUnscaledConversion<0b00, 0b110, FPR64, GPR64, asm> {
3330 let Inst{31} = 1; // 64-bit GPR flag
3331 let Inst{22} = 1; // 64-bit FPR flag
3334 def XDHighr : BaseUnscaledConversionToHigh<0b01, 0b111, GPR64, V128,
3340 def DXHighr : BaseUnscaledConversionFromHigh<0b01, 0b110, V128, GPR64,
3348 // Floating point conversion
3351 class BaseFPConversion<bits<2> type, bits<2> opcode, RegisterClass dstType,
3352 RegisterClass srcType, string asm, list<dag> pattern>
3353 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "", pattern>,
3354 Sched<[WriteFCvt]> {
3357 let Inst{31-24} = 0b00011110;
3358 let Inst{23-22} = type;
3359 let Inst{21-17} = 0b10001;
3360 let Inst{16-15} = opcode;
3361 let Inst{14-10} = 0b10000;
3366 multiclass FPConversion<string asm> {
3367 // Double-precision to Half-precision
3368 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3369 def HDr : BaseFPConversion<0b01, 0b11, FPR16, FPR64, asm, []>;
3371 // Double-precision to Single-precision
3372 def SDr : BaseFPConversion<0b01, 0b00, FPR32, FPR64, asm,
3373 [(set FPR32:$Rd, (fround FPR64:$Rn))]>;
3375 // Half-precision to Double-precision
3376 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3377 def DHr : BaseFPConversion<0b11, 0b01, FPR64, FPR16, asm, []>;
3379 // Half-precision to Single-precision
3380 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3381 def SHr : BaseFPConversion<0b11, 0b00, FPR32, FPR16, asm, []>;
3383 // Single-precision to Double-precision
3384 def DSr : BaseFPConversion<0b00, 0b01, FPR64, FPR32, asm,
3385 [(set FPR64:$Rd, (fextend FPR32:$Rn))]>;
3387 // Single-precision to Half-precision
3388 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3389 def HSr : BaseFPConversion<0b00, 0b11, FPR16, FPR32, asm, []>;
3393 // Single operand floating point data processing
3396 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3397 class BaseSingleOperandFPData<bits<4> opcode, RegisterClass regtype,
3398 ValueType vt, string asm, SDPatternOperator node>
3399 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
3400 [(set (vt regtype:$Rd), (node (vt regtype:$Rn)))]>,
3404 let Inst{31-23} = 0b000111100;
3405 let Inst{21-19} = 0b100;
3406 let Inst{18-15} = opcode;
3407 let Inst{14-10} = 0b10000;
3412 multiclass SingleOperandFPData<bits<4> opcode, string asm,
3413 SDPatternOperator node = null_frag> {
3414 def Sr : BaseSingleOperandFPData<opcode, FPR32, f32, asm, node> {
3415 let Inst{22} = 0; // 32-bit size flag
3418 def Dr : BaseSingleOperandFPData<opcode, FPR64, f64, asm, node> {
3419 let Inst{22} = 1; // 64-bit size flag
3424 // Two operand floating point data processing
3427 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3428 class BaseTwoOperandFPData<bits<4> opcode, RegisterClass regtype,
3429 string asm, list<dag> pat>
3430 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
3431 asm, "\t$Rd, $Rn, $Rm", "", pat>,
3436 let Inst{31-23} = 0b000111100;
3438 let Inst{20-16} = Rm;
3439 let Inst{15-12} = opcode;
3440 let Inst{11-10} = 0b10;
3445 multiclass TwoOperandFPData<bits<4> opcode, string asm,
3446 SDPatternOperator node = null_frag> {
3447 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
3448 [(set (f32 FPR32:$Rd),
3449 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]> {
3450 let Inst{22} = 0; // 32-bit size flag
3453 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
3454 [(set (f64 FPR64:$Rd),
3455 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]> {
3456 let Inst{22} = 1; // 64-bit size flag
3460 multiclass TwoOperandFPDataNeg<bits<4> opcode, string asm, SDNode node> {
3461 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
3462 [(set FPR32:$Rd, (fneg (node FPR32:$Rn, (f32 FPR32:$Rm))))]> {
3463 let Inst{22} = 0; // 32-bit size flag
3466 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
3467 [(set FPR64:$Rd, (fneg (node FPR64:$Rn, (f64 FPR64:$Rm))))]> {
3468 let Inst{22} = 1; // 64-bit size flag
3474 // Three operand floating point data processing
3477 class BaseThreeOperandFPData<bit isNegated, bit isSub,
3478 RegisterClass regtype, string asm, list<dag> pat>
3479 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, regtype: $Ra),
3480 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pat>,
3481 Sched<[WriteFMul]> {
3486 let Inst{31-23} = 0b000111110;
3487 let Inst{21} = isNegated;
3488 let Inst{20-16} = Rm;
3489 let Inst{15} = isSub;
3490 let Inst{14-10} = Ra;
3495 multiclass ThreeOperandFPData<bit isNegated, bit isSub,string asm,
3496 SDPatternOperator node> {
3497 def Srrr : BaseThreeOperandFPData<isNegated, isSub, FPR32, asm,
3499 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm), (f32 FPR32:$Ra)))]> {
3500 let Inst{22} = 0; // 32-bit size flag
3503 def Drrr : BaseThreeOperandFPData<isNegated, isSub, FPR64, asm,
3505 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm), (f64 FPR64:$Ra)))]> {
3506 let Inst{22} = 1; // 64-bit size flag
3511 // Floating point data comparisons
3514 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3515 class BaseOneOperandFPComparison<bit signalAllNans,
3516 RegisterClass regtype, string asm,
3518 : I<(outs), (ins regtype:$Rn), asm, "\t$Rn, #0.0", "", pat>,
3519 Sched<[WriteFCmp]> {
3521 let Inst{31-23} = 0b000111100;
3524 let Inst{15-10} = 0b001000;
3526 let Inst{4} = signalAllNans;
3527 let Inst{3-0} = 0b1000;
3529 // Rm should be 0b00000 canonically, but we need to accept any value.
3530 let PostEncoderMethod = "fixOneOperandFPComparison";
3533 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3534 class BaseTwoOperandFPComparison<bit signalAllNans, RegisterClass regtype,
3535 string asm, list<dag> pat>
3536 : I<(outs), (ins regtype:$Rn, regtype:$Rm), asm, "\t$Rn, $Rm", "", pat>,
3537 Sched<[WriteFCmp]> {
3540 let Inst{31-23} = 0b000111100;
3542 let Inst{20-16} = Rm;
3543 let Inst{15-10} = 0b001000;
3545 let Inst{4} = signalAllNans;
3546 let Inst{3-0} = 0b0000;
3549 multiclass FPComparison<bit signalAllNans, string asm,
3550 SDPatternOperator OpNode = null_frag> {
3551 let Defs = [CPSR] in {
3552 def Srr : BaseTwoOperandFPComparison<signalAllNans, FPR32, asm,
3553 [(OpNode FPR32:$Rn, (f32 FPR32:$Rm)), (implicit CPSR)]> {
3557 def Sri : BaseOneOperandFPComparison<signalAllNans, FPR32, asm,
3558 [(OpNode (f32 FPR32:$Rn), fpimm0), (implicit CPSR)]> {
3562 def Drr : BaseTwoOperandFPComparison<signalAllNans, FPR64, asm,
3563 [(OpNode FPR64:$Rn, (f64 FPR64:$Rm)), (implicit CPSR)]> {
3567 def Dri : BaseOneOperandFPComparison<signalAllNans, FPR64, asm,
3568 [(OpNode (f64 FPR64:$Rn), fpimm0), (implicit CPSR)]> {
3575 // Floating point conditional comparisons
3578 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3579 class BaseFPCondComparison<bit signalAllNans,
3580 RegisterClass regtype, string asm>
3581 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm0_15:$nzcv, ccode:$cond),
3582 asm, "\t$Rn, $Rm, $nzcv, $cond", "", []>,
3583 Sched<[WriteFCmp]> {
3589 let Inst{31-23} = 0b000111100;
3591 let Inst{20-16} = Rm;
3592 let Inst{15-12} = cond;
3593 let Inst{11-10} = 0b01;
3595 let Inst{4} = signalAllNans;
3596 let Inst{3-0} = nzcv;
3599 multiclass FPCondComparison<bit signalAllNans, string asm> {
3600 let Defs = [CPSR], Uses = [CPSR] in {
3601 def Srr : BaseFPCondComparison<signalAllNans, FPR32, asm> {
3605 def Drr : BaseFPCondComparison<signalAllNans, FPR64, asm> {
3608 } // Defs = [CPSR], Uses = [CPSR]
3612 // Floating point conditional select
3615 class BaseFPCondSelect<RegisterClass regtype, ValueType vt, string asm>
3616 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
3617 asm, "\t$Rd, $Rn, $Rm, $cond", "",
3619 (ARM64csel (vt regtype:$Rn), regtype:$Rm,
3620 (i32 imm:$cond), CPSR))]>,
3627 let Inst{31-23} = 0b000111100;
3629 let Inst{20-16} = Rm;
3630 let Inst{15-12} = cond;
3631 let Inst{11-10} = 0b11;
3636 multiclass FPCondSelect<string asm> {
3637 let Uses = [CPSR] in {
3638 def Srrr : BaseFPCondSelect<FPR32, f32, asm> {
3642 def Drrr : BaseFPCondSelect<FPR64, f64, asm> {
3649 // Floating move immediate
3652 class BaseFPMoveImmediate<RegisterClass regtype, Operand fpimmtype, string asm>
3653 : I<(outs regtype:$Rd), (ins fpimmtype:$imm), asm, "\t$Rd, $imm", "",
3654 [(set regtype:$Rd, fpimmtype:$imm)]>,
3655 Sched<[WriteFImm]> {
3658 let Inst{31-23} = 0b000111100;
3660 let Inst{20-13} = imm;
3661 let Inst{12-5} = 0b10000000;
3665 multiclass FPMoveImmediate<string asm> {
3666 def Si : BaseFPMoveImmediate<FPR32, fpimm32, asm> {
3670 def Di : BaseFPMoveImmediate<FPR64, fpimm64, asm> {
3675 //----------------------------------------------------------------------------
3677 //----------------------------------------------------------------------------
3679 def VectorIndexBOperand : AsmOperandClass { let Name = "VectorIndexB"; }
3680 def VectorIndexHOperand : AsmOperandClass { let Name = "VectorIndexH"; }
3681 def VectorIndexSOperand : AsmOperandClass { let Name = "VectorIndexS"; }
3682 def VectorIndexDOperand : AsmOperandClass { let Name = "VectorIndexD"; }
3683 def VectorIndexB : Operand<i64>, ImmLeaf<i64, [{
3684 return ((uint64_t)Imm) < 16;
3686 let ParserMatchClass = VectorIndexBOperand;
3687 let PrintMethod = "printVectorIndex";
3688 let MIOperandInfo = (ops i64imm);
3690 def VectorIndexH : Operand<i64>, ImmLeaf<i64, [{
3691 return ((uint64_t)Imm) < 8;
3693 let ParserMatchClass = VectorIndexHOperand;
3694 let PrintMethod = "printVectorIndex";
3695 let MIOperandInfo = (ops i64imm);
3697 def VectorIndexS : Operand<i64>, ImmLeaf<i64, [{
3698 return ((uint64_t)Imm) < 4;
3700 let ParserMatchClass = VectorIndexSOperand;
3701 let PrintMethod = "printVectorIndex";
3702 let MIOperandInfo = (ops i64imm);
3704 def VectorIndexD : Operand<i64>, ImmLeaf<i64, [{
3705 return ((uint64_t)Imm) < 2;
3707 let ParserMatchClass = VectorIndexDOperand;
3708 let PrintMethod = "printVectorIndex";
3709 let MIOperandInfo = (ops i64imm);
3712 //----------------------------------------------------------------------------
3713 // AdvSIMD three register vector instructions
3714 //----------------------------------------------------------------------------
3716 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3717 class BaseSIMDThreeSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
3718 RegisterOperand regtype, string asm, string kind,
3720 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
3721 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
3722 "|" # kind # "\t$Rd, $Rn, $Rm|}", "", pattern>,
3730 let Inst{28-24} = 0b01110;
3731 let Inst{23-22} = size;
3733 let Inst{20-16} = Rm;
3734 let Inst{15-11} = opcode;
3740 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3741 class BaseSIMDThreeSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
3742 RegisterOperand regtype, string asm, string kind,
3744 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn, regtype:$Rm), asm,
3745 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
3746 "|" # kind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
3754 let Inst{28-24} = 0b01110;
3755 let Inst{23-22} = size;
3757 let Inst{20-16} = Rm;
3758 let Inst{15-11} = opcode;
3764 // All operand sizes distinguished in the encoding.
3765 multiclass SIMDThreeSameVector<bit U, bits<5> opc, string asm,
3766 SDPatternOperator OpNode> {
3767 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3769 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3770 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3772 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3773 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3775 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3776 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3778 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3779 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3781 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3782 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3784 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3785 def v2i64 : BaseSIMDThreeSameVector<1, U, 0b11, opc, V128,
3787 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
3790 // As above, but D sized elements unsupported.
3791 multiclass SIMDThreeSameVectorBHS<bit U, bits<5> opc, string asm,
3792 SDPatternOperator OpNode> {
3793 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3795 [(set V64:$Rd, (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))]>;
3796 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3798 [(set V128:$Rd, (v16i8 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm))))]>;
3799 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3801 [(set V64:$Rd, (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))]>;
3802 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3804 [(set V128:$Rd, (v8i16 (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm))))]>;
3805 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3807 [(set V64:$Rd, (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))]>;
3808 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3810 [(set V128:$Rd, (v4i32 (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm))))]>;
3813 multiclass SIMDThreeSameVectorBHSTied<bit U, bits<5> opc, string asm,
3814 SDPatternOperator OpNode> {
3815 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, 0b00, opc, V64,
3817 [(set (v8i8 V64:$dst),
3818 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3819 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, 0b00, opc, V128,
3821 [(set (v16i8 V128:$dst),
3822 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3823 def v4i16 : BaseSIMDThreeSameVectorTied<0, U, 0b01, opc, V64,
3825 [(set (v4i16 V64:$dst),
3826 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3827 def v8i16 : BaseSIMDThreeSameVectorTied<1, U, 0b01, opc, V128,
3829 [(set (v8i16 V128:$dst),
3830 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3831 def v2i32 : BaseSIMDThreeSameVectorTied<0, U, 0b10, opc, V64,
3833 [(set (v2i32 V64:$dst),
3834 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3835 def v4i32 : BaseSIMDThreeSameVectorTied<1, U, 0b10, opc, V128,
3837 [(set (v4i32 V128:$dst),
3838 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3841 // As above, but only B sized elements supported.
3842 multiclass SIMDThreeSameVectorB<bit U, bits<5> opc, string asm,
3843 SDPatternOperator OpNode> {
3844 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3846 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3847 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3849 [(set (v16i8 V128:$Rd),
3850 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3853 // As above, but only S and D sized floating point elements supported.
3854 multiclass SIMDThreeSameVectorFP<bit U, bit S, bits<5> opc,
3855 string asm, SDPatternOperator OpNode> {
3856 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0}, opc, V64,
3858 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3859 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0}, opc, V128,
3861 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3862 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,1}, opc, V128,
3864 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3867 multiclass SIMDThreeSameVectorFPCmp<bit U, bit S, bits<5> opc,
3869 SDPatternOperator OpNode> {
3870 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0}, opc, V64,
3872 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3873 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0}, opc, V128,
3875 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3876 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,1}, opc, V128,
3878 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3881 multiclass SIMDThreeSameVectorFPTied<bit U, bit S, bits<5> opc,
3882 string asm, SDPatternOperator OpNode> {
3883 def v2f32 : BaseSIMDThreeSameVectorTied<0, U, {S,0}, opc, V64,
3885 [(set (v2f32 V64:$dst),
3886 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3887 def v4f32 : BaseSIMDThreeSameVectorTied<1, U, {S,0}, opc, V128,
3889 [(set (v4f32 V128:$dst),
3890 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3891 def v2f64 : BaseSIMDThreeSameVectorTied<1, U, {S,1}, opc, V128,
3893 [(set (v2f64 V128:$dst),
3894 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3897 // As above, but D and B sized elements unsupported.
3898 multiclass SIMDThreeSameVectorHS<bit U, bits<5> opc, string asm,
3899 SDPatternOperator OpNode> {
3900 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3902 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3903 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3905 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3906 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3908 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3909 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3911 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3914 // Logical three vector ops share opcode bits, and only use B sized elements.
3915 multiclass SIMDLogicalThreeVector<bit U, bits<2> size, string asm,
3916 SDPatternOperator OpNode = null_frag> {
3917 def v8i8 : BaseSIMDThreeSameVector<0, U, size, 0b00011, V64,
3919 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn, V64:$Rm))]>;
3920 def v16i8 : BaseSIMDThreeSameVector<1, U, size, 0b00011, V128,
3922 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn, V128:$Rm))]>;
3924 def : Pat<(v4i16 (OpNode V64:$LHS, V64:$RHS)),
3925 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
3926 def : Pat<(v2i32 (OpNode V64:$LHS, V64:$RHS)),
3927 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
3928 def : Pat<(v1i64 (OpNode V64:$LHS, V64:$RHS)),
3929 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
3931 def : Pat<(v8i16 (OpNode V128:$LHS, V128:$RHS)),
3932 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
3933 def : Pat<(v4i32 (OpNode V128:$LHS, V128:$RHS)),
3934 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
3935 def : Pat<(v2i64 (OpNode V128:$LHS, V128:$RHS)),
3936 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
3939 multiclass SIMDLogicalThreeVectorTied<bit U, bits<2> size,
3940 string asm, SDPatternOperator OpNode> {
3941 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, size, 0b00011, V64,
3943 [(set (v8i8 V64:$dst),
3944 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3945 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, size, 0b00011, V128,
3947 [(set (v16i8 V128:$dst),
3948 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
3949 (v16i8 V128:$Rm)))]>;
3951 def : Pat<(v4i16 (OpNode (v4i16 V64:$LHS), (v4i16 V64:$MHS),
3953 (!cast<Instruction>(NAME#"v8i8")
3954 V64:$LHS, V64:$MHS, V64:$RHS)>;
3955 def : Pat<(v2i32 (OpNode (v2i32 V64:$LHS), (v2i32 V64:$MHS),
3957 (!cast<Instruction>(NAME#"v8i8")
3958 V64:$LHS, V64:$MHS, V64:$RHS)>;
3959 def : Pat<(v1i64 (OpNode (v1i64 V64:$LHS), (v1i64 V64:$MHS),
3961 (!cast<Instruction>(NAME#"v8i8")
3962 V64:$LHS, V64:$MHS, V64:$RHS)>;
3964 def : Pat<(v8i16 (OpNode (v8i16 V128:$LHS), (v8i16 V128:$MHS),
3965 (v8i16 V128:$RHS))),
3966 (!cast<Instruction>(NAME#"v16i8")
3967 V128:$LHS, V128:$MHS, V128:$RHS)>;
3968 def : Pat<(v4i32 (OpNode (v4i32 V128:$LHS), (v4i32 V128:$MHS),
3969 (v4i32 V128:$RHS))),
3970 (!cast<Instruction>(NAME#"v16i8")
3971 V128:$LHS, V128:$MHS, V128:$RHS)>;
3972 def : Pat<(v2i64 (OpNode (v2i64 V128:$LHS), (v2i64 V128:$MHS),
3973 (v2i64 V128:$RHS))),
3974 (!cast<Instruction>(NAME#"v16i8")
3975 V128:$LHS, V128:$MHS, V128:$RHS)>;
3979 //----------------------------------------------------------------------------
3980 // AdvSIMD two register vector instructions.
3981 //----------------------------------------------------------------------------
3983 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3984 class BaseSIMDTwoSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
3985 RegisterOperand regtype, string asm, string dstkind,
3986 string srckind, list<dag> pattern>
3987 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
3988 "{\t$Rd" # dstkind # ", $Rn" # srckind #
3989 "|" # dstkind # "\t$Rd, $Rn}", "", pattern>,
3996 let Inst{28-24} = 0b01110;
3997 let Inst{23-22} = size;
3998 let Inst{21-17} = 0b10000;
3999 let Inst{16-12} = opcode;
4000 let Inst{11-10} = 0b10;
4005 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4006 class BaseSIMDTwoSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4007 RegisterOperand regtype, string asm, string dstkind,
4008 string srckind, list<dag> pattern>
4009 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn), asm,
4010 "{\t$Rd" # dstkind # ", $Rn" # srckind #
4011 "|" # dstkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
4018 let Inst{28-24} = 0b01110;
4019 let Inst{23-22} = size;
4020 let Inst{21-17} = 0b10000;
4021 let Inst{16-12} = opcode;
4022 let Inst{11-10} = 0b10;
4027 // Supports B, H, and S element sizes.
4028 multiclass SIMDTwoVectorBHS<bit U, bits<5> opc, string asm,
4029 SDPatternOperator OpNode> {
4030 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4032 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4033 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4034 asm, ".16b", ".16b",
4035 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4036 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4038 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4039 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4041 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4042 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4044 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4045 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4047 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4050 class BaseSIMDVectorLShiftLongBySize<bit Q, bits<2> size,
4051 RegisterOperand regtype, string asm, string dstkind,
4052 string srckind, string amount>
4053 : I<(outs V128:$Rd), (ins regtype:$Rn), asm,
4054 "{\t$Rd" # dstkind # ", $Rn" # srckind # ", #" # amount #
4055 "|" # dstkind # "\t$Rd, $Rn, #" # amount # "}", "", []>,
4061 let Inst{29-24} = 0b101110;
4062 let Inst{23-22} = size;
4063 let Inst{21-10} = 0b100001001110;
4068 multiclass SIMDVectorLShiftLongBySizeBHS {
4069 let neverHasSideEffects = 1 in {
4070 def v8i8 : BaseSIMDVectorLShiftLongBySize<0, 0b00, V64,
4071 "shll", ".8h", ".8b", "8">;
4072 def v16i8 : BaseSIMDVectorLShiftLongBySize<1, 0b00, V128,
4073 "shll2", ".8h", ".16b", "8">;
4074 def v4i16 : BaseSIMDVectorLShiftLongBySize<0, 0b01, V64,
4075 "shll", ".4s", ".4h", "16">;
4076 def v8i16 : BaseSIMDVectorLShiftLongBySize<1, 0b01, V128,
4077 "shll2", ".4s", ".8h", "16">;
4078 def v2i32 : BaseSIMDVectorLShiftLongBySize<0, 0b10, V64,
4079 "shll", ".2d", ".2s", "32">;
4080 def v4i32 : BaseSIMDVectorLShiftLongBySize<1, 0b10, V128,
4081 "shll2", ".2d", ".4s", "32">;
4085 // Supports all element sizes.
4086 multiclass SIMDLongTwoVector<bit U, bits<5> opc, string asm,
4087 SDPatternOperator OpNode> {
4088 def v8i8_v4i16 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4090 [(set (v4i16 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4091 def v16i8_v8i16 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4093 [(set (v8i16 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4094 def v4i16_v2i32 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4096 [(set (v2i32 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4097 def v8i16_v4i32 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4099 [(set (v4i32 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4100 def v2i32_v1i64 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4102 [(set (v1i64 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4103 def v4i32_v2i64 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4105 [(set (v2i64 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4108 multiclass SIMDLongTwoVectorTied<bit U, bits<5> opc, string asm,
4109 SDPatternOperator OpNode> {
4110 def v8i8_v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, V64,
4112 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd),
4114 def v16i8_v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, V128,
4116 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd),
4117 (v16i8 V128:$Rn)))]>;
4118 def v4i16_v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, V64,
4120 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd),
4121 (v4i16 V64:$Rn)))]>;
4122 def v8i16_v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, V128,
4124 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd),
4125 (v8i16 V128:$Rn)))]>;
4126 def v2i32_v1i64 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, V64,
4128 [(set (v1i64 V64:$dst), (OpNode (v1i64 V64:$Rd),
4129 (v2i32 V64:$Rn)))]>;
4130 def v4i32_v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, V128,
4132 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd),
4133 (v4i32 V128:$Rn)))]>;
4136 // Supports all element sizes, except 1xD.
4137 multiclass SIMDTwoVectorBHSDTied<bit U, bits<5> opc, string asm,
4138 SDPatternOperator OpNode> {
4139 def v8i8 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, V64,
4141 [(set (v8i8 V64:$dst), (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn)))]>;
4142 def v16i8 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, V128,
4143 asm, ".16b", ".16b",
4144 [(set (v16i8 V128:$dst), (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
4145 def v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, V64,
4147 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn)))]>;
4148 def v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, V128,
4150 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn)))]>;
4151 def v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, V64,
4153 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn)))]>;
4154 def v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, V128,
4156 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
4157 def v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b11, opc, V128,
4159 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn)))]>;
4162 multiclass SIMDTwoVectorBHSD<bit U, bits<5> opc, string asm,
4163 SDPatternOperator OpNode = null_frag> {
4164 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4166 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4167 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4168 asm, ".16b", ".16b",
4169 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4170 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4172 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4173 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4175 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4176 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4178 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4179 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4181 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4182 def v2i64 : BaseSIMDTwoSameVector<1, U, 0b11, opc, V128,
4184 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4188 // Supports only B element sizes.
4189 multiclass SIMDTwoVectorB<bit U, bits<2> size, bits<5> opc, string asm,
4190 SDPatternOperator OpNode> {
4191 def v8i8 : BaseSIMDTwoSameVector<0, U, size, opc, V64,
4193 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4194 def v16i8 : BaseSIMDTwoSameVector<1, U, size, opc, V128,
4195 asm, ".16b", ".16b",
4196 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4200 // Supports only B and H element sizes.
4201 multiclass SIMDTwoVectorBH<bit U, bits<5> opc, string asm,
4202 SDPatternOperator OpNode> {
4203 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4205 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn))]>;
4206 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4207 asm, ".16b", ".16b",
4208 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn))]>;
4209 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4211 [(set (v4i16 V64:$Rd), (OpNode V64:$Rn))]>;
4212 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4214 [(set (v8i16 V128:$Rd), (OpNode V128:$Rn))]>;
4217 // Supports only S and D element sizes, uses high bit of the size field
4218 // as an extra opcode bit.
4219 multiclass SIMDTwoVectorFP<bit U, bit S, bits<5> opc, string asm,
4220 SDPatternOperator OpNode> {
4221 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4223 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
4224 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4226 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
4227 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4229 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4232 // Supports only S element size.
4233 multiclass SIMDTwoVectorS<bit U, bit S, bits<5> opc, string asm,
4234 SDPatternOperator OpNode> {
4235 def v2i32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4237 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4238 def v4i32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4240 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4244 multiclass SIMDTwoVectorFPToInt<bit U, bit S, bits<5> opc, string asm,
4245 SDPatternOperator OpNode> {
4246 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4248 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
4249 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4251 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
4252 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4254 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4257 multiclass SIMDTwoVectorIntToFP<bit U, bit S, bits<5> opc, string asm,
4258 SDPatternOperator OpNode> {
4259 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4261 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4262 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4264 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4265 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4267 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4271 class BaseSIMDMixedTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4272 RegisterOperand inreg, RegisterOperand outreg,
4273 string asm, string outkind, string inkind,
4275 : I<(outs outreg:$Rd), (ins inreg:$Rn), asm,
4276 "{\t$Rd" # outkind # ", $Rn" # inkind #
4277 "|" # outkind # "\t$Rd, $Rn}", "", pattern>,
4284 let Inst{28-24} = 0b01110;
4285 let Inst{23-22} = size;
4286 let Inst{21-17} = 0b10000;
4287 let Inst{16-12} = opcode;
4288 let Inst{11-10} = 0b10;
4293 class BaseSIMDMixedTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4294 RegisterOperand inreg, RegisterOperand outreg,
4295 string asm, string outkind, string inkind,
4297 : I<(outs outreg:$dst), (ins outreg:$Rd, inreg:$Rn), asm,
4298 "{\t$Rd" # outkind # ", $Rn" # inkind #
4299 "|" # outkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
4306 let Inst{28-24} = 0b01110;
4307 let Inst{23-22} = size;
4308 let Inst{21-17} = 0b10000;
4309 let Inst{16-12} = opcode;
4310 let Inst{11-10} = 0b10;
4315 multiclass SIMDMixedTwoVector<bit U, bits<5> opc, string asm,
4316 SDPatternOperator OpNode> {
4317 def v8i8 : BaseSIMDMixedTwoVector<0, U, 0b00, opc, V128, V64,
4319 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4320 def v16i8 : BaseSIMDMixedTwoVectorTied<1, U, 0b00, opc, V128, V128,
4321 asm#"2", ".16b", ".8h", []>;
4322 def v4i16 : BaseSIMDMixedTwoVector<0, U, 0b01, opc, V128, V64,
4324 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4325 def v8i16 : BaseSIMDMixedTwoVectorTied<1, U, 0b01, opc, V128, V128,
4326 asm#"2", ".8h", ".4s", []>;
4327 def v2i32 : BaseSIMDMixedTwoVector<0, U, 0b10, opc, V128, V64,
4329 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4330 def v4i32 : BaseSIMDMixedTwoVectorTied<1, U, 0b10, opc, V128, V128,
4331 asm#"2", ".4s", ".2d", []>;
4333 def : Pat<(concat_vectors (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn))),
4334 (!cast<Instruction>(NAME # "v16i8")
4335 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4336 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn))),
4337 (!cast<Instruction>(NAME # "v8i16")
4338 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4339 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn))),
4340 (!cast<Instruction>(NAME # "v4i32")
4341 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4344 class BaseSIMDCmpTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4345 RegisterOperand regtype, string asm, string kind,
4346 ValueType dty, ValueType sty, SDNode OpNode>
4347 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
4348 "{\t$Rd" # kind # ", $Rn" # kind # ", #0" #
4349 "|" # kind # "\t$Rd, $Rn, #0}", "",
4350 [(set (dty regtype:$Rd), (OpNode (sty regtype:$Rn)))]>,
4357 let Inst{28-24} = 0b01110;
4358 let Inst{23-22} = size;
4359 let Inst{21-17} = 0b10000;
4360 let Inst{16-12} = opcode;
4361 let Inst{11-10} = 0b10;
4366 // Comparisons support all element sizes, except 1xD.
4367 multiclass SIMDCmpTwoVector<bit U, bits<5> opc, string asm,
4369 def v8i8rz : BaseSIMDCmpTwoVector<0, U, 0b00, opc, V64,
4371 v8i8, v8i8, OpNode>;
4372 def v16i8rz : BaseSIMDCmpTwoVector<1, U, 0b00, opc, V128,
4374 v16i8, v16i8, OpNode>;
4375 def v4i16rz : BaseSIMDCmpTwoVector<0, U, 0b01, opc, V64,
4377 v4i16, v4i16, OpNode>;
4378 def v8i16rz : BaseSIMDCmpTwoVector<1, U, 0b01, opc, V128,
4380 v8i16, v8i16, OpNode>;
4381 def v2i32rz : BaseSIMDCmpTwoVector<0, U, 0b10, opc, V64,
4383 v2i32, v2i32, OpNode>;
4384 def v4i32rz : BaseSIMDCmpTwoVector<1, U, 0b10, opc, V128,
4386 v4i32, v4i32, OpNode>;
4387 def v2i64rz : BaseSIMDCmpTwoVector<1, U, 0b11, opc, V128,
4389 v2i64, v2i64, OpNode>;
4392 // FP Comparisons support only S and D element sizes.
4393 multiclass SIMDFPCmpTwoVector<bit U, bit S, bits<5> opc,
4394 string asm, SDNode OpNode> {
4395 def v2i32rz : BaseSIMDCmpTwoVector<0, U, {S,0}, opc, V64,
4397 v2i32, v2f32, OpNode>;
4398 def v4i32rz : BaseSIMDCmpTwoVector<1, U, {S,0}, opc, V128,
4400 v4i32, v4f32, OpNode>;
4401 def v2i64rz : BaseSIMDCmpTwoVector<1, U, {S,1}, opc, V128,
4403 v2i64, v2f64, OpNode>;
4406 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4407 class BaseSIMDFPCvtTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4408 RegisterOperand outtype, RegisterOperand intype,
4409 string asm, string VdTy, string VnTy,
4411 : I<(outs outtype:$Rd), (ins intype:$Rn), asm,
4412 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "", pattern>,
4419 let Inst{28-24} = 0b01110;
4420 let Inst{23-22} = size;
4421 let Inst{21-17} = 0b10000;
4422 let Inst{16-12} = opcode;
4423 let Inst{11-10} = 0b10;
4428 class BaseSIMDFPCvtTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4429 RegisterOperand outtype, RegisterOperand intype,
4430 string asm, string VdTy, string VnTy,
4432 : I<(outs outtype:$dst), (ins outtype:$Rd, intype:$Rn), asm,
4433 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "$Rd = $dst", pattern>,
4440 let Inst{28-24} = 0b01110;
4441 let Inst{23-22} = size;
4442 let Inst{21-17} = 0b10000;
4443 let Inst{16-12} = opcode;
4444 let Inst{11-10} = 0b10;
4449 multiclass SIMDFPWidenTwoVector<bit U, bit S, bits<5> opc, string asm> {
4450 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V128, V64,
4451 asm, ".4s", ".4h", []>;
4452 def v8i16 : BaseSIMDFPCvtTwoVector<1, U, {S,0}, opc, V128, V128,
4453 asm#"2", ".4s", ".8h", []>;
4454 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V128, V64,
4455 asm, ".2d", ".2s", []>;
4456 def v4i32 : BaseSIMDFPCvtTwoVector<1, U, {S,1}, opc, V128, V128,
4457 asm#"2", ".2d", ".4s", []>;
4460 multiclass SIMDFPNarrowTwoVector<bit U, bit S, bits<5> opc, string asm> {
4461 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V64, V128,
4462 asm, ".4h", ".4s", []>;
4463 def v8i16 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,0}, opc, V128, V128,
4464 asm#"2", ".8h", ".4s", []>;
4465 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
4466 asm, ".2s", ".2d", []>;
4467 def v4i32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
4468 asm#"2", ".4s", ".2d", []>;
4471 multiclass SIMDFPInexactCvtTwoVector<bit U, bit S, bits<5> opc, string asm,
4473 def v2f32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
4475 [(set (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4476 def v4f32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
4477 asm#"2", ".4s", ".2d", []>;
4479 def : Pat<(concat_vectors (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn))),
4480 (!cast<Instruction>(NAME # "v4f32")
4481 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4484 //----------------------------------------------------------------------------
4485 // AdvSIMD three register different-size vector instructions.
4486 //----------------------------------------------------------------------------
4488 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4489 class BaseSIMDDifferentThreeVector<bit U, bits<3> size, bits<4> opcode,
4490 RegisterOperand outtype, RegisterOperand intype1,
4491 RegisterOperand intype2, string asm,
4492 string outkind, string inkind1, string inkind2,
4494 : I<(outs outtype:$Rd), (ins intype1:$Rn, intype2:$Rm), asm,
4495 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
4496 "|" # outkind # "\t$Rd, $Rn, $Rm}", "", pattern>,
4502 let Inst{30} = size{0};
4504 let Inst{28-24} = 0b01110;
4505 let Inst{23-22} = size{2-1};
4507 let Inst{20-16} = Rm;
4508 let Inst{15-12} = opcode;
4509 let Inst{11-10} = 0b00;
4514 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4515 class BaseSIMDDifferentThreeVectorTied<bit U, bits<3> size, bits<4> opcode,
4516 RegisterOperand outtype, RegisterOperand intype1,
4517 RegisterOperand intype2, string asm,
4518 string outkind, string inkind1, string inkind2,
4520 : I<(outs outtype:$dst), (ins outtype:$Rd, intype1:$Rn, intype2:$Rm), asm,
4521 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
4522 "|" # outkind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
4528 let Inst{30} = size{0};
4530 let Inst{28-24} = 0b01110;
4531 let Inst{23-22} = size{2-1};
4533 let Inst{20-16} = Rm;
4534 let Inst{15-12} = opcode;
4535 let Inst{11-10} = 0b00;
4540 // FIXME: TableGen doesn't know how to deal with expanded types that also
4541 // change the element count (in this case, placing the results in
4542 // the high elements of the result register rather than the low
4543 // elements). Until that's fixed, we can't code-gen those.
4544 multiclass SIMDNarrowThreeVectorBHS<bit U, bits<4> opc, string asm,
4546 def v8i16_v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4548 asm, ".8b", ".8h", ".8h",
4549 [(set (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
4550 def v8i16_v16i8 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4552 asm#"2", ".16b", ".8h", ".8h",
4554 def v4i32_v4i16 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4556 asm, ".4h", ".4s", ".4s",
4557 [(set (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
4558 def v4i32_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4560 asm#"2", ".8h", ".4s", ".4s",
4562 def v2i64_v2i32 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4564 asm, ".2s", ".2d", ".2d",
4565 [(set (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
4566 def v2i64_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4568 asm#"2", ".4s", ".2d", ".2d",
4572 // Patterns for the '2' variants involve INSERT_SUBREG, which you can't put in
4573 // a version attached to an instruction.
4574 def : Pat<(concat_vectors (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn),
4576 (!cast<Instruction>(NAME # "v8i16_v16i8")
4577 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4578 V128:$Rn, V128:$Rm)>;
4579 def : Pat<(concat_vectors (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn),
4581 (!cast<Instruction>(NAME # "v4i32_v8i16")
4582 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4583 V128:$Rn, V128:$Rm)>;
4584 def : Pat<(concat_vectors (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn),
4586 (!cast<Instruction>(NAME # "v2i64_v4i32")
4587 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4588 V128:$Rn, V128:$Rm)>;
4591 multiclass SIMDDifferentThreeVectorBD<bit U, bits<4> opc, string asm,
4593 def v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4595 asm, ".8h", ".8b", ".8b",
4596 [(set (v8i16 V128:$Rd), (IntOp (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4597 def v16i8 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4599 asm#"2", ".8h", ".16b", ".16b", []>;
4600 def v1i64 : BaseSIMDDifferentThreeVector<U, 0b110, opc,
4602 asm, ".1q", ".1d", ".1d", []>;
4603 def v2i64 : BaseSIMDDifferentThreeVector<U, 0b111, opc,
4605 asm#"2", ".1q", ".2d", ".2d", []>;
4607 def : Pat<(v8i16 (IntOp (v8i8 (extract_high_v16i8 V128:$Rn)),
4608 (v8i8 (extract_high_v16i8 V128:$Rm)))),
4609 (!cast<Instruction>(NAME#"v16i8") V128:$Rn, V128:$Rm)>;
4612 multiclass SIMDLongThreeVectorHS<bit U, bits<4> opc, string asm,
4613 SDPatternOperator OpNode> {
4614 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4616 asm, ".4s", ".4h", ".4h",
4617 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4618 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4620 asm#"2", ".4s", ".8h", ".8h",
4621 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
4622 (extract_high_v8i16 V128:$Rm)))]>;
4623 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4625 asm, ".2d", ".2s", ".2s",
4626 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4627 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4629 asm#"2", ".2d", ".4s", ".4s",
4630 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
4631 (extract_high_v4i32 V128:$Rm)))]>;
4634 multiclass SIMDLongThreeVectorBHSabdl<bit U, bits<4> opc, string asm,
4635 SDPatternOperator OpNode = null_frag> {
4636 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4638 asm, ".8h", ".8b", ".8b",
4639 [(set (v8i16 V128:$Rd),
4640 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))))]>;
4641 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4643 asm#"2", ".8h", ".16b", ".16b",
4644 [(set (v8i16 V128:$Rd),
4645 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
4646 (extract_high_v16i8 V128:$Rm)))))]>;
4647 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4649 asm, ".4s", ".4h", ".4h",
4650 [(set (v4i32 V128:$Rd),
4651 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))))]>;
4652 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4654 asm#"2", ".4s", ".8h", ".8h",
4655 [(set (v4i32 V128:$Rd),
4656 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
4657 (extract_high_v8i16 V128:$Rm)))))]>;
4658 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4660 asm, ".2d", ".2s", ".2s",
4661 [(set (v2i64 V128:$Rd),
4662 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))))]>;
4663 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4665 asm#"2", ".2d", ".4s", ".4s",
4666 [(set (v2i64 V128:$Rd),
4667 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
4668 (extract_high_v4i32 V128:$Rm)))))]>;
4671 multiclass SIMDLongThreeVectorTiedBHSabal<bit U, bits<4> opc,
4673 SDPatternOperator OpNode> {
4674 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
4676 asm, ".8h", ".8b", ".8b",
4677 [(set (v8i16 V128:$dst),
4678 (add (v8i16 V128:$Rd),
4679 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))))]>;
4680 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4682 asm#"2", ".8h", ".16b", ".16b",
4683 [(set (v8i16 V128:$dst),
4684 (add (v8i16 V128:$Rd),
4685 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
4686 (extract_high_v16i8 V128:$Rm))))))]>;
4687 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4689 asm, ".4s", ".4h", ".4h",
4690 [(set (v4i32 V128:$dst),
4691 (add (v4i32 V128:$Rd),
4692 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))))]>;
4693 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4695 asm#"2", ".4s", ".8h", ".8h",
4696 [(set (v4i32 V128:$dst),
4697 (add (v4i32 V128:$Rd),
4698 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
4699 (extract_high_v8i16 V128:$Rm))))))]>;
4700 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4702 asm, ".2d", ".2s", ".2s",
4703 [(set (v2i64 V128:$dst),
4704 (add (v2i64 V128:$Rd),
4705 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))))]>;
4706 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4708 asm#"2", ".2d", ".4s", ".4s",
4709 [(set (v2i64 V128:$dst),
4710 (add (v2i64 V128:$Rd),
4711 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
4712 (extract_high_v4i32 V128:$Rm))))))]>;
4715 multiclass SIMDLongThreeVectorBHS<bit U, bits<4> opc, string asm,
4716 SDPatternOperator OpNode = null_frag> {
4717 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4719 asm, ".8h", ".8b", ".8b",
4720 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4721 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4723 asm#"2", ".8h", ".16b", ".16b",
4724 [(set (v8i16 V128:$Rd), (OpNode (extract_high_v16i8 V128:$Rn),
4725 (extract_high_v16i8 V128:$Rm)))]>;
4726 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4728 asm, ".4s", ".4h", ".4h",
4729 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4730 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4732 asm#"2", ".4s", ".8h", ".8h",
4733 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
4734 (extract_high_v8i16 V128:$Rm)))]>;
4735 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4737 asm, ".2d", ".2s", ".2s",
4738 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4739 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4741 asm#"2", ".2d", ".4s", ".4s",
4742 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
4743 (extract_high_v4i32 V128:$Rm)))]>;
4746 multiclass SIMDLongThreeVectorTiedBHS<bit U, bits<4> opc,
4748 SDPatternOperator OpNode> {
4749 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
4751 asm, ".8h", ".8b", ".8b",
4752 [(set (v8i16 V128:$dst),
4753 (OpNode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4754 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4756 asm#"2", ".8h", ".16b", ".16b",
4757 [(set (v8i16 V128:$dst),
4758 (OpNode (v8i16 V128:$Rd),
4759 (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:$dst),
4765 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4766 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4768 asm#"2", ".4s", ".8h", ".8h",
4769 [(set (v4i32 V128:$dst),
4770 (OpNode (v4i32 V128:$Rd),
4771 (extract_high_v8i16 V128:$Rn),
4772 (extract_high_v8i16 V128:$Rm)))]>;
4773 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4775 asm, ".2d", ".2s", ".2s",
4776 [(set (v2i64 V128:$dst),
4777 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4778 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4780 asm#"2", ".2d", ".4s", ".4s",
4781 [(set (v2i64 V128:$dst),
4782 (OpNode (v2i64 V128:$Rd),
4783 (extract_high_v4i32 V128:$Rn),
4784 (extract_high_v4i32 V128:$Rm)))]>;
4787 multiclass SIMDLongThreeVectorSQDMLXTiedHS<bit U, bits<4> opc, string asm,
4788 SDPatternOperator Accum> {
4789 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4791 asm, ".4s", ".4h", ".4h",
4792 [(set (v4i32 V128:$dst),
4793 (Accum (v4i32 V128:$Rd),
4794 (v4i32 (int_arm64_neon_sqdmull (v4i16 V64:$Rn),
4795 (v4i16 V64:$Rm)))))]>;
4796 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4798 asm#"2", ".4s", ".8h", ".8h",
4799 [(set (v4i32 V128:$dst),
4800 (Accum (v4i32 V128:$Rd),
4801 (v4i32 (int_arm64_neon_sqdmull (extract_high_v8i16 V128:$Rn),
4802 (extract_high_v8i16 V128:$Rm)))))]>;
4803 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4805 asm, ".2d", ".2s", ".2s",
4806 [(set (v2i64 V128:$dst),
4807 (Accum (v2i64 V128:$Rd),
4808 (v2i64 (int_arm64_neon_sqdmull (v2i32 V64:$Rn),
4809 (v2i32 V64:$Rm)))))]>;
4810 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4812 asm#"2", ".2d", ".4s", ".4s",
4813 [(set (v2i64 V128:$dst),
4814 (Accum (v2i64 V128:$Rd),
4815 (v2i64 (int_arm64_neon_sqdmull (extract_high_v4i32 V128:$Rn),
4816 (extract_high_v4i32 V128:$Rm)))))]>;
4819 multiclass SIMDWideThreeVectorBHS<bit U, bits<4> opc, string asm,
4820 SDPatternOperator OpNode> {
4821 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4823 asm, ".8h", ".8h", ".8b",
4824 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i8 V64:$Rm)))]>;
4825 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4827 asm#"2", ".8h", ".8h", ".16b",
4828 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
4829 (extract_high_v16i8 V128:$Rm)))]>;
4830 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4832 asm, ".4s", ".4s", ".4h",
4833 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i16 V64:$Rm)))]>;
4834 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4836 asm#"2", ".4s", ".4s", ".8h",
4837 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
4838 (extract_high_v8i16 V128:$Rm)))]>;
4839 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4841 asm, ".2d", ".2d", ".2s",
4842 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i32 V64:$Rm)))]>;
4843 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4845 asm#"2", ".2d", ".2d", ".4s",
4846 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
4847 (extract_high_v4i32 V128:$Rm)))]>;
4850 //----------------------------------------------------------------------------
4851 // AdvSIMD bitwise extract from vector
4852 //----------------------------------------------------------------------------
4854 class BaseSIMDBitwiseExtract<bit size, RegisterOperand regtype, ValueType vty,
4855 string asm, string kind>
4856 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, i32imm:$imm), asm,
4857 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $imm" #
4858 "|" # kind # "\t$Rd, $Rn, $Rm, $imm}", "",
4859 [(set (vty regtype:$Rd),
4860 (ARM64ext regtype:$Rn, regtype:$Rm, (i32 imm:$imm)))]>,
4867 let Inst{30} = size;
4868 let Inst{29-21} = 0b101110000;
4869 let Inst{20-16} = Rm;
4871 let Inst{14-11} = imm;
4878 multiclass SIMDBitwiseExtract<string asm> {
4879 def v8i8 : BaseSIMDBitwiseExtract<0, V64, v8i8, asm, ".8b"> {
4882 def v16i8 : BaseSIMDBitwiseExtract<1, V128, v16i8, asm, ".16b">;
4885 //----------------------------------------------------------------------------
4886 // AdvSIMD zip vector
4887 //----------------------------------------------------------------------------
4889 class BaseSIMDZipVector<bits<3> size, bits<3> opc, RegisterOperand regtype,
4890 string asm, string kind, SDNode OpNode, ValueType valty>
4891 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
4892 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
4893 "|" # kind # "\t$Rd, $Rn, $Rm}", "",
4894 [(set (valty regtype:$Rd), (OpNode regtype:$Rn, regtype:$Rm))]>,
4900 let Inst{30} = size{0};
4901 let Inst{29-24} = 0b001110;
4902 let Inst{23-22} = size{2-1};
4904 let Inst{20-16} = Rm;
4906 let Inst{14-12} = opc;
4907 let Inst{11-10} = 0b10;
4912 multiclass SIMDZipVector<bits<3>opc, string asm,
4914 def v8i8 : BaseSIMDZipVector<0b000, opc, V64,
4915 asm, ".8b", OpNode, v8i8>;
4916 def v16i8 : BaseSIMDZipVector<0b001, opc, V128,
4917 asm, ".16b", OpNode, v16i8>;
4918 def v4i16 : BaseSIMDZipVector<0b010, opc, V64,
4919 asm, ".4h", OpNode, v4i16>;
4920 def v8i16 : BaseSIMDZipVector<0b011, opc, V128,
4921 asm, ".8h", OpNode, v8i16>;
4922 def v2i32 : BaseSIMDZipVector<0b100, opc, V64,
4923 asm, ".2s", OpNode, v2i32>;
4924 def v4i32 : BaseSIMDZipVector<0b101, opc, V128,
4925 asm, ".4s", OpNode, v4i32>;
4926 def v2i64 : BaseSIMDZipVector<0b111, opc, V128,
4927 asm, ".2d", OpNode, v2i64>;
4929 def : Pat<(v2f32 (OpNode V64:$Rn, V64:$Rm)),
4930 (!cast<Instruction>(NAME#"v2i32") V64:$Rn, V64:$Rm)>;
4931 def : Pat<(v4f32 (OpNode V128:$Rn, V128:$Rm)),
4932 (!cast<Instruction>(NAME#"v4i32") V128:$Rn, V128:$Rm)>;
4933 def : Pat<(v2f64 (OpNode V128:$Rn, V128:$Rm)),
4934 (!cast<Instruction>(NAME#"v2i64") V128:$Rn, V128:$Rm)>;
4937 //----------------------------------------------------------------------------
4938 // AdvSIMD three register scalar instructions
4939 //----------------------------------------------------------------------------
4941 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
4942 class BaseSIMDThreeScalar<bit U, bits<2> size, bits<5> opcode,
4943 RegisterClass regtype, string asm,
4945 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
4946 "\t$Rd, $Rn, $Rm", "", pattern>,
4951 let Inst{31-30} = 0b01;
4953 let Inst{28-24} = 0b11110;
4954 let Inst{23-22} = size;
4956 let Inst{20-16} = Rm;
4957 let Inst{15-11} = opcode;
4963 multiclass SIMDThreeScalarD<bit U, bits<5> opc, string asm,
4964 SDPatternOperator OpNode> {
4965 def v1i64 : BaseSIMDThreeScalar<U, 0b11, opc, FPR64, asm,
4966 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
4969 multiclass SIMDThreeScalarBHSD<bit U, bits<5> opc, string asm,
4970 SDPatternOperator OpNode> {
4971 def v1i64 : BaseSIMDThreeScalar<U, 0b11, opc, FPR64, asm,
4972 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
4973 def v1i32 : BaseSIMDThreeScalar<U, 0b10, opc, FPR32, asm, []>;
4974 def v1i16 : BaseSIMDThreeScalar<U, 0b01, opc, FPR16, asm, []>;
4975 def v1i8 : BaseSIMDThreeScalar<U, 0b00, opc, FPR8 , asm, []>;
4977 def : Pat<(i64 (OpNode (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4978 (!cast<Instruction>(NAME#"v1i64") FPR64:$Rn, FPR64:$Rm)>;
4979 def : Pat<(i32 (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm))),
4980 (!cast<Instruction>(NAME#"v1i32") FPR32:$Rn, FPR32:$Rm)>;
4983 multiclass SIMDThreeScalarHS<bit U, bits<5> opc, string asm,
4984 SDPatternOperator OpNode> {
4985 def v1i32 : BaseSIMDThreeScalar<U, 0b10, opc, FPR32, asm,
4986 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
4987 def v1i16 : BaseSIMDThreeScalar<U, 0b01, opc, FPR16, asm, []>;
4990 multiclass SIMDThreeScalarSD<bit U, bit S, bits<5> opc, string asm,
4991 SDPatternOperator OpNode = null_frag> {
4992 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
4993 def #NAME#64 : BaseSIMDThreeScalar<U, {S,1}, opc, FPR64, asm,
4994 [(set (f64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
4995 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0}, opc, FPR32, asm,
4996 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
4999 def : Pat<(v1f64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
5000 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
5003 multiclass SIMDThreeScalarFPCmp<bit U, bit S, bits<5> opc, string asm,
5004 SDPatternOperator OpNode = null_frag> {
5005 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5006 def #NAME#64 : BaseSIMDThreeScalar<U, {S,1}, opc, FPR64, asm,
5007 [(set (i64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
5008 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0}, opc, FPR32, asm,
5009 [(set (i32 FPR32:$Rd), (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]>;
5012 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
5013 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
5016 class BaseSIMDThreeScalarMixed<bit U, bits<2> size, bits<5> opcode,
5017 dag oops, dag iops, string asm, string cstr, list<dag> pat>
5018 : I<oops, iops, asm,
5019 "\t$Rd, $Rn, $Rm", cstr, pat>,
5024 let Inst{31-30} = 0b01;
5026 let Inst{28-24} = 0b11110;
5027 let Inst{23-22} = size;
5029 let Inst{20-16} = Rm;
5030 let Inst{15-11} = opcode;
5036 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5037 multiclass SIMDThreeScalarMixedHS<bit U, bits<5> opc, string asm,
5038 SDPatternOperator OpNode = null_frag> {
5039 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
5041 (ins FPR16:$Rn, FPR16:$Rm), asm, "", []>;
5042 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
5044 (ins FPR32:$Rn, FPR32:$Rm), asm, "",
5045 [(set (i64 FPR64:$Rd), (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
5048 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5049 multiclass SIMDThreeScalarMixedTiedHS<bit U, bits<5> opc, string asm,
5050 SDPatternOperator OpNode = null_frag> {
5051 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
5053 (ins FPR32:$Rd, FPR16:$Rn, FPR16:$Rm),
5054 asm, "$Rd = $dst", []>;
5055 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
5057 (ins FPR64:$Rd, FPR32:$Rn, FPR32:$Rm),
5059 [(set (i64 FPR64:$dst),
5060 (OpNode (i64 FPR64:$Rd), (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
5063 //----------------------------------------------------------------------------
5064 // AdvSIMD two register scalar instructions
5065 //----------------------------------------------------------------------------
5067 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5068 class BaseSIMDTwoScalar<bit U, bits<2> size, bits<5> opcode,
5069 RegisterClass regtype, RegisterClass regtype2,
5070 string asm, list<dag> pat>
5071 : I<(outs regtype:$Rd), (ins regtype2:$Rn), asm,
5072 "\t$Rd, $Rn", "", pat>,
5076 let Inst{31-30} = 0b01;
5078 let Inst{28-24} = 0b11110;
5079 let Inst{23-22} = size;
5080 let Inst{21-17} = 0b10000;
5081 let Inst{16-12} = opcode;
5082 let Inst{11-10} = 0b10;
5087 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5088 class BaseSIMDTwoScalarTied<bit U, bits<2> size, bits<5> opcode,
5089 RegisterClass regtype, RegisterClass regtype2,
5090 string asm, list<dag> pat>
5091 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype2:$Rn), asm,
5092 "\t$Rd, $Rn", "$Rd = $dst", pat>,
5096 let Inst{31-30} = 0b01;
5098 let Inst{28-24} = 0b11110;
5099 let Inst{23-22} = size;
5100 let Inst{21-17} = 0b10000;
5101 let Inst{16-12} = opcode;
5102 let Inst{11-10} = 0b10;
5108 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5109 class BaseSIMDCmpTwoScalar<bit U, bits<2> size, bits<5> opcode,
5110 RegisterClass regtype, string asm>
5111 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5112 "\t$Rd, $Rn, #0", "", []>,
5116 let Inst{31-30} = 0b01;
5118 let Inst{28-24} = 0b11110;
5119 let Inst{23-22} = size;
5120 let Inst{21-17} = 0b10000;
5121 let Inst{16-12} = opcode;
5122 let Inst{11-10} = 0b10;
5127 class SIMDInexactCvtTwoScalar<bits<5> opcode, string asm>
5128 : I<(outs FPR32:$Rd), (ins FPR64:$Rn), asm, "\t$Rd, $Rn", "",
5129 [(set (f32 FPR32:$Rd), (int_arm64_sisd_fcvtxn (f64 FPR64:$Rn)))]>,
5133 let Inst{31-17} = 0b011111100110000;
5134 let Inst{16-12} = opcode;
5135 let Inst{11-10} = 0b10;
5140 multiclass SIMDCmpTwoScalarD<bit U, bits<5> opc, string asm,
5141 SDPatternOperator OpNode> {
5142 def v1i64rz : BaseSIMDCmpTwoScalar<U, 0b11, opc, FPR64, asm>;
5144 def : Pat<(v1i64 (OpNode FPR64:$Rn)),
5145 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
5148 multiclass SIMDCmpTwoScalarSD<bit U, bit S, bits<5> opc, string asm,
5149 SDPatternOperator OpNode> {
5150 def v1i64rz : BaseSIMDCmpTwoScalar<U, {S,1}, opc, FPR64, asm>;
5151 def v1i32rz : BaseSIMDCmpTwoScalar<U, {S,0}, opc, FPR32, asm>;
5153 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn))),
5154 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
5157 multiclass SIMDTwoScalarD<bit U, bits<5> opc, string asm,
5158 SDPatternOperator OpNode = null_frag> {
5159 def v1i64 : BaseSIMDTwoScalar<U, 0b11, opc, FPR64, FPR64, asm,
5160 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn)))]>;
5162 def : Pat<(i64 (OpNode (i64 FPR64:$Rn))),
5163 (!cast<Instruction>(NAME # "v1i64") FPR64:$Rn)>;
5166 multiclass SIMDTwoScalarSD<bit U, bit S, bits<5> opc, string asm> {
5167 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, opc, FPR64, FPR64, asm,[]>;
5168 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, opc, FPR32, FPR32, asm,[]>;
5171 multiclass SIMDTwoScalarCVTSD<bit U, bit S, bits<5> opc, string asm,
5172 SDPatternOperator OpNode> {
5173 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, opc, FPR64, FPR64, asm,
5174 [(set FPR64:$Rd, (OpNode (f64 FPR64:$Rn)))]>;
5175 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, opc, FPR32, FPR32, asm,
5176 [(set FPR32:$Rd, (OpNode (f32 FPR32:$Rn)))]>;
5179 multiclass SIMDTwoScalarBHSD<bit U, bits<5> opc, string asm,
5180 SDPatternOperator OpNode = null_frag> {
5181 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5182 def v1i64 : BaseSIMDTwoScalar<U, 0b11, opc, FPR64, FPR64, asm,
5183 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
5184 def v1i32 : BaseSIMDTwoScalar<U, 0b10, opc, FPR32, FPR32, asm,
5185 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
5186 def v1i16 : BaseSIMDTwoScalar<U, 0b01, opc, FPR16, FPR16, asm, []>;
5187 def v1i8 : BaseSIMDTwoScalar<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
5190 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn))),
5191 (!cast<Instruction>(NAME # v1i64) FPR64:$Rn)>;
5194 multiclass SIMDTwoScalarBHSDTied<bit U, bits<5> opc, string asm,
5196 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5197 def v1i64 : BaseSIMDTwoScalarTied<U, 0b11, opc, FPR64, FPR64, asm,
5198 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn)))]>;
5199 def v1i32 : BaseSIMDTwoScalarTied<U, 0b10, opc, FPR32, FPR32, asm,
5200 [(set (i32 FPR32:$dst), (OpNode (i32 FPR32:$Rd), (i32 FPR32:$Rn)))]>;
5201 def v1i16 : BaseSIMDTwoScalarTied<U, 0b01, opc, FPR16, FPR16, asm, []>;
5202 def v1i8 : BaseSIMDTwoScalarTied<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
5205 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn))),
5206 (!cast<Instruction>(NAME # v1i64) FPR64:$Rd, FPR64:$Rn)>;
5211 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5212 multiclass SIMDTwoScalarMixedBHS<bit U, bits<5> opc, string asm,
5213 SDPatternOperator OpNode = null_frag> {
5214 def v1i32 : BaseSIMDTwoScalar<U, 0b10, opc, FPR32, FPR64, asm,
5215 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
5216 def v1i16 : BaseSIMDTwoScalar<U, 0b01, opc, FPR16, FPR32, asm, []>;
5217 def v1i8 : BaseSIMDTwoScalar<U, 0b00, opc, FPR8 , FPR16, asm, []>;
5220 //----------------------------------------------------------------------------
5221 // AdvSIMD scalar pairwise instructions
5222 //----------------------------------------------------------------------------
5224 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5225 class BaseSIMDPairwiseScalar<bit U, bits<2> size, bits<5> opcode,
5226 RegisterOperand regtype, RegisterOperand vectype,
5227 string asm, string kind>
5228 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
5229 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", []>,
5233 let Inst{31-30} = 0b01;
5235 let Inst{28-24} = 0b11110;
5236 let Inst{23-22} = size;
5237 let Inst{21-17} = 0b11000;
5238 let Inst{16-12} = opcode;
5239 let Inst{11-10} = 0b10;
5244 multiclass SIMDPairwiseScalarD<bit U, bits<5> opc, string asm> {
5245 def v2i64p : BaseSIMDPairwiseScalar<U, 0b11, opc, FPR64Op, V128,
5249 multiclass SIMDPairwiseScalarSD<bit U, bit S, bits<5> opc, string asm> {
5250 def v2i32p : BaseSIMDPairwiseScalar<U, {S,0}, opc, FPR32Op, V64,
5252 def v2i64p : BaseSIMDPairwiseScalar<U, {S,1}, opc, FPR64Op, V128,
5256 //----------------------------------------------------------------------------
5257 // AdvSIMD across lanes instructions
5258 //----------------------------------------------------------------------------
5260 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5261 class BaseSIMDAcrossLanes<bit Q, bit U, bits<2> size, bits<5> opcode,
5262 RegisterClass regtype, RegisterOperand vectype,
5263 string asm, string kind, list<dag> pattern>
5264 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
5265 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", pattern>,
5272 let Inst{28-24} = 0b01110;
5273 let Inst{23-22} = size;
5274 let Inst{21-17} = 0b11000;
5275 let Inst{16-12} = opcode;
5276 let Inst{11-10} = 0b10;
5281 multiclass SIMDAcrossLanesBHS<bit U, bits<5> opcode,
5283 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR8, V64,
5285 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR8, V128,
5287 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR16, V64,
5289 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR16, V128,
5291 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR32, V128,
5295 multiclass SIMDAcrossLanesHSD<bit U, bits<5> opcode, string asm> {
5296 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR16, V64,
5298 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR16, V128,
5300 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR32, V64,
5302 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR32, V128,
5304 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR64, V128,
5308 multiclass SIMDAcrossLanesS<bits<5> opcode, bit sz1, string asm,
5310 def v4i32v : BaseSIMDAcrossLanes<1, 1, {sz1, 0}, opcode, FPR32, V128,
5312 [(set FPR32:$Rd, (intOp (v4f32 V128:$Rn)))]>;
5315 //----------------------------------------------------------------------------
5316 // AdvSIMD INS/DUP instructions
5317 //----------------------------------------------------------------------------
5319 // FIXME: There has got to be a better way to factor these. ugh.
5321 class BaseSIMDInsDup<bit Q, bit op, dag outs, dag ins, string asm,
5322 string operands, string constraints, list<dag> pattern>
5323 : I<outs, ins, asm, operands, constraints, pattern>,
5330 let Inst{28-21} = 0b01110000;
5337 class SIMDDupFromMain<bit Q, bits<5> imm5, string size, ValueType vectype,
5338 RegisterOperand vecreg, RegisterClass regtype>
5339 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins regtype:$Rn), "dup",
5340 "{\t$Rd" # size # ", $Rn" #
5341 "|" # size # "\t$Rd, $Rn}", "",
5342 [(set (vectype vecreg:$Rd), (ARM64dup regtype:$Rn))]> {
5343 let Inst{20-16} = imm5;
5344 let Inst{14-11} = 0b0001;
5347 class SIMDDupFromElement<bit Q, string dstkind, string srckind,
5348 ValueType vectype, ValueType insreg,
5349 RegisterOperand vecreg, Operand idxtype,
5350 ValueType elttype, SDNode OpNode>
5351 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins V128:$Rn, idxtype:$idx), "dup",
5352 "{\t$Rd" # dstkind # ", $Rn" # srckind # "$idx" #
5353 "|" # dstkind # "\t$Rd, $Rn$idx}", "",
5354 [(set (vectype vecreg:$Rd),
5355 (OpNode (insreg V128:$Rn), idxtype:$idx))]> {
5356 let Inst{14-11} = 0b0000;
5359 class SIMDDup64FromElement
5360 : SIMDDupFromElement<1, ".2d", ".d", v2i64, v2i64, V128,
5361 VectorIndexD, i64, ARM64duplane64> {
5364 let Inst{19-16} = 0b1000;
5367 class SIMDDup32FromElement<bit Q, string size, ValueType vectype,
5368 RegisterOperand vecreg>
5369 : SIMDDupFromElement<Q, size, ".s", vectype, v4i32, vecreg,
5370 VectorIndexS, i64, ARM64duplane32> {
5372 let Inst{20-19} = idx;
5373 let Inst{18-16} = 0b100;
5376 class SIMDDup16FromElement<bit Q, string size, ValueType vectype,
5377 RegisterOperand vecreg>
5378 : SIMDDupFromElement<Q, size, ".h", vectype, v8i16, vecreg,
5379 VectorIndexH, i64, ARM64duplane16> {
5381 let Inst{20-18} = idx;
5382 let Inst{17-16} = 0b10;
5385 class SIMDDup8FromElement<bit Q, string size, ValueType vectype,
5386 RegisterOperand vecreg>
5387 : SIMDDupFromElement<Q, size, ".b", vectype, v16i8, vecreg,
5388 VectorIndexB, i64, ARM64duplane8> {
5390 let Inst{20-17} = idx;
5394 class BaseSIMDMov<bit Q, string size, bits<4> imm4, RegisterClass regtype,
5395 Operand idxtype, string asm, list<dag> pattern>
5396 : BaseSIMDInsDup<Q, 0, (outs regtype:$Rd), (ins V128:$Rn, idxtype:$idx), asm,
5397 "{\t$Rd, $Rn" # size # "$idx" #
5398 "|" # size # "\t$Rd, $Rn$idx}", "", pattern> {
5399 let Inst{14-11} = imm4;
5402 class SIMDSMov<bit Q, string size, RegisterClass regtype,
5404 : BaseSIMDMov<Q, size, 0b0101, regtype, idxtype, "smov", []>;
5405 class SIMDUMov<bit Q, string size, ValueType vectype, RegisterClass regtype,
5407 : BaseSIMDMov<Q, size, 0b0111, regtype, idxtype, "umov",
5408 [(set regtype:$Rd, (vector_extract (vectype V128:$Rn), idxtype:$idx))]>;
5410 class SIMDMovAlias<string asm, string size, Instruction inst,
5411 RegisterClass regtype, Operand idxtype>
5412 : InstAlias<asm#"{\t$dst, $src"#size#"$idx" #
5413 "|" # size # "\t$dst, $src$idx}",
5414 (inst regtype:$dst, V128:$src, idxtype:$idx)>;
5417 def vi8to32 : SIMDSMov<0, ".b", GPR32, VectorIndexB> {
5419 let Inst{20-17} = idx;
5422 def vi8to64 : SIMDSMov<1, ".b", GPR64, VectorIndexB> {
5424 let Inst{20-17} = idx;
5427 def vi16to32 : SIMDSMov<0, ".h", GPR32, VectorIndexH> {
5429 let Inst{20-18} = idx;
5430 let Inst{17-16} = 0b10;
5432 def vi16to64 : SIMDSMov<1, ".h", GPR64, VectorIndexH> {
5434 let Inst{20-18} = idx;
5435 let Inst{17-16} = 0b10;
5437 def vi32to64 : SIMDSMov<1, ".s", GPR64, VectorIndexS> {
5439 let Inst{20-19} = idx;
5440 let Inst{18-16} = 0b100;
5445 def vi8 : SIMDUMov<0, ".b", v16i8, GPR32, VectorIndexB> {
5447 let Inst{20-17} = idx;
5450 def vi16 : SIMDUMov<0, ".h", v8i16, GPR32, VectorIndexH> {
5452 let Inst{20-18} = idx;
5453 let Inst{17-16} = 0b10;
5455 def vi32 : SIMDUMov<0, ".s", v4i32, GPR32, VectorIndexS> {
5457 let Inst{20-19} = idx;
5458 let Inst{18-16} = 0b100;
5460 def vi64 : SIMDUMov<1, ".d", v2i64, GPR64, VectorIndexD> {
5463 let Inst{19-16} = 0b1000;
5465 def : SIMDMovAlias<"mov", ".s",
5466 !cast<Instruction>(NAME#"vi32"),
5467 GPR32, VectorIndexS>;
5468 def : SIMDMovAlias<"mov", ".d",
5469 !cast<Instruction>(NAME#"vi64"),
5470 GPR64, VectorIndexD>;
5473 class SIMDInsFromMain<string size, ValueType vectype,
5474 RegisterClass regtype, Operand idxtype>
5475 : BaseSIMDInsDup<1, 0, (outs V128:$dst),
5476 (ins V128:$Rd, idxtype:$idx, regtype:$Rn), "ins",
5477 "{\t$Rd" # size # "$idx, $Rn" #
5478 "|" # size # "\t$Rd$idx, $Rn}",
5481 (vector_insert (vectype V128:$Rd), regtype:$Rn, idxtype:$idx))]> {
5482 let Inst{14-11} = 0b0011;
5485 class SIMDInsFromElement<string size, ValueType vectype,
5486 ValueType elttype, Operand idxtype>
5487 : BaseSIMDInsDup<1, 1, (outs V128:$dst),
5488 (ins V128:$Rd, idxtype:$idx, V128:$Rn, idxtype:$idx2), "ins",
5489 "{\t$Rd" # size # "$idx, $Rn" # size # "$idx2" #
5490 "|" # size # "\t$Rd$idx, $Rn$idx2}",
5495 (elttype (vector_extract (vectype V128:$Rn), idxtype:$idx2)),
5498 class SIMDInsMainMovAlias<string size, Instruction inst,
5499 RegisterClass regtype, Operand idxtype>
5500 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" #
5501 "|" # size #"\t$dst$idx, $src}",
5502 (inst V128:$dst, idxtype:$idx, regtype:$src)>;
5503 class SIMDInsElementMovAlias<string size, Instruction inst,
5505 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" # size # "$idx2" #
5506 # "|" # size #" $dst$idx, $src$idx2}",
5507 (inst V128:$dst, idxtype:$idx, V128:$src, idxtype:$idx2)>;
5510 multiclass SIMDIns {
5511 def vi8gpr : SIMDInsFromMain<".b", v16i8, GPR32, VectorIndexB> {
5513 let Inst{20-17} = idx;
5516 def vi16gpr : SIMDInsFromMain<".h", v8i16, GPR32, VectorIndexH> {
5518 let Inst{20-18} = idx;
5519 let Inst{17-16} = 0b10;
5521 def vi32gpr : SIMDInsFromMain<".s", v4i32, GPR32, VectorIndexS> {
5523 let Inst{20-19} = idx;
5524 let Inst{18-16} = 0b100;
5526 def vi64gpr : SIMDInsFromMain<".d", v2i64, GPR64, VectorIndexD> {
5529 let Inst{19-16} = 0b1000;
5532 def vi8lane : SIMDInsFromElement<".b", v16i8, i32, VectorIndexB> {
5535 let Inst{20-17} = idx;
5537 let Inst{14-11} = idx2;
5539 def vi16lane : SIMDInsFromElement<".h", v8i16, i32, VectorIndexH> {
5542 let Inst{20-18} = idx;
5543 let Inst{17-16} = 0b10;
5544 let Inst{14-12} = idx2;
5547 def vi32lane : SIMDInsFromElement<".s", v4i32, i32, VectorIndexS> {
5550 let Inst{20-19} = idx;
5551 let Inst{18-16} = 0b100;
5552 let Inst{14-13} = idx2;
5553 let Inst{12-11} = 0;
5555 def vi64lane : SIMDInsFromElement<".d", v2i64, i64, VectorIndexD> {
5559 let Inst{19-16} = 0b1000;
5560 let Inst{14} = idx2;
5561 let Inst{13-11} = 0;
5564 // For all forms of the INS instruction, the "mov" mnemonic is the
5565 // preferred alias. Why they didn't just call the instruction "mov" in
5566 // the first place is a very good question indeed...
5567 def : SIMDInsMainMovAlias<".b", !cast<Instruction>(NAME#"vi8gpr"),
5568 GPR32, VectorIndexB>;
5569 def : SIMDInsMainMovAlias<".h", !cast<Instruction>(NAME#"vi16gpr"),
5570 GPR32, VectorIndexH>;
5571 def : SIMDInsMainMovAlias<".s", !cast<Instruction>(NAME#"vi32gpr"),
5572 GPR32, VectorIndexS>;
5573 def : SIMDInsMainMovAlias<".d", !cast<Instruction>(NAME#"vi64gpr"),
5574 GPR64, VectorIndexD>;
5576 def : SIMDInsElementMovAlias<".b", !cast<Instruction>(NAME#"vi8lane"),
5578 def : SIMDInsElementMovAlias<".h", !cast<Instruction>(NAME#"vi16lane"),
5580 def : SIMDInsElementMovAlias<".s", !cast<Instruction>(NAME#"vi32lane"),
5582 def : SIMDInsElementMovAlias<".d", !cast<Instruction>(NAME#"vi64lane"),
5586 //----------------------------------------------------------------------------
5588 //----------------------------------------------------------------------------
5590 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
5591 class BaseSIMDTableLookup<bit Q, bits<2> len, bit op, RegisterOperand vectype,
5592 RegisterOperand listtype, string asm, string kind>
5593 : I<(outs vectype:$Vd), (ins listtype:$Vn, vectype:$Vm), asm,
5594 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "", []>,
5601 let Inst{29-21} = 0b001110000;
5602 let Inst{20-16} = Vm;
5604 let Inst{14-13} = len;
5606 let Inst{11-10} = 0b00;
5611 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
5612 class BaseSIMDTableLookupTied<bit Q, bits<2> len, bit op, RegisterOperand vectype,
5613 RegisterOperand listtype, string asm, string kind>
5614 : I<(outs vectype:$dst), (ins vectype:$Vd, listtype:$Vn, vectype:$Vm), asm,
5615 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "$Vd = $dst", []>,
5622 let Inst{29-21} = 0b001110000;
5623 let Inst{20-16} = Vm;
5625 let Inst{14-13} = len;
5627 let Inst{11-10} = 0b00;
5632 class SIMDTableLookupAlias<string asm, Instruction inst,
5633 RegisterOperand vectype, RegisterOperand listtype>
5634 : InstAlias<!strconcat(asm, "\t$dst, $lst, $index"),
5635 (inst vectype:$dst, listtype:$lst, vectype:$index), 0>;
5637 multiclass SIMDTableLookup<bit op, string asm> {
5638 def v8i8One : BaseSIMDTableLookup<0, 0b00, op, V64, VecListOne16b,
5640 def v8i8Two : BaseSIMDTableLookup<0, 0b01, op, V64, VecListTwo16b,
5642 def v8i8Three : BaseSIMDTableLookup<0, 0b10, op, V64, VecListThree16b,
5644 def v8i8Four : BaseSIMDTableLookup<0, 0b11, op, V64, VecListFour16b,
5646 def v16i8One : BaseSIMDTableLookup<1, 0b00, op, V128, VecListOne16b,
5648 def v16i8Two : BaseSIMDTableLookup<1, 0b01, op, V128, VecListTwo16b,
5650 def v16i8Three: BaseSIMDTableLookup<1, 0b10, op, V128, VecListThree16b,
5652 def v16i8Four : BaseSIMDTableLookup<1, 0b11, op, V128, VecListFour16b,
5655 def : SIMDTableLookupAlias<asm # ".8b",
5656 !cast<Instruction>(NAME#"v8i8One"),
5657 V64, VecListOne128>;
5658 def : SIMDTableLookupAlias<asm # ".8b",
5659 !cast<Instruction>(NAME#"v8i8Two"),
5660 V64, VecListTwo128>;
5661 def : SIMDTableLookupAlias<asm # ".8b",
5662 !cast<Instruction>(NAME#"v8i8Three"),
5663 V64, VecListThree128>;
5664 def : SIMDTableLookupAlias<asm # ".8b",
5665 !cast<Instruction>(NAME#"v8i8Four"),
5666 V64, VecListFour128>;
5667 def : SIMDTableLookupAlias<asm # ".16b",
5668 !cast<Instruction>(NAME#"v16i8One"),
5669 V128, VecListOne128>;
5670 def : SIMDTableLookupAlias<asm # ".16b",
5671 !cast<Instruction>(NAME#"v16i8Two"),
5672 V128, VecListTwo128>;
5673 def : SIMDTableLookupAlias<asm # ".16b",
5674 !cast<Instruction>(NAME#"v16i8Three"),
5675 V128, VecListThree128>;
5676 def : SIMDTableLookupAlias<asm # ".16b",
5677 !cast<Instruction>(NAME#"v16i8Four"),
5678 V128, VecListFour128>;
5681 multiclass SIMDTableLookupTied<bit op, string asm> {
5682 def v8i8One : BaseSIMDTableLookupTied<0, 0b00, op, V64, VecListOne16b,
5684 def v8i8Two : BaseSIMDTableLookupTied<0, 0b01, op, V64, VecListTwo16b,
5686 def v8i8Three : BaseSIMDTableLookupTied<0, 0b10, op, V64, VecListThree16b,
5688 def v8i8Four : BaseSIMDTableLookupTied<0, 0b11, op, V64, VecListFour16b,
5690 def v16i8One : BaseSIMDTableLookupTied<1, 0b00, op, V128, VecListOne16b,
5692 def v16i8Two : BaseSIMDTableLookupTied<1, 0b01, op, V128, VecListTwo16b,
5694 def v16i8Three: BaseSIMDTableLookupTied<1, 0b10, op, V128, VecListThree16b,
5696 def v16i8Four : BaseSIMDTableLookupTied<1, 0b11, op, V128, VecListFour16b,
5699 def : SIMDTableLookupAlias<asm # ".8b",
5700 !cast<Instruction>(NAME#"v8i8One"),
5701 V64, VecListOne128>;
5702 def : SIMDTableLookupAlias<asm # ".8b",
5703 !cast<Instruction>(NAME#"v8i8Two"),
5704 V64, VecListTwo128>;
5705 def : SIMDTableLookupAlias<asm # ".8b",
5706 !cast<Instruction>(NAME#"v8i8Three"),
5707 V64, VecListThree128>;
5708 def : SIMDTableLookupAlias<asm # ".8b",
5709 !cast<Instruction>(NAME#"v8i8Four"),
5710 V64, VecListFour128>;
5711 def : SIMDTableLookupAlias<asm # ".16b",
5712 !cast<Instruction>(NAME#"v16i8One"),
5713 V128, VecListOne128>;
5714 def : SIMDTableLookupAlias<asm # ".16b",
5715 !cast<Instruction>(NAME#"v16i8Two"),
5716 V128, VecListTwo128>;
5717 def : SIMDTableLookupAlias<asm # ".16b",
5718 !cast<Instruction>(NAME#"v16i8Three"),
5719 V128, VecListThree128>;
5720 def : SIMDTableLookupAlias<asm # ".16b",
5721 !cast<Instruction>(NAME#"v16i8Four"),
5722 V128, VecListFour128>;
5726 //----------------------------------------------------------------------------
5727 // AdvSIMD scalar CPY
5728 //----------------------------------------------------------------------------
5729 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5730 class BaseSIMDScalarCPY<RegisterClass regtype, RegisterOperand vectype,
5731 string kind, Operand idxtype>
5732 : I<(outs regtype:$dst), (ins vectype:$src, idxtype:$idx), "mov",
5733 "{\t$dst, $src" # kind # "$idx" #
5734 "|\t$dst, $src$idx}", "", []>,
5738 let Inst{31-21} = 0b01011110000;
5739 let Inst{15-10} = 0b000001;
5740 let Inst{9-5} = src;
5741 let Inst{4-0} = dst;
5744 class SIMDScalarCPYAlias<string asm, string size, Instruction inst,
5745 RegisterClass regtype, RegisterOperand vectype, Operand idxtype>
5746 : InstAlias<asm # "{\t$dst, $src" # size # "$index" #
5747 # "|\t$dst, $src$index}",
5748 (inst regtype:$dst, vectype:$src, idxtype:$index)>;
5751 multiclass SIMDScalarCPY<string asm> {
5752 def i8 : BaseSIMDScalarCPY<FPR8, V128, ".b", VectorIndexB> {
5754 let Inst{20-17} = idx;
5757 def i16 : BaseSIMDScalarCPY<FPR16, V128, ".h", VectorIndexH> {
5759 let Inst{20-18} = idx;
5760 let Inst{17-16} = 0b10;
5762 def i32 : BaseSIMDScalarCPY<FPR32, V128, ".s", VectorIndexS> {
5764 let Inst{20-19} = idx;
5765 let Inst{18-16} = 0b100;
5767 def i64 : BaseSIMDScalarCPY<FPR64, V128, ".d", VectorIndexD> {
5770 let Inst{19-16} = 0b1000;
5773 // 'DUP' mnemonic aliases.
5774 def : SIMDScalarCPYAlias<"dup", ".b",
5775 !cast<Instruction>(NAME#"i8"),
5776 FPR8, V128, VectorIndexB>;
5777 def : SIMDScalarCPYAlias<"dup", ".h",
5778 !cast<Instruction>(NAME#"i16"),
5779 FPR16, V128, VectorIndexH>;
5780 def : SIMDScalarCPYAlias<"dup", ".s",
5781 !cast<Instruction>(NAME#"i32"),
5782 FPR32, V128, VectorIndexS>;
5783 def : SIMDScalarCPYAlias<"dup", ".d",
5784 !cast<Instruction>(NAME#"i64"),
5785 FPR64, V128, VectorIndexD>;
5788 //----------------------------------------------------------------------------
5789 // AdvSIMD modified immediate instructions
5790 //----------------------------------------------------------------------------
5792 class BaseSIMDModifiedImm<bit Q, bit op, dag oops, dag iops,
5793 string asm, string op_string,
5794 string cstr, list<dag> pattern>
5795 : I<oops, iops, asm, op_string, cstr, pattern>,
5802 let Inst{28-19} = 0b0111100000;
5803 let Inst{18-16} = imm8{7-5};
5804 let Inst{11-10} = 0b01;
5805 let Inst{9-5} = imm8{4-0};
5809 class BaseSIMDModifiedImmVector<bit Q, bit op, RegisterOperand vectype,
5810 Operand immtype, dag opt_shift_iop,
5811 string opt_shift, string asm, string kind,
5813 : BaseSIMDModifiedImm<Q, op, (outs vectype:$Rd),
5814 !con((ins immtype:$imm8), opt_shift_iop), asm,
5815 "{\t$Rd" # kind # ", $imm8" # opt_shift #
5816 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
5818 let DecoderMethod = "DecodeModImmInstruction";
5821 class BaseSIMDModifiedImmVectorTied<bit Q, bit op, RegisterOperand vectype,
5822 Operand immtype, dag opt_shift_iop,
5823 string opt_shift, string asm, string kind,
5825 : BaseSIMDModifiedImm<Q, op, (outs vectype:$dst),
5826 !con((ins vectype:$Rd, immtype:$imm8), opt_shift_iop),
5827 asm, "{\t$Rd" # kind # ", $imm8" # opt_shift #
5828 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
5829 "$Rd = $dst", pattern> {
5830 let DecoderMethod = "DecodeModImmTiedInstruction";
5833 class BaseSIMDModifiedImmVectorShift<bit Q, bit op, bits<2> b15_b12,
5834 RegisterOperand vectype, string asm,
5835 string kind, list<dag> pattern>
5836 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5837 (ins logical_vec_shift:$shift),
5838 "$shift", asm, kind, pattern> {
5840 let Inst{15} = b15_b12{1};
5841 let Inst{14-13} = shift;
5842 let Inst{12} = b15_b12{0};
5845 class BaseSIMDModifiedImmVectorShiftTied<bit Q, bit op, bits<2> b15_b12,
5846 RegisterOperand vectype, string asm,
5847 string kind, list<dag> pattern>
5848 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
5849 (ins logical_vec_shift:$shift),
5850 "$shift", asm, kind, pattern> {
5852 let Inst{15} = b15_b12{1};
5853 let Inst{14-13} = shift;
5854 let Inst{12} = b15_b12{0};
5858 class BaseSIMDModifiedImmVectorShiftHalf<bit Q, bit op, bits<2> b15_b12,
5859 RegisterOperand vectype, string asm,
5860 string kind, list<dag> pattern>
5861 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5862 (ins logical_vec_hw_shift:$shift),
5863 "$shift", asm, kind, pattern> {
5865 let Inst{15} = b15_b12{1};
5867 let Inst{13} = shift{0};
5868 let Inst{12} = b15_b12{0};
5871 class BaseSIMDModifiedImmVectorShiftHalfTied<bit Q, bit op, bits<2> b15_b12,
5872 RegisterOperand vectype, string asm,
5873 string kind, list<dag> pattern>
5874 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
5875 (ins logical_vec_hw_shift:$shift),
5876 "$shift", asm, kind, pattern> {
5878 let Inst{15} = b15_b12{1};
5880 let Inst{13} = shift{0};
5881 let Inst{12} = b15_b12{0};
5884 multiclass SIMDModifiedImmVectorShift<bit op, bits<2> hw_cmode, bits<2> w_cmode,
5886 def v4i16 : BaseSIMDModifiedImmVectorShiftHalf<0, op, hw_cmode, V64,
5888 def v8i16 : BaseSIMDModifiedImmVectorShiftHalf<1, op, hw_cmode, V128,
5891 def v2i32 : BaseSIMDModifiedImmVectorShift<0, op, w_cmode, V64,
5893 def v4i32 : BaseSIMDModifiedImmVectorShift<1, op, w_cmode, V128,
5897 multiclass SIMDModifiedImmVectorShiftTied<bit op, bits<2> hw_cmode,
5898 bits<2> w_cmode, string asm,
5900 def v4i16 : BaseSIMDModifiedImmVectorShiftHalfTied<0, op, hw_cmode, V64,
5902 [(set (v4i16 V64:$dst), (OpNode V64:$Rd,
5904 (i32 imm:$shift)))]>;
5905 def v8i16 : BaseSIMDModifiedImmVectorShiftHalfTied<1, op, hw_cmode, V128,
5907 [(set (v8i16 V128:$dst), (OpNode V128:$Rd,
5909 (i32 imm:$shift)))]>;
5911 def v2i32 : BaseSIMDModifiedImmVectorShiftTied<0, op, w_cmode, V64,
5913 [(set (v2i32 V64:$dst), (OpNode V64:$Rd,
5915 (i32 imm:$shift)))]>;
5916 def v4i32 : BaseSIMDModifiedImmVectorShiftTied<1, op, w_cmode, V128,
5918 [(set (v4i32 V128:$dst), (OpNode V128:$Rd,
5920 (i32 imm:$shift)))]>;
5923 class SIMDModifiedImmMoveMSL<bit Q, bit op, bits<4> cmode,
5924 RegisterOperand vectype, string asm,
5925 string kind, list<dag> pattern>
5926 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5927 (ins move_vec_shift:$shift),
5928 "$shift", asm, kind, pattern> {
5930 let Inst{15-13} = cmode{3-1};
5931 let Inst{12} = shift;
5934 class SIMDModifiedImmVectorNoShift<bit Q, bit op, bits<4> cmode,
5935 RegisterOperand vectype,
5936 Operand imm_type, string asm,
5937 string kind, list<dag> pattern>
5938 : BaseSIMDModifiedImmVector<Q, op, vectype, imm_type, (ins), "",
5939 asm, kind, pattern> {
5940 let Inst{15-12} = cmode;
5943 class SIMDModifiedImmScalarNoShift<bit Q, bit op, bits<4> cmode, string asm,
5945 : BaseSIMDModifiedImm<Q, op, (outs FPR64:$Rd), (ins simdimmtype10:$imm8), asm,
5946 "\t$Rd, $imm8", "", pattern> {
5947 let Inst{15-12} = cmode;
5948 let DecoderMethod = "DecodeModImmInstruction";
5951 //----------------------------------------------------------------------------
5952 // AdvSIMD indexed element
5953 //----------------------------------------------------------------------------
5955 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5956 class BaseSIMDIndexed<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
5957 RegisterOperand dst_reg, RegisterOperand lhs_reg,
5958 RegisterOperand rhs_reg, Operand vec_idx, string asm,
5959 string apple_kind, string dst_kind, string lhs_kind,
5960 string rhs_kind, list<dag> pattern>
5961 : I<(outs dst_reg:$Rd), (ins lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx),
5963 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
5964 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "", pattern>,
5973 let Inst{28} = Scalar;
5974 let Inst{27-24} = 0b1111;
5975 let Inst{23-22} = size;
5976 // Bit 21 must be set by the derived class.
5977 let Inst{20-16} = Rm;
5978 let Inst{15-12} = opc;
5979 // Bit 11 must be set by the derived class.
5985 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5986 class BaseSIMDIndexedTied<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
5987 RegisterOperand dst_reg, RegisterOperand lhs_reg,
5988 RegisterOperand rhs_reg, Operand vec_idx, string asm,
5989 string apple_kind, string dst_kind, string lhs_kind,
5990 string rhs_kind, list<dag> pattern>
5991 : I<(outs dst_reg:$dst),
5992 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx), asm,
5993 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
5994 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "$Rd = $dst", pattern>,
6003 let Inst{28} = Scalar;
6004 let Inst{27-24} = 0b1111;
6005 let Inst{23-22} = size;
6006 // Bit 21 must be set by the derived class.
6007 let Inst{20-16} = Rm;
6008 let Inst{15-12} = opc;
6009 // Bit 11 must be set by the derived class.
6015 multiclass SIMDFPIndexedSD<bit U, bits<4> opc, string asm,
6016 SDPatternOperator OpNode> {
6017 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6020 asm, ".2s", ".2s", ".2s", ".s",
6021 [(set (v2f32 V64:$Rd),
6022 (OpNode (v2f32 V64:$Rn),
6023 (v2f32 (ARM64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
6025 let Inst{11} = idx{1};
6026 let Inst{21} = idx{0};
6029 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6032 asm, ".4s", ".4s", ".4s", ".s",
6033 [(set (v4f32 V128:$Rd),
6034 (OpNode (v4f32 V128:$Rn),
6035 (v4f32 (ARM64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
6037 let Inst{11} = idx{1};
6038 let Inst{21} = idx{0};
6041 def v2i64_indexed : BaseSIMDIndexed<1, U, 0, 0b11, opc,
6044 asm, ".2d", ".2d", ".2d", ".d",
6045 [(set (v2f64 V128:$Rd),
6046 (OpNode (v2f64 V128:$Rn),
6047 (v2f64 (ARM64duplane64 (v2f64 V128:$Rm), VectorIndexD:$idx))))]> {
6049 let Inst{11} = idx{0};
6053 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6054 FPR32Op, FPR32Op, V128, VectorIndexS,
6055 asm, ".s", "", "", ".s",
6056 [(set (f32 FPR32Op:$Rd),
6057 (OpNode (f32 FPR32Op:$Rn),
6058 (f32 (vector_extract (v4f32 V128:$Rm),
6059 VectorIndexS:$idx))))]> {
6061 let Inst{11} = idx{1};
6062 let Inst{21} = idx{0};
6065 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b11, opc,
6066 FPR64Op, FPR64Op, V128, VectorIndexD,
6067 asm, ".d", "", "", ".d",
6068 [(set (f64 FPR64Op:$Rd),
6069 (OpNode (f64 FPR64Op:$Rn),
6070 (f64 (vector_extract (v2f64 V128:$Rm),
6071 VectorIndexD:$idx))))]> {
6073 let Inst{11} = idx{0};
6078 multiclass SIMDFPIndexedSDTiedPatterns<string INST, SDPatternOperator OpNode> {
6079 // 2 variants for the .2s version: DUPLANE from 128-bit and DUP scalar.
6080 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6081 (ARM64duplane32 (v4f32 V128:$Rm),
6082 VectorIndexS:$idx))),
6083 (!cast<Instruction>(INST # v2i32_indexed)
6084 V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6085 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6086 (ARM64dup (f32 FPR32Op:$Rm)))),
6087 (!cast<Instruction>(INST # "v2i32_indexed") V64:$Rd, V64:$Rn,
6088 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6091 // 2 variants for the .4s version: DUPLANE from 128-bit and DUP scalar.
6092 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6093 (ARM64duplane32 (v4f32 V128:$Rm),
6094 VectorIndexS:$idx))),
6095 (!cast<Instruction>(INST # "v4i32_indexed")
6096 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6097 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6098 (ARM64dup (f32 FPR32Op:$Rm)))),
6099 (!cast<Instruction>(INST # "v4i32_indexed") V128:$Rd, V128:$Rn,
6100 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6102 // 2 variants for the .2d version: DUPLANE from 128-bit and DUP scalar.
6103 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6104 (ARM64duplane64 (v2f64 V128:$Rm),
6105 VectorIndexD:$idx))),
6106 (!cast<Instruction>(INST # "v2i64_indexed")
6107 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6108 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6109 (ARM64dup (f64 FPR64Op:$Rm)))),
6110 (!cast<Instruction>(INST # "v2i64_indexed") V128:$Rd, V128:$Rn,
6111 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
6113 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
6114 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6115 (vector_extract (v4f32 V128:$Rm), VectorIndexS:$idx))),
6116 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
6117 V128:$Rm, VectorIndexS:$idx)>;
6118 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6119 (vector_extract (v2f32 V64:$Rm), VectorIndexS:$idx))),
6120 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
6121 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
6123 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
6124 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
6125 (vector_extract (v2f64 V128:$Rm), VectorIndexD:$idx))),
6126 (!cast<Instruction>(INST # "v1i64_indexed") FPR64:$Rd, FPR64:$Rn,
6127 V128:$Rm, VectorIndexD:$idx)>;
6130 multiclass SIMDFPIndexedSDTied<bit U, bits<4> opc, string asm> {
6131 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc, V64, V64,
6133 asm, ".2s", ".2s", ".2s", ".s", []> {
6135 let Inst{11} = idx{1};
6136 let Inst{21} = idx{0};
6139 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6142 asm, ".4s", ".4s", ".4s", ".s", []> {
6144 let Inst{11} = idx{1};
6145 let Inst{21} = idx{0};
6148 def v2i64_indexed : BaseSIMDIndexedTied<1, U, 0, 0b11, opc,
6151 asm, ".2d", ".2d", ".2d", ".d", []> {
6153 let Inst{11} = idx{0};
6158 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
6159 FPR32Op, FPR32Op, V128, VectorIndexS,
6160 asm, ".s", "", "", ".s", []> {
6162 let Inst{11} = idx{1};
6163 let Inst{21} = idx{0};
6166 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b11, opc,
6167 FPR64Op, FPR64Op, V128, VectorIndexD,
6168 asm, ".d", "", "", ".d", []> {
6170 let Inst{11} = idx{0};
6175 multiclass SIMDIndexedHS<bit U, bits<4> opc, string asm,
6176 SDPatternOperator OpNode> {
6177 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc, V64, V64,
6178 V128_lo, VectorIndexH,
6179 asm, ".4h", ".4h", ".4h", ".h",
6180 [(set (v4i16 V64:$Rd),
6181 (OpNode (v4i16 V64:$Rn),
6182 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6184 let Inst{11} = idx{2};
6185 let Inst{21} = idx{1};
6186 let Inst{20} = idx{0};
6189 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6191 V128_lo, VectorIndexH,
6192 asm, ".8h", ".8h", ".8h", ".h",
6193 [(set (v8i16 V128:$Rd),
6194 (OpNode (v8i16 V128:$Rn),
6195 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6197 let Inst{11} = idx{2};
6198 let Inst{21} = idx{1};
6199 let Inst{20} = idx{0};
6202 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6205 asm, ".2s", ".2s", ".2s", ".s",
6206 [(set (v2i32 V64:$Rd),
6207 (OpNode (v2i32 V64:$Rn),
6208 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6210 let Inst{11} = idx{1};
6211 let Inst{21} = idx{0};
6214 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6217 asm, ".4s", ".4s", ".4s", ".s",
6218 [(set (v4i32 V128:$Rd),
6219 (OpNode (v4i32 V128:$Rn),
6220 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6222 let Inst{11} = idx{1};
6223 let Inst{21} = idx{0};
6226 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
6227 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
6228 asm, ".h", "", "", ".h", []> {
6230 let Inst{11} = idx{2};
6231 let Inst{21} = idx{1};
6232 let Inst{20} = idx{0};
6235 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6236 FPR32Op, FPR32Op, V128, VectorIndexS,
6237 asm, ".s", "", "", ".s",
6238 [(set (i32 FPR32Op:$Rd),
6239 (OpNode FPR32Op:$Rn,
6240 (i32 (vector_extract (v4i32 V128:$Rm),
6241 VectorIndexS:$idx))))]> {
6243 let Inst{11} = idx{1};
6244 let Inst{21} = idx{0};
6248 multiclass SIMDVectorIndexedHS<bit U, bits<4> opc, string asm,
6249 SDPatternOperator OpNode> {
6250 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6252 V128_lo, VectorIndexH,
6253 asm, ".4h", ".4h", ".4h", ".h",
6254 [(set (v4i16 V64:$Rd),
6255 (OpNode (v4i16 V64:$Rn),
6256 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6258 let Inst{11} = idx{2};
6259 let Inst{21} = idx{1};
6260 let Inst{20} = idx{0};
6263 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6265 V128_lo, VectorIndexH,
6266 asm, ".8h", ".8h", ".8h", ".h",
6267 [(set (v8i16 V128:$Rd),
6268 (OpNode (v8i16 V128:$Rn),
6269 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6271 let Inst{11} = idx{2};
6272 let Inst{21} = idx{1};
6273 let Inst{20} = idx{0};
6276 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6279 asm, ".2s", ".2s", ".2s", ".s",
6280 [(set (v2i32 V64:$Rd),
6281 (OpNode (v2i32 V64:$Rn),
6282 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6284 let Inst{11} = idx{1};
6285 let Inst{21} = idx{0};
6288 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6291 asm, ".4s", ".4s", ".4s", ".s",
6292 [(set (v4i32 V128:$Rd),
6293 (OpNode (v4i32 V128:$Rn),
6294 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6296 let Inst{11} = idx{1};
6297 let Inst{21} = idx{0};
6301 multiclass SIMDVectorIndexedHSTied<bit U, bits<4> opc, string asm,
6302 SDPatternOperator OpNode> {
6303 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc, V64, V64,
6304 V128_lo, VectorIndexH,
6305 asm, ".4h", ".4h", ".4h", ".h",
6306 [(set (v4i16 V64:$dst),
6307 (OpNode (v4i16 V64:$Rd),(v4i16 V64:$Rn),
6308 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6310 let Inst{11} = idx{2};
6311 let Inst{21} = idx{1};
6312 let Inst{20} = idx{0};
6315 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6317 V128_lo, VectorIndexH,
6318 asm, ".8h", ".8h", ".8h", ".h",
6319 [(set (v8i16 V128:$dst),
6320 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
6321 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6323 let Inst{11} = idx{2};
6324 let Inst{21} = idx{1};
6325 let Inst{20} = idx{0};
6328 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6331 asm, ".2s", ".2s", ".2s", ".s",
6332 [(set (v2i32 V64:$dst),
6333 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
6334 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6336 let Inst{11} = idx{1};
6337 let Inst{21} = idx{0};
6340 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6343 asm, ".4s", ".4s", ".4s", ".s",
6344 [(set (v4i32 V128:$dst),
6345 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
6346 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6348 let Inst{11} = idx{1};
6349 let Inst{21} = idx{0};
6353 multiclass SIMDIndexedLongSD<bit U, bits<4> opc, string asm,
6354 SDPatternOperator OpNode> {
6355 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6357 V128_lo, VectorIndexH,
6358 asm, ".4s", ".4s", ".4h", ".h",
6359 [(set (v4i32 V128:$Rd),
6360 (OpNode (v4i16 V64:$Rn),
6361 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6363 let Inst{11} = idx{2};
6364 let Inst{21} = idx{1};
6365 let Inst{20} = idx{0};
6368 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6370 V128_lo, VectorIndexH,
6371 asm#"2", ".4s", ".4s", ".8h", ".h",
6372 [(set (v4i32 V128:$Rd),
6373 (OpNode (extract_high_v8i16 V128:$Rn),
6374 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6375 VectorIndexH:$idx))))]> {
6378 let Inst{11} = idx{2};
6379 let Inst{21} = idx{1};
6380 let Inst{20} = idx{0};
6383 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6386 asm, ".2d", ".2d", ".2s", ".s",
6387 [(set (v2i64 V128:$Rd),
6388 (OpNode (v2i32 V64:$Rn),
6389 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6391 let Inst{11} = idx{1};
6392 let Inst{21} = idx{0};
6395 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6398 asm#"2", ".2d", ".2d", ".4s", ".s",
6399 [(set (v2i64 V128:$Rd),
6400 (OpNode (extract_high_v4i32 V128:$Rn),
6401 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6402 VectorIndexS:$idx))))]> {
6404 let Inst{11} = idx{1};
6405 let Inst{21} = idx{0};
6408 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
6409 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
6410 asm, ".h", "", "", ".h", []> {
6412 let Inst{11} = idx{2};
6413 let Inst{21} = idx{1};
6414 let Inst{20} = idx{0};
6417 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6418 FPR64Op, FPR32Op, V128, VectorIndexS,
6419 asm, ".s", "", "", ".s", []> {
6421 let Inst{11} = idx{1};
6422 let Inst{21} = idx{0};
6426 multiclass SIMDIndexedLongSQDMLXSDTied<bit U, bits<4> opc, string asm,
6427 SDPatternOperator Accum> {
6428 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
6430 V128_lo, VectorIndexH,
6431 asm, ".4s", ".4s", ".4h", ".h",
6432 [(set (v4i32 V128:$dst),
6433 (Accum (v4i32 V128:$Rd),
6434 (v4i32 (int_arm64_neon_sqdmull
6436 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6437 VectorIndexH:$idx))))))]> {
6439 let Inst{11} = idx{2};
6440 let Inst{21} = idx{1};
6441 let Inst{20} = idx{0};
6444 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but an
6445 // intermediate EXTRACT_SUBREG would be untyped.
6446 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
6447 (i32 (vector_extract (v4i32
6448 (int_arm64_neon_sqdmull (v4i16 V64:$Rn),
6449 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6450 VectorIndexH:$idx)))),
6453 (!cast<Instruction>(NAME # v4i16_indexed)
6454 (SUBREG_TO_REG (i32 0), FPR32Op:$Rd, ssub), V64:$Rn,
6455 V128_lo:$Rm, VectorIndexH:$idx),
6458 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6460 V128_lo, VectorIndexH,
6461 asm#"2", ".4s", ".4s", ".8h", ".h",
6462 [(set (v4i32 V128:$dst),
6463 (Accum (v4i32 V128:$Rd),
6464 (v4i32 (int_arm64_neon_sqdmull
6465 (extract_high_v8i16 V128:$Rn),
6467 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6468 VectorIndexH:$idx))))))]> {
6470 let Inst{11} = idx{2};
6471 let Inst{21} = idx{1};
6472 let Inst{20} = idx{0};
6475 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6478 asm, ".2d", ".2d", ".2s", ".s",
6479 [(set (v2i64 V128:$dst),
6480 (Accum (v2i64 V128:$Rd),
6481 (v2i64 (int_arm64_neon_sqdmull
6483 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm),
6484 VectorIndexS:$idx))))))]> {
6486 let Inst{11} = idx{1};
6487 let Inst{21} = idx{0};
6490 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6493 asm#"2", ".2d", ".2d", ".4s", ".s",
6494 [(set (v2i64 V128:$dst),
6495 (Accum (v2i64 V128:$Rd),
6496 (v2i64 (int_arm64_neon_sqdmull
6497 (extract_high_v4i32 V128:$Rn),
6499 (ARM64duplane32 (v4i32 V128:$Rm),
6500 VectorIndexS:$idx))))))]> {
6502 let Inst{11} = idx{1};
6503 let Inst{21} = idx{0};
6506 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
6507 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
6508 asm, ".h", "", "", ".h", []> {
6510 let Inst{11} = idx{2};
6511 let Inst{21} = idx{1};
6512 let Inst{20} = idx{0};
6516 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
6517 FPR64Op, FPR32Op, V128, VectorIndexS,
6518 asm, ".s", "", "", ".s",
6519 [(set (i64 FPR64Op:$dst),
6520 (Accum (i64 FPR64Op:$Rd),
6521 (i64 (int_arm64_neon_sqdmulls_scalar
6523 (i32 (vector_extract (v4i32 V128:$Rm),
6524 VectorIndexS:$idx))))))]> {
6527 let Inst{11} = idx{1};
6528 let Inst{21} = idx{0};
6532 multiclass SIMDVectorIndexedLongSD<bit U, bits<4> opc, string asm,
6533 SDPatternOperator OpNode> {
6534 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6535 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6537 V128_lo, VectorIndexH,
6538 asm, ".4s", ".4s", ".4h", ".h",
6539 [(set (v4i32 V128:$Rd),
6540 (OpNode (v4i16 V64:$Rn),
6541 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6543 let Inst{11} = idx{2};
6544 let Inst{21} = idx{1};
6545 let Inst{20} = idx{0};
6548 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6550 V128_lo, VectorIndexH,
6551 asm#"2", ".4s", ".4s", ".8h", ".h",
6552 [(set (v4i32 V128:$Rd),
6553 (OpNode (extract_high_v8i16 V128:$Rn),
6554 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6555 VectorIndexH:$idx))))]> {
6558 let Inst{11} = idx{2};
6559 let Inst{21} = idx{1};
6560 let Inst{20} = idx{0};
6563 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6566 asm, ".2d", ".2d", ".2s", ".s",
6567 [(set (v2i64 V128:$Rd),
6568 (OpNode (v2i32 V64:$Rn),
6569 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6571 let Inst{11} = idx{1};
6572 let Inst{21} = idx{0};
6575 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6578 asm#"2", ".2d", ".2d", ".4s", ".s",
6579 [(set (v2i64 V128:$Rd),
6580 (OpNode (extract_high_v4i32 V128:$Rn),
6581 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6582 VectorIndexS:$idx))))]> {
6584 let Inst{11} = idx{1};
6585 let Inst{21} = idx{0};
6590 multiclass SIMDVectorIndexedLongSDTied<bit U, bits<4> opc, string asm,
6591 SDPatternOperator OpNode> {
6592 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6593 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
6595 V128_lo, VectorIndexH,
6596 asm, ".4s", ".4s", ".4h", ".h",
6597 [(set (v4i32 V128:$dst),
6598 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn),
6599 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6601 let Inst{11} = idx{2};
6602 let Inst{21} = idx{1};
6603 let Inst{20} = idx{0};
6606 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6608 V128_lo, VectorIndexH,
6609 asm#"2", ".4s", ".4s", ".8h", ".h",
6610 [(set (v4i32 V128:$dst),
6611 (OpNode (v4i32 V128:$Rd),
6612 (extract_high_v8i16 V128:$Rn),
6613 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6614 VectorIndexH:$idx))))]> {
6616 let Inst{11} = idx{2};
6617 let Inst{21} = idx{1};
6618 let Inst{20} = idx{0};
6621 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6624 asm, ".2d", ".2d", ".2s", ".s",
6625 [(set (v2i64 V128:$dst),
6626 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn),
6627 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6629 let Inst{11} = idx{1};
6630 let Inst{21} = idx{0};
6633 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6636 asm#"2", ".2d", ".2d", ".4s", ".s",
6637 [(set (v2i64 V128:$dst),
6638 (OpNode (v2i64 V128:$Rd),
6639 (extract_high_v4i32 V128:$Rn),
6640 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6641 VectorIndexS:$idx))))]> {
6643 let Inst{11} = idx{1};
6644 let Inst{21} = idx{0};
6649 //----------------------------------------------------------------------------
6650 // AdvSIMD scalar shift by immediate
6651 //----------------------------------------------------------------------------
6653 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6654 class BaseSIMDScalarShift<bit U, bits<5> opc, bits<7> fixed_imm,
6655 RegisterClass regtype1, RegisterClass regtype2,
6656 Operand immtype, string asm, list<dag> pattern>
6657 : I<(outs regtype1:$Rd), (ins regtype2:$Rn, immtype:$imm),
6658 asm, "\t$Rd, $Rn, $imm", "", pattern>,
6663 let Inst{31-30} = 0b01;
6665 let Inst{28-23} = 0b111110;
6666 let Inst{22-16} = fixed_imm;
6667 let Inst{15-11} = opc;
6673 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6674 class BaseSIMDScalarShiftTied<bit U, bits<5> opc, bits<7> fixed_imm,
6675 RegisterClass regtype1, RegisterClass regtype2,
6676 Operand immtype, string asm, list<dag> pattern>
6677 : I<(outs regtype1:$dst), (ins regtype1:$Rd, regtype2:$Rn, immtype:$imm),
6678 asm, "\t$Rd, $Rn, $imm", "$Rd = $dst", pattern>,
6683 let Inst{31-30} = 0b01;
6685 let Inst{28-23} = 0b111110;
6686 let Inst{22-16} = fixed_imm;
6687 let Inst{15-11} = opc;
6694 multiclass SIMDScalarRShiftSD<bit U, bits<5> opc, string asm> {
6695 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6696 FPR32, FPR32, vecshiftR32, asm, []> {
6697 let Inst{20-16} = imm{4-0};
6700 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6701 FPR64, FPR64, vecshiftR64, asm, []> {
6702 let Inst{21-16} = imm{5-0};
6706 multiclass SIMDScalarRShiftD<bit U, bits<5> opc, string asm,
6707 SDPatternOperator OpNode> {
6708 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6709 FPR64, FPR64, vecshiftR64, asm,
6710 [(set (i64 FPR64:$Rd),
6711 (OpNode (i64 FPR64:$Rn), (i32 vecshiftR64:$imm)))]> {
6712 let Inst{21-16} = imm{5-0};
6715 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftR64:$imm))),
6716 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftR64:$imm)>;
6719 multiclass SIMDScalarRShiftDTied<bit U, bits<5> opc, string asm,
6720 SDPatternOperator OpNode = null_frag> {
6721 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
6722 FPR64, FPR64, vecshiftR64, asm,
6723 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn),
6724 (i32 vecshiftR64:$imm)))]> {
6725 let Inst{21-16} = imm{5-0};
6728 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
6729 (i32 vecshiftR64:$imm))),
6730 (!cast<Instruction>(NAME # "d") FPR64:$Rd, FPR64:$Rn,
6734 multiclass SIMDScalarLShiftD<bit U, bits<5> opc, string asm,
6735 SDPatternOperator OpNode> {
6736 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6737 FPR64, FPR64, vecshiftL64, asm,
6738 [(set (v1i64 FPR64:$Rd),
6739 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
6740 let Inst{21-16} = imm{5-0};
6744 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6745 multiclass SIMDScalarLShiftDTied<bit U, bits<5> opc, string asm> {
6746 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
6747 FPR64, FPR64, vecshiftL64, asm, []> {
6748 let Inst{21-16} = imm{5-0};
6752 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6753 multiclass SIMDScalarRShiftBHS<bit U, bits<5> opc, string asm,
6754 SDPatternOperator OpNode = null_frag> {
6755 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6756 FPR8, FPR16, vecshiftR8, asm, []> {
6757 let Inst{18-16} = imm{2-0};
6760 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6761 FPR16, FPR32, vecshiftR16, asm, []> {
6762 let Inst{19-16} = imm{3-0};
6765 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6766 FPR32, FPR64, vecshiftR32, asm,
6767 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn), vecshiftR32:$imm))]> {
6768 let Inst{20-16} = imm{4-0};
6772 multiclass SIMDScalarLShiftBHSD<bit U, bits<5> opc, string asm,
6773 SDPatternOperator OpNode> {
6774 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6775 FPR8, FPR8, vecshiftL8, asm, []> {
6776 let Inst{18-16} = imm{2-0};
6779 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6780 FPR16, FPR16, vecshiftL16, asm, []> {
6781 let Inst{19-16} = imm{3-0};
6784 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6785 FPR32, FPR32, vecshiftL32, asm,
6786 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn), (i32 vecshiftL32:$imm)))]> {
6787 let Inst{20-16} = imm{4-0};
6790 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6791 FPR64, FPR64, vecshiftL64, asm,
6792 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn),
6793 (i32 vecshiftL64:$imm)))]> {
6794 let Inst{21-16} = imm{5-0};
6798 multiclass SIMDScalarRShiftBHSD<bit U, bits<5> opc, string asm> {
6799 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6800 FPR8, FPR8, vecshiftR8, asm, []> {
6801 let Inst{18-16} = imm{2-0};
6804 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6805 FPR16, FPR16, vecshiftR16, asm, []> {
6806 let Inst{19-16} = imm{3-0};
6809 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6810 FPR32, FPR32, vecshiftR32, asm, []> {
6811 let Inst{20-16} = imm{4-0};
6814 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6815 FPR64, FPR64, vecshiftR64, asm, []> {
6816 let Inst{21-16} = imm{5-0};
6820 //----------------------------------------------------------------------------
6821 // AdvSIMD vector x indexed element
6822 //----------------------------------------------------------------------------
6824 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6825 class BaseSIMDVectorShift<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
6826 RegisterOperand dst_reg, RegisterOperand src_reg,
6828 string asm, string dst_kind, string src_kind,
6830 : I<(outs dst_reg:$Rd), (ins src_reg:$Rn, immtype:$imm),
6831 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
6832 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "", pattern>,
6839 let Inst{28-23} = 0b011110;
6840 let Inst{22-16} = fixed_imm;
6841 let Inst{15-11} = opc;
6847 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6848 class BaseSIMDVectorShiftTied<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
6849 RegisterOperand vectype1, RegisterOperand vectype2,
6851 string asm, string dst_kind, string src_kind,
6853 : I<(outs vectype1:$dst), (ins vectype1:$Rd, vectype2:$Rn, immtype:$imm),
6854 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
6855 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "$Rd = $dst", pattern>,
6862 let Inst{28-23} = 0b011110;
6863 let Inst{22-16} = fixed_imm;
6864 let Inst{15-11} = opc;
6870 multiclass SIMDVectorRShiftSD<bit U, bits<5> opc, string asm,
6872 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6873 V64, V64, vecshiftR32,
6875 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (i32 imm:$imm)))]> {
6877 let Inst{20-16} = imm;
6880 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
6881 V128, V128, vecshiftR32,
6883 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (i32 imm:$imm)))]> {
6885 let Inst{20-16} = imm;
6888 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
6889 V128, V128, vecshiftR64,
6891 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (i32 imm:$imm)))]> {
6893 let Inst{21-16} = imm;
6897 multiclass SIMDVectorRShiftSDToFP<bit U, bits<5> opc, string asm,
6899 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6900 V64, V64, vecshiftR32,
6902 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (i32 imm:$imm)))]> {
6904 let Inst{20-16} = imm;
6907 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
6908 V128, V128, vecshiftR32,
6910 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (i32 imm:$imm)))]> {
6912 let Inst{20-16} = imm;
6915 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
6916 V128, V128, vecshiftR64,
6918 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (i32 imm:$imm)))]> {
6920 let Inst{21-16} = imm;
6924 multiclass SIMDVectorRShiftNarrowBHS<bit U, bits<5> opc, string asm,
6925 SDPatternOperator OpNode> {
6926 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
6927 V64, V128, vecshiftR16Narrow,
6929 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))]> {
6931 let Inst{18-16} = imm;
6934 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
6935 V128, V128, vecshiftR16Narrow,
6936 asm#"2", ".16b", ".8h", []> {
6938 let Inst{18-16} = imm;
6939 let hasSideEffects = 0;
6942 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
6943 V64, V128, vecshiftR32Narrow,
6945 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))]> {
6947 let Inst{19-16} = imm;
6950 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
6951 V128, V128, vecshiftR32Narrow,
6952 asm#"2", ".8h", ".4s", []> {
6954 let Inst{19-16} = imm;
6955 let hasSideEffects = 0;
6958 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6959 V64, V128, vecshiftR64Narrow,
6961 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))]> {
6963 let Inst{20-16} = imm;
6966 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
6967 V128, V128, vecshiftR64Narrow,
6968 asm#"2", ".4s", ".2d", []> {
6970 let Inst{20-16} = imm;
6971 let hasSideEffects = 0;
6974 // TableGen doesn't like patters w/ INSERT_SUBREG on the instructions
6975 // themselves, so put them here instead.
6977 // Patterns involving what's effectively an insert high and a normal
6978 // intrinsic, represented by CONCAT_VECTORS.
6979 def : Pat<(concat_vectors (v8i8 V64:$Rd),(OpNode (v8i16 V128:$Rn),
6980 vecshiftR16Narrow:$imm)),
6981 (!cast<Instruction>(NAME # "v16i8_shift")
6982 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6983 V128:$Rn, vecshiftR16Narrow:$imm)>;
6984 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn),
6985 vecshiftR32Narrow:$imm)),
6986 (!cast<Instruction>(NAME # "v8i16_shift")
6987 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6988 V128:$Rn, vecshiftR32Narrow:$imm)>;
6989 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn),
6990 vecshiftR64Narrow:$imm)),
6991 (!cast<Instruction>(NAME # "v4i32_shift")
6992 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6993 V128:$Rn, vecshiftR64Narrow:$imm)>;
6996 multiclass SIMDVectorLShiftBHSD<bit U, bits<5> opc, string asm,
6997 SDPatternOperator OpNode> {
6998 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
6999 V64, V64, vecshiftL8,
7001 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
7002 (i32 vecshiftL8:$imm)))]> {
7004 let Inst{18-16} = imm;
7007 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7008 V128, V128, vecshiftL8,
7009 asm, ".16b", ".16b",
7010 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
7011 (i32 vecshiftL8:$imm)))]> {
7013 let Inst{18-16} = imm;
7016 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7017 V64, V64, vecshiftL16,
7019 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
7020 (i32 vecshiftL16:$imm)))]> {
7022 let Inst{19-16} = imm;
7025 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7026 V128, V128, vecshiftL16,
7028 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
7029 (i32 vecshiftL16:$imm)))]> {
7031 let Inst{19-16} = imm;
7034 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7035 V64, V64, vecshiftL32,
7037 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
7038 (i32 vecshiftL32:$imm)))]> {
7040 let Inst{20-16} = imm;
7043 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7044 V128, V128, vecshiftL32,
7046 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
7047 (i32 vecshiftL32:$imm)))]> {
7049 let Inst{20-16} = imm;
7052 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
7053 V128, V128, vecshiftL64,
7055 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
7056 (i32 vecshiftL64:$imm)))]> {
7058 let Inst{21-16} = imm;
7062 multiclass SIMDVectorRShiftBHSD<bit U, bits<5> opc, string asm,
7063 SDPatternOperator OpNode> {
7064 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7065 V64, V64, vecshiftR8,
7067 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
7068 (i32 vecshiftR8:$imm)))]> {
7070 let Inst{18-16} = imm;
7073 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7074 V128, V128, vecshiftR8,
7075 asm, ".16b", ".16b",
7076 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
7077 (i32 vecshiftR8:$imm)))]> {
7079 let Inst{18-16} = imm;
7082 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7083 V64, V64, vecshiftR16,
7085 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
7086 (i32 vecshiftR16:$imm)))]> {
7088 let Inst{19-16} = imm;
7091 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7092 V128, V128, vecshiftR16,
7094 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
7095 (i32 vecshiftR16:$imm)))]> {
7097 let Inst{19-16} = imm;
7100 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7101 V64, V64, vecshiftR32,
7103 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
7104 (i32 vecshiftR32:$imm)))]> {
7106 let Inst{20-16} = imm;
7109 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7110 V128, V128, vecshiftR32,
7112 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
7113 (i32 vecshiftR32:$imm)))]> {
7115 let Inst{20-16} = imm;
7118 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
7119 V128, V128, vecshiftR64,
7121 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
7122 (i32 vecshiftR64:$imm)))]> {
7124 let Inst{21-16} = imm;
7128 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7129 multiclass SIMDVectorRShiftBHSDTied<bit U, bits<5> opc, string asm,
7130 SDPatternOperator OpNode = null_frag> {
7131 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
7132 V64, V64, vecshiftR8, asm, ".8b", ".8b",
7133 [(set (v8i8 V64:$dst),
7134 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
7135 (i32 vecshiftR8:$imm)))]> {
7137 let Inst{18-16} = imm;
7140 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
7141 V128, V128, vecshiftR8, asm, ".16b", ".16b",
7142 [(set (v16i8 V128:$dst),
7143 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
7144 (i32 vecshiftR8:$imm)))]> {
7146 let Inst{18-16} = imm;
7149 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
7150 V64, V64, vecshiftR16, asm, ".4h", ".4h",
7151 [(set (v4i16 V64:$dst),
7152 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
7153 (i32 vecshiftR16:$imm)))]> {
7155 let Inst{19-16} = imm;
7158 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
7159 V128, V128, vecshiftR16, asm, ".8h", ".8h",
7160 [(set (v8i16 V128:$dst),
7161 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7162 (i32 vecshiftR16:$imm)))]> {
7164 let Inst{19-16} = imm;
7167 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
7168 V64, V64, vecshiftR32, asm, ".2s", ".2s",
7169 [(set (v2i32 V64:$dst),
7170 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7171 (i32 vecshiftR32:$imm)))]> {
7173 let Inst{20-16} = imm;
7176 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
7177 V128, V128, vecshiftR32, asm, ".4s", ".4s",
7178 [(set (v4i32 V128:$dst),
7179 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7180 (i32 vecshiftR32:$imm)))]> {
7182 let Inst{20-16} = imm;
7185 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
7186 V128, V128, vecshiftR64,
7187 asm, ".2d", ".2d", [(set (v2i64 V128:$dst),
7188 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
7189 (i32 vecshiftR64:$imm)))]> {
7191 let Inst{21-16} = imm;
7195 multiclass SIMDVectorLShiftBHSDTied<bit U, bits<5> opc, string asm,
7196 SDPatternOperator OpNode = null_frag> {
7197 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
7198 V64, V64, vecshiftL8,
7200 [(set (v8i8 V64:$dst),
7201 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
7202 (i32 vecshiftL8:$imm)))]> {
7204 let Inst{18-16} = imm;
7207 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
7208 V128, V128, vecshiftL8,
7209 asm, ".16b", ".16b",
7210 [(set (v16i8 V128:$dst),
7211 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
7212 (i32 vecshiftL8:$imm)))]> {
7214 let Inst{18-16} = imm;
7217 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
7218 V64, V64, vecshiftL16,
7220 [(set (v4i16 V64:$dst),
7221 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
7222 (i32 vecshiftL16:$imm)))]> {
7224 let Inst{19-16} = imm;
7227 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
7228 V128, V128, vecshiftL16,
7230 [(set (v8i16 V128:$dst),
7231 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7232 (i32 vecshiftL16:$imm)))]> {
7234 let Inst{19-16} = imm;
7237 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
7238 V64, V64, vecshiftL32,
7240 [(set (v2i32 V64:$dst),
7241 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7242 (i32 vecshiftL32:$imm)))]> {
7244 let Inst{20-16} = imm;
7247 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
7248 V128, V128, vecshiftL32,
7250 [(set (v4i32 V128:$dst),
7251 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7252 (i32 vecshiftL32:$imm)))]> {
7254 let Inst{20-16} = imm;
7257 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
7258 V128, V128, vecshiftL64,
7260 [(set (v2i64 V128:$dst),
7261 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
7262 (i32 vecshiftL64:$imm)))]> {
7264 let Inst{21-16} = imm;
7268 multiclass SIMDVectorLShiftLongBHSD<bit U, bits<5> opc, string asm,
7269 SDPatternOperator OpNode> {
7270 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7271 V128, V64, vecshiftL8, asm, ".8h", ".8b",
7272 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), vecshiftL8:$imm))]> {
7274 let Inst{18-16} = imm;
7277 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7278 V128, V128, vecshiftL8,
7279 asm#"2", ".8h", ".16b",
7280 [(set (v8i16 V128:$Rd),
7281 (OpNode (extract_high_v16i8 V128:$Rn), vecshiftL8:$imm))]> {
7283 let Inst{18-16} = imm;
7286 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7287 V128, V64, vecshiftL16, asm, ".4s", ".4h",
7288 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), vecshiftL16:$imm))]> {
7290 let Inst{19-16} = imm;
7293 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7294 V128, V128, vecshiftL16,
7295 asm#"2", ".4s", ".8h",
7296 [(set (v4i32 V128:$Rd),
7297 (OpNode (extract_high_v8i16 V128:$Rn), vecshiftL16:$imm))]> {
7300 let Inst{19-16} = imm;
7303 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7304 V128, V64, vecshiftL32, asm, ".2d", ".2s",
7305 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), vecshiftL32:$imm))]> {
7307 let Inst{20-16} = imm;
7310 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7311 V128, V128, vecshiftL32,
7312 asm#"2", ".2d", ".4s",
7313 [(set (v2i64 V128:$Rd),
7314 (OpNode (extract_high_v4i32 V128:$Rn), vecshiftL32:$imm))]> {
7316 let Inst{20-16} = imm;
7322 // Vector load/store
7324 // SIMD ldX/stX no-index memory references don't allow the optional
7325 // ", #0" constant and handle post-indexing explicitly, so we use
7326 // a more specialized parse method for them. Otherwise, it's the same as
7327 // the general am_noindex handling.
7328 def MemorySIMDNoIndexOperand : AsmOperandClass {
7329 let Name = "MemorySIMDNoIndex";
7330 let ParserMethod = "tryParseNoIndexMemory";
7332 def am_simdnoindex : Operand<i64>,
7333 ComplexPattern<i64, 1, "SelectAddrModeNoIndex", []> {
7334 let PrintMethod = "printAMNoIndex";
7335 let ParserMatchClass = MemorySIMDNoIndexOperand;
7336 let MIOperandInfo = (ops GPR64sp:$base);
7337 let DecoderMethod = "DecodeGPR64spRegisterClass";
7340 class BaseSIMDLdSt<bit Q, bit L, bits<4> opcode, bits<2> size,
7341 string asm, dag oops, dag iops, list<dag> pattern>
7342 : I<oops, iops, asm, "\t$Vt, $vaddr", "", pattern> {
7347 let Inst{29-23} = 0b0011000;
7349 let Inst{21-16} = 0b000000;
7350 let Inst{15-12} = opcode;
7351 let Inst{11-10} = size;
7352 let Inst{9-5} = vaddr;
7356 class BaseSIMDLdStPost<bit Q, bit L, bits<4> opcode, bits<2> size,
7357 string asm, dag oops, dag iops>
7358 : I<oops, iops, asm, "\t$Vt, $vaddr, $Xm", "", []> {
7364 let Inst{29-23} = 0b0011001;
7367 let Inst{20-16} = Xm;
7368 let Inst{15-12} = opcode;
7369 let Inst{11-10} = size;
7370 let Inst{9-5} = vaddr;
7372 let DecoderMethod = "DecodeSIMDLdStPost";
7375 // The immediate form of AdvSIMD post-indexed addressing is encoded with
7376 // register post-index addressing from the zero register.
7377 multiclass SIMDLdStAliases<string asm, string layout, string Count,
7378 int Offset, int Size> {
7379 // E.g. "ld1 { v0.8b, v1.8b }, [x1], #16"
7380 // "ld1\t$Vt, $vaddr, #16"
7381 // may get mapped to
7382 // (LD1Twov8b_POST VecListTwo8b:$Vt, am_simdnoindex:$vaddr, XZR)
7383 def : InstAlias<asm # "\t$Vt, $vaddr, #" # Offset,
7384 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7385 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
7386 am_simdnoindex:$vaddr, XZR), 1>;
7388 // E.g. "ld1.8b { v0, v1 }, [x1], #16"
7389 // "ld1.8b\t$Vt, $vaddr, #16"
7390 // may get mapped to
7391 // (LD1Twov8b_POST VecListTwo64:$Vt, am_simdnoindex:$vaddr, XZR)
7392 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, #" # Offset,
7393 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7394 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7395 am_simdnoindex:$vaddr, XZR), 0>;
7397 // E.g. "ld1.8b { v0, v1 }, [x1]"
7398 // "ld1\t$Vt, $vaddr"
7399 // may get mapped to
7400 // (LD1Twov8b VecListTwo64:$Vt, am_simdnoindex:$vaddr)
7401 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr",
7402 (!cast<Instruction>(NAME # Count # "v" # layout)
7403 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7404 am_simdnoindex:$vaddr), 0>;
7406 // E.g. "ld1.8b { v0, v1 }, [x1], x2"
7407 // "ld1\t$Vt, $vaddr, $Xm"
7408 // may get mapped to
7409 // (LD1Twov8b_POST VecListTwo64:$Vt, am_simdnoindex:$vaddr, GPR64pi8:$Xm)
7410 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, $Xm",
7411 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7412 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7413 am_simdnoindex:$vaddr,
7414 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
7417 multiclass BaseSIMDLdN<string Count, string asm, string veclist, int Offset128,
7418 int Offset64, bits<4> opcode> {
7419 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
7420 def v16b: BaseSIMDLdSt<1, 1, opcode, 0b00, asm,
7421 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
7422 (ins am_simdnoindex:$vaddr), []>;
7423 def v8h : BaseSIMDLdSt<1, 1, opcode, 0b01, asm,
7424 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
7425 (ins am_simdnoindex:$vaddr), []>;
7426 def v4s : BaseSIMDLdSt<1, 1, opcode, 0b10, asm,
7427 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
7428 (ins am_simdnoindex:$vaddr), []>;
7429 def v2d : BaseSIMDLdSt<1, 1, opcode, 0b11, asm,
7430 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
7431 (ins am_simdnoindex:$vaddr), []>;
7432 def v8b : BaseSIMDLdSt<0, 1, opcode, 0b00, asm,
7433 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
7434 (ins am_simdnoindex:$vaddr), []>;
7435 def v4h : BaseSIMDLdSt<0, 1, opcode, 0b01, asm,
7436 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
7437 (ins am_simdnoindex:$vaddr), []>;
7438 def v2s : BaseSIMDLdSt<0, 1, opcode, 0b10, asm,
7439 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
7440 (ins am_simdnoindex:$vaddr), []>;
7443 def v16b_POST: BaseSIMDLdStPost<1, 1, opcode, 0b00, asm,
7444 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
7445 (ins am_simdnoindex:$vaddr,
7446 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7447 def v8h_POST : BaseSIMDLdStPost<1, 1, opcode, 0b01, asm,
7448 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
7449 (ins am_simdnoindex:$vaddr,
7450 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7451 def v4s_POST : BaseSIMDLdStPost<1, 1, opcode, 0b10, asm,
7452 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
7453 (ins am_simdnoindex:$vaddr,
7454 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7455 def v2d_POST : BaseSIMDLdStPost<1, 1, opcode, 0b11, asm,
7456 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
7457 (ins am_simdnoindex:$vaddr,
7458 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7459 def v8b_POST : BaseSIMDLdStPost<0, 1, opcode, 0b00, asm,
7460 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
7461 (ins am_simdnoindex:$vaddr,
7462 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7463 def v4h_POST : BaseSIMDLdStPost<0, 1, opcode, 0b01, asm,
7464 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
7465 (ins am_simdnoindex:$vaddr,
7466 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7467 def v2s_POST : BaseSIMDLdStPost<0, 1, opcode, 0b10, asm,
7468 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
7469 (ins am_simdnoindex:$vaddr,
7470 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7473 defm : SIMDLdStAliases<asm, "16b", Count, Offset128, 128>;
7474 defm : SIMDLdStAliases<asm, "8h", Count, Offset128, 128>;
7475 defm : SIMDLdStAliases<asm, "4s", Count, Offset128, 128>;
7476 defm : SIMDLdStAliases<asm, "2d", Count, Offset128, 128>;
7477 defm : SIMDLdStAliases<asm, "8b", Count, Offset64, 64>;
7478 defm : SIMDLdStAliases<asm, "4h", Count, Offset64, 64>;
7479 defm : SIMDLdStAliases<asm, "2s", Count, Offset64, 64>;
7482 // Only ld1/st1 has a v1d version.
7483 multiclass BaseSIMDStN<string Count, string asm, string veclist, int Offset128,
7484 int Offset64, bits<4> opcode> {
7485 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in {
7486 def v16b : BaseSIMDLdSt<1, 0, opcode, 0b00, asm, (outs),
7487 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
7488 am_simdnoindex:$vaddr), []>;
7489 def v8h : BaseSIMDLdSt<1, 0, opcode, 0b01, asm, (outs),
7490 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
7491 am_simdnoindex:$vaddr), []>;
7492 def v4s : BaseSIMDLdSt<1, 0, opcode, 0b10, asm, (outs),
7493 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
7494 am_simdnoindex:$vaddr), []>;
7495 def v2d : BaseSIMDLdSt<1, 0, opcode, 0b11, asm, (outs),
7496 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
7497 am_simdnoindex:$vaddr), []>;
7498 def v8b : BaseSIMDLdSt<0, 0, opcode, 0b00, asm, (outs),
7499 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
7500 am_simdnoindex:$vaddr), []>;
7501 def v4h : BaseSIMDLdSt<0, 0, opcode, 0b01, asm, (outs),
7502 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
7503 am_simdnoindex:$vaddr), []>;
7504 def v2s : BaseSIMDLdSt<0, 0, opcode, 0b10, asm, (outs),
7505 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
7506 am_simdnoindex:$vaddr), []>;
7508 def v16b_POST : BaseSIMDLdStPost<1, 0, opcode, 0b00, asm, (outs),
7509 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
7510 am_simdnoindex:$vaddr,
7511 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7512 def v8h_POST : BaseSIMDLdStPost<1, 0, opcode, 0b01, asm, (outs),
7513 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
7514 am_simdnoindex:$vaddr,
7515 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7516 def v4s_POST : BaseSIMDLdStPost<1, 0, opcode, 0b10, asm, (outs),
7517 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
7518 am_simdnoindex:$vaddr,
7519 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7520 def v2d_POST : BaseSIMDLdStPost<1, 0, opcode, 0b11, asm, (outs),
7521 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
7522 am_simdnoindex:$vaddr,
7523 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7524 def v8b_POST : BaseSIMDLdStPost<0, 0, opcode, 0b00, asm, (outs),
7525 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
7526 am_simdnoindex:$vaddr,
7527 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7528 def v4h_POST : BaseSIMDLdStPost<0, 0, opcode, 0b01, asm, (outs),
7529 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
7530 am_simdnoindex:$vaddr,
7531 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7532 def v2s_POST : BaseSIMDLdStPost<0, 0, opcode, 0b10, asm, (outs),
7533 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
7534 am_simdnoindex:$vaddr,
7535 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7538 defm : SIMDLdStAliases<asm, "16b", Count, Offset128, 128>;
7539 defm : SIMDLdStAliases<asm, "8h", Count, Offset128, 128>;
7540 defm : SIMDLdStAliases<asm, "4s", Count, Offset128, 128>;
7541 defm : SIMDLdStAliases<asm, "2d", Count, Offset128, 128>;
7542 defm : SIMDLdStAliases<asm, "8b", Count, Offset64, 64>;
7543 defm : SIMDLdStAliases<asm, "4h", Count, Offset64, 64>;
7544 defm : SIMDLdStAliases<asm, "2s", Count, Offset64, 64>;
7547 multiclass BaseSIMDLd1<string Count, string asm, string veclist,
7548 int Offset128, int Offset64, bits<4> opcode>
7549 : BaseSIMDLdN<Count, asm, veclist, Offset128, Offset64, opcode> {
7551 // LD1 instructions have extra "1d" variants.
7552 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
7553 def v1d : BaseSIMDLdSt<0, 1, opcode, 0b11, asm,
7554 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
7555 (ins am_simdnoindex:$vaddr), []>;
7557 def v1d_POST : BaseSIMDLdStPost<0, 1, opcode, 0b11, asm,
7558 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
7559 (ins am_simdnoindex:$vaddr,
7560 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7563 defm : SIMDLdStAliases<asm, "1d", Count, Offset64, 64>;
7566 multiclass BaseSIMDSt1<string Count, string asm, string veclist,
7567 int Offset128, int Offset64, bits<4> opcode>
7568 : BaseSIMDStN<Count, asm, veclist, Offset128, Offset64, opcode> {
7570 // ST1 instructions have extra "1d" variants.
7571 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in {
7572 def v1d : BaseSIMDLdSt<0, 0, opcode, 0b11, asm, (outs),
7573 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
7574 am_simdnoindex:$vaddr), []>;
7576 def v1d_POST : BaseSIMDLdStPost<0, 0, opcode, 0b11, asm, (outs),
7577 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
7578 am_simdnoindex:$vaddr,
7579 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7582 defm : SIMDLdStAliases<asm, "1d", Count, Offset64, 64>;
7585 multiclass SIMDLd1Multiple<string asm> {
7586 defm One : BaseSIMDLd1<"One", asm, "VecListOne", 16, 8, 0b0111>;
7587 defm Two : BaseSIMDLd1<"Two", asm, "VecListTwo", 32, 16, 0b1010>;
7588 defm Three : BaseSIMDLd1<"Three", asm, "VecListThree", 48, 24, 0b0110>;
7589 defm Four : BaseSIMDLd1<"Four", asm, "VecListFour", 64, 32, 0b0010>;
7592 multiclass SIMDSt1Multiple<string asm> {
7593 defm One : BaseSIMDSt1<"One", asm, "VecListOne", 16, 8, 0b0111>;
7594 defm Two : BaseSIMDSt1<"Two", asm, "VecListTwo", 32, 16, 0b1010>;
7595 defm Three : BaseSIMDSt1<"Three", asm, "VecListThree", 48, 24, 0b0110>;
7596 defm Four : BaseSIMDSt1<"Four", asm, "VecListFour", 64, 32, 0b0010>;
7599 multiclass SIMDLd2Multiple<string asm> {
7600 defm Two : BaseSIMDLdN<"Two", asm, "VecListTwo", 32, 16, 0b1000>;
7603 multiclass SIMDSt2Multiple<string asm> {
7604 defm Two : BaseSIMDStN<"Two", asm, "VecListTwo", 32, 16, 0b1000>;
7607 multiclass SIMDLd3Multiple<string asm> {
7608 defm Three : BaseSIMDLdN<"Three", asm, "VecListThree", 48, 24, 0b0100>;
7611 multiclass SIMDSt3Multiple<string asm> {
7612 defm Three : BaseSIMDStN<"Three", asm, "VecListThree", 48, 24, 0b0100>;
7615 multiclass SIMDLd4Multiple<string asm> {
7616 defm Four : BaseSIMDLdN<"Four", asm, "VecListFour", 64, 32, 0b0000>;
7619 multiclass SIMDSt4Multiple<string asm> {
7620 defm Four : BaseSIMDStN<"Four", asm, "VecListFour", 64, 32, 0b0000>;
7624 // AdvSIMD Load/store single-element
7627 class BaseSIMDLdStSingle<bit L, bit R, bits<3> opcode,
7628 string asm, string operands, dag oops, dag iops,
7630 : I<oops, iops, asm, operands, "", pattern> {
7634 let Inst{29-24} = 0b001101;
7637 let Inst{15-13} = opcode;
7638 let Inst{9-5} = vaddr;
7640 let DecoderMethod = "DecodeSIMDLdStSingle";
7643 class BaseSIMDLdStSingleTied<bit L, bit R, bits<3> opcode,
7644 string asm, string operands, dag oops, dag iops,
7646 : I<oops, iops, asm, operands, "$Vt = $dst", pattern> {
7650 let Inst{29-24} = 0b001101;
7653 let Inst{15-13} = opcode;
7654 let Inst{9-5} = vaddr;
7656 let DecoderMethod = "DecodeSIMDLdStSingleTied";
7660 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7661 class BaseSIMDLdR<bit Q, bit R, bits<3> opcode, bit S, bits<2> size, string asm,
7663 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, $vaddr",
7664 (outs listtype:$Vt), (ins am_simdnoindex:$vaddr), []> {
7667 let Inst{20-16} = 0b00000;
7669 let Inst{11-10} = size;
7671 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7672 class BaseSIMDLdRPost<bit Q, bit R, bits<3> opcode, bit S, bits<2> size,
7673 string asm, Operand listtype, Operand GPR64pi>
7674 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, $vaddr, $Xm",
7675 (outs listtype:$Vt),
7676 (ins am_simdnoindex:$vaddr, GPR64pi:$Xm), []> {
7680 let Inst{20-16} = Xm;
7682 let Inst{11-10} = size;
7685 multiclass SIMDLdrAliases<string asm, string layout, string Count,
7686 int Offset, int Size> {
7687 // E.g. "ld1r { v0.8b }, [x1], #1"
7688 // "ld1r.8b\t$Vt, $vaddr, #1"
7689 // may get mapped to
7690 // (LD1Rv8b_POST VecListOne8b:$Vt, am_simdnoindex:$vaddr, XZR)
7691 def : InstAlias<asm # "\t$Vt, $vaddr, #" # Offset,
7692 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7693 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
7694 am_simdnoindex:$vaddr, XZR), 1>;
7696 // E.g. "ld1r.8b { v0 }, [x1], #1"
7697 // "ld1r.8b\t$Vt, $vaddr, #1"
7698 // may get mapped to
7699 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, XZR)
7700 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, #" # Offset,
7701 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7702 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7703 am_simdnoindex:$vaddr, XZR), 0>;
7705 // E.g. "ld1r.8b { v0 }, [x1]"
7706 // "ld1r.8b\t$Vt, $vaddr"
7707 // may get mapped to
7708 // (LD1Rv8b VecListOne64:$Vt, am_simdnoindex:$vaddr)
7709 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr",
7710 (!cast<Instruction>(NAME # "v" # layout)
7711 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7712 am_simdnoindex:$vaddr), 0>;
7714 // E.g. "ld1r.8b { v0 }, [x1], x2"
7715 // "ld1r.8b\t$Vt, $vaddr, $Xm"
7716 // may get mapped to
7717 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, GPR64pi1:$Xm)
7718 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, $Xm",
7719 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7720 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7721 am_simdnoindex:$vaddr,
7722 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
7725 multiclass SIMDLdR<bit R, bits<3> opcode, bit S, string asm, string Count,
7726 int Offset1, int Offset2, int Offset4, int Offset8> {
7727 def v8b : BaseSIMDLdR<0, R, opcode, S, 0b00, asm,
7728 !cast<Operand>("VecList" # Count # "8b")>;
7729 def v16b: BaseSIMDLdR<1, R, opcode, S, 0b00, asm,
7730 !cast<Operand>("VecList" # Count #"16b")>;
7731 def v4h : BaseSIMDLdR<0, R, opcode, S, 0b01, asm,
7732 !cast<Operand>("VecList" # Count #"4h")>;
7733 def v8h : BaseSIMDLdR<1, R, opcode, S, 0b01, asm,
7734 !cast<Operand>("VecList" # Count #"8h")>;
7735 def v2s : BaseSIMDLdR<0, R, opcode, S, 0b10, asm,
7736 !cast<Operand>("VecList" # Count #"2s")>;
7737 def v4s : BaseSIMDLdR<1, R, opcode, S, 0b10, asm,
7738 !cast<Operand>("VecList" # Count #"4s")>;
7739 def v1d : BaseSIMDLdR<0, R, opcode, S, 0b11, asm,
7740 !cast<Operand>("VecList" # Count #"1d")>;
7741 def v2d : BaseSIMDLdR<1, R, opcode, S, 0b11, asm,
7742 !cast<Operand>("VecList" # Count #"2d")>;
7744 def v8b_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b00, asm,
7745 !cast<Operand>("VecList" # Count # "8b"),
7746 !cast<Operand>("GPR64pi" # Offset1)>;
7747 def v16b_POST: BaseSIMDLdRPost<1, R, opcode, S, 0b00, asm,
7748 !cast<Operand>("VecList" # Count # "16b"),
7749 !cast<Operand>("GPR64pi" # Offset1)>;
7750 def v4h_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b01, asm,
7751 !cast<Operand>("VecList" # Count # "4h"),
7752 !cast<Operand>("GPR64pi" # Offset2)>;
7753 def v8h_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b01, asm,
7754 !cast<Operand>("VecList" # Count # "8h"),
7755 !cast<Operand>("GPR64pi" # Offset2)>;
7756 def v2s_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b10, asm,
7757 !cast<Operand>("VecList" # Count # "2s"),
7758 !cast<Operand>("GPR64pi" # Offset4)>;
7759 def v4s_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b10, asm,
7760 !cast<Operand>("VecList" # Count # "4s"),
7761 !cast<Operand>("GPR64pi" # Offset4)>;
7762 def v1d_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b11, asm,
7763 !cast<Operand>("VecList" # Count # "1d"),
7764 !cast<Operand>("GPR64pi" # Offset8)>;
7765 def v2d_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b11, asm,
7766 !cast<Operand>("VecList" # Count # "2d"),
7767 !cast<Operand>("GPR64pi" # Offset8)>;
7769 defm : SIMDLdrAliases<asm, "8b", Count, Offset1, 64>;
7770 defm : SIMDLdrAliases<asm, "16b", Count, Offset1, 128>;
7771 defm : SIMDLdrAliases<asm, "4h", Count, Offset2, 64>;
7772 defm : SIMDLdrAliases<asm, "8h", Count, Offset2, 128>;
7773 defm : SIMDLdrAliases<asm, "2s", Count, Offset4, 64>;
7774 defm : SIMDLdrAliases<asm, "4s", Count, Offset4, 128>;
7775 defm : SIMDLdrAliases<asm, "1d", Count, Offset8, 64>;
7776 defm : SIMDLdrAliases<asm, "2d", Count, Offset8, 128>;
7779 class SIMDLdStSingleB<bit L, bit R, bits<3> opcode, string asm,
7780 dag oops, dag iops, list<dag> pattern>
7781 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7783 // idx encoded in Q:S:size fields.
7785 let Inst{30} = idx{3};
7787 let Inst{20-16} = 0b00000;
7788 let Inst{12} = idx{2};
7789 let Inst{11-10} = idx{1-0};
7791 class SIMDLdStSingleBTied<bit L, bit R, bits<3> opcode, string asm,
7792 dag oops, dag iops, list<dag> pattern>
7793 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7795 // idx encoded in Q:S:size fields.
7797 let Inst{30} = idx{3};
7799 let Inst{20-16} = 0b00000;
7800 let Inst{12} = idx{2};
7801 let Inst{11-10} = idx{1-0};
7803 class SIMDLdStSingleBPost<bit L, bit R, bits<3> opcode, string asm,
7805 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7807 // idx encoded in Q:S:size fields.
7810 let Inst{30} = idx{3};
7812 let Inst{20-16} = Xm;
7813 let Inst{12} = idx{2};
7814 let Inst{11-10} = idx{1-0};
7816 class SIMDLdStSingleBTiedPost<bit L, bit R, bits<3> opcode, string asm,
7818 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7820 // idx encoded in Q:S:size fields.
7823 let Inst{30} = idx{3};
7825 let Inst{20-16} = Xm;
7826 let Inst{12} = idx{2};
7827 let Inst{11-10} = idx{1-0};
7830 class SIMDLdStSingleH<bit L, bit R, bits<3> opcode, bit size, string asm,
7831 dag oops, dag iops, list<dag> pattern>
7832 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7834 // idx encoded in Q:S:size<1> fields.
7836 let Inst{30} = idx{2};
7838 let Inst{20-16} = 0b00000;
7839 let Inst{12} = idx{1};
7840 let Inst{11} = idx{0};
7841 let Inst{10} = size;
7843 class SIMDLdStSingleHTied<bit L, bit R, bits<3> opcode, bit size, string asm,
7844 dag oops, dag iops, list<dag> pattern>
7845 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7847 // idx encoded in Q:S:size<1> fields.
7849 let Inst{30} = idx{2};
7851 let Inst{20-16} = 0b00000;
7852 let Inst{12} = idx{1};
7853 let Inst{11} = idx{0};
7854 let Inst{10} = size;
7857 class SIMDLdStSingleHPost<bit L, bit R, bits<3> opcode, bit size, string asm,
7859 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7861 // idx encoded in Q:S:size<1> fields.
7864 let Inst{30} = idx{2};
7866 let Inst{20-16} = Xm;
7867 let Inst{12} = idx{1};
7868 let Inst{11} = idx{0};
7869 let Inst{10} = size;
7871 class SIMDLdStSingleHTiedPost<bit L, bit R, bits<3> opcode, bit size, string asm,
7873 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7875 // idx encoded in Q:S:size<1> fields.
7878 let Inst{30} = idx{2};
7880 let Inst{20-16} = Xm;
7881 let Inst{12} = idx{1};
7882 let Inst{11} = idx{0};
7883 let Inst{10} = size;
7885 class SIMDLdStSingleS<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7886 dag oops, dag iops, list<dag> pattern>
7887 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7889 // idx encoded in Q:S fields.
7891 let Inst{30} = idx{1};
7893 let Inst{20-16} = 0b00000;
7894 let Inst{12} = idx{0};
7895 let Inst{11-10} = size;
7897 class SIMDLdStSingleSTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7898 dag oops, dag iops, list<dag> pattern>
7899 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7901 // idx encoded in Q:S fields.
7903 let Inst{30} = idx{1};
7905 let Inst{20-16} = 0b00000;
7906 let Inst{12} = idx{0};
7907 let Inst{11-10} = size;
7909 class SIMDLdStSingleSPost<bit L, bit R, bits<3> opcode, bits<2> size,
7910 string asm, dag oops, dag iops>
7911 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7913 // idx encoded in Q:S fields.
7916 let Inst{30} = idx{1};
7918 let Inst{20-16} = Xm;
7919 let Inst{12} = idx{0};
7920 let Inst{11-10} = size;
7922 class SIMDLdStSingleSTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
7923 string asm, dag oops, dag iops>
7924 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7926 // idx encoded in Q:S fields.
7929 let Inst{30} = idx{1};
7931 let Inst{20-16} = Xm;
7932 let Inst{12} = idx{0};
7933 let Inst{11-10} = size;
7935 class SIMDLdStSingleD<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7936 dag oops, dag iops, list<dag> pattern>
7937 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7939 // idx encoded in Q field.
7943 let Inst{20-16} = 0b00000;
7945 let Inst{11-10} = size;
7947 class SIMDLdStSingleDTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7948 dag oops, dag iops, list<dag> pattern>
7949 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7951 // idx encoded in Q field.
7955 let Inst{20-16} = 0b00000;
7957 let Inst{11-10} = size;
7959 class SIMDLdStSingleDPost<bit L, bit R, bits<3> opcode, bits<2> size,
7960 string asm, dag oops, dag iops>
7961 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7963 // idx encoded in Q field.
7968 let Inst{20-16} = Xm;
7970 let Inst{11-10} = size;
7972 class SIMDLdStSingleDTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
7973 string asm, dag oops, dag iops>
7974 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7976 // idx encoded in Q field.
7981 let Inst{20-16} = Xm;
7983 let Inst{11-10} = size;
7986 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7987 multiclass SIMDLdSingleBTied<bit R, bits<3> opcode, string asm,
7988 RegisterOperand listtype,
7989 RegisterOperand GPR64pi> {
7990 def i8 : SIMDLdStSingleBTied<1, R, opcode, asm,
7991 (outs listtype:$dst),
7992 (ins listtype:$Vt, VectorIndexB:$idx,
7993 am_simdnoindex:$vaddr), []>;
7995 def i8_POST : SIMDLdStSingleBTiedPost<1, R, opcode, asm,
7996 (outs listtype:$dst),
7997 (ins listtype:$Vt, VectorIndexB:$idx,
7998 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8000 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8001 multiclass SIMDLdSingleHTied<bit R, bits<3> opcode, bit size, string asm,
8002 RegisterOperand listtype,
8003 RegisterOperand GPR64pi> {
8004 def i16 : SIMDLdStSingleHTied<1, R, opcode, size, asm,
8005 (outs listtype:$dst),
8006 (ins listtype:$Vt, VectorIndexH:$idx,
8007 am_simdnoindex:$vaddr), []>;
8009 def i16_POST : SIMDLdStSingleHTiedPost<1, R, opcode, size, asm,
8010 (outs listtype:$dst),
8011 (ins listtype:$Vt, VectorIndexH:$idx,
8012 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8014 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8015 multiclass SIMDLdSingleSTied<bit R, bits<3> opcode, bits<2> size,string asm,
8016 RegisterOperand listtype,
8017 RegisterOperand GPR64pi> {
8018 def i32 : SIMDLdStSingleSTied<1, R, opcode, size, asm,
8019 (outs listtype:$dst),
8020 (ins listtype:$Vt, VectorIndexS:$idx,
8021 am_simdnoindex:$vaddr), []>;
8023 def i32_POST : SIMDLdStSingleSTiedPost<1, R, opcode, size, asm,
8024 (outs listtype:$dst),
8025 (ins listtype:$Vt, VectorIndexS:$idx,
8026 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8028 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8029 multiclass SIMDLdSingleDTied<bit R, bits<3> opcode, bits<2> size, string asm,
8030 RegisterOperand listtype, RegisterOperand GPR64pi> {
8031 def i64 : SIMDLdStSingleDTied<1, R, opcode, size, asm,
8032 (outs listtype:$dst),
8033 (ins listtype:$Vt, VectorIndexD:$idx,
8034 am_simdnoindex:$vaddr), []>;
8036 def i64_POST : SIMDLdStSingleDTiedPost<1, R, opcode, size, asm,
8037 (outs listtype:$dst),
8038 (ins listtype:$Vt, VectorIndexD:$idx,
8039 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8041 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8042 multiclass SIMDStSingleB<bit R, bits<3> opcode, string asm,
8043 RegisterOperand listtype, RegisterOperand GPR64pi> {
8044 def i8 : SIMDLdStSingleB<0, R, opcode, asm,
8045 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
8046 am_simdnoindex:$vaddr), []>;
8048 def i8_POST : SIMDLdStSingleBPost<0, R, opcode, asm,
8049 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
8050 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8052 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8053 multiclass SIMDStSingleH<bit R, bits<3> opcode, bit size, string asm,
8054 RegisterOperand listtype, RegisterOperand GPR64pi> {
8055 def i16 : SIMDLdStSingleH<0, R, opcode, size, asm,
8056 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
8057 am_simdnoindex:$vaddr), []>;
8059 def i16_POST : SIMDLdStSingleHPost<0, R, opcode, size, asm,
8060 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
8061 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8063 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8064 multiclass SIMDStSingleS<bit R, bits<3> opcode, bits<2> size,string asm,
8065 RegisterOperand listtype, RegisterOperand GPR64pi> {
8066 def i32 : SIMDLdStSingleS<0, R, opcode, size, asm,
8067 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
8068 am_simdnoindex:$vaddr), []>;
8070 def i32_POST : SIMDLdStSingleSPost<0, R, opcode, size, asm,
8071 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
8072 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8074 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8075 multiclass SIMDStSingleD<bit R, bits<3> opcode, bits<2> size, string asm,
8076 RegisterOperand listtype, RegisterOperand GPR64pi> {
8077 def i64 : SIMDLdStSingleD<0, R, opcode, size, asm,
8078 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
8079 am_simdnoindex:$vaddr), []>;
8081 def i64_POST : SIMDLdStSingleDPost<0, R, opcode, size, asm,
8082 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
8083 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8086 multiclass SIMDLdStSingleAliases<string asm, string layout, string Type,
8087 string Count, int Offset, Operand idxtype> {
8088 // E.g. "ld1 { v0.8b }[0], [x1], #1"
8089 // "ld1\t$Vt, $vaddr, #1"
8090 // may get mapped to
8091 // (LD1Rv8b_POST VecListOne8b:$Vt, am_simdnoindex:$vaddr, XZR)
8092 def : InstAlias<asm # "\t$Vt$idx, $vaddr, #" # Offset,
8093 (!cast<Instruction>(NAME # Type # "_POST")
8094 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
8095 idxtype:$idx, am_simdnoindex:$vaddr, XZR), 1>;
8097 // E.g. "ld1.8b { v0 }[0], [x1], #1"
8098 // "ld1.8b\t$Vt, $vaddr, #1"
8099 // may get mapped to
8100 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, XZR)
8101 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr, #" # Offset,
8102 (!cast<Instruction>(NAME # Type # "_POST")
8103 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8104 idxtype:$idx, am_simdnoindex:$vaddr, XZR), 0>;
8106 // E.g. "ld1.8b { v0 }[0], [x1]"
8107 // "ld1.8b\t$Vt, $vaddr"
8108 // may get mapped to
8109 // (LD1Rv8b VecListOne64:$Vt, am_simdnoindex:$vaddr)
8110 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr",
8111 (!cast<Instruction>(NAME # Type)
8112 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8113 idxtype:$idx, am_simdnoindex:$vaddr), 0>;
8115 // E.g. "ld1.8b { v0 }[0], [x1], x2"
8116 // "ld1.8b\t$Vt, $vaddr, $Xm"
8117 // may get mapped to
8118 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, GPR64pi1:$Xm)
8119 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr, $Xm",
8120 (!cast<Instruction>(NAME # Type # "_POST")
8121 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8122 idxtype:$idx, am_simdnoindex:$vaddr,
8123 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
8126 multiclass SIMDLdSt1SingleAliases<string asm> {
8127 defm : SIMDLdStSingleAliases<asm, "b", "i8", "One", 1, VectorIndexB>;
8128 defm : SIMDLdStSingleAliases<asm, "h", "i16", "One", 2, VectorIndexH>;
8129 defm : SIMDLdStSingleAliases<asm, "s", "i32", "One", 4, VectorIndexS>;
8130 defm : SIMDLdStSingleAliases<asm, "d", "i64", "One", 8, VectorIndexD>;
8133 multiclass SIMDLdSt2SingleAliases<string asm> {
8134 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Two", 2, VectorIndexB>;
8135 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Two", 4, VectorIndexH>;
8136 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Two", 8, VectorIndexS>;
8137 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Two", 16, VectorIndexD>;
8140 multiclass SIMDLdSt3SingleAliases<string asm> {
8141 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Three", 3, VectorIndexB>;
8142 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Three", 6, VectorIndexH>;
8143 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Three", 12, VectorIndexS>;
8144 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Three", 24, VectorIndexD>;
8147 multiclass SIMDLdSt4SingleAliases<string asm> {
8148 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Four", 4, VectorIndexB>;
8149 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Four", 8, VectorIndexH>;
8150 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Four", 16, VectorIndexS>;
8151 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Four", 32, VectorIndexD>;
8154 //----------------------------------------------------------------------------
8155 // Crypto extensions
8156 //----------------------------------------------------------------------------
8158 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8159 class AESBase<bits<4> opc, string asm, dag outs, dag ins, string cstr,
8161 : I<outs, ins, asm, "{\t$Rd.16b, $Rn.16b|.16b\t$Rd, $Rn}", cstr, pat>,
8165 let Inst{31-16} = 0b0100111000101000;
8166 let Inst{15-12} = opc;
8167 let Inst{11-10} = 0b10;
8172 class AESInst<bits<4> opc, string asm, Intrinsic OpNode>
8173 : AESBase<opc, asm, (outs V128:$Rd), (ins V128:$Rn), "",
8174 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
8176 class AESTiedInst<bits<4> opc, string asm, Intrinsic OpNode>
8177 : AESBase<opc, asm, (outs V128:$dst), (ins V128:$Rd, V128:$Rn),
8179 [(set (v16i8 V128:$dst),
8180 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
8182 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8183 class SHA3OpTiedInst<bits<3> opc, string asm, string dst_lhs_kind,
8184 dag oops, dag iops, list<dag> pat>
8185 : I<oops, iops, asm,
8186 "{\t$Rd" # dst_lhs_kind # ", $Rn" # dst_lhs_kind # ", $Rm.4s" #
8187 "|.4s\t$Rd, $Rn, $Rm}", "$Rd = $dst", pat>,
8192 let Inst{31-21} = 0b01011110000;
8193 let Inst{20-16} = Rm;
8195 let Inst{14-12} = opc;
8196 let Inst{11-10} = 0b00;
8201 class SHATiedInstQSV<bits<3> opc, string asm, Intrinsic OpNode>
8202 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
8203 (ins FPR128:$Rd, FPR32:$Rn, V128:$Rm),
8204 [(set (v4i32 FPR128:$dst),
8205 (OpNode (v4i32 FPR128:$Rd), (i32 FPR32:$Rn),
8206 (v4i32 V128:$Rm)))]>;
8208 class SHATiedInstVVV<bits<3> opc, string asm, Intrinsic OpNode>
8209 : SHA3OpTiedInst<opc, asm, ".4s", (outs V128:$dst),
8210 (ins V128:$Rd, V128:$Rn, V128:$Rm),
8211 [(set (v4i32 V128:$dst),
8212 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8213 (v4i32 V128:$Rm)))]>;
8215 class SHATiedInstQQV<bits<3> opc, string asm, Intrinsic OpNode>
8216 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
8217 (ins FPR128:$Rd, FPR128:$Rn, V128:$Rm),
8218 [(set (v4i32 FPR128:$dst),
8219 (OpNode (v4i32 FPR128:$Rd), (v4i32 FPR128:$Rn),
8220 (v4i32 V128:$Rm)))]>;
8222 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8223 class SHA2OpInst<bits<4> opc, string asm, string kind,
8224 string cstr, dag oops, dag iops,
8226 : I<oops, iops, asm, "{\t$Rd" # kind # ", $Rn" # kind #
8227 "|" # kind # "\t$Rd, $Rn}", cstr, pat>,
8231 let Inst{31-16} = 0b0101111000101000;
8232 let Inst{15-12} = opc;
8233 let Inst{11-10} = 0b10;
8238 class SHATiedInstVV<bits<4> opc, string asm, Intrinsic OpNode>
8239 : SHA2OpInst<opc, asm, ".4s", "$Rd = $dst", (outs V128:$dst),
8240 (ins V128:$Rd, V128:$Rn),
8241 [(set (v4i32 V128:$dst),
8242 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
8244 class SHAInstSS<bits<4> opc, string asm, Intrinsic OpNode>
8245 : SHA2OpInst<opc, asm, "", "", (outs FPR32:$Rd), (ins FPR32:$Rn),
8246 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
8248 // Allow the size specifier tokens to be upper case, not just lower.
8249 def : TokenAlias<".8B", ".8b">;
8250 def : TokenAlias<".4H", ".4h">;
8251 def : TokenAlias<".2S", ".2s">;
8252 def : TokenAlias<".1D", ".1d">;
8253 def : TokenAlias<".16B", ".16b">;
8254 def : TokenAlias<".8H", ".8h">;
8255 def : TokenAlias<".4S", ".4s">;
8256 def : TokenAlias<".2D", ".2d">;
8257 def : TokenAlias<".1Q", ".1q">;
8258 def : TokenAlias<".B", ".b">;
8259 def : TokenAlias<".H", ".h">;
8260 def : TokenAlias<".S", ".s">;
8261 def : TokenAlias<".D", ".d">;
8262 def : TokenAlias<".Q", ".q">;