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 multiclass CondSelectOp<bit op, bits<2> op2, string asm, PatFrag frag> {
1940 def Wr : BaseCondSelectOp<op, op2, GPR32, asm, frag> {
1943 def Xr : BaseCondSelectOp<op, op2, GPR64, asm, frag> {
1949 // Special Mask Value
1951 def maski8_or_more : Operand<i32>,
1952 ImmLeaf<i32, [{ return (Imm & 0xff) == 0xff; }]> {
1954 def maski16_or_more : Operand<i32>,
1955 ImmLeaf<i32, [{ return (Imm & 0xffff) == 0xffff; }]> {
1963 // (unsigned immediate)
1964 // Indexed for 8-bit registers. offset is in range [0,4095].
1965 def MemoryIndexed8Operand : AsmOperandClass {
1966 let Name = "MemoryIndexed8";
1967 let DiagnosticType = "InvalidMemoryIndexed8";
1969 def am_indexed8 : Operand<i64>,
1970 ComplexPattern<i64, 2, "SelectAddrModeIndexed8", []> {
1971 let PrintMethod = "printAMIndexed8";
1973 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale1>";
1974 let ParserMatchClass = MemoryIndexed8Operand;
1975 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1978 // Indexed for 16-bit registers. offset is multiple of 2 in range [0,8190],
1979 // stored as immval/2 (the 12-bit literal that encodes directly into the insn).
1980 def MemoryIndexed16Operand : AsmOperandClass {
1981 let Name = "MemoryIndexed16";
1982 let DiagnosticType = "InvalidMemoryIndexed16";
1984 def am_indexed16 : Operand<i64>,
1985 ComplexPattern<i64, 2, "SelectAddrModeIndexed16", []> {
1986 let PrintMethod = "printAMIndexed16";
1988 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale2>";
1989 let ParserMatchClass = MemoryIndexed16Operand;
1990 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
1993 // Indexed for 32-bit registers. offset is multiple of 4 in range [0,16380],
1994 // stored as immval/4 (the 12-bit literal that encodes directly into the insn).
1995 def MemoryIndexed32Operand : AsmOperandClass {
1996 let Name = "MemoryIndexed32";
1997 let DiagnosticType = "InvalidMemoryIndexed32";
1999 def am_indexed32 : Operand<i64>,
2000 ComplexPattern<i64, 2, "SelectAddrModeIndexed32", []> {
2001 let PrintMethod = "printAMIndexed32";
2003 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale4>";
2004 let ParserMatchClass = MemoryIndexed32Operand;
2005 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2008 // Indexed for 64-bit registers. offset is multiple of 8 in range [0,32760],
2009 // stored as immval/8 (the 12-bit literal that encodes directly into the insn).
2010 def MemoryIndexed64Operand : AsmOperandClass {
2011 let Name = "MemoryIndexed64";
2012 let DiagnosticType = "InvalidMemoryIndexed64";
2014 def am_indexed64 : Operand<i64>,
2015 ComplexPattern<i64, 2, "SelectAddrModeIndexed64", []> {
2016 let PrintMethod = "printAMIndexed64";
2018 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale8>";
2019 let ParserMatchClass = MemoryIndexed64Operand;
2020 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2023 // Indexed for 128-bit registers. offset is multiple of 16 in range [0,65520],
2024 // stored as immval/16 (the 12-bit literal that encodes directly into the insn).
2025 def MemoryIndexed128Operand : AsmOperandClass {
2026 let Name = "MemoryIndexed128";
2027 let DiagnosticType = "InvalidMemoryIndexed128";
2029 def am_indexed128 : Operand<i64>,
2030 ComplexPattern<i64, 2, "SelectAddrModeIndexed128", []> {
2031 let PrintMethod = "printAMIndexed128";
2033 = "getAMIndexed8OpValue<ARM64::fixup_arm64_ldst_imm12_scale16>";
2034 let ParserMatchClass = MemoryIndexed128Operand;
2035 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2039 def MemoryNoIndexOperand : AsmOperandClass { let Name = "MemoryNoIndex"; }
2040 def am_noindex : Operand<i64>,
2041 ComplexPattern<i64, 1, "SelectAddrModeNoIndex", []> {
2042 let PrintMethod = "printAMNoIndex";
2043 let ParserMatchClass = MemoryNoIndexOperand;
2044 let MIOperandInfo = (ops GPR64sp:$base);
2047 class BaseLoadStoreUI<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2048 string asm, list<dag> pattern>
2049 : I<oops, iops, asm, "\t$Rt, $addr", "", pattern> {
2053 bits<5> base = addr{4-0};
2054 bits<12> offset = addr{16-5};
2056 let Inst{31-30} = sz;
2057 let Inst{29-27} = 0b111;
2059 let Inst{25-24} = 0b01;
2060 let Inst{23-22} = opc;
2061 let Inst{21-10} = offset;
2062 let Inst{9-5} = base;
2063 let Inst{4-0} = dst;
2065 let DecoderMethod = "DecodeUnsignedLdStInstruction";
2068 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2069 class LoadUI<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2070 Operand indextype, string asm, list<dag> pattern>
2071 : BaseLoadStoreUI<sz, V, opc,
2072 (outs regtype:$Rt), (ins indextype:$addr), asm, pattern>,
2075 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2076 class StoreUI<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2077 Operand indextype, string asm, list<dag> pattern>
2078 : BaseLoadStoreUI<sz, V, opc,
2079 (outs), (ins regtype:$Rt, indextype:$addr), asm, pattern>,
2082 def PrefetchOperand : AsmOperandClass {
2083 let Name = "Prefetch";
2084 let ParserMethod = "tryParsePrefetch";
2086 def prfop : Operand<i32> {
2087 let PrintMethod = "printPrefetchOp";
2088 let ParserMatchClass = PrefetchOperand;
2091 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2092 class PrefetchUI<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2093 : BaseLoadStoreUI<sz, V, opc,
2094 (outs), (ins prfop:$Rt, am_indexed64:$addr), asm, pat>,
2101 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2102 class LoadLiteral<bits<2> opc, bit V, RegisterClass regtype, string asm>
2103 : I<(outs regtype:$Rt), (ins am_brcond:$label),
2104 asm, "\t$Rt, $label", "", []>,
2108 let Inst{31-30} = opc;
2109 let Inst{29-27} = 0b011;
2111 let Inst{25-24} = 0b00;
2112 let Inst{23-5} = label;
2116 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2117 class PrefetchLiteral<bits<2> opc, bit V, string asm, list<dag> pat>
2118 : I<(outs), (ins prfop:$Rt, am_brcond:$label),
2119 asm, "\t$Rt, $label", "", pat>,
2123 let Inst{31-30} = opc;
2124 let Inst{29-27} = 0b011;
2126 let Inst{25-24} = 0b00;
2127 let Inst{23-5} = label;
2132 // Load/store register offset
2135 class MemROAsmOperand<int sz> : AsmOperandClass {
2136 let Name = "MemoryRegisterOffset"#sz;
2139 def MemROAsmOperand8 : MemROAsmOperand<8>;
2140 def MemROAsmOperand16 : MemROAsmOperand<16>;
2141 def MemROAsmOperand32 : MemROAsmOperand<32>;
2142 def MemROAsmOperand64 : MemROAsmOperand<64>;
2143 def MemROAsmOperand128 : MemROAsmOperand<128>;
2145 class ro_indexed<int sz> : Operand<i64> { // ComplexPattern<...>
2146 let PrintMethod = "printMemoryRegOffset"#sz;
2147 let MIOperandInfo = (ops GPR64sp:$base, GPR64:$offset, i32imm:$extend);
2150 def ro_indexed8 : ro_indexed<8>, ComplexPattern<i64, 3, "SelectAddrModeRO8", []> {
2151 let ParserMatchClass = MemROAsmOperand8;
2154 def ro_indexed16 : ro_indexed<16>, ComplexPattern<i64, 3, "SelectAddrModeRO16", []> {
2155 let ParserMatchClass = MemROAsmOperand16;
2158 def ro_indexed32 : ro_indexed<32>, ComplexPattern<i64, 3, "SelectAddrModeRO32", []> {
2159 let ParserMatchClass = MemROAsmOperand32;
2162 def ro_indexed64 : ro_indexed<64>, ComplexPattern<i64, 3, "SelectAddrModeRO64", []> {
2163 let ParserMatchClass = MemROAsmOperand64;
2166 def ro_indexed128 : ro_indexed<128>, ComplexPattern<i64, 3, "SelectAddrModeRO128", []> {
2167 let ParserMatchClass = MemROAsmOperand128;
2170 class LoadStore8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2171 string asm, dag ins, dag outs, list<dag> pat>
2172 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2173 // The operands are in order to match the 'addr' MI operands, so we
2174 // don't need an encoder method and by-name matching. Just use the default
2175 // in-order handling. Since we're using by-order, make sure the names
2181 let Inst{31-30} = sz;
2182 let Inst{29-27} = 0b111;
2184 let Inst{25-24} = 0b00;
2185 let Inst{23-22} = opc;
2187 let Inst{20-16} = offset;
2188 let Inst{15-13} = extend{3-1};
2190 let Inst{12} = extend{0};
2191 let Inst{11-10} = 0b10;
2192 let Inst{9-5} = base;
2193 let Inst{4-0} = dst;
2195 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2198 class Load8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2199 string asm, list<dag> pat>
2200 : LoadStore8RO<sz, V, opc, regtype, asm,
2201 (outs regtype:$Rt), (ins ro_indexed8:$addr), pat>,
2202 Sched<[WriteLDIdx, ReadAdrBase]>;
2204 class Store8RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2205 string asm, list<dag> pat>
2206 : LoadStore8RO<sz, V, opc, regtype, asm,
2207 (outs), (ins regtype:$Rt, ro_indexed8:$addr), pat>,
2208 Sched<[WriteSTIdx, ReadAdrBase]>;
2210 class LoadStore16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2211 string asm, dag ins, dag outs, list<dag> pat>
2212 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2213 // The operands are in order to match the 'addr' MI operands, so we
2214 // don't need an encoder method and by-name matching. Just use the default
2215 // in-order handling. Since we're using by-order, make sure the names
2221 let Inst{31-30} = sz;
2222 let Inst{29-27} = 0b111;
2224 let Inst{25-24} = 0b00;
2225 let Inst{23-22} = opc;
2227 let Inst{20-16} = offset;
2228 let Inst{15-13} = extend{3-1};
2230 let Inst{12} = extend{0};
2231 let Inst{11-10} = 0b10;
2232 let Inst{9-5} = base;
2233 let Inst{4-0} = dst;
2235 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2238 class Load16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2239 string asm, list<dag> pat>
2240 : LoadStore16RO<sz, V, opc, regtype, asm,
2241 (outs regtype:$Rt), (ins ro_indexed16:$addr), pat>,
2242 Sched<[WriteLDIdx, ReadAdrBase]>;
2244 class Store16RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2245 string asm, list<dag> pat>
2246 : LoadStore16RO<sz, V, opc, regtype, asm,
2247 (outs), (ins regtype:$Rt, ro_indexed16:$addr), pat>,
2248 Sched<[WriteSTIdx, ReadAdrBase]>;
2250 class LoadStore32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2251 string asm, dag ins, dag outs, list<dag> pat>
2252 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2253 // The operands are in order to match the 'addr' MI operands, so we
2254 // don't need an encoder method and by-name matching. Just use the default
2255 // in-order handling. Since we're using by-order, make sure the names
2261 let Inst{31-30} = sz;
2262 let Inst{29-27} = 0b111;
2264 let Inst{25-24} = 0b00;
2265 let Inst{23-22} = opc;
2267 let Inst{20-16} = offset;
2268 let Inst{15-13} = extend{3-1};
2270 let Inst{12} = extend{0};
2271 let Inst{11-10} = 0b10;
2272 let Inst{9-5} = base;
2273 let Inst{4-0} = dst;
2275 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2278 class Load32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2279 string asm, list<dag> pat>
2280 : LoadStore32RO<sz, V, opc, regtype, asm,
2281 (outs regtype:$Rt), (ins ro_indexed32:$addr), pat>,
2282 Sched<[WriteLDIdx, ReadAdrBase]>;
2284 class Store32RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2285 string asm, list<dag> pat>
2286 : LoadStore32RO<sz, V, opc, regtype, asm,
2287 (outs), (ins regtype:$Rt, ro_indexed32:$addr), pat>,
2288 Sched<[WriteSTIdx, ReadAdrBase]>;
2290 class LoadStore64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2291 string asm, dag ins, dag outs, list<dag> pat>
2292 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2293 // The operands are in order to match the 'addr' MI operands, so we
2294 // don't need an encoder method and by-name matching. Just use the default
2295 // in-order handling. Since we're using by-order, make sure the names
2301 let Inst{31-30} = sz;
2302 let Inst{29-27} = 0b111;
2304 let Inst{25-24} = 0b00;
2305 let Inst{23-22} = opc;
2307 let Inst{20-16} = offset;
2308 let Inst{15-13} = extend{3-1};
2310 let Inst{12} = extend{0};
2311 let Inst{11-10} = 0b10;
2312 let Inst{9-5} = base;
2313 let Inst{4-0} = dst;
2315 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2318 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2319 class Load64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2320 string asm, list<dag> pat>
2321 : LoadStore64RO<sz, V, opc, regtype, asm,
2322 (outs regtype:$Rt), (ins ro_indexed64:$addr), pat>,
2323 Sched<[WriteLDIdx, ReadAdrBase]>;
2325 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2326 class Store64RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2327 string asm, list<dag> pat>
2328 : LoadStore64RO<sz, V, opc, regtype, asm,
2329 (outs), (ins regtype:$Rt, ro_indexed64:$addr), pat>,
2330 Sched<[WriteSTIdx, ReadAdrBase]>;
2333 class LoadStore128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2334 string asm, dag ins, dag outs, list<dag> pat>
2335 : I<ins, outs, asm, "\t$Rt, $addr", "", pat> {
2336 // The operands are in order to match the 'addr' MI operands, so we
2337 // don't need an encoder method and by-name matching. Just use the default
2338 // in-order handling. Since we're using by-order, make sure the names
2344 let Inst{31-30} = sz;
2345 let Inst{29-27} = 0b111;
2347 let Inst{25-24} = 0b00;
2348 let Inst{23-22} = opc;
2350 let Inst{20-16} = offset;
2351 let Inst{15-13} = extend{3-1};
2353 let Inst{12} = extend{0};
2354 let Inst{11-10} = 0b10;
2355 let Inst{9-5} = base;
2356 let Inst{4-0} = dst;
2358 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2361 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
2362 class Load128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2363 string asm, list<dag> pat>
2364 : LoadStore128RO<sz, V, opc, regtype, asm,
2365 (outs regtype:$Rt), (ins ro_indexed128:$addr), pat>,
2366 Sched<[WriteLDIdx, ReadAdrBase]>;
2368 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
2369 class Store128RO<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2370 string asm, list<dag> pat>
2371 : LoadStore128RO<sz, V, opc, regtype, asm,
2372 (outs), (ins regtype:$Rt, ro_indexed128:$addr), pat>,
2373 Sched<[WriteSTIdx, ReadAdrBase]>;
2375 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2376 class PrefetchRO<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2377 : I<(outs), (ins prfop:$Rt, ro_indexed64:$addr), asm,
2378 "\t$Rt, $addr", "", pat>,
2380 // The operands are in order to match the 'addr' MI operands, so we
2381 // don't need an encoder method and by-name matching. Just use the default
2382 // in-order handling. Since we're using by-order, make sure the names
2388 let Inst{31-30} = sz;
2389 let Inst{29-27} = 0b111;
2391 let Inst{25-24} = 0b00;
2392 let Inst{23-22} = opc;
2394 let Inst{20-16} = offset;
2395 let Inst{15-13} = extend{3-1};
2397 let Inst{12} = extend{0};
2398 let Inst{11-10} = 0b10;
2399 let Inst{9-5} = base;
2400 let Inst{4-0} = dst;
2402 let DecoderMethod = "DecodeRegOffsetLdStInstruction";
2406 // Load/store unscaled immediate
2409 def MemoryUnscaledOperand : AsmOperandClass {
2410 let Name = "MemoryUnscaled";
2411 let DiagnosticType = "InvalidMemoryIndexedSImm9";
2413 class am_unscaled_operand : Operand<i64> {
2414 let PrintMethod = "printAMUnscaled";
2415 let ParserMatchClass = MemoryUnscaledOperand;
2416 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2418 class am_unscaled_wb_operand : Operand<i64> {
2419 let PrintMethod = "printAMUnscaledWB";
2420 let ParserMatchClass = MemoryUnscaledOperand;
2421 let MIOperandInfo = (ops GPR64sp:$base, i64imm:$offset);
2423 def am_unscaled : am_unscaled_operand;
2424 def am_unscaled_wb: am_unscaled_wb_operand;
2425 def am_unscaled8 : am_unscaled_operand,
2426 ComplexPattern<i64, 2, "SelectAddrModeUnscaled8", []>;
2427 def am_unscaled16 : am_unscaled_operand,
2428 ComplexPattern<i64, 2, "SelectAddrModeUnscaled16", []>;
2429 def am_unscaled32 : am_unscaled_operand,
2430 ComplexPattern<i64, 2, "SelectAddrModeUnscaled32", []>;
2431 def am_unscaled64 : am_unscaled_operand,
2432 ComplexPattern<i64, 2, "SelectAddrModeUnscaled64", []>;
2433 def am_unscaled128 : am_unscaled_operand,
2434 ComplexPattern<i64, 2, "SelectAddrModeUnscaled128", []>;
2436 class BaseLoadStoreUnscale<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2437 string asm, list<dag> pattern>
2438 : I<oops, iops, asm, "\t$Rt, $addr", "", pattern> {
2439 // The operands are in order to match the 'addr' MI operands, so we
2440 // don't need an encoder method and by-name matching. Just use the default
2441 // in-order handling. Since we're using by-order, make sure the names
2446 let Inst{31-30} = sz;
2447 let Inst{29-27} = 0b111;
2449 let Inst{25-24} = 0b00;
2450 let Inst{23-22} = opc;
2452 let Inst{20-12} = offset;
2453 let Inst{11-10} = 0b00;
2454 let Inst{9-5} = base;
2455 let Inst{4-0} = dst;
2457 let DecoderMethod = "DecodeSignedLdStInstruction";
2460 let AddedComplexity = 1 in // try this before LoadUI
2461 class LoadUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2462 Operand amtype, string asm, list<dag> pattern>
2463 : BaseLoadStoreUnscale<sz, V, opc, (outs regtype:$Rt),
2464 (ins amtype:$addr), asm, pattern>,
2467 let AddedComplexity = 1 in // try this before StoreUI
2468 class StoreUnscaled<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2469 Operand amtype, string asm, list<dag> pattern>
2470 : BaseLoadStoreUnscale<sz, V, opc, (outs),
2471 (ins regtype:$Rt, amtype:$addr), asm, pattern>,
2474 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
2475 class PrefetchUnscaled<bits<2> sz, bit V, bits<2> opc, string asm, list<dag> pat>
2476 : BaseLoadStoreUnscale<sz, V, opc, (outs),
2477 (ins prfop:$Rt, am_unscaled:$addr), asm, pat>,
2481 // Load/store unscaled immediate, unprivileged
2484 class BaseLoadStoreUnprivileged<bits<2> sz, bit V, bits<2> opc,
2485 dag oops, dag iops, string asm>
2486 : I<oops, iops, asm, "\t$Rt, $addr", "", []> {
2487 // The operands are in order to match the 'addr' MI operands, so we
2488 // don't need an encoder method and by-name matching. Just use the default
2489 // in-order handling. Since we're using by-order, make sure the names
2494 let Inst{31-30} = sz;
2495 let Inst{29-27} = 0b111;
2497 let Inst{25-24} = 0b00;
2498 let Inst{23-22} = opc;
2500 let Inst{20-12} = offset;
2501 let Inst{11-10} = 0b10;
2502 let Inst{9-5} = base;
2503 let Inst{4-0} = dst;
2505 let DecoderMethod = "DecodeSignedLdStInstruction";
2508 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in {
2509 class LoadUnprivileged<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2511 : BaseLoadStoreUnprivileged<sz, V, opc,
2512 (outs regtype:$Rt), (ins am_unscaled:$addr), asm>,
2516 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in {
2517 class StoreUnprivileged<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2519 : BaseLoadStoreUnprivileged<sz, V, opc,
2520 (outs), (ins regtype:$Rt, am_unscaled:$addr), asm>,
2525 // Load/store pre-indexed
2528 class BaseLoadStorePreIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2529 string asm, string cstr>
2530 : I<oops, iops, asm, "\t$Rt, $addr!", cstr, []> {
2531 // The operands are in order to match the 'addr' MI operands, so we
2532 // don't need an encoder method and by-name matching. Just use the default
2533 // in-order handling.
2537 let Inst{31-30} = sz;
2538 let Inst{29-27} = 0b111;
2540 let Inst{25-24} = 0;
2541 let Inst{23-22} = opc;
2543 let Inst{20-12} = offset;
2544 let Inst{11-10} = 0b11;
2545 let Inst{9-5} = base;
2546 let Inst{4-0} = dst;
2548 let DecoderMethod = "DecodeSignedLdStInstruction";
2551 let hasSideEffects = 0 in {
2552 let mayStore = 0, mayLoad = 1 in
2553 // FIXME: Modeling the write-back of these instructions for isel is tricky.
2554 // we need the complex addressing mode for the memory reference, but
2555 // we also need the write-back specified as a tied operand to the
2556 // base register. That combination does not play nicely with
2557 // the asm matcher and friends.
2558 class LoadPreIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2560 : BaseLoadStorePreIdx<sz, V, opc,
2561 (outs regtype:$Rt/*, GPR64sp:$wback*/),
2562 (ins am_unscaled_wb:$addr), asm, ""/*"$addr.base = $wback"*/>,
2563 Sched<[WriteLD, WriteAdr]>;
2565 let mayStore = 1, mayLoad = 0 in
2566 class StorePreIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2568 : BaseLoadStorePreIdx<sz, V, opc,
2569 (outs/* GPR64sp:$wback*/),
2570 (ins regtype:$Rt, am_unscaled_wb:$addr),
2571 asm, ""/*"$addr.base = $wback"*/>,
2572 Sched<[WriteAdr, WriteST]>;
2573 } // hasSideEffects = 0
2575 // ISel pseudo-instructions which have the tied operands. When the MC lowering
2576 // logic finally gets smart enough to strip off tied operands that are just
2577 // for isel convenience, we can get rid of these pseudos and just reference
2578 // the real instructions directly.
2580 // Ironically, also because of the writeback operands, we can't put the
2581 // matcher pattern directly on the instruction, but need to define it
2584 // Loads aren't matched with patterns here at all, but rather in C++
2586 let mayStore = 0, mayLoad = 1, hasSideEffects = 0 in {
2587 class LoadPreIdxPseudo<RegisterClass regtype>
2588 : Pseudo<(outs regtype:$Rt, GPR64sp:$wback),
2589 (ins am_noindex:$addr, simm9:$offset), [],
2590 "$addr.base = $wback,@earlyclobber $wback">,
2591 Sched<[WriteLD, WriteAdr]>;
2592 class LoadPostIdxPseudo<RegisterClass regtype>
2593 : Pseudo<(outs regtype:$Rt, GPR64sp:$wback),
2594 (ins am_noindex:$addr, simm9:$offset), [],
2595 "$addr.base = $wback,@earlyclobber $wback">,
2596 Sched<[WriteLD, WriteI]>;
2598 multiclass StorePreIdxPseudo<RegisterClass regtype, ValueType Ty,
2599 SDPatternOperator OpNode> {
2600 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
2601 def _isel: Pseudo<(outs GPR64sp:$wback),
2602 (ins regtype:$Rt, am_noindex:$addr, simm9:$offset), [],
2603 "$addr.base = $wback,@earlyclobber $wback">,
2604 Sched<[WriteAdr, WriteST]>;
2606 def : Pat<(OpNode (Ty regtype:$Rt), am_noindex:$addr, simm9:$offset),
2607 (!cast<Instruction>(NAME#_isel) regtype:$Rt, am_noindex:$addr,
2612 // Load/store post-indexed
2615 // (pre-index) load/stores.
2616 class BaseLoadStorePostIdx<bits<2> sz, bit V, bits<2> opc, dag oops, dag iops,
2617 string asm, string cstr>
2618 : I<oops, iops, asm, "\t$Rt, $addr, $idx", cstr, []> {
2619 // The operands are in order to match the 'addr' MI operands, so we
2620 // don't need an encoder method and by-name matching. Just use the default
2621 // in-order handling.
2625 let Inst{31-30} = sz;
2626 let Inst{29-27} = 0b111;
2628 let Inst{25-24} = 0b00;
2629 let Inst{23-22} = opc;
2631 let Inst{20-12} = offset;
2632 let Inst{11-10} = 0b01;
2633 let Inst{9-5} = base;
2634 let Inst{4-0} = dst;
2636 let DecoderMethod = "DecodeSignedLdStInstruction";
2639 let hasSideEffects = 0 in {
2640 let mayStore = 0, mayLoad = 1 in
2641 // FIXME: Modeling the write-back of these instructions for isel is tricky.
2642 // we need the complex addressing mode for the memory reference, but
2643 // we also need the write-back specified as a tied operand to the
2644 // base register. That combination does not play nicely with
2645 // the asm matcher and friends.
2646 class LoadPostIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2648 : BaseLoadStorePostIdx<sz, V, opc,
2649 (outs regtype:$Rt/*, GPR64sp:$wback*/),
2650 (ins am_noindex:$addr, simm9:$idx),
2651 asm, ""/*"$addr.base = $wback"*/>,
2652 Sched<[WriteLD, WriteI]>;
2654 let mayStore = 1, mayLoad = 0 in
2655 class StorePostIdx<bits<2> sz, bit V, bits<2> opc, RegisterClass regtype,
2657 : BaseLoadStorePostIdx<sz, V, opc,
2658 (outs/* GPR64sp:$wback*/),
2659 (ins regtype:$Rt, am_noindex:$addr, simm9:$idx),
2660 asm, ""/*"$addr.base = $wback"*/>,
2661 Sched<[WriteAdr, WriteST, ReadAdrBase]>;
2662 } // hasSideEffects = 0
2664 // ISel pseudo-instructions which have the tied operands. When the MC lowering
2665 // logic finally gets smart enough to strip off tied operands that are just
2666 // for isel convenience, we can get rid of these pseudos and just reference
2667 // the real instructions directly.
2669 // Ironically, also because of the writeback operands, we can't put the
2670 // matcher pattern directly on the instruction, but need to define it
2672 multiclass StorePostIdxPseudo<RegisterClass regtype, ValueType Ty,
2673 SDPatternOperator OpNode, Instruction Insn> {
2674 let mayStore = 1, mayLoad = 0, hasSideEffects = 0 in
2675 def _isel: Pseudo<(outs GPR64sp:$wback),
2676 (ins regtype:$Rt, am_noindex:$addr, simm9:$idx), [],
2677 "$addr.base = $wback,@earlyclobber $wback">,
2678 PseudoInstExpansion<(Insn regtype:$Rt, am_noindex:$addr, simm9:$idx)>,
2679 Sched<[WriteAdr, WriteST, ReadAdrBase]>;
2681 def : Pat<(OpNode (Ty regtype:$Rt), am_noindex:$addr, simm9:$idx),
2682 (!cast<Instruction>(NAME#_isel) regtype:$Rt, am_noindex:$addr,
2690 // (indexed, offset)
2692 class BaseLoadStorePairOffset<bits<2> opc, bit V, bit L, dag oops, dag iops,
2694 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr", "", []> {
2695 // The operands are in order to match the 'addr' MI operands, so we
2696 // don't need an encoder method and by-name matching. Just use the default
2697 // in-order handling. Since we're using by-order, make sure the names
2703 let Inst{31-30} = opc;
2704 let Inst{29-27} = 0b101;
2706 let Inst{25-23} = 0b010;
2708 let Inst{21-15} = offset;
2709 let Inst{14-10} = dst2;
2710 let Inst{9-5} = base;
2711 let Inst{4-0} = dst;
2713 let DecoderMethod = "DecodePairLdStInstruction";
2716 let hasSideEffects = 0 in {
2717 let mayStore = 0, mayLoad = 1 in
2718 class LoadPairOffset<bits<2> opc, bit V, RegisterClass regtype,
2719 Operand indextype, string asm>
2720 : BaseLoadStorePairOffset<opc, V, 1,
2721 (outs regtype:$Rt, regtype:$Rt2),
2722 (ins indextype:$addr), asm>,
2723 Sched<[WriteLD, WriteLDHi]>;
2725 let mayLoad = 0, mayStore = 1 in
2726 class StorePairOffset<bits<2> opc, bit V, RegisterClass regtype,
2727 Operand indextype, string asm>
2728 : BaseLoadStorePairOffset<opc, V, 0, (outs),
2729 (ins regtype:$Rt, regtype:$Rt2, indextype:$addr),
2732 } // hasSideEffects = 0
2736 def MemoryIndexed32SImm7 : AsmOperandClass {
2737 let Name = "MemoryIndexed32SImm7";
2738 let DiagnosticType = "InvalidMemoryIndexed32SImm7";
2740 def am_indexed32simm7 : Operand<i32> { // ComplexPattern<...>
2741 let PrintMethod = "printAMIndexed32";
2742 let ParserMatchClass = MemoryIndexed32SImm7;
2743 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2745 def am_indexed32simm7_wb : Operand<i32> { // ComplexPattern<...>
2746 let PrintMethod = "printAMIndexed32WB";
2747 let ParserMatchClass = MemoryIndexed32SImm7;
2748 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2751 def MemoryIndexed64SImm7 : AsmOperandClass {
2752 let Name = "MemoryIndexed64SImm7";
2753 let DiagnosticType = "InvalidMemoryIndexed64SImm7";
2755 def am_indexed64simm7 : Operand<i32> { // ComplexPattern<...>
2756 let PrintMethod = "printAMIndexed64";
2757 let ParserMatchClass = MemoryIndexed64SImm7;
2758 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2760 def am_indexed64simm7_wb : Operand<i32> { // ComplexPattern<...>
2761 let PrintMethod = "printAMIndexed64WB";
2762 let ParserMatchClass = MemoryIndexed64SImm7;
2763 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2766 def MemoryIndexed128SImm7 : AsmOperandClass {
2767 let Name = "MemoryIndexed128SImm7";
2768 let DiagnosticType = "InvalidMemoryIndexed128SImm7";
2770 def am_indexed128simm7 : Operand<i32> { // ComplexPattern<...>
2771 let PrintMethod = "printAMIndexed128";
2772 let ParserMatchClass = MemoryIndexed128SImm7;
2773 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2775 def am_indexed128simm7_wb : Operand<i32> { // ComplexPattern<...>
2776 let PrintMethod = "printAMIndexed128WB";
2777 let ParserMatchClass = MemoryIndexed128SImm7;
2778 let MIOperandInfo = (ops GPR64sp:$base, i32imm:$offset);
2781 class BaseLoadStorePairPreIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
2783 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr!", "", []> {
2784 // The operands are in order to match the 'addr' MI operands, so we
2785 // don't need an encoder method and by-name matching. Just use the default
2786 // in-order handling. Since we're using by-order, make sure the names
2792 let Inst{31-30} = opc;
2793 let Inst{29-27} = 0b101;
2795 let Inst{25-23} = 0b011;
2797 let Inst{21-15} = offset;
2798 let Inst{14-10} = dst2;
2799 let Inst{9-5} = base;
2800 let Inst{4-0} = dst;
2802 let DecoderMethod = "DecodePairLdStInstruction";
2805 let hasSideEffects = 0 in {
2806 let mayStore = 0, mayLoad = 1 in
2807 class LoadPairPreIdx<bits<2> opc, bit V, RegisterClass regtype,
2808 Operand addrmode, string asm>
2809 : BaseLoadStorePairPreIdx<opc, V, 1,
2810 (outs regtype:$Rt, regtype:$Rt2),
2811 (ins addrmode:$addr), asm>,
2812 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
2814 let mayStore = 1, mayLoad = 0 in
2815 class StorePairPreIdx<bits<2> opc, bit V, RegisterClass regtype,
2816 Operand addrmode, string asm>
2817 : BaseLoadStorePairPreIdx<opc, V, 0, (outs),
2818 (ins regtype:$Rt, regtype:$Rt2, addrmode:$addr),
2820 Sched<[WriteAdr, WriteSTP]>;
2821 } // hasSideEffects = 0
2825 class BaseLoadStorePairPostIdx<bits<2> opc, bit V, bit L, dag oops, dag iops,
2827 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr, $idx", "", []> {
2828 // The operands are in order to match the 'addr' MI operands, so we
2829 // don't need an encoder method and by-name matching. Just use the default
2830 // in-order handling. Since we're using by-order, make sure the names
2836 let Inst{31-30} = opc;
2837 let Inst{29-27} = 0b101;
2839 let Inst{25-23} = 0b001;
2841 let Inst{21-15} = offset;
2842 let Inst{14-10} = dst2;
2843 let Inst{9-5} = base;
2844 let Inst{4-0} = dst;
2846 let DecoderMethod = "DecodePairLdStInstruction";
2849 let hasSideEffects = 0 in {
2850 let mayStore = 0, mayLoad = 1 in
2851 class LoadPairPostIdx<bits<2> opc, bit V, RegisterClass regtype,
2852 Operand idxtype, string asm>
2853 : BaseLoadStorePairPostIdx<opc, V, 1,
2854 (outs regtype:$Rt, regtype:$Rt2),
2855 (ins am_noindex:$addr, idxtype:$idx), asm>,
2856 Sched<[WriteLD, WriteLDHi, WriteAdr]>;
2858 let mayStore = 1, mayLoad = 0 in
2859 class StorePairPostIdx<bits<2> opc, bit V, RegisterClass regtype,
2860 Operand idxtype, string asm>
2861 : BaseLoadStorePairPostIdx<opc, V, 0, (outs),
2862 (ins regtype:$Rt, regtype:$Rt2,
2863 am_noindex:$addr, idxtype:$idx),
2865 Sched<[WriteAdr, WriteSTP]>;
2866 } // hasSideEffects = 0
2870 class BaseLoadStorePairNoAlloc<bits<2> opc, bit V, bit L, dag oops, dag iops,
2872 : I<oops, iops, asm, "\t$Rt, $Rt2, $addr", "", []> {
2873 // The operands are in order to match the 'addr' MI operands, so we
2874 // don't need an encoder method and by-name matching. Just use the default
2875 // in-order handling. Since we're using by-order, make sure the names
2881 let Inst{31-30} = opc;
2882 let Inst{29-27} = 0b101;
2884 let Inst{25-23} = 0b000;
2886 let Inst{21-15} = offset;
2887 let Inst{14-10} = dst2;
2888 let Inst{9-5} = base;
2889 let Inst{4-0} = dst;
2891 let DecoderMethod = "DecodePairLdStInstruction";
2894 let hasSideEffects = 0 in {
2895 let mayStore = 0, mayLoad = 1 in
2896 class LoadPairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
2897 Operand indextype, string asm>
2898 : BaseLoadStorePairNoAlloc<opc, V, 1,
2899 (outs regtype:$Rt, regtype:$Rt2),
2900 (ins indextype:$addr), asm>,
2901 Sched<[WriteLD, WriteLDHi]>;
2903 let mayStore = 1, mayLoad = 0 in
2904 class StorePairNoAlloc<bits<2> opc, bit V, RegisterClass regtype,
2905 Operand indextype, string asm>
2906 : BaseLoadStorePairNoAlloc<opc, V, 0, (outs),
2907 (ins regtype:$Rt, regtype:$Rt2, indextype:$addr),
2910 } // hasSideEffects = 0
2913 // Load/store exclusive
2916 // True exclusive operations write to and/or read from the system's exclusive
2917 // monitors, which as far as a compiler is concerned can be modelled as a
2918 // random shared memory address. Hence LoadExclusive mayStore.
2920 // Since these instructions have the undefined register bits set to 1 in
2921 // their canonical form, we need a post encoder method to set those bits
2922 // to 1 when encoding these instructions. We do this using the
2923 // fixLoadStoreExclusive function. This function has template parameters:
2925 // fixLoadStoreExclusive<int hasRs, int hasRt2>
2927 // hasRs indicates that the instruction uses the Rs field, so we won't set
2928 // it to 1 (and the same for Rt2). We don't need template parameters for
2929 // the other register fields since Rt and Rn are always used.
2931 let hasSideEffects = 1, mayLoad = 1, mayStore = 1 in
2932 class BaseLoadStoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2933 dag oops, dag iops, string asm, string operands>
2934 : I<oops, iops, asm, operands, "", []> {
2935 let Inst{31-30} = sz;
2936 let Inst{29-24} = 0b001000;
2942 let DecoderMethod = "DecodeExclusiveLdStInstruction";
2945 // Neither Rs nor Rt2 operands.
2946 class LoadStoreExclusiveSimple<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2947 dag oops, dag iops, string asm, string operands>
2948 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, oops, iops, asm, operands> {
2951 let Inst{9-5} = base;
2952 let Inst{4-0} = reg;
2954 let PostEncoderMethod = "fixLoadStoreExclusive<0,0>";
2957 // Simple load acquires don't set the exclusive monitor
2958 let mayLoad = 1, mayStore = 0 in
2959 class LoadAcquire<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2960 RegisterClass regtype, string asm>
2961 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
2962 (ins am_noindex:$addr), asm, "\t$Rt, $addr">,
2965 class LoadExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2966 RegisterClass regtype, string asm>
2967 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs regtype:$Rt),
2968 (ins am_noindex:$addr), asm, "\t$Rt, $addr">,
2971 class LoadExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2972 RegisterClass regtype, string asm>
2973 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
2974 (outs regtype:$Rt, regtype:$Rt2),
2975 (ins am_noindex:$addr), asm,
2976 "\t$Rt, $Rt2, $addr">,
2977 Sched<[WriteLD, WriteLDHi]> {
2981 let Inst{14-10} = dst2;
2982 let Inst{9-5} = base;
2983 let Inst{4-0} = dst1;
2985 let PostEncoderMethod = "fixLoadStoreExclusive<0,1>";
2988 // Simple store release operations do not check the exclusive monitor.
2989 let mayLoad = 0, mayStore = 1 in
2990 class StoreRelease<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2991 RegisterClass regtype, string asm>
2992 : LoadStoreExclusiveSimple<sz, o2, L, o1, o0, (outs),
2993 (ins regtype:$Rt, am_noindex:$addr),
2994 asm, "\t$Rt, $addr">,
2997 let mayLoad = 1, mayStore = 1 in
2998 class StoreExclusive<bits<2> sz, bit o2, bit L, bit o1, bit o0,
2999 RegisterClass regtype, string asm>
3000 : BaseLoadStoreExclusive<sz, o2, L, o1, o0, (outs GPR32:$Ws),
3001 (ins regtype:$Rt, am_noindex:$addr),
3002 asm, "\t$Ws, $Rt, $addr">,
3007 let Inst{20-16} = status;
3008 let Inst{9-5} = base;
3009 let Inst{4-0} = reg;
3011 let Constraints = "@earlyclobber $Ws";
3012 let PostEncoderMethod = "fixLoadStoreExclusive<1,0>";
3015 class StoreExclusivePair<bits<2> sz, bit o2, bit L, bit o1, bit o0,
3016 RegisterClass regtype, string asm>
3017 : BaseLoadStoreExclusive<sz, o2, L, o1, o0,
3019 (ins regtype:$Rt, regtype:$Rt2, am_noindex:$addr),
3020 asm, "\t$Ws, $Rt, $Rt2, $addr">,
3026 let Inst{20-16} = status;
3027 let Inst{14-10} = dst2;
3028 let Inst{9-5} = base;
3029 let Inst{4-0} = dst1;
3031 let Constraints = "@earlyclobber $Ws";
3035 // Exception generation
3038 let mayLoad = 0, mayStore = 0, hasSideEffects = 1 in
3039 class ExceptionGeneration<bits<3> op1, bits<2> ll, string asm>
3040 : I<(outs), (ins imm0_65535:$imm), asm, "\t$imm", "", []>,
3043 let Inst{31-24} = 0b11010100;
3044 let Inst{23-21} = op1;
3045 let Inst{20-5} = imm;
3046 let Inst{4-2} = 0b000;
3051 // Floating point to integer conversion
3054 class BaseFPToIntegerUnscaled<bits<2> type, bits<2> rmode, bits<3> opcode,
3055 RegisterClass srcType, RegisterClass dstType,
3056 string asm, list<dag> pattern>
3057 : I<(outs dstType:$Rd), (ins srcType:$Rn),
3058 asm, "\t$Rd, $Rn", "", pattern>,
3059 Sched<[WriteFCvt]> {
3062 let Inst{30-29} = 0b00;
3063 let Inst{28-24} = 0b11110;
3064 let Inst{23-22} = type;
3066 let Inst{20-19} = rmode;
3067 let Inst{18-16} = opcode;
3068 let Inst{15-10} = 0;
3073 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3074 class BaseFPToInteger<bits<2> type, bits<2> rmode, bits<3> opcode,
3075 RegisterClass srcType, RegisterClass dstType,
3076 Operand immType, string asm>
3077 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
3078 asm, "\t$Rd, $Rn, $scale", "", []>,
3079 Sched<[WriteFCvt]> {
3083 let Inst{30-29} = 0b00;
3084 let Inst{28-24} = 0b11110;
3085 let Inst{23-22} = type;
3087 let Inst{20-19} = rmode;
3088 let Inst{18-16} = opcode;
3089 let Inst{15-10} = scale;
3094 multiclass FPToIntegerUnscaled<bits<2> rmode, bits<3> opcode, string asm,
3095 SDPatternOperator OpN> {
3096 // Unscaled single-precision to 32-bit
3097 def UWSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR32, asm,
3098 [(set GPR32:$Rd, (OpN FPR32:$Rn))]> {
3099 let Inst{31} = 0; // 32-bit GPR flag
3102 // Unscaled single-precision to 64-bit
3103 def UXSr : BaseFPToIntegerUnscaled<0b00, rmode, opcode, FPR32, GPR64, asm,
3104 [(set GPR64:$Rd, (OpN FPR32:$Rn))]> {
3105 let Inst{31} = 1; // 64-bit GPR flag
3108 // Unscaled double-precision to 32-bit
3109 def UWDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR32, asm,
3110 [(set GPR32:$Rd, (OpN (f64 FPR64:$Rn)))]> {
3111 let Inst{31} = 0; // 32-bit GPR flag
3114 // Unscaled double-precision to 64-bit
3115 def UXDr : BaseFPToIntegerUnscaled<0b01, rmode, opcode, FPR64, GPR64, asm,
3116 [(set GPR64:$Rd, (OpN (f64 FPR64:$Rn)))]> {
3117 let Inst{31} = 1; // 64-bit GPR flag
3121 multiclass FPToIntegerScaled<bits<2> rmode, bits<3> opcode, string asm,
3122 SDPatternOperator OpN> {
3123 // Scaled single-precision to 32-bit
3124 def SWSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR32,
3125 fixedpoint32, asm> {
3126 let Inst{31} = 0; // 32-bit GPR flag
3130 // Scaled single-precision to 64-bit
3131 def SXSri : BaseFPToInteger<0b00, rmode, opcode, FPR32, GPR64,
3132 fixedpoint64, asm> {
3133 let Inst{31} = 1; // 64-bit GPR flag
3136 // Scaled double-precision to 32-bit
3137 def SWDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR32,
3138 fixedpoint32, asm> {
3139 let Inst{31} = 0; // 32-bit GPR flag
3143 // Scaled double-precision to 64-bit
3144 def SXDri : BaseFPToInteger<0b01, rmode, opcode, FPR64, GPR64,
3145 fixedpoint64, asm> {
3146 let Inst{31} = 1; // 64-bit GPR flag
3151 // Integer to floating point conversion
3154 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
3155 class BaseIntegerToFP<bit isUnsigned,
3156 RegisterClass srcType, RegisterClass dstType,
3157 Operand immType, string asm>
3158 : I<(outs dstType:$Rd), (ins srcType:$Rn, immType:$scale),
3159 asm, "\t$Rd, $Rn, $scale", "", []>,
3160 Sched<[WriteFCvt]> {
3164 let Inst{30-23} = 0b00111100;
3165 let Inst{21-17} = 0b00001;
3166 let Inst{16} = isUnsigned;
3167 let Inst{15-10} = scale;
3172 class BaseIntegerToFPUnscaled<bit isUnsigned,
3173 RegisterClass srcType, RegisterClass dstType,
3174 ValueType dvt, string asm, SDNode node>
3175 : I<(outs dstType:$Rd), (ins srcType:$Rn),
3176 asm, "\t$Rd, $Rn", "", [(set (dvt dstType:$Rd), (node srcType:$Rn))]>,
3177 Sched<[WriteFCvt]> {
3181 let Inst{30-23} = 0b00111100;
3182 let Inst{21-17} = 0b10001;
3183 let Inst{16} = isUnsigned;
3184 let Inst{15-10} = 0b000000;
3189 multiclass IntegerToFP<bit isUnsigned, string asm, SDNode node> {
3191 def UWSri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR32, f32, asm, node> {
3192 let Inst{31} = 0; // 32-bit GPR flag
3193 let Inst{22} = 0; // 32-bit FPR flag
3196 def UWDri: BaseIntegerToFPUnscaled<isUnsigned, GPR32, FPR64, f64, asm, node> {
3197 let Inst{31} = 0; // 32-bit GPR flag
3198 let Inst{22} = 1; // 64-bit FPR flag
3201 def UXSri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR32, f32, asm, node> {
3202 let Inst{31} = 1; // 64-bit GPR flag
3203 let Inst{22} = 0; // 32-bit FPR flag
3206 def UXDri: BaseIntegerToFPUnscaled<isUnsigned, GPR64, FPR64, f64, asm, node> {
3207 let Inst{31} = 1; // 64-bit GPR flag
3208 let Inst{22} = 1; // 64-bit FPR flag
3212 def SWSri: BaseIntegerToFP<isUnsigned, GPR32, FPR32, fixedpoint32, asm> {
3213 let Inst{31} = 0; // 32-bit GPR flag
3214 let Inst{22} = 0; // 32-bit FPR flag
3218 def SWDri: BaseIntegerToFP<isUnsigned, GPR32, FPR64, fixedpoint32, asm> {
3219 let Inst{31} = 0; // 32-bit GPR flag
3220 let Inst{22} = 1; // 64-bit FPR flag
3224 def SXSri: BaseIntegerToFP<isUnsigned, GPR64, FPR32, fixedpoint64, asm> {
3225 let Inst{31} = 1; // 64-bit GPR flag
3226 let Inst{22} = 0; // 32-bit FPR flag
3229 def SXDri: BaseIntegerToFP<isUnsigned, GPR64, FPR64, fixedpoint64, asm> {
3230 let Inst{31} = 1; // 64-bit GPR flag
3231 let Inst{22} = 1; // 64-bit FPR flag
3236 // Unscaled integer <-> floating point conversion (i.e. FMOV)
3239 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3240 class BaseUnscaledConversion<bits<2> rmode, bits<3> opcode,
3241 RegisterClass srcType, RegisterClass dstType,
3243 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "",
3244 // We use COPY_TO_REGCLASS for these bitconvert operations.
3245 // copyPhysReg() expands the resultant COPY instructions after
3246 // regalloc is done. This gives greater freedom for the allocator
3247 // and related passes (coalescing, copy propagation, et. al.) to
3248 // be more effective.
3249 [/*(set (dvt dstType:$Rd), (bitconvert (svt srcType:$Rn)))*/]>,
3250 Sched<[WriteFCopy]> {
3253 let Inst{30-23} = 0b00111100;
3255 let Inst{20-19} = rmode;
3256 let Inst{18-16} = opcode;
3257 let Inst{15-10} = 0b000000;
3262 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3263 class BaseUnscaledConversionToHigh<bits<2> rmode, bits<3> opcode,
3264 RegisterClass srcType, RegisterOperand dstType, string asm,
3266 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm,
3267 "{\t$Rd"#kind#"[1], $Rn|"#kind#"\t$Rd[1], $Rn}", "", []>,
3268 Sched<[WriteFCopy]> {
3271 let Inst{30-23} = 0b00111101;
3273 let Inst{20-19} = rmode;
3274 let Inst{18-16} = opcode;
3275 let Inst{15-10} = 0b000000;
3280 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3281 class BaseUnscaledConversionFromHigh<bits<2> rmode, bits<3> opcode,
3282 RegisterOperand srcType, RegisterClass dstType, string asm,
3284 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm,
3285 "{\t$Rd, $Rn"#kind#"[1]|"#kind#"\t$Rd, $Rn[1]}", "", []>,
3286 Sched<[WriteFCopy]> {
3289 let Inst{30-23} = 0b00111101;
3291 let Inst{20-19} = rmode;
3292 let Inst{18-16} = opcode;
3293 let Inst{15-10} = 0b000000;
3300 multiclass UnscaledConversion<string asm> {
3301 def WSr : BaseUnscaledConversion<0b00, 0b111, GPR32, FPR32, asm> {
3302 let Inst{31} = 0; // 32-bit GPR flag
3303 let Inst{22} = 0; // 32-bit FPR flag
3306 def XDr : BaseUnscaledConversion<0b00, 0b111, GPR64, FPR64, asm> {
3307 let Inst{31} = 1; // 64-bit GPR flag
3308 let Inst{22} = 1; // 64-bit FPR flag
3311 def SWr : BaseUnscaledConversion<0b00, 0b110, FPR32, GPR32, asm> {
3312 let Inst{31} = 0; // 32-bit GPR flag
3313 let Inst{22} = 0; // 32-bit FPR flag
3316 def DXr : BaseUnscaledConversion<0b00, 0b110, FPR64, GPR64, asm> {
3317 let Inst{31} = 1; // 64-bit GPR flag
3318 let Inst{22} = 1; // 64-bit FPR flag
3321 def XDHighr : BaseUnscaledConversionToHigh<0b01, 0b111, GPR64, V128,
3327 def DXHighr : BaseUnscaledConversionFromHigh<0b01, 0b110, V128, GPR64,
3335 // Floating point conversion
3338 class BaseFPConversion<bits<2> type, bits<2> opcode, RegisterClass dstType,
3339 RegisterClass srcType, string asm, list<dag> pattern>
3340 : I<(outs dstType:$Rd), (ins srcType:$Rn), asm, "\t$Rd, $Rn", "", pattern>,
3341 Sched<[WriteFCvt]> {
3344 let Inst{31-24} = 0b00011110;
3345 let Inst{23-22} = type;
3346 let Inst{21-17} = 0b10001;
3347 let Inst{16-15} = opcode;
3348 let Inst{14-10} = 0b10000;
3353 multiclass FPConversion<string asm> {
3354 // Double-precision to Half-precision
3355 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3356 def HDr : BaseFPConversion<0b01, 0b11, FPR16, FPR64, asm, []>;
3358 // Double-precision to Single-precision
3359 def SDr : BaseFPConversion<0b01, 0b00, FPR32, FPR64, asm,
3360 [(set FPR32:$Rd, (fround FPR64:$Rn))]>;
3362 // Half-precision to Double-precision
3363 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3364 def DHr : BaseFPConversion<0b11, 0b01, FPR64, FPR16, asm, []>;
3366 // Half-precision to Single-precision
3367 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3368 def SHr : BaseFPConversion<0b11, 0b00, FPR32, FPR16, asm, []>;
3370 // Single-precision to Double-precision
3371 def DSr : BaseFPConversion<0b00, 0b01, FPR64, FPR32, asm,
3372 [(set FPR64:$Rd, (fextend FPR32:$Rn))]>;
3374 // Single-precision to Half-precision
3375 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3376 def HSr : BaseFPConversion<0b00, 0b11, FPR16, FPR32, asm, []>;
3380 // Single operand floating point data processing
3383 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3384 class BaseSingleOperandFPData<bits<4> opcode, RegisterClass regtype,
3385 ValueType vt, string asm, SDPatternOperator node>
3386 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm, "\t$Rd, $Rn", "",
3387 [(set (vt regtype:$Rd), (node (vt regtype:$Rn)))]>,
3391 let Inst{31-23} = 0b000111100;
3392 let Inst{21-19} = 0b100;
3393 let Inst{18-15} = opcode;
3394 let Inst{14-10} = 0b10000;
3399 multiclass SingleOperandFPData<bits<4> opcode, string asm,
3400 SDPatternOperator node = null_frag> {
3401 def Sr : BaseSingleOperandFPData<opcode, FPR32, f32, asm, node> {
3402 let Inst{22} = 0; // 32-bit size flag
3405 def Dr : BaseSingleOperandFPData<opcode, FPR64, f64, asm, node> {
3406 let Inst{22} = 1; // 64-bit size flag
3411 // Two operand floating point data processing
3414 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3415 class BaseTwoOperandFPData<bits<4> opcode, RegisterClass regtype,
3416 string asm, list<dag> pat>
3417 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm),
3418 asm, "\t$Rd, $Rn, $Rm", "", pat>,
3423 let Inst{31-23} = 0b000111100;
3425 let Inst{20-16} = Rm;
3426 let Inst{15-12} = opcode;
3427 let Inst{11-10} = 0b10;
3432 multiclass TwoOperandFPData<bits<4> opcode, string asm,
3433 SDPatternOperator node = null_frag> {
3434 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
3435 [(set (f32 FPR32:$Rd),
3436 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]> {
3437 let Inst{22} = 0; // 32-bit size flag
3440 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
3441 [(set (f64 FPR64:$Rd),
3442 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]> {
3443 let Inst{22} = 1; // 64-bit size flag
3447 multiclass TwoOperandFPDataNeg<bits<4> opcode, string asm, SDNode node> {
3448 def Srr : BaseTwoOperandFPData<opcode, FPR32, asm,
3449 [(set FPR32:$Rd, (fneg (node FPR32:$Rn, (f32 FPR32:$Rm))))]> {
3450 let Inst{22} = 0; // 32-bit size flag
3453 def Drr : BaseTwoOperandFPData<opcode, FPR64, asm,
3454 [(set FPR64:$Rd, (fneg (node FPR64:$Rn, (f64 FPR64:$Rm))))]> {
3455 let Inst{22} = 1; // 64-bit size flag
3461 // Three operand floating point data processing
3464 class BaseThreeOperandFPData<bit isNegated, bit isSub,
3465 RegisterClass regtype, string asm, list<dag> pat>
3466 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, regtype: $Ra),
3467 asm, "\t$Rd, $Rn, $Rm, $Ra", "", pat>,
3468 Sched<[WriteFMul]> {
3473 let Inst{31-23} = 0b000111110;
3474 let Inst{21} = isNegated;
3475 let Inst{20-16} = Rm;
3476 let Inst{15} = isSub;
3477 let Inst{14-10} = Ra;
3482 multiclass ThreeOperandFPData<bit isNegated, bit isSub,string asm,
3483 SDPatternOperator node> {
3484 def Srrr : BaseThreeOperandFPData<isNegated, isSub, FPR32, asm,
3486 (node (f32 FPR32:$Rn), (f32 FPR32:$Rm), (f32 FPR32:$Ra)))]> {
3487 let Inst{22} = 0; // 32-bit size flag
3490 def Drrr : BaseThreeOperandFPData<isNegated, isSub, FPR64, asm,
3492 (node (f64 FPR64:$Rn), (f64 FPR64:$Rm), (f64 FPR64:$Ra)))]> {
3493 let Inst{22} = 1; // 64-bit size flag
3498 // Floating point data comparisons
3501 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3502 class BaseOneOperandFPComparison<bit signalAllNans,
3503 RegisterClass regtype, string asm,
3505 : I<(outs), (ins regtype:$Rn), asm, "\t$Rn, #0.0", "", pat>,
3506 Sched<[WriteFCmp]> {
3508 let Inst{31-23} = 0b000111100;
3511 let Inst{15-10} = 0b001000;
3513 let Inst{4} = signalAllNans;
3514 let Inst{3-0} = 0b1000;
3516 // Rm should be 0b00000 canonically, but we need to accept any value.
3517 let PostEncoderMethod = "fixOneOperandFPComparison";
3520 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3521 class BaseTwoOperandFPComparison<bit signalAllNans, RegisterClass regtype,
3522 string asm, list<dag> pat>
3523 : I<(outs), (ins regtype:$Rn, regtype:$Rm), asm, "\t$Rn, $Rm", "", pat>,
3524 Sched<[WriteFCmp]> {
3527 let Inst{31-23} = 0b000111100;
3529 let Inst{20-16} = Rm;
3530 let Inst{15-10} = 0b001000;
3532 let Inst{4} = signalAllNans;
3533 let Inst{3-0} = 0b0000;
3536 multiclass FPComparison<bit signalAllNans, string asm,
3537 SDPatternOperator OpNode = null_frag> {
3538 let Defs = [CPSR] in {
3539 def Srr : BaseTwoOperandFPComparison<signalAllNans, FPR32, asm,
3540 [(OpNode FPR32:$Rn, (f32 FPR32:$Rm)), (implicit CPSR)]> {
3544 def Sri : BaseOneOperandFPComparison<signalAllNans, FPR32, asm,
3545 [(OpNode (f32 FPR32:$Rn), fpimm0), (implicit CPSR)]> {
3549 def Drr : BaseTwoOperandFPComparison<signalAllNans, FPR64, asm,
3550 [(OpNode FPR64:$Rn, (f64 FPR64:$Rm)), (implicit CPSR)]> {
3554 def Dri : BaseOneOperandFPComparison<signalAllNans, FPR64, asm,
3555 [(OpNode (f64 FPR64:$Rn), fpimm0), (implicit CPSR)]> {
3562 // Floating point conditional comparisons
3565 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3566 class BaseFPCondComparison<bit signalAllNans,
3567 RegisterClass regtype, string asm>
3568 : I<(outs), (ins regtype:$Rn, regtype:$Rm, imm0_15:$nzcv, ccode:$cond),
3569 asm, "\t$Rn, $Rm, $nzcv, $cond", "", []>,
3570 Sched<[WriteFCmp]> {
3576 let Inst{31-23} = 0b000111100;
3578 let Inst{20-16} = Rm;
3579 let Inst{15-12} = cond;
3580 let Inst{11-10} = 0b01;
3582 let Inst{4} = signalAllNans;
3583 let Inst{3-0} = nzcv;
3586 multiclass FPCondComparison<bit signalAllNans, string asm> {
3587 let Defs = [CPSR], Uses = [CPSR] in {
3588 def Srr : BaseFPCondComparison<signalAllNans, FPR32, asm> {
3592 def Drr : BaseFPCondComparison<signalAllNans, FPR64, asm> {
3595 } // Defs = [CPSR], Uses = [CPSR]
3599 // Floating point conditional select
3602 class BaseFPCondSelect<RegisterClass regtype, ValueType vt, string asm>
3603 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, ccode:$cond),
3604 asm, "\t$Rd, $Rn, $Rm, $cond", "",
3606 (ARM64csel (vt regtype:$Rn), regtype:$Rm,
3607 (i32 imm:$cond), CPSR))]>,
3614 let Inst{31-23} = 0b000111100;
3616 let Inst{20-16} = Rm;
3617 let Inst{15-12} = cond;
3618 let Inst{11-10} = 0b11;
3623 multiclass FPCondSelect<string asm> {
3624 let Uses = [CPSR] in {
3625 def Srrr : BaseFPCondSelect<FPR32, f32, asm> {
3629 def Drrr : BaseFPCondSelect<FPR64, f64, asm> {
3636 // Floating move immediate
3639 class BaseFPMoveImmediate<RegisterClass regtype, Operand fpimmtype, string asm>
3640 : I<(outs regtype:$Rd), (ins fpimmtype:$imm), asm, "\t$Rd, $imm", "",
3641 [(set regtype:$Rd, fpimmtype:$imm)]>,
3642 Sched<[WriteFImm]> {
3645 let Inst{31-23} = 0b000111100;
3647 let Inst{20-13} = imm;
3648 let Inst{12-5} = 0b10000000;
3652 multiclass FPMoveImmediate<string asm> {
3653 def Si : BaseFPMoveImmediate<FPR32, fpimm32, asm> {
3657 def Di : BaseFPMoveImmediate<FPR64, fpimm64, asm> {
3662 //----------------------------------------------------------------------------
3664 //----------------------------------------------------------------------------
3666 def VectorIndexBOperand : AsmOperandClass { let Name = "VectorIndexB"; }
3667 def VectorIndexHOperand : AsmOperandClass { let Name = "VectorIndexH"; }
3668 def VectorIndexSOperand : AsmOperandClass { let Name = "VectorIndexS"; }
3669 def VectorIndexDOperand : AsmOperandClass { let Name = "VectorIndexD"; }
3670 def VectorIndexB : Operand<i64>, ImmLeaf<i64, [{
3671 return ((uint64_t)Imm) < 16;
3673 let ParserMatchClass = VectorIndexBOperand;
3674 let PrintMethod = "printVectorIndex";
3675 let MIOperandInfo = (ops i64imm);
3677 def VectorIndexH : Operand<i64>, ImmLeaf<i64, [{
3678 return ((uint64_t)Imm) < 8;
3680 let ParserMatchClass = VectorIndexHOperand;
3681 let PrintMethod = "printVectorIndex";
3682 let MIOperandInfo = (ops i64imm);
3684 def VectorIndexS : Operand<i64>, ImmLeaf<i64, [{
3685 return ((uint64_t)Imm) < 4;
3687 let ParserMatchClass = VectorIndexSOperand;
3688 let PrintMethod = "printVectorIndex";
3689 let MIOperandInfo = (ops i64imm);
3691 def VectorIndexD : Operand<i64>, ImmLeaf<i64, [{
3692 return ((uint64_t)Imm) < 2;
3694 let ParserMatchClass = VectorIndexDOperand;
3695 let PrintMethod = "printVectorIndex";
3696 let MIOperandInfo = (ops i64imm);
3699 //----------------------------------------------------------------------------
3700 // AdvSIMD three register vector instructions
3701 //----------------------------------------------------------------------------
3703 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3704 class BaseSIMDThreeSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
3705 RegisterOperand regtype, string asm, string kind,
3707 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
3708 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
3709 "|" # kind # "\t$Rd, $Rn, $Rm|}", "", pattern>,
3717 let Inst{28-24} = 0b01110;
3718 let Inst{23-22} = size;
3720 let Inst{20-16} = Rm;
3721 let Inst{15-11} = opcode;
3727 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3728 class BaseSIMDThreeSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
3729 RegisterOperand regtype, string asm, string kind,
3731 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn, regtype:$Rm), asm,
3732 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
3733 "|" # kind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
3741 let Inst{28-24} = 0b01110;
3742 let Inst{23-22} = size;
3744 let Inst{20-16} = Rm;
3745 let Inst{15-11} = opcode;
3751 // All operand sizes distinguished in the encoding.
3752 multiclass SIMDThreeSameVector<bit U, bits<5> opc, string asm,
3753 SDPatternOperator OpNode> {
3754 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3756 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3757 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3759 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3760 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3762 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3763 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3765 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3766 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3768 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3769 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3771 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3772 def v2i64 : BaseSIMDThreeSameVector<1, U, 0b11, opc, V128,
3774 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
3777 // As above, but D sized elements unsupported.
3778 multiclass SIMDThreeSameVectorBHS<bit U, bits<5> opc, string asm,
3779 SDPatternOperator OpNode> {
3780 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3782 [(set V64:$Rd, (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))]>;
3783 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3785 [(set V128:$Rd, (v16i8 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm))))]>;
3786 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3788 [(set V64:$Rd, (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))]>;
3789 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3791 [(set V128:$Rd, (v8i16 (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm))))]>;
3792 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3794 [(set V64:$Rd, (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))]>;
3795 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3797 [(set V128:$Rd, (v4i32 (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm))))]>;
3800 multiclass SIMDThreeSameVectorBHSTied<bit U, bits<5> opc, string asm,
3801 SDPatternOperator OpNode> {
3802 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, 0b00, opc, V64,
3804 [(set (v8i8 V64:$dst),
3805 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3806 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, 0b00, opc, V128,
3808 [(set (v16i8 V128:$dst),
3809 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3810 def v4i16 : BaseSIMDThreeSameVectorTied<0, U, 0b01, opc, V64,
3812 [(set (v4i16 V64:$dst),
3813 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3814 def v8i16 : BaseSIMDThreeSameVectorTied<1, U, 0b01, opc, V128,
3816 [(set (v8i16 V128:$dst),
3817 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3818 def v2i32 : BaseSIMDThreeSameVectorTied<0, U, 0b10, opc, V64,
3820 [(set (v2i32 V64:$dst),
3821 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3822 def v4i32 : BaseSIMDThreeSameVectorTied<1, U, 0b10, opc, V128,
3824 [(set (v4i32 V128:$dst),
3825 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3828 // As above, but only B sized elements supported.
3829 multiclass SIMDThreeSameVectorB<bit U, bits<5> opc, string asm,
3830 SDPatternOperator OpNode> {
3831 def v8i8 : BaseSIMDThreeSameVector<0, U, 0b00, opc, V64,
3833 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3834 def v16i8 : BaseSIMDThreeSameVector<1, U, 0b00, opc, V128,
3836 [(set (v16i8 V128:$Rd),
3837 (OpNode (v16i8 V128:$Rn), (v16i8 V128:$Rm)))]>;
3840 // As above, but only S and D sized floating point elements supported.
3841 multiclass SIMDThreeSameVectorFP<bit U, bit S, bits<5> opc,
3842 string asm, SDPatternOperator OpNode> {
3843 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0}, opc, V64,
3845 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3846 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0}, opc, V128,
3848 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3849 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,1}, opc, V128,
3851 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3854 multiclass SIMDThreeSameVectorFPCmp<bit U, bit S, bits<5> opc,
3856 SDPatternOperator OpNode> {
3857 def v2f32 : BaseSIMDThreeSameVector<0, U, {S,0}, opc, V64,
3859 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3860 def v4f32 : BaseSIMDThreeSameVector<1, U, {S,0}, opc, V128,
3862 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3863 def v2f64 : BaseSIMDThreeSameVector<1, U, {S,1}, opc, V128,
3865 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3868 multiclass SIMDThreeSameVectorFPTied<bit U, bit S, bits<5> opc,
3869 string asm, SDPatternOperator OpNode> {
3870 def v2f32 : BaseSIMDThreeSameVectorTied<0, U, {S,0}, opc, V64,
3872 [(set (v2f32 V64:$dst),
3873 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn), (v2f32 V64:$Rm)))]>;
3874 def v4f32 : BaseSIMDThreeSameVectorTied<1, U, {S,0}, opc, V128,
3876 [(set (v4f32 V128:$dst),
3877 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn), (v4f32 V128:$Rm)))]>;
3878 def v2f64 : BaseSIMDThreeSameVectorTied<1, U, {S,1}, opc, V128,
3880 [(set (v2f64 V128:$dst),
3881 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn), (v2f64 V128:$Rm)))]>;
3884 // As above, but D and B sized elements unsupported.
3885 multiclass SIMDThreeSameVectorHS<bit U, bits<5> opc, string asm,
3886 SDPatternOperator OpNode> {
3887 def v4i16 : BaseSIMDThreeSameVector<0, U, 0b01, opc, V64,
3889 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
3890 def v8i16 : BaseSIMDThreeSameVector<1, U, 0b01, opc, V128,
3892 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
3893 def v2i32 : BaseSIMDThreeSameVector<0, U, 0b10, opc, V64,
3895 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
3896 def v4i32 : BaseSIMDThreeSameVector<1, U, 0b10, opc, V128,
3898 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
3901 // Logical three vector ops share opcode bits, and only use B sized elements.
3902 multiclass SIMDLogicalThreeVector<bit U, bits<2> size, string asm,
3903 SDPatternOperator OpNode = null_frag> {
3904 def v8i8 : BaseSIMDThreeSameVector<0, U, size, 0b00011, V64,
3906 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn, V64:$Rm))]>;
3907 def v16i8 : BaseSIMDThreeSameVector<1, U, size, 0b00011, V128,
3909 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn, V128:$Rm))]>;
3911 def : Pat<(v4i16 (OpNode V64:$LHS, V64:$RHS)),
3912 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
3913 def : Pat<(v2i32 (OpNode V64:$LHS, V64:$RHS)),
3914 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
3915 def : Pat<(v1i64 (OpNode V64:$LHS, V64:$RHS)),
3916 (!cast<Instruction>(NAME#"v8i8") V64:$LHS, V64:$RHS)>;
3918 def : Pat<(v8i16 (OpNode V128:$LHS, V128:$RHS)),
3919 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
3920 def : Pat<(v4i32 (OpNode V128:$LHS, V128:$RHS)),
3921 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
3922 def : Pat<(v2i64 (OpNode V128:$LHS, V128:$RHS)),
3923 (!cast<Instruction>(NAME#"v16i8") V128:$LHS, V128:$RHS)>;
3926 multiclass SIMDLogicalThreeVectorTied<bit U, bits<2> size,
3927 string asm, SDPatternOperator OpNode> {
3928 def v8i8 : BaseSIMDThreeSameVectorTied<0, U, size, 0b00011, V64,
3930 [(set (v8i8 V64:$dst),
3931 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
3932 def v16i8 : BaseSIMDThreeSameVectorTied<1, U, size, 0b00011, V128,
3934 [(set (v16i8 V128:$dst),
3935 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
3936 (v16i8 V128:$Rm)))]>;
3938 def : Pat<(v4i16 (OpNode (v4i16 V64:$LHS), (v4i16 V64:$MHS),
3940 (!cast<Instruction>(NAME#"v8i8")
3941 V64:$LHS, V64:$MHS, V64:$RHS)>;
3942 def : Pat<(v2i32 (OpNode (v2i32 V64:$LHS), (v2i32 V64:$MHS),
3944 (!cast<Instruction>(NAME#"v8i8")
3945 V64:$LHS, V64:$MHS, V64:$RHS)>;
3946 def : Pat<(v1i64 (OpNode (v1i64 V64:$LHS), (v1i64 V64:$MHS),
3948 (!cast<Instruction>(NAME#"v8i8")
3949 V64:$LHS, V64:$MHS, V64:$RHS)>;
3951 def : Pat<(v8i16 (OpNode (v8i16 V128:$LHS), (v8i16 V128:$MHS),
3952 (v8i16 V128:$RHS))),
3953 (!cast<Instruction>(NAME#"v16i8")
3954 V128:$LHS, V128:$MHS, V128:$RHS)>;
3955 def : Pat<(v4i32 (OpNode (v4i32 V128:$LHS), (v4i32 V128:$MHS),
3956 (v4i32 V128:$RHS))),
3957 (!cast<Instruction>(NAME#"v16i8")
3958 V128:$LHS, V128:$MHS, V128:$RHS)>;
3959 def : Pat<(v2i64 (OpNode (v2i64 V128:$LHS), (v2i64 V128:$MHS),
3960 (v2i64 V128:$RHS))),
3961 (!cast<Instruction>(NAME#"v16i8")
3962 V128:$LHS, V128:$MHS, V128:$RHS)>;
3966 //----------------------------------------------------------------------------
3967 // AdvSIMD two register vector instructions.
3968 //----------------------------------------------------------------------------
3970 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3971 class BaseSIMDTwoSameVector<bit Q, bit U, bits<2> size, bits<5> opcode,
3972 RegisterOperand regtype, string asm, string dstkind,
3973 string srckind, list<dag> pattern>
3974 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
3975 "{\t$Rd" # dstkind # ", $Rn" # srckind #
3976 "|" # dstkind # "\t$Rd, $Rn}", "", pattern>,
3983 let Inst{28-24} = 0b01110;
3984 let Inst{23-22} = size;
3985 let Inst{21-17} = 0b10000;
3986 let Inst{16-12} = opcode;
3987 let Inst{11-10} = 0b10;
3992 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
3993 class BaseSIMDTwoSameVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
3994 RegisterOperand regtype, string asm, string dstkind,
3995 string srckind, list<dag> pattern>
3996 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype:$Rn), asm,
3997 "{\t$Rd" # dstkind # ", $Rn" # srckind #
3998 "|" # dstkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
4005 let Inst{28-24} = 0b01110;
4006 let Inst{23-22} = size;
4007 let Inst{21-17} = 0b10000;
4008 let Inst{16-12} = opcode;
4009 let Inst{11-10} = 0b10;
4014 // Supports B, H, and S element sizes.
4015 multiclass SIMDTwoVectorBHS<bit U, bits<5> opc, string asm,
4016 SDPatternOperator OpNode> {
4017 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4019 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4020 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4021 asm, ".16b", ".16b",
4022 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4023 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4025 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4026 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4028 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4029 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4031 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4032 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4034 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4037 class BaseSIMDVectorLShiftLongBySize<bit Q, bits<2> size,
4038 RegisterOperand regtype, string asm, string dstkind,
4039 string srckind, string amount>
4040 : I<(outs V128:$Rd), (ins regtype:$Rn), asm,
4041 "{\t$Rd" # dstkind # ", $Rn" # srckind # ", #" # amount #
4042 "|" # dstkind # "\t$Rd, $Rn, #" # amount # "}", "", []>,
4048 let Inst{29-24} = 0b101110;
4049 let Inst{23-22} = size;
4050 let Inst{21-10} = 0b100001001110;
4055 multiclass SIMDVectorLShiftLongBySizeBHS {
4056 let neverHasSideEffects = 1 in {
4057 def v8i8 : BaseSIMDVectorLShiftLongBySize<0, 0b00, V64,
4058 "shll", ".8h", ".8b", "8">;
4059 def v16i8 : BaseSIMDVectorLShiftLongBySize<1, 0b00, V128,
4060 "shll2", ".8h", ".16b", "8">;
4061 def v4i16 : BaseSIMDVectorLShiftLongBySize<0, 0b01, V64,
4062 "shll", ".4s", ".4h", "16">;
4063 def v8i16 : BaseSIMDVectorLShiftLongBySize<1, 0b01, V128,
4064 "shll2", ".4s", ".8h", "16">;
4065 def v2i32 : BaseSIMDVectorLShiftLongBySize<0, 0b10, V64,
4066 "shll", ".2d", ".2s", "32">;
4067 def v4i32 : BaseSIMDVectorLShiftLongBySize<1, 0b10, V128,
4068 "shll2", ".2d", ".4s", "32">;
4072 // Supports all element sizes.
4073 multiclass SIMDLongTwoVector<bit U, bits<5> opc, string asm,
4074 SDPatternOperator OpNode> {
4075 def v8i8_v4i16 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4077 [(set (v4i16 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4078 def v16i8_v8i16 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4080 [(set (v8i16 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4081 def v4i16_v2i32 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4083 [(set (v2i32 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4084 def v8i16_v4i32 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4086 [(set (v4i32 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4087 def v2i32_v1i64 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4089 [(set (v1i64 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4090 def v4i32_v2i64 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4092 [(set (v2i64 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4095 multiclass SIMDLongTwoVectorTied<bit U, bits<5> opc, string asm,
4096 SDPatternOperator OpNode> {
4097 def v8i8_v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, V64,
4099 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd),
4101 def v16i8_v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, V128,
4103 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd),
4104 (v16i8 V128:$Rn)))]>;
4105 def v4i16_v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, V64,
4107 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd),
4108 (v4i16 V64:$Rn)))]>;
4109 def v8i16_v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, V128,
4111 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd),
4112 (v8i16 V128:$Rn)))]>;
4113 def v2i32_v1i64 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, V64,
4115 [(set (v1i64 V64:$dst), (OpNode (v1i64 V64:$Rd),
4116 (v2i32 V64:$Rn)))]>;
4117 def v4i32_v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, V128,
4119 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd),
4120 (v4i32 V128:$Rn)))]>;
4123 // Supports all element sizes, except 1xD.
4124 multiclass SIMDTwoVectorBHSDTied<bit U, bits<5> opc, string asm,
4125 SDPatternOperator OpNode> {
4126 def v8i8 : BaseSIMDTwoSameVectorTied<0, U, 0b00, opc, V64,
4128 [(set (v8i8 V64:$dst), (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn)))]>;
4129 def v16i8 : BaseSIMDTwoSameVectorTied<1, U, 0b00, opc, V128,
4130 asm, ".16b", ".16b",
4131 [(set (v16i8 V128:$dst), (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
4132 def v4i16 : BaseSIMDTwoSameVectorTied<0, U, 0b01, opc, V64,
4134 [(set (v4i16 V64:$dst), (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn)))]>;
4135 def v8i16 : BaseSIMDTwoSameVectorTied<1, U, 0b01, opc, V128,
4137 [(set (v8i16 V128:$dst), (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn)))]>;
4138 def v2i32 : BaseSIMDTwoSameVectorTied<0, U, 0b10, opc, V64,
4140 [(set (v2i32 V64:$dst), (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn)))]>;
4141 def v4i32 : BaseSIMDTwoSameVectorTied<1, U, 0b10, opc, V128,
4143 [(set (v4i32 V128:$dst), (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
4144 def v2i64 : BaseSIMDTwoSameVectorTied<1, U, 0b11, opc, V128,
4146 [(set (v2i64 V128:$dst), (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn)))]>;
4149 multiclass SIMDTwoVectorBHSD<bit U, bits<5> opc, string asm,
4150 SDPatternOperator OpNode = null_frag> {
4151 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4153 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4154 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4155 asm, ".16b", ".16b",
4156 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4157 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4159 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn)))]>;
4160 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4162 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4163 def v2i32 : BaseSIMDTwoSameVector<0, U, 0b10, opc, V64,
4165 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4166 def v4i32 : BaseSIMDTwoSameVector<1, U, 0b10, opc, V128,
4168 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4169 def v2i64 : BaseSIMDTwoSameVector<1, U, 0b11, opc, V128,
4171 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4175 // Supports only B element sizes.
4176 multiclass SIMDTwoVectorB<bit U, bits<2> size, bits<5> opc, string asm,
4177 SDPatternOperator OpNode> {
4178 def v8i8 : BaseSIMDTwoSameVector<0, U, size, opc, V64,
4180 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn)))]>;
4181 def v16i8 : BaseSIMDTwoSameVector<1, U, size, opc, V128,
4182 asm, ".16b", ".16b",
4183 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
4187 // Supports only B and H element sizes.
4188 multiclass SIMDTwoVectorBH<bit U, bits<5> opc, string asm,
4189 SDPatternOperator OpNode> {
4190 def v8i8 : BaseSIMDTwoSameVector<0, U, 0b00, opc, V64,
4192 [(set (v8i8 V64:$Rd), (OpNode V64:$Rn))]>;
4193 def v16i8 : BaseSIMDTwoSameVector<1, U, 0b00, opc, V128,
4194 asm, ".16b", ".16b",
4195 [(set (v16i8 V128:$Rd), (OpNode V128:$Rn))]>;
4196 def v4i16 : BaseSIMDTwoSameVector<0, U, 0b01, opc, V64,
4198 [(set (v4i16 V64:$Rd), (OpNode V64:$Rn))]>;
4199 def v8i16 : BaseSIMDTwoSameVector<1, U, 0b01, opc, V128,
4201 [(set (v8i16 V128:$Rd), (OpNode V128:$Rn))]>;
4204 // Supports only S and D element sizes, uses high bit of the size field
4205 // as an extra opcode bit.
4206 multiclass SIMDTwoVectorFP<bit U, bit S, bits<5> opc, string asm,
4207 SDPatternOperator OpNode> {
4208 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4210 [(set (v2f32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
4211 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4213 [(set (v4f32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
4214 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4216 [(set (v2f64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4219 // Supports only S element size.
4220 multiclass SIMDTwoVectorS<bit U, bit S, bits<5> opc, string asm,
4221 SDPatternOperator OpNode> {
4222 def v2i32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4224 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4225 def v4i32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4227 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4231 multiclass SIMDTwoVectorFPToInt<bit U, bit S, bits<5> opc, string asm,
4232 SDPatternOperator OpNode> {
4233 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4235 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn)))]>;
4236 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4238 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn)))]>;
4239 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4241 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4244 multiclass SIMDTwoVectorIntToFP<bit U, bit S, bits<5> opc, string asm,
4245 SDPatternOperator OpNode> {
4246 def v2f32 : BaseSIMDTwoSameVector<0, U, {S,0}, opc, V64,
4248 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn)))]>;
4249 def v4f32 : BaseSIMDTwoSameVector<1, U, {S,0}, opc, V128,
4251 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4252 def v2f64 : BaseSIMDTwoSameVector<1, U, {S,1}, opc, V128,
4254 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4258 class BaseSIMDMixedTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4259 RegisterOperand inreg, RegisterOperand outreg,
4260 string asm, string outkind, string inkind,
4262 : I<(outs outreg:$Rd), (ins inreg:$Rn), asm,
4263 "{\t$Rd" # outkind # ", $Rn" # inkind #
4264 "|" # outkind # "\t$Rd, $Rn}", "", pattern>,
4271 let Inst{28-24} = 0b01110;
4272 let Inst{23-22} = size;
4273 let Inst{21-17} = 0b10000;
4274 let Inst{16-12} = opcode;
4275 let Inst{11-10} = 0b10;
4280 class BaseSIMDMixedTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4281 RegisterOperand inreg, RegisterOperand outreg,
4282 string asm, string outkind, string inkind,
4284 : I<(outs outreg:$dst), (ins outreg:$Rd, inreg:$Rn), asm,
4285 "{\t$Rd" # outkind # ", $Rn" # inkind #
4286 "|" # outkind # "\t$Rd, $Rn}", "$Rd = $dst", pattern>,
4293 let Inst{28-24} = 0b01110;
4294 let Inst{23-22} = size;
4295 let Inst{21-17} = 0b10000;
4296 let Inst{16-12} = opcode;
4297 let Inst{11-10} = 0b10;
4302 multiclass SIMDMixedTwoVector<bit U, bits<5> opc, string asm,
4303 SDPatternOperator OpNode> {
4304 def v8i8 : BaseSIMDMixedTwoVector<0, U, 0b00, opc, V128, V64,
4306 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn)))]>;
4307 def v16i8 : BaseSIMDMixedTwoVectorTied<1, U, 0b00, opc, V128, V128,
4308 asm#"2", ".16b", ".8h", []>;
4309 def v4i16 : BaseSIMDMixedTwoVector<0, U, 0b01, opc, V128, V64,
4311 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn)))]>;
4312 def v8i16 : BaseSIMDMixedTwoVectorTied<1, U, 0b01, opc, V128, V128,
4313 asm#"2", ".8h", ".4s", []>;
4314 def v2i32 : BaseSIMDMixedTwoVector<0, U, 0b10, opc, V128, V64,
4316 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn)))]>;
4317 def v4i32 : BaseSIMDMixedTwoVectorTied<1, U, 0b10, opc, V128, V128,
4318 asm#"2", ".4s", ".2d", []>;
4320 def : Pat<(concat_vectors (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn))),
4321 (!cast<Instruction>(NAME # "v16i8")
4322 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4323 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn))),
4324 (!cast<Instruction>(NAME # "v8i16")
4325 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4326 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn))),
4327 (!cast<Instruction>(NAME # "v4i32")
4328 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4331 class BaseSIMDCmpTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4332 RegisterOperand regtype, string asm, string kind,
4333 ValueType dty, ValueType sty, SDNode OpNode>
4334 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
4335 "{\t$Rd" # kind # ", $Rn" # kind # ", #0" #
4336 "|" # kind # "\t$Rd, $Rn, #0}", "",
4337 [(set (dty regtype:$Rd), (OpNode (sty regtype:$Rn)))]>,
4344 let Inst{28-24} = 0b01110;
4345 let Inst{23-22} = size;
4346 let Inst{21-17} = 0b10000;
4347 let Inst{16-12} = opcode;
4348 let Inst{11-10} = 0b10;
4353 // Comparisons support all element sizes, except 1xD.
4354 multiclass SIMDCmpTwoVector<bit U, bits<5> opc, string asm,
4356 def v8i8rz : BaseSIMDCmpTwoVector<0, U, 0b00, opc, V64,
4358 v8i8, v8i8, OpNode>;
4359 def v16i8rz : BaseSIMDCmpTwoVector<1, U, 0b00, opc, V128,
4361 v16i8, v16i8, OpNode>;
4362 def v4i16rz : BaseSIMDCmpTwoVector<0, U, 0b01, opc, V64,
4364 v4i16, v4i16, OpNode>;
4365 def v8i16rz : BaseSIMDCmpTwoVector<1, U, 0b01, opc, V128,
4367 v8i16, v8i16, OpNode>;
4368 def v2i32rz : BaseSIMDCmpTwoVector<0, U, 0b10, opc, V64,
4370 v2i32, v2i32, OpNode>;
4371 def v4i32rz : BaseSIMDCmpTwoVector<1, U, 0b10, opc, V128,
4373 v4i32, v4i32, OpNode>;
4374 def v2i64rz : BaseSIMDCmpTwoVector<1, U, 0b11, opc, V128,
4376 v2i64, v2i64, OpNode>;
4379 // FP Comparisons support only S and D element sizes.
4380 multiclass SIMDFPCmpTwoVector<bit U, bit S, bits<5> opc,
4381 string asm, SDNode OpNode> {
4382 def v2i32rz : BaseSIMDCmpTwoVector<0, U, {S,0}, opc, V64,
4384 v2i32, v2f32, OpNode>;
4385 def v4i32rz : BaseSIMDCmpTwoVector<1, U, {S,0}, opc, V128,
4387 v4i32, v4f32, OpNode>;
4388 def v2i64rz : BaseSIMDCmpTwoVector<1, U, {S,1}, opc, V128,
4390 v2i64, v2f64, OpNode>;
4393 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4394 class BaseSIMDFPCvtTwoVector<bit Q, bit U, bits<2> size, bits<5> opcode,
4395 RegisterOperand outtype, RegisterOperand intype,
4396 string asm, string VdTy, string VnTy,
4398 : I<(outs outtype:$Rd), (ins intype:$Rn), asm,
4399 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "", pattern>,
4406 let Inst{28-24} = 0b01110;
4407 let Inst{23-22} = size;
4408 let Inst{21-17} = 0b10000;
4409 let Inst{16-12} = opcode;
4410 let Inst{11-10} = 0b10;
4415 class BaseSIMDFPCvtTwoVectorTied<bit Q, bit U, bits<2> size, bits<5> opcode,
4416 RegisterOperand outtype, RegisterOperand intype,
4417 string asm, string VdTy, string VnTy,
4419 : I<(outs outtype:$dst), (ins outtype:$Rd, intype:$Rn), asm,
4420 !strconcat("\t$Rd", VdTy, ", $Rn", VnTy), "$Rd = $dst", pattern>,
4427 let Inst{28-24} = 0b01110;
4428 let Inst{23-22} = size;
4429 let Inst{21-17} = 0b10000;
4430 let Inst{16-12} = opcode;
4431 let Inst{11-10} = 0b10;
4436 multiclass SIMDFPWidenTwoVector<bit U, bit S, bits<5> opc, string asm> {
4437 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V128, V64,
4438 asm, ".4s", ".4h", []>;
4439 def v8i16 : BaseSIMDFPCvtTwoVector<1, U, {S,0}, opc, V128, V128,
4440 asm#"2", ".4s", ".8h", []>;
4441 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V128, V64,
4442 asm, ".2d", ".2s", []>;
4443 def v4i32 : BaseSIMDFPCvtTwoVector<1, U, {S,1}, opc, V128, V128,
4444 asm#"2", ".2d", ".4s", []>;
4447 multiclass SIMDFPNarrowTwoVector<bit U, bit S, bits<5> opc, string asm> {
4448 def v4i16 : BaseSIMDFPCvtTwoVector<0, U, {S,0}, opc, V64, V128,
4449 asm, ".4h", ".4s", []>;
4450 def v8i16 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,0}, opc, V128, V128,
4451 asm#"2", ".8h", ".4s", []>;
4452 def v2i32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
4453 asm, ".2s", ".2d", []>;
4454 def v4i32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
4455 asm#"2", ".4s", ".2d", []>;
4458 multiclass SIMDFPInexactCvtTwoVector<bit U, bit S, bits<5> opc, string asm,
4460 def v2f32 : BaseSIMDFPCvtTwoVector<0, U, {S,1}, opc, V64, V128,
4462 [(set (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn)))]>;
4463 def v4f32 : BaseSIMDFPCvtTwoVectorTied<1, U, {S,1}, opc, V128, V128,
4464 asm#"2", ".4s", ".2d", []>;
4466 def : Pat<(concat_vectors (v2f32 V64:$Rd), (OpNode (v2f64 V128:$Rn))),
4467 (!cast<Instruction>(NAME # "v4f32")
4468 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub), V128:$Rn)>;
4471 //----------------------------------------------------------------------------
4472 // AdvSIMD three register different-size vector instructions.
4473 //----------------------------------------------------------------------------
4475 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4476 class BaseSIMDDifferentThreeVector<bit U, bits<3> size, bits<4> opcode,
4477 RegisterOperand outtype, RegisterOperand intype1,
4478 RegisterOperand intype2, string asm,
4479 string outkind, string inkind1, string inkind2,
4481 : I<(outs outtype:$Rd), (ins intype1:$Rn, intype2:$Rm), asm,
4482 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
4483 "|" # outkind # "\t$Rd, $Rn, $Rm}", "", pattern>,
4489 let Inst{30} = size{0};
4491 let Inst{28-24} = 0b01110;
4492 let Inst{23-22} = size{2-1};
4494 let Inst{20-16} = Rm;
4495 let Inst{15-12} = opcode;
4496 let Inst{11-10} = 0b00;
4501 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
4502 class BaseSIMDDifferentThreeVectorTied<bit U, bits<3> size, bits<4> opcode,
4503 RegisterOperand outtype, RegisterOperand intype1,
4504 RegisterOperand intype2, string asm,
4505 string outkind, string inkind1, string inkind2,
4507 : I<(outs outtype:$dst), (ins outtype:$Rd, intype1:$Rn, intype2:$Rm), asm,
4508 "{\t$Rd" # outkind # ", $Rn" # inkind1 # ", $Rm" # inkind2 #
4509 "|" # outkind # "\t$Rd, $Rn, $Rm}", "$Rd = $dst", pattern>,
4515 let Inst{30} = size{0};
4517 let Inst{28-24} = 0b01110;
4518 let Inst{23-22} = size{2-1};
4520 let Inst{20-16} = Rm;
4521 let Inst{15-12} = opcode;
4522 let Inst{11-10} = 0b00;
4527 // FIXME: TableGen doesn't know how to deal with expanded types that also
4528 // change the element count (in this case, placing the results in
4529 // the high elements of the result register rather than the low
4530 // elements). Until that's fixed, we can't code-gen those.
4531 multiclass SIMDNarrowThreeVectorBHS<bit U, bits<4> opc, string asm,
4533 def v8i16_v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4535 asm, ".8b", ".8h", ".8h",
4536 [(set (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn), (v8i16 V128:$Rm)))]>;
4537 def v8i16_v16i8 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4539 asm#"2", ".16b", ".8h", ".8h",
4541 def v4i32_v4i16 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4543 asm, ".4h", ".4s", ".4s",
4544 [(set (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn), (v4i32 V128:$Rm)))]>;
4545 def v4i32_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4547 asm#"2", ".8h", ".4s", ".4s",
4549 def v2i64_v2i32 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4551 asm, ".2s", ".2d", ".2d",
4552 [(set (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn), (v2i64 V128:$Rm)))]>;
4553 def v2i64_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4555 asm#"2", ".4s", ".2d", ".2d",
4559 // Patterns for the '2' variants involve INSERT_SUBREG, which you can't put in
4560 // a version attached to an instruction.
4561 def : Pat<(concat_vectors (v8i8 V64:$Rd), (IntOp (v8i16 V128:$Rn),
4563 (!cast<Instruction>(NAME # "v8i16_v16i8")
4564 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4565 V128:$Rn, V128:$Rm)>;
4566 def : Pat<(concat_vectors (v4i16 V64:$Rd), (IntOp (v4i32 V128:$Rn),
4568 (!cast<Instruction>(NAME # "v4i32_v8i16")
4569 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4570 V128:$Rn, V128:$Rm)>;
4571 def : Pat<(concat_vectors (v2i32 V64:$Rd), (IntOp (v2i64 V128:$Rn),
4573 (!cast<Instruction>(NAME # "v2i64_v4i32")
4574 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
4575 V128:$Rn, V128:$Rm)>;
4578 multiclass SIMDDifferentThreeVectorBD<bit U, bits<4> opc, string asm,
4580 def v8i8 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4582 asm, ".8h", ".8b", ".8b",
4583 [(set (v8i16 V128:$Rd), (IntOp (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4584 def v16i8 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4586 asm#"2", ".8h", ".16b", ".16b", []>;
4587 def v1i64 : BaseSIMDDifferentThreeVector<U, 0b110, opc,
4589 asm, ".1q", ".1d", ".1d", []>;
4590 def v2i64 : BaseSIMDDifferentThreeVector<U, 0b111, opc,
4592 asm#"2", ".1q", ".2d", ".2d", []>;
4594 def : Pat<(v8i16 (IntOp (v8i8 (extract_high_v16i8 V128:$Rn)),
4595 (v8i8 (extract_high_v16i8 V128:$Rm)))),
4596 (!cast<Instruction>(NAME#"v16i8") V128:$Rn, V128:$Rm)>;
4599 multiclass SIMDLongThreeVectorHS<bit U, bits<4> opc, string asm,
4600 SDPatternOperator OpNode> {
4601 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4603 asm, ".4s", ".4h", ".4h",
4604 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4605 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4607 asm#"2", ".4s", ".8h", ".8h",
4608 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
4609 (extract_high_v8i16 V128:$Rm)))]>;
4610 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4612 asm, ".2d", ".2s", ".2s",
4613 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4614 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4616 asm#"2", ".2d", ".4s", ".4s",
4617 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
4618 (extract_high_v4i32 V128:$Rm)))]>;
4621 multiclass SIMDLongThreeVectorBHSabdl<bit U, bits<4> opc, string asm,
4622 SDPatternOperator OpNode = null_frag> {
4623 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4625 asm, ".8h", ".8b", ".8b",
4626 [(set (v8i16 V128:$Rd),
4627 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))))]>;
4628 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4630 asm#"2", ".8h", ".16b", ".16b",
4631 [(set (v8i16 V128:$Rd),
4632 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
4633 (extract_high_v16i8 V128:$Rm)))))]>;
4634 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4636 asm, ".4s", ".4h", ".4h",
4637 [(set (v4i32 V128:$Rd),
4638 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))))]>;
4639 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4641 asm#"2", ".4s", ".8h", ".8h",
4642 [(set (v4i32 V128:$Rd),
4643 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
4644 (extract_high_v8i16 V128:$Rm)))))]>;
4645 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4647 asm, ".2d", ".2s", ".2s",
4648 [(set (v2i64 V128:$Rd),
4649 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))))]>;
4650 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4652 asm#"2", ".2d", ".4s", ".4s",
4653 [(set (v2i64 V128:$Rd),
4654 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
4655 (extract_high_v4i32 V128:$Rm)))))]>;
4658 multiclass SIMDLongThreeVectorTiedBHSabal<bit U, bits<4> opc,
4660 SDPatternOperator OpNode> {
4661 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
4663 asm, ".8h", ".8b", ".8b",
4664 [(set (v8i16 V128:$dst),
4665 (add (v8i16 V128:$Rd),
4666 (zext (v8i8 (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm))))))]>;
4667 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4669 asm#"2", ".8h", ".16b", ".16b",
4670 [(set (v8i16 V128:$dst),
4671 (add (v8i16 V128:$Rd),
4672 (zext (v8i8 (OpNode (extract_high_v16i8 V128:$Rn),
4673 (extract_high_v16i8 V128:$Rm))))))]>;
4674 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4676 asm, ".4s", ".4h", ".4h",
4677 [(set (v4i32 V128:$dst),
4678 (add (v4i32 V128:$Rd),
4679 (zext (v4i16 (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm))))))]>;
4680 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4682 asm#"2", ".4s", ".8h", ".8h",
4683 [(set (v4i32 V128:$dst),
4684 (add (v4i32 V128:$Rd),
4685 (zext (v4i16 (OpNode (extract_high_v8i16 V128:$Rn),
4686 (extract_high_v8i16 V128:$Rm))))))]>;
4687 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4689 asm, ".2d", ".2s", ".2s",
4690 [(set (v2i64 V128:$dst),
4691 (add (v2i64 V128:$Rd),
4692 (zext (v2i32 (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm))))))]>;
4693 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4695 asm#"2", ".2d", ".4s", ".4s",
4696 [(set (v2i64 V128:$dst),
4697 (add (v2i64 V128:$Rd),
4698 (zext (v2i32 (OpNode (extract_high_v4i32 V128:$Rn),
4699 (extract_high_v4i32 V128:$Rm))))))]>;
4702 multiclass SIMDLongThreeVectorBHS<bit U, bits<4> opc, string asm,
4703 SDPatternOperator OpNode = null_frag> {
4704 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4706 asm, ".8h", ".8b", ".8b",
4707 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4708 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4710 asm#"2", ".8h", ".16b", ".16b",
4711 [(set (v8i16 V128:$Rd), (OpNode (extract_high_v16i8 V128:$Rn),
4712 (extract_high_v16i8 V128:$Rm)))]>;
4713 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4715 asm, ".4s", ".4h", ".4h",
4716 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4717 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4719 asm#"2", ".4s", ".8h", ".8h",
4720 [(set (v4i32 V128:$Rd), (OpNode (extract_high_v8i16 V128:$Rn),
4721 (extract_high_v8i16 V128:$Rm)))]>;
4722 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4724 asm, ".2d", ".2s", ".2s",
4725 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4726 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4728 asm#"2", ".2d", ".4s", ".4s",
4729 [(set (v2i64 V128:$Rd), (OpNode (extract_high_v4i32 V128:$Rn),
4730 (extract_high_v4i32 V128:$Rm)))]>;
4733 multiclass SIMDLongThreeVectorTiedBHS<bit U, bits<4> opc,
4735 SDPatternOperator OpNode> {
4736 def v8i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b000, opc,
4738 asm, ".8h", ".8b", ".8b",
4739 [(set (v8i16 V128:$dst),
4740 (OpNode (v8i16 V128:$Rd), (v8i8 V64:$Rn), (v8i8 V64:$Rm)))]>;
4741 def v16i8_v8i16 : BaseSIMDDifferentThreeVectorTied<U, 0b001, opc,
4743 asm#"2", ".8h", ".16b", ".16b",
4744 [(set (v8i16 V128:$dst),
4745 (OpNode (v8i16 V128:$Rd),
4746 (extract_high_v16i8 V128:$Rn),
4747 (extract_high_v16i8 V128:$Rm)))]>;
4748 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4750 asm, ".4s", ".4h", ".4h",
4751 [(set (v4i32 V128:$dst),
4752 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn), (v4i16 V64:$Rm)))]>;
4753 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4755 asm#"2", ".4s", ".8h", ".8h",
4756 [(set (v4i32 V128:$dst),
4757 (OpNode (v4i32 V128:$Rd),
4758 (extract_high_v8i16 V128:$Rn),
4759 (extract_high_v8i16 V128:$Rm)))]>;
4760 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4762 asm, ".2d", ".2s", ".2s",
4763 [(set (v2i64 V128:$dst),
4764 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn), (v2i32 V64:$Rm)))]>;
4765 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4767 asm#"2", ".2d", ".4s", ".4s",
4768 [(set (v2i64 V128:$dst),
4769 (OpNode (v2i64 V128:$Rd),
4770 (extract_high_v4i32 V128:$Rn),
4771 (extract_high_v4i32 V128:$Rm)))]>;
4774 multiclass SIMDLongThreeVectorSQDMLXTiedHS<bit U, bits<4> opc, string asm,
4775 SDPatternOperator Accum> {
4776 def v4i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b010, opc,
4778 asm, ".4s", ".4h", ".4h",
4779 [(set (v4i32 V128:$dst),
4780 (Accum (v4i32 V128:$Rd),
4781 (v4i32 (int_arm64_neon_sqdmull (v4i16 V64:$Rn),
4782 (v4i16 V64:$Rm)))))]>;
4783 def v8i16_v4i32 : BaseSIMDDifferentThreeVectorTied<U, 0b011, opc,
4785 asm#"2", ".4s", ".8h", ".8h",
4786 [(set (v4i32 V128:$dst),
4787 (Accum (v4i32 V128:$Rd),
4788 (v4i32 (int_arm64_neon_sqdmull (extract_high_v8i16 V128:$Rn),
4789 (extract_high_v8i16 V128:$Rm)))))]>;
4790 def v2i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b100, opc,
4792 asm, ".2d", ".2s", ".2s",
4793 [(set (v2i64 V128:$dst),
4794 (Accum (v2i64 V128:$Rd),
4795 (v2i64 (int_arm64_neon_sqdmull (v2i32 V64:$Rn),
4796 (v2i32 V64:$Rm)))))]>;
4797 def v4i32_v2i64 : BaseSIMDDifferentThreeVectorTied<U, 0b101, opc,
4799 asm#"2", ".2d", ".4s", ".4s",
4800 [(set (v2i64 V128:$dst),
4801 (Accum (v2i64 V128:$Rd),
4802 (v2i64 (int_arm64_neon_sqdmull (extract_high_v4i32 V128:$Rn),
4803 (extract_high_v4i32 V128:$Rm)))))]>;
4806 multiclass SIMDWideThreeVectorBHS<bit U, bits<4> opc, string asm,
4807 SDPatternOperator OpNode> {
4808 def v8i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b000, opc,
4810 asm, ".8h", ".8h", ".8b",
4811 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn), (v8i8 V64:$Rm)))]>;
4812 def v16i8_v8i16 : BaseSIMDDifferentThreeVector<U, 0b001, opc,
4814 asm#"2", ".8h", ".8h", ".16b",
4815 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
4816 (extract_high_v16i8 V128:$Rm)))]>;
4817 def v4i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b010, opc,
4819 asm, ".4s", ".4s", ".4h",
4820 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (v4i16 V64:$Rm)))]>;
4821 def v8i16_v4i32 : BaseSIMDDifferentThreeVector<U, 0b011, opc,
4823 asm#"2", ".4s", ".4s", ".8h",
4824 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
4825 (extract_high_v8i16 V128:$Rm)))]>;
4826 def v2i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b100, opc,
4828 asm, ".2d", ".2d", ".2s",
4829 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (v2i32 V64:$Rm)))]>;
4830 def v4i32_v2i64 : BaseSIMDDifferentThreeVector<U, 0b101, opc,
4832 asm#"2", ".2d", ".2d", ".4s",
4833 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
4834 (extract_high_v4i32 V128:$Rm)))]>;
4837 //----------------------------------------------------------------------------
4838 // AdvSIMD bitwise extract from vector
4839 //----------------------------------------------------------------------------
4841 class BaseSIMDBitwiseExtract<bit size, RegisterOperand regtype, ValueType vty,
4842 string asm, string kind>
4843 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm, i32imm:$imm), asm,
4844 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind # ", $imm" #
4845 "|" # kind # "\t$Rd, $Rn, $Rm, $imm}", "",
4846 [(set (vty regtype:$Rd),
4847 (ARM64ext regtype:$Rn, regtype:$Rm, (i32 imm:$imm)))]>,
4854 let Inst{30} = size;
4855 let Inst{29-21} = 0b101110000;
4856 let Inst{20-16} = Rm;
4858 let Inst{14-11} = imm;
4865 multiclass SIMDBitwiseExtract<string asm> {
4866 def v8i8 : BaseSIMDBitwiseExtract<0, V64, v8i8, asm, ".8b"> {
4869 def v16i8 : BaseSIMDBitwiseExtract<1, V128, v16i8, asm, ".16b">;
4872 //----------------------------------------------------------------------------
4873 // AdvSIMD zip vector
4874 //----------------------------------------------------------------------------
4876 class BaseSIMDZipVector<bits<3> size, bits<3> opc, RegisterOperand regtype,
4877 string asm, string kind, SDNode OpNode, ValueType valty>
4878 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
4879 "{\t$Rd" # kind # ", $Rn" # kind # ", $Rm" # kind #
4880 "|" # kind # "\t$Rd, $Rn, $Rm}", "",
4881 [(set (valty regtype:$Rd), (OpNode regtype:$Rn, regtype:$Rm))]>,
4887 let Inst{30} = size{0};
4888 let Inst{29-24} = 0b001110;
4889 let Inst{23-22} = size{2-1};
4891 let Inst{20-16} = Rm;
4893 let Inst{14-12} = opc;
4894 let Inst{11-10} = 0b10;
4899 multiclass SIMDZipVector<bits<3>opc, string asm,
4901 def v8i8 : BaseSIMDZipVector<0b000, opc, V64,
4902 asm, ".8b", OpNode, v8i8>;
4903 def v16i8 : BaseSIMDZipVector<0b001, opc, V128,
4904 asm, ".16b", OpNode, v16i8>;
4905 def v4i16 : BaseSIMDZipVector<0b010, opc, V64,
4906 asm, ".4h", OpNode, v4i16>;
4907 def v8i16 : BaseSIMDZipVector<0b011, opc, V128,
4908 asm, ".8h", OpNode, v8i16>;
4909 def v2i32 : BaseSIMDZipVector<0b100, opc, V64,
4910 asm, ".2s", OpNode, v2i32>;
4911 def v4i32 : BaseSIMDZipVector<0b101, opc, V128,
4912 asm, ".4s", OpNode, v4i32>;
4913 def v2i64 : BaseSIMDZipVector<0b111, opc, V128,
4914 asm, ".2d", OpNode, v2i64>;
4916 def : Pat<(v2f32 (OpNode V64:$Rn, V64:$Rm)),
4917 (!cast<Instruction>(NAME#"v2i32") V64:$Rn, V64:$Rm)>;
4918 def : Pat<(v4f32 (OpNode V128:$Rn, V128:$Rm)),
4919 (!cast<Instruction>(NAME#"v4i32") V128:$Rn, V128:$Rm)>;
4920 def : Pat<(v2f64 (OpNode V128:$Rn, V128:$Rm)),
4921 (!cast<Instruction>(NAME#"v2i64") V128:$Rn, V128:$Rm)>;
4924 //----------------------------------------------------------------------------
4925 // AdvSIMD three register scalar instructions
4926 //----------------------------------------------------------------------------
4928 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
4929 class BaseSIMDThreeScalar<bit U, bits<2> size, bits<5> opcode,
4930 RegisterClass regtype, string asm,
4932 : I<(outs regtype:$Rd), (ins regtype:$Rn, regtype:$Rm), asm,
4933 "\t$Rd, $Rn, $Rm", "", pattern>,
4938 let Inst{31-30} = 0b01;
4940 let Inst{28-24} = 0b11110;
4941 let Inst{23-22} = size;
4943 let Inst{20-16} = Rm;
4944 let Inst{15-11} = opcode;
4950 multiclass SIMDThreeScalarD<bit U, bits<5> opc, string asm,
4951 SDPatternOperator OpNode> {
4952 def v1i64 : BaseSIMDThreeScalar<U, 0b11, opc, FPR64, asm,
4953 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
4956 multiclass SIMDThreeScalarBHSD<bit U, bits<5> opc, string asm,
4957 SDPatternOperator OpNode> {
4958 def v1i64 : BaseSIMDThreeScalar<U, 0b11, opc, FPR64, asm,
4959 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn), (v1i64 FPR64:$Rm)))]>;
4960 def v1i32 : BaseSIMDThreeScalar<U, 0b10, opc, FPR32, asm, []>;
4961 def v1i16 : BaseSIMDThreeScalar<U, 0b01, opc, FPR16, asm, []>;
4962 def v1i8 : BaseSIMDThreeScalar<U, 0b00, opc, FPR8 , asm, []>;
4964 def : Pat<(i64 (OpNode (i64 FPR64:$Rn), (i64 FPR64:$Rm))),
4965 (!cast<Instruction>(NAME#"v1i64") FPR64:$Rn, FPR64:$Rm)>;
4966 def : Pat<(i32 (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm))),
4967 (!cast<Instruction>(NAME#"v1i32") FPR32:$Rn, FPR32:$Rm)>;
4970 multiclass SIMDThreeScalarHS<bit U, bits<5> opc, string asm,
4971 SDPatternOperator OpNode> {
4972 def v1i32 : BaseSIMDThreeScalar<U, 0b10, opc, FPR32, asm,
4973 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
4974 def v1i16 : BaseSIMDThreeScalar<U, 0b01, opc, FPR16, asm, []>;
4977 multiclass SIMDThreeScalarSD<bit U, bit S, bits<5> opc, string asm,
4978 SDPatternOperator OpNode = null_frag> {
4979 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
4980 def #NAME#64 : BaseSIMDThreeScalar<U, {S,1}, opc, FPR64, asm,
4981 [(set (f64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
4982 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0}, opc, FPR32, asm,
4983 [(set FPR32:$Rd, (OpNode FPR32:$Rn, FPR32:$Rm))]>;
4986 def : Pat<(v1f64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
4987 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
4990 multiclass SIMDThreeScalarFPCmp<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 (i64 FPR64:$Rd), (OpNode (f64 FPR64:$Rn), (f64 FPR64:$Rm)))]>;
4995 def #NAME#32 : BaseSIMDThreeScalar<U, {S,0}, opc, FPR32, asm,
4996 [(set (i32 FPR32:$Rd), (OpNode (f32 FPR32:$Rn), (f32 FPR32:$Rm)))]>;
4999 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn), (v1f64 FPR64:$Rm))),
5000 (!cast<Instruction>(NAME # "64") FPR64:$Rn, FPR64:$Rm)>;
5003 class BaseSIMDThreeScalarMixed<bit U, bits<2> size, bits<5> opcode,
5004 dag oops, dag iops, string asm, string cstr, list<dag> pat>
5005 : I<oops, iops, asm,
5006 "\t$Rd, $Rn, $Rm", cstr, pat>,
5011 let Inst{31-30} = 0b01;
5013 let Inst{28-24} = 0b11110;
5014 let Inst{23-22} = size;
5016 let Inst{20-16} = Rm;
5017 let Inst{15-11} = opcode;
5023 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5024 multiclass SIMDThreeScalarMixedHS<bit U, bits<5> opc, string asm,
5025 SDPatternOperator OpNode = null_frag> {
5026 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
5028 (ins FPR16:$Rn, FPR16:$Rm), asm, "", []>;
5029 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
5031 (ins FPR32:$Rn, FPR32:$Rm), asm, "",
5032 [(set (i64 FPR64:$Rd), (OpNode (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
5035 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5036 multiclass SIMDThreeScalarMixedTiedHS<bit U, bits<5> opc, string asm,
5037 SDPatternOperator OpNode = null_frag> {
5038 def i16 : BaseSIMDThreeScalarMixed<U, 0b01, opc,
5040 (ins FPR32:$Rd, FPR16:$Rn, FPR16:$Rm),
5041 asm, "$Rd = $dst", []>;
5042 def i32 : BaseSIMDThreeScalarMixed<U, 0b10, opc,
5044 (ins FPR64:$Rd, FPR32:$Rn, FPR32:$Rm),
5046 [(set (i64 FPR64:$dst),
5047 (OpNode (i64 FPR64:$Rd), (i32 FPR32:$Rn), (i32 FPR32:$Rm)))]>;
5050 //----------------------------------------------------------------------------
5051 // AdvSIMD two register scalar instructions
5052 //----------------------------------------------------------------------------
5054 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5055 class BaseSIMDTwoScalar<bit U, bits<2> size, bits<5> opcode,
5056 RegisterClass regtype, RegisterClass regtype2,
5057 string asm, list<dag> pat>
5058 : I<(outs regtype:$Rd), (ins regtype2:$Rn), asm,
5059 "\t$Rd, $Rn", "", pat>,
5063 let Inst{31-30} = 0b01;
5065 let Inst{28-24} = 0b11110;
5066 let Inst{23-22} = size;
5067 let Inst{21-17} = 0b10000;
5068 let Inst{16-12} = opcode;
5069 let Inst{11-10} = 0b10;
5074 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5075 class BaseSIMDTwoScalarTied<bit U, bits<2> size, bits<5> opcode,
5076 RegisterClass regtype, RegisterClass regtype2,
5077 string asm, list<dag> pat>
5078 : I<(outs regtype:$dst), (ins regtype:$Rd, regtype2:$Rn), asm,
5079 "\t$Rd, $Rn", "$Rd = $dst", pat>,
5083 let Inst{31-30} = 0b01;
5085 let Inst{28-24} = 0b11110;
5086 let Inst{23-22} = size;
5087 let Inst{21-17} = 0b10000;
5088 let Inst{16-12} = opcode;
5089 let Inst{11-10} = 0b10;
5095 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5096 class BaseSIMDCmpTwoScalar<bit U, bits<2> size, bits<5> opcode,
5097 RegisterClass regtype, string asm>
5098 : I<(outs regtype:$Rd), (ins regtype:$Rn), asm,
5099 "\t$Rd, $Rn, #0", "", []>,
5103 let Inst{31-30} = 0b01;
5105 let Inst{28-24} = 0b11110;
5106 let Inst{23-22} = size;
5107 let Inst{21-17} = 0b10000;
5108 let Inst{16-12} = opcode;
5109 let Inst{11-10} = 0b10;
5114 class SIMDInexactCvtTwoScalar<bits<5> opcode, string asm>
5115 : I<(outs FPR32:$Rd), (ins FPR64:$Rn), asm, "\t$Rd, $Rn", "",
5116 [(set (f32 FPR32:$Rd), (int_arm64_sisd_fcvtxn (f64 FPR64:$Rn)))]>,
5120 let Inst{31-17} = 0b011111100110000;
5121 let Inst{16-12} = opcode;
5122 let Inst{11-10} = 0b10;
5127 multiclass SIMDCmpTwoScalarD<bit U, bits<5> opc, string asm,
5128 SDPatternOperator OpNode> {
5129 def v1i64rz : BaseSIMDCmpTwoScalar<U, 0b11, opc, FPR64, asm>;
5131 def : Pat<(v1i64 (OpNode FPR64:$Rn)),
5132 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
5135 multiclass SIMDCmpTwoScalarSD<bit U, bit S, bits<5> opc, string asm,
5136 SDPatternOperator OpNode> {
5137 def v1i64rz : BaseSIMDCmpTwoScalar<U, {S,1}, opc, FPR64, asm>;
5138 def v1i32rz : BaseSIMDCmpTwoScalar<U, {S,0}, opc, FPR32, asm>;
5140 def : Pat<(v1i64 (OpNode (v1f64 FPR64:$Rn))),
5141 (!cast<Instruction>(NAME # v1i64rz) FPR64:$Rn)>;
5144 multiclass SIMDTwoScalarD<bit U, bits<5> opc, string asm,
5145 SDPatternOperator OpNode = null_frag> {
5146 def v1i64 : BaseSIMDTwoScalar<U, 0b11, opc, FPR64, FPR64, asm,
5147 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn)))]>;
5149 def : Pat<(i64 (OpNode (i64 FPR64:$Rn))),
5150 (!cast<Instruction>(NAME # "v1i64") FPR64:$Rn)>;
5153 multiclass SIMDTwoScalarSD<bit U, bit S, bits<5> opc, string asm> {
5154 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, opc, FPR64, FPR64, asm,[]>;
5155 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, opc, FPR32, FPR32, asm,[]>;
5158 multiclass SIMDTwoScalarCVTSD<bit U, bit S, bits<5> opc, string asm,
5159 SDPatternOperator OpNode> {
5160 def v1i64 : BaseSIMDTwoScalar<U, {S,1}, opc, FPR64, FPR64, asm,
5161 [(set FPR64:$Rd, (OpNode (f64 FPR64:$Rn)))]>;
5162 def v1i32 : BaseSIMDTwoScalar<U, {S,0}, opc, FPR32, FPR32, asm,
5163 [(set FPR32:$Rd, (OpNode (f32 FPR32:$Rn)))]>;
5166 multiclass SIMDTwoScalarBHSD<bit U, bits<5> opc, string asm,
5167 SDPatternOperator OpNode = null_frag> {
5168 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5169 def v1i64 : BaseSIMDTwoScalar<U, 0b11, opc, FPR64, FPR64, asm,
5170 [(set (i64 FPR64:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
5171 def v1i32 : BaseSIMDTwoScalar<U, 0b10, opc, FPR32, FPR32, asm,
5172 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
5173 def v1i16 : BaseSIMDTwoScalar<U, 0b01, opc, FPR16, FPR16, asm, []>;
5174 def v1i8 : BaseSIMDTwoScalar<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
5177 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn))),
5178 (!cast<Instruction>(NAME # v1i64) FPR64:$Rn)>;
5181 multiclass SIMDTwoScalarBHSDTied<bit U, bits<5> opc, string asm,
5183 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
5184 def v1i64 : BaseSIMDTwoScalarTied<U, 0b11, opc, FPR64, FPR64, asm,
5185 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn)))]>;
5186 def v1i32 : BaseSIMDTwoScalarTied<U, 0b10, opc, FPR32, FPR32, asm,
5187 [(set (i32 FPR32:$dst), (OpNode (i32 FPR32:$Rd), (i32 FPR32:$Rn)))]>;
5188 def v1i16 : BaseSIMDTwoScalarTied<U, 0b01, opc, FPR16, FPR16, asm, []>;
5189 def v1i8 : BaseSIMDTwoScalarTied<U, 0b00, opc, FPR8 , FPR8 , asm, []>;
5192 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn))),
5193 (!cast<Instruction>(NAME # v1i64) FPR64:$Rd, FPR64:$Rn)>;
5198 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5199 multiclass SIMDTwoScalarMixedBHS<bit U, bits<5> opc, string asm,
5200 SDPatternOperator OpNode = null_frag> {
5201 def v1i32 : BaseSIMDTwoScalar<U, 0b10, opc, FPR32, FPR64, asm,
5202 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn)))]>;
5203 def v1i16 : BaseSIMDTwoScalar<U, 0b01, opc, FPR16, FPR32, asm, []>;
5204 def v1i8 : BaseSIMDTwoScalar<U, 0b00, opc, FPR8 , FPR16, asm, []>;
5207 //----------------------------------------------------------------------------
5208 // AdvSIMD scalar pairwise instructions
5209 //----------------------------------------------------------------------------
5211 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5212 class BaseSIMDPairwiseScalar<bit U, bits<2> size, bits<5> opcode,
5213 RegisterOperand regtype, RegisterOperand vectype,
5214 string asm, string kind>
5215 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
5216 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", []>,
5220 let Inst{31-30} = 0b01;
5222 let Inst{28-24} = 0b11110;
5223 let Inst{23-22} = size;
5224 let Inst{21-17} = 0b11000;
5225 let Inst{16-12} = opcode;
5226 let Inst{11-10} = 0b10;
5231 multiclass SIMDPairwiseScalarD<bit U, bits<5> opc, string asm> {
5232 def v2i64p : BaseSIMDPairwiseScalar<U, 0b11, opc, FPR64Op, V128,
5236 multiclass SIMDPairwiseScalarSD<bit U, bit S, bits<5> opc, string asm> {
5237 def v2i32p : BaseSIMDPairwiseScalar<U, {S,0}, opc, FPR32Op, V64,
5239 def v2i64p : BaseSIMDPairwiseScalar<U, {S,1}, opc, FPR64Op, V128,
5243 //----------------------------------------------------------------------------
5244 // AdvSIMD across lanes instructions
5245 //----------------------------------------------------------------------------
5247 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5248 class BaseSIMDAcrossLanes<bit Q, bit U, bits<2> size, bits<5> opcode,
5249 RegisterClass regtype, RegisterOperand vectype,
5250 string asm, string kind, list<dag> pattern>
5251 : I<(outs regtype:$Rd), (ins vectype:$Rn), asm,
5252 "{\t$Rd, $Rn" # kind # "|" # kind # "\t$Rd, $Rn}", "", pattern>,
5259 let Inst{28-24} = 0b01110;
5260 let Inst{23-22} = size;
5261 let Inst{21-17} = 0b11000;
5262 let Inst{16-12} = opcode;
5263 let Inst{11-10} = 0b10;
5268 multiclass SIMDAcrossLanesBHS<bit U, bits<5> opcode,
5270 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR8, V64,
5272 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR8, V128,
5274 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR16, V64,
5276 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR16, V128,
5278 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR32, V128,
5282 multiclass SIMDAcrossLanesHSD<bit U, bits<5> opcode, string asm> {
5283 def v8i8v : BaseSIMDAcrossLanes<0, U, 0b00, opcode, FPR16, V64,
5285 def v16i8v : BaseSIMDAcrossLanes<1, U, 0b00, opcode, FPR16, V128,
5287 def v4i16v : BaseSIMDAcrossLanes<0, U, 0b01, opcode, FPR32, V64,
5289 def v8i16v : BaseSIMDAcrossLanes<1, U, 0b01, opcode, FPR32, V128,
5291 def v4i32v : BaseSIMDAcrossLanes<1, U, 0b10, opcode, FPR64, V128,
5295 multiclass SIMDAcrossLanesS<bits<5> opcode, bit sz1, string asm,
5297 def v4i32v : BaseSIMDAcrossLanes<1, 1, {sz1, 0}, opcode, FPR32, V128,
5299 [(set FPR32:$Rd, (intOp (v4f32 V128:$Rn)))]>;
5302 //----------------------------------------------------------------------------
5303 // AdvSIMD INS/DUP instructions
5304 //----------------------------------------------------------------------------
5306 // FIXME: There has got to be a better way to factor these. ugh.
5308 class BaseSIMDInsDup<bit Q, bit op, dag outs, dag ins, string asm,
5309 string operands, string constraints, list<dag> pattern>
5310 : I<outs, ins, asm, operands, constraints, pattern>,
5317 let Inst{28-21} = 0b01110000;
5324 class SIMDDupFromMain<bit Q, bits<5> imm5, string size, ValueType vectype,
5325 RegisterOperand vecreg, RegisterClass regtype>
5326 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins regtype:$Rn), "dup",
5327 "{\t$Rd" # size # ", $Rn" #
5328 "|" # size # "\t$Rd, $Rn}", "",
5329 [(set (vectype vecreg:$Rd), (ARM64dup regtype:$Rn))]> {
5330 let Inst{20-16} = imm5;
5331 let Inst{14-11} = 0b0001;
5334 class SIMDDupFromElement<bit Q, string dstkind, string srckind,
5335 ValueType vectype, ValueType insreg,
5336 RegisterOperand vecreg, Operand idxtype,
5337 ValueType elttype, SDNode OpNode>
5338 : BaseSIMDInsDup<Q, 0, (outs vecreg:$Rd), (ins V128:$Rn, idxtype:$idx), "dup",
5339 "{\t$Rd" # dstkind # ", $Rn" # srckind # "$idx" #
5340 "|" # dstkind # "\t$Rd, $Rn$idx}", "",
5341 [(set (vectype vecreg:$Rd),
5342 (OpNode (insreg V128:$Rn), idxtype:$idx))]> {
5343 let Inst{14-11} = 0b0000;
5346 class SIMDDup64FromElement
5347 : SIMDDupFromElement<1, ".2d", ".d", v2i64, v2i64, V128,
5348 VectorIndexD, i64, ARM64duplane64> {
5351 let Inst{19-16} = 0b1000;
5354 class SIMDDup32FromElement<bit Q, string size, ValueType vectype,
5355 RegisterOperand vecreg>
5356 : SIMDDupFromElement<Q, size, ".s", vectype, v4i32, vecreg,
5357 VectorIndexS, i64, ARM64duplane32> {
5359 let Inst{20-19} = idx;
5360 let Inst{18-16} = 0b100;
5363 class SIMDDup16FromElement<bit Q, string size, ValueType vectype,
5364 RegisterOperand vecreg>
5365 : SIMDDupFromElement<Q, size, ".h", vectype, v8i16, vecreg,
5366 VectorIndexH, i64, ARM64duplane16> {
5368 let Inst{20-18} = idx;
5369 let Inst{17-16} = 0b10;
5372 class SIMDDup8FromElement<bit Q, string size, ValueType vectype,
5373 RegisterOperand vecreg>
5374 : SIMDDupFromElement<Q, size, ".b", vectype, v16i8, vecreg,
5375 VectorIndexB, i64, ARM64duplane8> {
5377 let Inst{20-17} = idx;
5381 class BaseSIMDMov<bit Q, string size, bits<4> imm4, RegisterClass regtype,
5382 Operand idxtype, string asm, list<dag> pattern>
5383 : BaseSIMDInsDup<Q, 0, (outs regtype:$Rd), (ins V128:$Rn, idxtype:$idx), asm,
5384 "{\t$Rd, $Rn" # size # "$idx" #
5385 "|" # size # "\t$Rd, $Rn$idx}", "", pattern> {
5386 let Inst{14-11} = imm4;
5389 class SIMDSMov<bit Q, string size, RegisterClass regtype,
5391 : BaseSIMDMov<Q, size, 0b0101, regtype, idxtype, "smov", []>;
5392 class SIMDUMov<bit Q, string size, ValueType vectype, RegisterClass regtype,
5394 : BaseSIMDMov<Q, size, 0b0111, regtype, idxtype, "umov",
5395 [(set regtype:$Rd, (vector_extract (vectype V128:$Rn), idxtype:$idx))]>;
5397 class SIMDMovAlias<string asm, string size, Instruction inst,
5398 RegisterClass regtype, Operand idxtype>
5399 : InstAlias<asm#"{\t$dst, $src"#size#"$idx" #
5400 "|" # size # "\t$dst, $src$idx}",
5401 (inst regtype:$dst, V128:$src, idxtype:$idx)>;
5404 def vi8to32 : SIMDSMov<0, ".b", GPR32, VectorIndexB> {
5406 let Inst{20-17} = idx;
5409 def vi8to64 : SIMDSMov<1, ".b", GPR64, VectorIndexB> {
5411 let Inst{20-17} = idx;
5414 def vi16to32 : SIMDSMov<0, ".h", GPR32, VectorIndexH> {
5416 let Inst{20-18} = idx;
5417 let Inst{17-16} = 0b10;
5419 def vi16to64 : SIMDSMov<1, ".h", GPR64, VectorIndexH> {
5421 let Inst{20-18} = idx;
5422 let Inst{17-16} = 0b10;
5424 def vi32to64 : SIMDSMov<1, ".s", GPR64, VectorIndexS> {
5426 let Inst{20-19} = idx;
5427 let Inst{18-16} = 0b100;
5432 def vi8 : SIMDUMov<0, ".b", v16i8, GPR32, VectorIndexB> {
5434 let Inst{20-17} = idx;
5437 def vi16 : SIMDUMov<0, ".h", v8i16, GPR32, VectorIndexH> {
5439 let Inst{20-18} = idx;
5440 let Inst{17-16} = 0b10;
5442 def vi32 : SIMDUMov<0, ".s", v4i32, GPR32, VectorIndexS> {
5444 let Inst{20-19} = idx;
5445 let Inst{18-16} = 0b100;
5447 def vi64 : SIMDUMov<1, ".d", v2i64, GPR64, VectorIndexD> {
5450 let Inst{19-16} = 0b1000;
5452 def : SIMDMovAlias<"mov", ".s",
5453 !cast<Instruction>(NAME#"vi32"),
5454 GPR32, VectorIndexS>;
5455 def : SIMDMovAlias<"mov", ".d",
5456 !cast<Instruction>(NAME#"vi64"),
5457 GPR64, VectorIndexD>;
5460 class SIMDInsFromMain<string size, ValueType vectype,
5461 RegisterClass regtype, Operand idxtype>
5462 : BaseSIMDInsDup<1, 0, (outs V128:$dst),
5463 (ins V128:$Rd, idxtype:$idx, regtype:$Rn), "ins",
5464 "{\t$Rd" # size # "$idx, $Rn" #
5465 "|" # size # "\t$Rd$idx, $Rn}",
5468 (vector_insert (vectype V128:$Rd), regtype:$Rn, idxtype:$idx))]> {
5469 let Inst{14-11} = 0b0011;
5472 class SIMDInsFromElement<string size, ValueType vectype,
5473 ValueType elttype, Operand idxtype>
5474 : BaseSIMDInsDup<1, 1, (outs V128:$dst),
5475 (ins V128:$Rd, idxtype:$idx, V128:$Rn, idxtype:$idx2), "ins",
5476 "{\t$Rd" # size # "$idx, $Rn" # size # "$idx2" #
5477 "|" # size # "\t$Rd$idx, $Rn$idx2}",
5482 (elttype (vector_extract (vectype V128:$Rn), idxtype:$idx2)),
5485 class SIMDInsMainMovAlias<string size, Instruction inst,
5486 RegisterClass regtype, Operand idxtype>
5487 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" #
5488 "|" # size #"\t$dst$idx, $src}",
5489 (inst V128:$dst, idxtype:$idx, regtype:$src)>;
5490 class SIMDInsElementMovAlias<string size, Instruction inst,
5492 : InstAlias<"mov" # "{\t$dst" # size # "$idx, $src" # size # "$idx2" #
5493 # "|" # size #" $dst$idx, $src$idx2}",
5494 (inst V128:$dst, idxtype:$idx, V128:$src, idxtype:$idx2)>;
5497 multiclass SIMDIns {
5498 def vi8gpr : SIMDInsFromMain<".b", v16i8, GPR32, VectorIndexB> {
5500 let Inst{20-17} = idx;
5503 def vi16gpr : SIMDInsFromMain<".h", v8i16, GPR32, VectorIndexH> {
5505 let Inst{20-18} = idx;
5506 let Inst{17-16} = 0b10;
5508 def vi32gpr : SIMDInsFromMain<".s", v4i32, GPR32, VectorIndexS> {
5510 let Inst{20-19} = idx;
5511 let Inst{18-16} = 0b100;
5513 def vi64gpr : SIMDInsFromMain<".d", v2i64, GPR64, VectorIndexD> {
5516 let Inst{19-16} = 0b1000;
5519 def vi8lane : SIMDInsFromElement<".b", v16i8, i32, VectorIndexB> {
5522 let Inst{20-17} = idx;
5524 let Inst{14-11} = idx2;
5526 def vi16lane : SIMDInsFromElement<".h", v8i16, i32, VectorIndexH> {
5529 let Inst{20-18} = idx;
5530 let Inst{17-16} = 0b10;
5531 let Inst{14-12} = idx2;
5534 def vi32lane : SIMDInsFromElement<".s", v4i32, i32, VectorIndexS> {
5537 let Inst{20-19} = idx;
5538 let Inst{18-16} = 0b100;
5539 let Inst{14-13} = idx2;
5540 let Inst{12-11} = 0;
5542 def vi64lane : SIMDInsFromElement<".d", v2i64, i64, VectorIndexD> {
5546 let Inst{19-16} = 0b1000;
5547 let Inst{14} = idx2;
5548 let Inst{13-11} = 0;
5551 // For all forms of the INS instruction, the "mov" mnemonic is the
5552 // preferred alias. Why they didn't just call the instruction "mov" in
5553 // the first place is a very good question indeed...
5554 def : SIMDInsMainMovAlias<".b", !cast<Instruction>(NAME#"vi8gpr"),
5555 GPR32, VectorIndexB>;
5556 def : SIMDInsMainMovAlias<".h", !cast<Instruction>(NAME#"vi16gpr"),
5557 GPR32, VectorIndexH>;
5558 def : SIMDInsMainMovAlias<".s", !cast<Instruction>(NAME#"vi32gpr"),
5559 GPR32, VectorIndexS>;
5560 def : SIMDInsMainMovAlias<".d", !cast<Instruction>(NAME#"vi64gpr"),
5561 GPR64, VectorIndexD>;
5563 def : SIMDInsElementMovAlias<".b", !cast<Instruction>(NAME#"vi8lane"),
5565 def : SIMDInsElementMovAlias<".h", !cast<Instruction>(NAME#"vi16lane"),
5567 def : SIMDInsElementMovAlias<".s", !cast<Instruction>(NAME#"vi32lane"),
5569 def : SIMDInsElementMovAlias<".d", !cast<Instruction>(NAME#"vi64lane"),
5573 //----------------------------------------------------------------------------
5575 //----------------------------------------------------------------------------
5577 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
5578 class BaseSIMDTableLookup<bit Q, bits<2> len, bit op, RegisterOperand vectype,
5579 RegisterOperand listtype, string asm, string kind>
5580 : I<(outs vectype:$Vd), (ins listtype:$Vn, vectype:$Vm), asm,
5581 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "", []>,
5588 let Inst{29-21} = 0b001110000;
5589 let Inst{20-16} = Vm;
5591 let Inst{14-13} = len;
5593 let Inst{11-10} = 0b00;
5598 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
5599 class BaseSIMDTableLookupTied<bit Q, bits<2> len, bit op, RegisterOperand vectype,
5600 RegisterOperand listtype, string asm, string kind>
5601 : I<(outs vectype:$dst), (ins vectype:$Vd, listtype:$Vn, vectype:$Vm), asm,
5602 "\t$Vd" # kind # ", $Vn, $Vm" # kind, "$Vd = $dst", []>,
5609 let Inst{29-21} = 0b001110000;
5610 let Inst{20-16} = Vm;
5612 let Inst{14-13} = len;
5614 let Inst{11-10} = 0b00;
5619 class SIMDTableLookupAlias<string asm, Instruction inst,
5620 RegisterOperand vectype, RegisterOperand listtype>
5621 : InstAlias<!strconcat(asm, "\t$dst, $lst, $index"),
5622 (inst vectype:$dst, listtype:$lst, vectype:$index), 0>;
5624 multiclass SIMDTableLookup<bit op, string asm> {
5625 def v8i8One : BaseSIMDTableLookup<0, 0b00, op, V64, VecListOne16b,
5627 def v8i8Two : BaseSIMDTableLookup<0, 0b01, op, V64, VecListTwo16b,
5629 def v8i8Three : BaseSIMDTableLookup<0, 0b10, op, V64, VecListThree16b,
5631 def v8i8Four : BaseSIMDTableLookup<0, 0b11, op, V64, VecListFour16b,
5633 def v16i8One : BaseSIMDTableLookup<1, 0b00, op, V128, VecListOne16b,
5635 def v16i8Two : BaseSIMDTableLookup<1, 0b01, op, V128, VecListTwo16b,
5637 def v16i8Three: BaseSIMDTableLookup<1, 0b10, op, V128, VecListThree16b,
5639 def v16i8Four : BaseSIMDTableLookup<1, 0b11, op, V128, VecListFour16b,
5642 def : SIMDTableLookupAlias<asm # ".8b",
5643 !cast<Instruction>(NAME#"v8i8One"),
5644 V64, VecListOne128>;
5645 def : SIMDTableLookupAlias<asm # ".8b",
5646 !cast<Instruction>(NAME#"v8i8Two"),
5647 V64, VecListTwo128>;
5648 def : SIMDTableLookupAlias<asm # ".8b",
5649 !cast<Instruction>(NAME#"v8i8Three"),
5650 V64, VecListThree128>;
5651 def : SIMDTableLookupAlias<asm # ".8b",
5652 !cast<Instruction>(NAME#"v8i8Four"),
5653 V64, VecListFour128>;
5654 def : SIMDTableLookupAlias<asm # ".16b",
5655 !cast<Instruction>(NAME#"v16i8One"),
5656 V128, VecListOne128>;
5657 def : SIMDTableLookupAlias<asm # ".16b",
5658 !cast<Instruction>(NAME#"v16i8Two"),
5659 V128, VecListTwo128>;
5660 def : SIMDTableLookupAlias<asm # ".16b",
5661 !cast<Instruction>(NAME#"v16i8Three"),
5662 V128, VecListThree128>;
5663 def : SIMDTableLookupAlias<asm # ".16b",
5664 !cast<Instruction>(NAME#"v16i8Four"),
5665 V128, VecListFour128>;
5668 multiclass SIMDTableLookupTied<bit op, string asm> {
5669 def v8i8One : BaseSIMDTableLookupTied<0, 0b00, op, V64, VecListOne16b,
5671 def v8i8Two : BaseSIMDTableLookupTied<0, 0b01, op, V64, VecListTwo16b,
5673 def v8i8Three : BaseSIMDTableLookupTied<0, 0b10, op, V64, VecListThree16b,
5675 def v8i8Four : BaseSIMDTableLookupTied<0, 0b11, op, V64, VecListFour16b,
5677 def v16i8One : BaseSIMDTableLookupTied<1, 0b00, op, V128, VecListOne16b,
5679 def v16i8Two : BaseSIMDTableLookupTied<1, 0b01, op, V128, VecListTwo16b,
5681 def v16i8Three: BaseSIMDTableLookupTied<1, 0b10, op, V128, VecListThree16b,
5683 def v16i8Four : BaseSIMDTableLookupTied<1, 0b11, op, V128, VecListFour16b,
5686 def : SIMDTableLookupAlias<asm # ".8b",
5687 !cast<Instruction>(NAME#"v8i8One"),
5688 V64, VecListOne128>;
5689 def : SIMDTableLookupAlias<asm # ".8b",
5690 !cast<Instruction>(NAME#"v8i8Two"),
5691 V64, VecListTwo128>;
5692 def : SIMDTableLookupAlias<asm # ".8b",
5693 !cast<Instruction>(NAME#"v8i8Three"),
5694 V64, VecListThree128>;
5695 def : SIMDTableLookupAlias<asm # ".8b",
5696 !cast<Instruction>(NAME#"v8i8Four"),
5697 V64, VecListFour128>;
5698 def : SIMDTableLookupAlias<asm # ".16b",
5699 !cast<Instruction>(NAME#"v16i8One"),
5700 V128, VecListOne128>;
5701 def : SIMDTableLookupAlias<asm # ".16b",
5702 !cast<Instruction>(NAME#"v16i8Two"),
5703 V128, VecListTwo128>;
5704 def : SIMDTableLookupAlias<asm # ".16b",
5705 !cast<Instruction>(NAME#"v16i8Three"),
5706 V128, VecListThree128>;
5707 def : SIMDTableLookupAlias<asm # ".16b",
5708 !cast<Instruction>(NAME#"v16i8Four"),
5709 V128, VecListFour128>;
5713 //----------------------------------------------------------------------------
5714 // AdvSIMD scalar CPY
5715 //----------------------------------------------------------------------------
5716 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5717 class BaseSIMDScalarCPY<RegisterClass regtype, RegisterOperand vectype,
5718 string kind, Operand idxtype>
5719 : I<(outs regtype:$dst), (ins vectype:$src, idxtype:$idx), "mov",
5720 "{\t$dst, $src" # kind # "$idx" #
5721 "|\t$dst, $src$idx}", "", []>,
5725 let Inst{31-21} = 0b01011110000;
5726 let Inst{15-10} = 0b000001;
5727 let Inst{9-5} = src;
5728 let Inst{4-0} = dst;
5731 class SIMDScalarCPYAlias<string asm, string size, Instruction inst,
5732 RegisterClass regtype, RegisterOperand vectype, Operand idxtype>
5733 : InstAlias<asm # "{\t$dst, $src" # size # "$index" #
5734 # "|\t$dst, $src$index}",
5735 (inst regtype:$dst, vectype:$src, idxtype:$index)>;
5738 multiclass SIMDScalarCPY<string asm> {
5739 def i8 : BaseSIMDScalarCPY<FPR8, V128, ".b", VectorIndexB> {
5741 let Inst{20-17} = idx;
5744 def i16 : BaseSIMDScalarCPY<FPR16, V128, ".h", VectorIndexH> {
5746 let Inst{20-18} = idx;
5747 let Inst{17-16} = 0b10;
5749 def i32 : BaseSIMDScalarCPY<FPR32, V128, ".s", VectorIndexS> {
5751 let Inst{20-19} = idx;
5752 let Inst{18-16} = 0b100;
5754 def i64 : BaseSIMDScalarCPY<FPR64, V128, ".d", VectorIndexD> {
5757 let Inst{19-16} = 0b1000;
5760 // 'DUP' mnemonic aliases.
5761 def : SIMDScalarCPYAlias<"dup", ".b",
5762 !cast<Instruction>(NAME#"i8"),
5763 FPR8, V128, VectorIndexB>;
5764 def : SIMDScalarCPYAlias<"dup", ".h",
5765 !cast<Instruction>(NAME#"i16"),
5766 FPR16, V128, VectorIndexH>;
5767 def : SIMDScalarCPYAlias<"dup", ".s",
5768 !cast<Instruction>(NAME#"i32"),
5769 FPR32, V128, VectorIndexS>;
5770 def : SIMDScalarCPYAlias<"dup", ".d",
5771 !cast<Instruction>(NAME#"i64"),
5772 FPR64, V128, VectorIndexD>;
5775 //----------------------------------------------------------------------------
5776 // AdvSIMD modified immediate instructions
5777 //----------------------------------------------------------------------------
5779 class BaseSIMDModifiedImm<bit Q, bit op, dag oops, dag iops,
5780 string asm, string op_string,
5781 string cstr, list<dag> pattern>
5782 : I<oops, iops, asm, op_string, cstr, pattern>,
5789 let Inst{28-19} = 0b0111100000;
5790 let Inst{18-16} = imm8{7-5};
5791 let Inst{11-10} = 0b01;
5792 let Inst{9-5} = imm8{4-0};
5796 class BaseSIMDModifiedImmVector<bit Q, bit op, RegisterOperand vectype,
5797 Operand immtype, dag opt_shift_iop,
5798 string opt_shift, string asm, string kind,
5800 : BaseSIMDModifiedImm<Q, op, (outs vectype:$Rd),
5801 !con((ins immtype:$imm8), opt_shift_iop), asm,
5802 "{\t$Rd" # kind # ", $imm8" # opt_shift #
5803 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
5805 let DecoderMethod = "DecodeModImmInstruction";
5808 class BaseSIMDModifiedImmVectorTied<bit Q, bit op, RegisterOperand vectype,
5809 Operand immtype, dag opt_shift_iop,
5810 string opt_shift, string asm, string kind,
5812 : BaseSIMDModifiedImm<Q, op, (outs vectype:$dst),
5813 !con((ins vectype:$Rd, immtype:$imm8), opt_shift_iop),
5814 asm, "{\t$Rd" # kind # ", $imm8" # opt_shift #
5815 "|" # kind # "\t$Rd, $imm8" # opt_shift # "}",
5816 "$Rd = $dst", pattern> {
5817 let DecoderMethod = "DecodeModImmTiedInstruction";
5820 class BaseSIMDModifiedImmVectorShift<bit Q, bit op, bits<2> b15_b12,
5821 RegisterOperand vectype, string asm,
5822 string kind, list<dag> pattern>
5823 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5824 (ins logical_vec_shift:$shift),
5825 "$shift", asm, kind, pattern> {
5827 let Inst{15} = b15_b12{1};
5828 let Inst{14-13} = shift;
5829 let Inst{12} = b15_b12{0};
5832 class BaseSIMDModifiedImmVectorShiftTied<bit Q, bit op, bits<2> b15_b12,
5833 RegisterOperand vectype, string asm,
5834 string kind, list<dag> pattern>
5835 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
5836 (ins logical_vec_shift:$shift),
5837 "$shift", asm, kind, pattern> {
5839 let Inst{15} = b15_b12{1};
5840 let Inst{14-13} = shift;
5841 let Inst{12} = b15_b12{0};
5845 class BaseSIMDModifiedImmVectorShiftHalf<bit Q, bit op, bits<2> b15_b12,
5846 RegisterOperand vectype, string asm,
5847 string kind, list<dag> pattern>
5848 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5849 (ins logical_vec_hw_shift:$shift),
5850 "$shift", asm, kind, pattern> {
5852 let Inst{15} = b15_b12{1};
5854 let Inst{13} = shift{0};
5855 let Inst{12} = b15_b12{0};
5858 class BaseSIMDModifiedImmVectorShiftHalfTied<bit Q, bit op, bits<2> b15_b12,
5859 RegisterOperand vectype, string asm,
5860 string kind, list<dag> pattern>
5861 : BaseSIMDModifiedImmVectorTied<Q, op, vectype, imm0_255,
5862 (ins logical_vec_hw_shift:$shift),
5863 "$shift", asm, kind, pattern> {
5865 let Inst{15} = b15_b12{1};
5867 let Inst{13} = shift{0};
5868 let Inst{12} = b15_b12{0};
5871 multiclass SIMDModifiedImmVectorShift<bit op, bits<2> hw_cmode, bits<2> w_cmode,
5873 def v4i16 : BaseSIMDModifiedImmVectorShiftHalf<0, op, hw_cmode, V64,
5875 def v8i16 : BaseSIMDModifiedImmVectorShiftHalf<1, op, hw_cmode, V128,
5878 def v2i32 : BaseSIMDModifiedImmVectorShift<0, op, w_cmode, V64,
5880 def v4i32 : BaseSIMDModifiedImmVectorShift<1, op, w_cmode, V128,
5884 multiclass SIMDModifiedImmVectorShiftTied<bit op, bits<2> hw_cmode,
5885 bits<2> w_cmode, string asm,
5887 def v4i16 : BaseSIMDModifiedImmVectorShiftHalfTied<0, op, hw_cmode, V64,
5889 [(set (v4i16 V64:$dst), (OpNode V64:$Rd,
5891 (i32 imm:$shift)))]>;
5892 def v8i16 : BaseSIMDModifiedImmVectorShiftHalfTied<1, op, hw_cmode, V128,
5894 [(set (v8i16 V128:$dst), (OpNode V128:$Rd,
5896 (i32 imm:$shift)))]>;
5898 def v2i32 : BaseSIMDModifiedImmVectorShiftTied<0, op, w_cmode, V64,
5900 [(set (v2i32 V64:$dst), (OpNode V64:$Rd,
5902 (i32 imm:$shift)))]>;
5903 def v4i32 : BaseSIMDModifiedImmVectorShiftTied<1, op, w_cmode, V128,
5905 [(set (v4i32 V128:$dst), (OpNode V128:$Rd,
5907 (i32 imm:$shift)))]>;
5910 class SIMDModifiedImmMoveMSL<bit Q, bit op, bits<4> cmode,
5911 RegisterOperand vectype, string asm,
5912 string kind, list<dag> pattern>
5913 : BaseSIMDModifiedImmVector<Q, op, vectype, imm0_255,
5914 (ins move_vec_shift:$shift),
5915 "$shift", asm, kind, pattern> {
5917 let Inst{15-13} = cmode{3-1};
5918 let Inst{12} = shift;
5921 class SIMDModifiedImmVectorNoShift<bit Q, bit op, bits<4> cmode,
5922 RegisterOperand vectype,
5923 Operand imm_type, string asm,
5924 string kind, list<dag> pattern>
5925 : BaseSIMDModifiedImmVector<Q, op, vectype, imm_type, (ins), "",
5926 asm, kind, pattern> {
5927 let Inst{15-12} = cmode;
5930 class SIMDModifiedImmScalarNoShift<bit Q, bit op, bits<4> cmode, string asm,
5932 : BaseSIMDModifiedImm<Q, op, (outs FPR64:$Rd), (ins simdimmtype10:$imm8), asm,
5933 "\t$Rd, $imm8", "", pattern> {
5934 let Inst{15-12} = cmode;
5935 let DecoderMethod = "DecodeModImmInstruction";
5938 //----------------------------------------------------------------------------
5939 // AdvSIMD indexed element
5940 //----------------------------------------------------------------------------
5942 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5943 class BaseSIMDIndexed<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
5944 RegisterOperand dst_reg, RegisterOperand lhs_reg,
5945 RegisterOperand rhs_reg, Operand vec_idx, string asm,
5946 string apple_kind, string dst_kind, string lhs_kind,
5947 string rhs_kind, list<dag> pattern>
5948 : I<(outs dst_reg:$Rd), (ins lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx),
5950 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
5951 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "", pattern>,
5960 let Inst{28} = Scalar;
5961 let Inst{27-24} = 0b1111;
5962 let Inst{23-22} = size;
5963 // Bit 21 must be set by the derived class.
5964 let Inst{20-16} = Rm;
5965 let Inst{15-12} = opc;
5966 // Bit 11 must be set by the derived class.
5972 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
5973 class BaseSIMDIndexedTied<bit Q, bit U, bit Scalar, bits<2> size, bits<4> opc,
5974 RegisterOperand dst_reg, RegisterOperand lhs_reg,
5975 RegisterOperand rhs_reg, Operand vec_idx, string asm,
5976 string apple_kind, string dst_kind, string lhs_kind,
5977 string rhs_kind, list<dag> pattern>
5978 : I<(outs dst_reg:$dst),
5979 (ins dst_reg:$Rd, lhs_reg:$Rn, rhs_reg:$Rm, vec_idx:$idx), asm,
5980 "{\t$Rd" # dst_kind # ", $Rn" # lhs_kind # ", $Rm" # rhs_kind # "$idx" #
5981 "|" # apple_kind # "\t$Rd, $Rn, $Rm$idx}", "$Rd = $dst", pattern>,
5990 let Inst{28} = Scalar;
5991 let Inst{27-24} = 0b1111;
5992 let Inst{23-22} = size;
5993 // Bit 21 must be set by the derived class.
5994 let Inst{20-16} = Rm;
5995 let Inst{15-12} = opc;
5996 // Bit 11 must be set by the derived class.
6002 multiclass SIMDFPIndexedSD<bit U, bits<4> opc, string asm,
6003 SDPatternOperator OpNode> {
6004 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6007 asm, ".2s", ".2s", ".2s", ".s",
6008 [(set (v2f32 V64:$Rd),
6009 (OpNode (v2f32 V64:$Rn),
6010 (v2f32 (ARM64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
6012 let Inst{11} = idx{1};
6013 let Inst{21} = idx{0};
6016 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6019 asm, ".4s", ".4s", ".4s", ".s",
6020 [(set (v4f32 V128:$Rd),
6021 (OpNode (v4f32 V128:$Rn),
6022 (v4f32 (ARM64duplane32 (v4f32 V128:$Rm), VectorIndexS:$idx))))]> {
6024 let Inst{11} = idx{1};
6025 let Inst{21} = idx{0};
6028 def v2i64_indexed : BaseSIMDIndexed<1, U, 0, 0b11, opc,
6031 asm, ".2d", ".2d", ".2d", ".d",
6032 [(set (v2f64 V128:$Rd),
6033 (OpNode (v2f64 V128:$Rn),
6034 (v2f64 (ARM64duplane64 (v2f64 V128:$Rm), VectorIndexD:$idx))))]> {
6036 let Inst{11} = idx{0};
6040 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6041 FPR32Op, FPR32Op, V128, VectorIndexS,
6042 asm, ".s", "", "", ".s",
6043 [(set (f32 FPR32Op:$Rd),
6044 (OpNode (f32 FPR32Op:$Rn),
6045 (f32 (vector_extract (v4f32 V128:$Rm),
6046 VectorIndexS:$idx))))]> {
6048 let Inst{11} = idx{1};
6049 let Inst{21} = idx{0};
6052 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b11, opc,
6053 FPR64Op, FPR64Op, V128, VectorIndexD,
6054 asm, ".d", "", "", ".d",
6055 [(set (f64 FPR64Op:$Rd),
6056 (OpNode (f64 FPR64Op:$Rn),
6057 (f64 (vector_extract (v2f64 V128:$Rm),
6058 VectorIndexD:$idx))))]> {
6060 let Inst{11} = idx{0};
6065 multiclass SIMDFPIndexedSDTiedPatterns<string INST, SDPatternOperator OpNode> {
6066 // 2 variants for the .2s version: DUPLANE from 128-bit and DUP scalar.
6067 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6068 (ARM64duplane32 (v4f32 V128:$Rm),
6069 VectorIndexS:$idx))),
6070 (!cast<Instruction>(INST # v2i32_indexed)
6071 V64:$Rd, V64:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6072 def : Pat<(v2f32 (OpNode (v2f32 V64:$Rd), (v2f32 V64:$Rn),
6073 (ARM64dup (f32 FPR32Op:$Rm)))),
6074 (!cast<Instruction>(INST # "v2i32_indexed") V64:$Rd, V64:$Rn,
6075 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6078 // 2 variants for the .4s version: DUPLANE from 128-bit and DUP scalar.
6079 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6080 (ARM64duplane32 (v4f32 V128:$Rm),
6081 VectorIndexS:$idx))),
6082 (!cast<Instruction>(INST # "v4i32_indexed")
6083 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6084 def : Pat<(v4f32 (OpNode (v4f32 V128:$Rd), (v4f32 V128:$Rn),
6085 (ARM64dup (f32 FPR32Op:$Rm)))),
6086 (!cast<Instruction>(INST # "v4i32_indexed") V128:$Rd, V128:$Rn,
6087 (SUBREG_TO_REG (i32 0), FPR32Op:$Rm, ssub), (i64 0))>;
6089 // 2 variants for the .2d version: DUPLANE from 128-bit and DUP scalar.
6090 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6091 (ARM64duplane64 (v2f64 V128:$Rm),
6092 VectorIndexD:$idx))),
6093 (!cast<Instruction>(INST # "v2i64_indexed")
6094 V128:$Rd, V128:$Rn, V128:$Rm, VectorIndexS:$idx)>;
6095 def : Pat<(v2f64 (OpNode (v2f64 V128:$Rd), (v2f64 V128:$Rn),
6096 (ARM64dup (f64 FPR64Op:$Rm)))),
6097 (!cast<Instruction>(INST # "v2i64_indexed") V128:$Rd, V128:$Rn,
6098 (SUBREG_TO_REG (i32 0), FPR64Op:$Rm, dsub), (i64 0))>;
6100 // 2 variants for 32-bit scalar version: extract from .2s or from .4s
6101 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6102 (vector_extract (v4f32 V128:$Rm), VectorIndexS:$idx))),
6103 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
6104 V128:$Rm, VectorIndexS:$idx)>;
6105 def : Pat<(f32 (OpNode (f32 FPR32:$Rd), (f32 FPR32:$Rn),
6106 (vector_extract (v2f32 V64:$Rm), VectorIndexS:$idx))),
6107 (!cast<Instruction>(INST # "v1i32_indexed") FPR32:$Rd, FPR32:$Rn,
6108 (SUBREG_TO_REG (i32 0), V64:$Rm, dsub), VectorIndexS:$idx)>;
6110 // 1 variant for 64-bit scalar version: extract from .1d or from .2d
6111 def : Pat<(f64 (OpNode (f64 FPR64:$Rd), (f64 FPR64:$Rn),
6112 (vector_extract (v2f64 V128:$Rm), VectorIndexD:$idx))),
6113 (!cast<Instruction>(INST # "v1i64_indexed") FPR64:$Rd, FPR64:$Rn,
6114 V128:$Rm, VectorIndexD:$idx)>;
6117 multiclass SIMDFPIndexedSDTied<bit U, bits<4> opc, string asm> {
6118 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc, V64, V64,
6120 asm, ".2s", ".2s", ".2s", ".s", []> {
6122 let Inst{11} = idx{1};
6123 let Inst{21} = idx{0};
6126 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6129 asm, ".4s", ".4s", ".4s", ".s", []> {
6131 let Inst{11} = idx{1};
6132 let Inst{21} = idx{0};
6135 def v2i64_indexed : BaseSIMDIndexedTied<1, U, 0, 0b11, opc,
6138 asm, ".2d", ".2d", ".2d", ".d", []> {
6140 let Inst{11} = idx{0};
6145 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
6146 FPR32Op, FPR32Op, V128, VectorIndexS,
6147 asm, ".s", "", "", ".s", []> {
6149 let Inst{11} = idx{1};
6150 let Inst{21} = idx{0};
6153 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b11, opc,
6154 FPR64Op, FPR64Op, V128, VectorIndexD,
6155 asm, ".d", "", "", ".d", []> {
6157 let Inst{11} = idx{0};
6162 multiclass SIMDIndexedHS<bit U, bits<4> opc, string asm,
6163 SDPatternOperator OpNode> {
6164 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc, V64, V64,
6165 V128_lo, VectorIndexH,
6166 asm, ".4h", ".4h", ".4h", ".h",
6167 [(set (v4i16 V64:$Rd),
6168 (OpNode (v4i16 V64:$Rn),
6169 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6171 let Inst{11} = idx{2};
6172 let Inst{21} = idx{1};
6173 let Inst{20} = idx{0};
6176 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6178 V128_lo, VectorIndexH,
6179 asm, ".8h", ".8h", ".8h", ".h",
6180 [(set (v8i16 V128:$Rd),
6181 (OpNode (v8i16 V128:$Rn),
6182 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6184 let Inst{11} = idx{2};
6185 let Inst{21} = idx{1};
6186 let Inst{20} = idx{0};
6189 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6192 asm, ".2s", ".2s", ".2s", ".s",
6193 [(set (v2i32 V64:$Rd),
6194 (OpNode (v2i32 V64:$Rn),
6195 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6197 let Inst{11} = idx{1};
6198 let Inst{21} = idx{0};
6201 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6204 asm, ".4s", ".4s", ".4s", ".s",
6205 [(set (v4i32 V128:$Rd),
6206 (OpNode (v4i32 V128:$Rn),
6207 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6209 let Inst{11} = idx{1};
6210 let Inst{21} = idx{0};
6213 def v1i16_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
6214 FPR16Op, FPR16Op, V128_lo, VectorIndexH,
6215 asm, ".h", "", "", ".h", []> {
6217 let Inst{11} = idx{2};
6218 let Inst{21} = idx{1};
6219 let Inst{20} = idx{0};
6222 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6223 FPR32Op, FPR32Op, V128, VectorIndexS,
6224 asm, ".s", "", "", ".s",
6225 [(set (i32 FPR32Op:$Rd),
6226 (OpNode FPR32Op:$Rn,
6227 (i32 (vector_extract (v4i32 V128:$Rm),
6228 VectorIndexS:$idx))))]> {
6230 let Inst{11} = idx{1};
6231 let Inst{21} = idx{0};
6235 multiclass SIMDVectorIndexedHS<bit U, bits<4> opc, string asm,
6236 SDPatternOperator OpNode> {
6237 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6239 V128_lo, VectorIndexH,
6240 asm, ".4h", ".4h", ".4h", ".h",
6241 [(set (v4i16 V64:$Rd),
6242 (OpNode (v4i16 V64:$Rn),
6243 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6245 let Inst{11} = idx{2};
6246 let Inst{21} = idx{1};
6247 let Inst{20} = idx{0};
6250 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6252 V128_lo, VectorIndexH,
6253 asm, ".8h", ".8h", ".8h", ".h",
6254 [(set (v8i16 V128:$Rd),
6255 (OpNode (v8i16 V128:$Rn),
6256 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6258 let Inst{11} = idx{2};
6259 let Inst{21} = idx{1};
6260 let Inst{20} = idx{0};
6263 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6266 asm, ".2s", ".2s", ".2s", ".s",
6267 [(set (v2i32 V64:$Rd),
6268 (OpNode (v2i32 V64:$Rn),
6269 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6271 let Inst{11} = idx{1};
6272 let Inst{21} = idx{0};
6275 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6278 asm, ".4s", ".4s", ".4s", ".s",
6279 [(set (v4i32 V128:$Rd),
6280 (OpNode (v4i32 V128:$Rn),
6281 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6283 let Inst{11} = idx{1};
6284 let Inst{21} = idx{0};
6288 multiclass SIMDVectorIndexedHSTied<bit U, bits<4> opc, string asm,
6289 SDPatternOperator OpNode> {
6290 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc, V64, V64,
6291 V128_lo, VectorIndexH,
6292 asm, ".4h", ".4h", ".4h", ".h",
6293 [(set (v4i16 V64:$dst),
6294 (OpNode (v4i16 V64:$Rd),(v4i16 V64:$Rn),
6295 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6297 let Inst{11} = idx{2};
6298 let Inst{21} = idx{1};
6299 let Inst{20} = idx{0};
6302 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6304 V128_lo, VectorIndexH,
6305 asm, ".8h", ".8h", ".8h", ".h",
6306 [(set (v8i16 V128:$dst),
6307 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
6308 (v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6310 let Inst{11} = idx{2};
6311 let Inst{21} = idx{1};
6312 let Inst{20} = idx{0};
6315 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6318 asm, ".2s", ".2s", ".2s", ".s",
6319 [(set (v2i32 V64:$dst),
6320 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
6321 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6323 let Inst{11} = idx{1};
6324 let Inst{21} = idx{0};
6327 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6330 asm, ".4s", ".4s", ".4s", ".s",
6331 [(set (v4i32 V128:$dst),
6332 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
6333 (v4i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6335 let Inst{11} = idx{1};
6336 let Inst{21} = idx{0};
6340 multiclass SIMDIndexedLongSD<bit U, bits<4> opc, string asm,
6341 SDPatternOperator OpNode> {
6342 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6344 V128_lo, VectorIndexH,
6345 asm, ".4s", ".4s", ".4h", ".h",
6346 [(set (v4i32 V128:$Rd),
6347 (OpNode (v4i16 V64:$Rn),
6348 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6350 let Inst{11} = idx{2};
6351 let Inst{21} = idx{1};
6352 let Inst{20} = idx{0};
6355 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6357 V128_lo, VectorIndexH,
6358 asm#"2", ".4s", ".4s", ".8h", ".h",
6359 [(set (v4i32 V128:$Rd),
6360 (OpNode (extract_high_v8i16 V128:$Rn),
6361 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6362 VectorIndexH:$idx))))]> {
6365 let Inst{11} = idx{2};
6366 let Inst{21} = idx{1};
6367 let Inst{20} = idx{0};
6370 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6373 asm, ".2d", ".2d", ".2s", ".s",
6374 [(set (v2i64 V128:$Rd),
6375 (OpNode (v2i32 V64:$Rn),
6376 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6378 let Inst{11} = idx{1};
6379 let Inst{21} = idx{0};
6382 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6385 asm#"2", ".2d", ".2d", ".4s", ".s",
6386 [(set (v2i64 V128:$Rd),
6387 (OpNode (extract_high_v4i32 V128:$Rn),
6388 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6389 VectorIndexS:$idx))))]> {
6391 let Inst{11} = idx{1};
6392 let Inst{21} = idx{0};
6395 def v1i32_indexed : BaseSIMDIndexed<1, U, 1, 0b01, opc,
6396 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
6397 asm, ".h", "", "", ".h", []> {
6399 let Inst{11} = idx{2};
6400 let Inst{21} = idx{1};
6401 let Inst{20} = idx{0};
6404 def v1i64_indexed : BaseSIMDIndexed<1, U, 1, 0b10, opc,
6405 FPR64Op, FPR32Op, V128, VectorIndexS,
6406 asm, ".s", "", "", ".s", []> {
6408 let Inst{11} = idx{1};
6409 let Inst{21} = idx{0};
6413 multiclass SIMDIndexedLongSQDMLXSDTied<bit U, bits<4> opc, string asm,
6414 SDPatternOperator Accum> {
6415 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
6417 V128_lo, VectorIndexH,
6418 asm, ".4s", ".4s", ".4h", ".h",
6419 [(set (v4i32 V128:$dst),
6420 (Accum (v4i32 V128:$Rd),
6421 (v4i32 (int_arm64_neon_sqdmull
6423 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6424 VectorIndexH:$idx))))))]> {
6426 let Inst{11} = idx{2};
6427 let Inst{21} = idx{1};
6428 let Inst{20} = idx{0};
6431 // FIXME: it would be nice to use the scalar (v1i32) instruction here, but an
6432 // intermediate EXTRACT_SUBREG would be untyped.
6433 def : Pat<(i32 (Accum (i32 FPR32Op:$Rd),
6434 (i32 (vector_extract (v4i32
6435 (int_arm64_neon_sqdmull (v4i16 V64:$Rn),
6436 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6437 VectorIndexH:$idx)))),
6440 (!cast<Instruction>(NAME # v4i16_indexed)
6441 (SUBREG_TO_REG (i32 0), FPR32Op:$Rd, ssub), V64:$Rn,
6442 V128_lo:$Rm, VectorIndexH:$idx),
6445 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6447 V128_lo, VectorIndexH,
6448 asm#"2", ".4s", ".4s", ".8h", ".h",
6449 [(set (v4i32 V128:$dst),
6450 (Accum (v4i32 V128:$Rd),
6451 (v4i32 (int_arm64_neon_sqdmull
6452 (extract_high_v8i16 V128:$Rn),
6454 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6455 VectorIndexH:$idx))))))]> {
6457 let Inst{11} = idx{2};
6458 let Inst{21} = idx{1};
6459 let Inst{20} = idx{0};
6462 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6465 asm, ".2d", ".2d", ".2s", ".s",
6466 [(set (v2i64 V128:$dst),
6467 (Accum (v2i64 V128:$Rd),
6468 (v2i64 (int_arm64_neon_sqdmull
6470 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm),
6471 VectorIndexS:$idx))))))]> {
6473 let Inst{11} = idx{1};
6474 let Inst{21} = idx{0};
6477 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6480 asm#"2", ".2d", ".2d", ".4s", ".s",
6481 [(set (v2i64 V128:$dst),
6482 (Accum (v2i64 V128:$Rd),
6483 (v2i64 (int_arm64_neon_sqdmull
6484 (extract_high_v4i32 V128:$Rn),
6486 (ARM64duplane32 (v4i32 V128:$Rm),
6487 VectorIndexS:$idx))))))]> {
6489 let Inst{11} = idx{1};
6490 let Inst{21} = idx{0};
6493 def v1i32_indexed : BaseSIMDIndexedTied<1, U, 1, 0b01, opc,
6494 FPR32Op, FPR16Op, V128_lo, VectorIndexH,
6495 asm, ".h", "", "", ".h", []> {
6497 let Inst{11} = idx{2};
6498 let Inst{21} = idx{1};
6499 let Inst{20} = idx{0};
6503 def v1i64_indexed : BaseSIMDIndexedTied<1, U, 1, 0b10, opc,
6504 FPR64Op, FPR32Op, V128, VectorIndexS,
6505 asm, ".s", "", "", ".s",
6506 [(set (i64 FPR64Op:$dst),
6507 (Accum (i64 FPR64Op:$Rd),
6508 (i64 (int_arm64_neon_sqdmulls_scalar
6510 (i32 (vector_extract (v4i32 V128:$Rm),
6511 VectorIndexS:$idx))))))]> {
6514 let Inst{11} = idx{1};
6515 let Inst{21} = idx{0};
6519 multiclass SIMDVectorIndexedLongSD<bit U, bits<4> opc, string asm,
6520 SDPatternOperator OpNode> {
6521 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6522 def v4i16_indexed : BaseSIMDIndexed<0, U, 0, 0b01, opc,
6524 V128_lo, VectorIndexH,
6525 asm, ".4s", ".4s", ".4h", ".h",
6526 [(set (v4i32 V128:$Rd),
6527 (OpNode (v4i16 V64:$Rn),
6528 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6530 let Inst{11} = idx{2};
6531 let Inst{21} = idx{1};
6532 let Inst{20} = idx{0};
6535 def v8i16_indexed : BaseSIMDIndexed<1, U, 0, 0b01, opc,
6537 V128_lo, VectorIndexH,
6538 asm#"2", ".4s", ".4s", ".8h", ".h",
6539 [(set (v4i32 V128:$Rd),
6540 (OpNode (extract_high_v8i16 V128:$Rn),
6541 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6542 VectorIndexH:$idx))))]> {
6545 let Inst{11} = idx{2};
6546 let Inst{21} = idx{1};
6547 let Inst{20} = idx{0};
6550 def v2i32_indexed : BaseSIMDIndexed<0, U, 0, 0b10, opc,
6553 asm, ".2d", ".2d", ".2s", ".s",
6554 [(set (v2i64 V128:$Rd),
6555 (OpNode (v2i32 V64:$Rn),
6556 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6558 let Inst{11} = idx{1};
6559 let Inst{21} = idx{0};
6562 def v4i32_indexed : BaseSIMDIndexed<1, U, 0, 0b10, opc,
6565 asm#"2", ".2d", ".2d", ".4s", ".s",
6566 [(set (v2i64 V128:$Rd),
6567 (OpNode (extract_high_v4i32 V128:$Rn),
6568 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6569 VectorIndexS:$idx))))]> {
6571 let Inst{11} = idx{1};
6572 let Inst{21} = idx{0};
6577 multiclass SIMDVectorIndexedLongSDTied<bit U, bits<4> opc, string asm,
6578 SDPatternOperator OpNode> {
6579 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in {
6580 def v4i16_indexed : BaseSIMDIndexedTied<0, U, 0, 0b01, opc,
6582 V128_lo, VectorIndexH,
6583 asm, ".4s", ".4s", ".4h", ".h",
6584 [(set (v4i32 V128:$dst),
6585 (OpNode (v4i32 V128:$Rd), (v4i16 V64:$Rn),
6586 (v4i16 (ARM64duplane16 (v8i16 V128_lo:$Rm), VectorIndexH:$idx))))]> {
6588 let Inst{11} = idx{2};
6589 let Inst{21} = idx{1};
6590 let Inst{20} = idx{0};
6593 def v8i16_indexed : BaseSIMDIndexedTied<1, U, 0, 0b01, opc,
6595 V128_lo, VectorIndexH,
6596 asm#"2", ".4s", ".4s", ".8h", ".h",
6597 [(set (v4i32 V128:$dst),
6598 (OpNode (v4i32 V128:$Rd),
6599 (extract_high_v8i16 V128:$Rn),
6600 (extract_high_v8i16 (ARM64duplane16 (v8i16 V128_lo:$Rm),
6601 VectorIndexH:$idx))))]> {
6603 let Inst{11} = idx{2};
6604 let Inst{21} = idx{1};
6605 let Inst{20} = idx{0};
6608 def v2i32_indexed : BaseSIMDIndexedTied<0, U, 0, 0b10, opc,
6611 asm, ".2d", ".2d", ".2s", ".s",
6612 [(set (v2i64 V128:$dst),
6613 (OpNode (v2i64 V128:$Rd), (v2i32 V64:$Rn),
6614 (v2i32 (ARM64duplane32 (v4i32 V128:$Rm), VectorIndexS:$idx))))]> {
6616 let Inst{11} = idx{1};
6617 let Inst{21} = idx{0};
6620 def v4i32_indexed : BaseSIMDIndexedTied<1, U, 0, 0b10, opc,
6623 asm#"2", ".2d", ".2d", ".4s", ".s",
6624 [(set (v2i64 V128:$dst),
6625 (OpNode (v2i64 V128:$Rd),
6626 (extract_high_v4i32 V128:$Rn),
6627 (extract_high_v4i32 (ARM64duplane32 (v4i32 V128:$Rm),
6628 VectorIndexS:$idx))))]> {
6630 let Inst{11} = idx{1};
6631 let Inst{21} = idx{0};
6636 //----------------------------------------------------------------------------
6637 // AdvSIMD scalar shift by immediate
6638 //----------------------------------------------------------------------------
6640 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6641 class BaseSIMDScalarShift<bit U, bits<5> opc, bits<7> fixed_imm,
6642 RegisterClass regtype1, RegisterClass regtype2,
6643 Operand immtype, string asm, list<dag> pattern>
6644 : I<(outs regtype1:$Rd), (ins regtype2:$Rn, immtype:$imm),
6645 asm, "\t$Rd, $Rn, $imm", "", pattern>,
6650 let Inst{31-30} = 0b01;
6652 let Inst{28-23} = 0b111110;
6653 let Inst{22-16} = fixed_imm;
6654 let Inst{15-11} = opc;
6660 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6661 class BaseSIMDScalarShiftTied<bit U, bits<5> opc, bits<7> fixed_imm,
6662 RegisterClass regtype1, RegisterClass regtype2,
6663 Operand immtype, string asm, list<dag> pattern>
6664 : I<(outs regtype1:$dst), (ins regtype1:$Rd, regtype2:$Rn, immtype:$imm),
6665 asm, "\t$Rd, $Rn, $imm", "$Rd = $dst", pattern>,
6670 let Inst{31-30} = 0b01;
6672 let Inst{28-23} = 0b111110;
6673 let Inst{22-16} = fixed_imm;
6674 let Inst{15-11} = opc;
6681 multiclass SIMDScalarRShiftSD<bit U, bits<5> opc, string asm> {
6682 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6683 FPR32, FPR32, vecshiftR32, asm, []> {
6684 let Inst{20-16} = imm{4-0};
6687 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6688 FPR64, FPR64, vecshiftR64, asm, []> {
6689 let Inst{21-16} = imm{5-0};
6693 multiclass SIMDScalarRShiftD<bit U, bits<5> opc, string asm,
6694 SDPatternOperator OpNode> {
6695 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6696 FPR64, FPR64, vecshiftR64, asm,
6697 [(set (i64 FPR64:$Rd),
6698 (OpNode (i64 FPR64:$Rn), (i32 vecshiftR64:$imm)))]> {
6699 let Inst{21-16} = imm{5-0};
6702 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftR64:$imm))),
6703 (!cast<Instruction>(NAME # "d") FPR64:$Rn, vecshiftR64:$imm)>;
6706 multiclass SIMDScalarRShiftDTied<bit U, bits<5> opc, string asm,
6707 SDPatternOperator OpNode = null_frag> {
6708 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
6709 FPR64, FPR64, vecshiftR64, asm,
6710 [(set (i64 FPR64:$dst), (OpNode (i64 FPR64:$Rd), (i64 FPR64:$Rn),
6711 (i32 vecshiftR64:$imm)))]> {
6712 let Inst{21-16} = imm{5-0};
6715 def : Pat<(v1i64 (OpNode (v1i64 FPR64:$Rd), (v1i64 FPR64:$Rn),
6716 (i32 vecshiftR64:$imm))),
6717 (!cast<Instruction>(NAME # "d") FPR64:$Rd, FPR64:$Rn,
6721 multiclass SIMDScalarLShiftD<bit U, bits<5> opc, string asm,
6722 SDPatternOperator OpNode> {
6723 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6724 FPR64, FPR64, vecshiftL64, asm,
6725 [(set (v1i64 FPR64:$Rd),
6726 (OpNode (v1i64 FPR64:$Rn), (i32 vecshiftL64:$imm)))]> {
6727 let Inst{21-16} = imm{5-0};
6731 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6732 multiclass SIMDScalarLShiftDTied<bit U, bits<5> opc, string asm> {
6733 def d : BaseSIMDScalarShiftTied<U, opc, {1,?,?,?,?,?,?},
6734 FPR64, FPR64, vecshiftL64, asm, []> {
6735 let Inst{21-16} = imm{5-0};
6739 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6740 multiclass SIMDScalarRShiftBHS<bit U, bits<5> opc, string asm,
6741 SDPatternOperator OpNode = null_frag> {
6742 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6743 FPR8, FPR16, vecshiftR8, asm, []> {
6744 let Inst{18-16} = imm{2-0};
6747 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6748 FPR16, FPR32, vecshiftR16, asm, []> {
6749 let Inst{19-16} = imm{3-0};
6752 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6753 FPR32, FPR64, vecshiftR32, asm,
6754 [(set (i32 FPR32:$Rd), (OpNode (i64 FPR64:$Rn), vecshiftR32:$imm))]> {
6755 let Inst{20-16} = imm{4-0};
6759 multiclass SIMDScalarLShiftBHSD<bit U, bits<5> opc, string asm,
6760 SDPatternOperator OpNode> {
6761 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6762 FPR8, FPR8, vecshiftL8, asm, []> {
6763 let Inst{18-16} = imm{2-0};
6766 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6767 FPR16, FPR16, vecshiftL16, asm, []> {
6768 let Inst{19-16} = imm{3-0};
6771 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6772 FPR32, FPR32, vecshiftL32, asm,
6773 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn), (i32 vecshiftL32:$imm)))]> {
6774 let Inst{20-16} = imm{4-0};
6777 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6778 FPR64, FPR64, vecshiftL64, asm,
6779 [(set (v1i64 FPR64:$Rd), (OpNode (v1i64 FPR64:$Rn),
6780 (i32 vecshiftL64:$imm)))]> {
6781 let Inst{21-16} = imm{5-0};
6785 multiclass SIMDScalarRShiftBHSD<bit U, bits<5> opc, string asm> {
6786 def b : BaseSIMDScalarShift<U, opc, {0,0,0,1,?,?,?},
6787 FPR8, FPR8, vecshiftR8, asm, []> {
6788 let Inst{18-16} = imm{2-0};
6791 def h : BaseSIMDScalarShift<U, opc, {0,0,1,?,?,?,?},
6792 FPR16, FPR16, vecshiftR16, asm, []> {
6793 let Inst{19-16} = imm{3-0};
6796 def s : BaseSIMDScalarShift<U, opc, {0,1,?,?,?,?,?},
6797 FPR32, FPR32, vecshiftR32, asm, []> {
6798 let Inst{20-16} = imm{4-0};
6801 def d : BaseSIMDScalarShift<U, opc, {1,?,?,?,?,?,?},
6802 FPR64, FPR64, vecshiftR64, asm, []> {
6803 let Inst{21-16} = imm{5-0};
6807 //----------------------------------------------------------------------------
6808 // AdvSIMD vector x indexed element
6809 //----------------------------------------------------------------------------
6811 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6812 class BaseSIMDVectorShift<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
6813 RegisterOperand dst_reg, RegisterOperand src_reg,
6815 string asm, string dst_kind, string src_kind,
6817 : I<(outs dst_reg:$Rd), (ins src_reg:$Rn, immtype:$imm),
6818 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
6819 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "", pattern>,
6826 let Inst{28-23} = 0b011110;
6827 let Inst{22-16} = fixed_imm;
6828 let Inst{15-11} = opc;
6834 let mayStore = 0, mayLoad = 0, hasSideEffects = 0 in
6835 class BaseSIMDVectorShiftTied<bit Q, bit U, bits<5> opc, bits<7> fixed_imm,
6836 RegisterOperand vectype1, RegisterOperand vectype2,
6838 string asm, string dst_kind, string src_kind,
6840 : I<(outs vectype1:$dst), (ins vectype1:$Rd, vectype2:$Rn, immtype:$imm),
6841 asm, "{\t$Rd" # dst_kind # ", $Rn" # src_kind # ", $imm" #
6842 "|" # dst_kind # "\t$Rd, $Rn, $imm}", "$Rd = $dst", pattern>,
6849 let Inst{28-23} = 0b011110;
6850 let Inst{22-16} = fixed_imm;
6851 let Inst{15-11} = opc;
6857 multiclass SIMDVectorRShiftSD<bit U, bits<5> opc, string asm,
6859 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6860 V64, V64, vecshiftR32,
6862 [(set (v2i32 V64:$Rd), (OpNode (v2f32 V64:$Rn), (i32 imm:$imm)))]> {
6864 let Inst{20-16} = imm;
6867 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
6868 V128, V128, vecshiftR32,
6870 [(set (v4i32 V128:$Rd), (OpNode (v4f32 V128:$Rn), (i32 imm:$imm)))]> {
6872 let Inst{20-16} = imm;
6875 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
6876 V128, V128, vecshiftR64,
6878 [(set (v2i64 V128:$Rd), (OpNode (v2f64 V128:$Rn), (i32 imm:$imm)))]> {
6880 let Inst{21-16} = imm;
6884 multiclass SIMDVectorRShiftSDToFP<bit U, bits<5> opc, string asm,
6886 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6887 V64, V64, vecshiftR32,
6889 [(set (v2f32 V64:$Rd), (OpNode (v2i32 V64:$Rn), (i32 imm:$imm)))]> {
6891 let Inst{20-16} = imm;
6894 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
6895 V128, V128, vecshiftR32,
6897 [(set (v4f32 V128:$Rd), (OpNode (v4i32 V128:$Rn), (i32 imm:$imm)))]> {
6899 let Inst{20-16} = imm;
6902 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
6903 V128, V128, vecshiftR64,
6905 [(set (v2f64 V128:$Rd), (OpNode (v2i64 V128:$Rn), (i32 imm:$imm)))]> {
6907 let Inst{21-16} = imm;
6911 multiclass SIMDVectorRShiftNarrowBHS<bit U, bits<5> opc, string asm,
6912 SDPatternOperator OpNode> {
6913 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
6914 V64, V128, vecshiftR16Narrow,
6916 [(set (v8i8 V64:$Rd), (OpNode (v8i16 V128:$Rn), vecshiftR16Narrow:$imm))]> {
6918 let Inst{18-16} = imm;
6921 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
6922 V128, V128, vecshiftR16Narrow,
6923 asm#"2", ".16b", ".8h", []> {
6925 let Inst{18-16} = imm;
6926 let hasSideEffects = 0;
6929 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
6930 V64, V128, vecshiftR32Narrow,
6932 [(set (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn), vecshiftR32Narrow:$imm))]> {
6934 let Inst{19-16} = imm;
6937 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
6938 V128, V128, vecshiftR32Narrow,
6939 asm#"2", ".8h", ".4s", []> {
6941 let Inst{19-16} = imm;
6942 let hasSideEffects = 0;
6945 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
6946 V64, V128, vecshiftR64Narrow,
6948 [(set (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn), vecshiftR64Narrow:$imm))]> {
6950 let Inst{20-16} = imm;
6953 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
6954 V128, V128, vecshiftR64Narrow,
6955 asm#"2", ".4s", ".2d", []> {
6957 let Inst{20-16} = imm;
6958 let hasSideEffects = 0;
6961 // TableGen doesn't like patters w/ INSERT_SUBREG on the instructions
6962 // themselves, so put them here instead.
6964 // Patterns involving what's effectively an insert high and a normal
6965 // intrinsic, represented by CONCAT_VECTORS.
6966 def : Pat<(concat_vectors (v8i8 V64:$Rd),(OpNode (v8i16 V128:$Rn),
6967 vecshiftR16Narrow:$imm)),
6968 (!cast<Instruction>(NAME # "v16i8_shift")
6969 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6970 V128:$Rn, vecshiftR16Narrow:$imm)>;
6971 def : Pat<(concat_vectors (v4i16 V64:$Rd), (OpNode (v4i32 V128:$Rn),
6972 vecshiftR32Narrow:$imm)),
6973 (!cast<Instruction>(NAME # "v8i16_shift")
6974 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6975 V128:$Rn, vecshiftR32Narrow:$imm)>;
6976 def : Pat<(concat_vectors (v2i32 V64:$Rd), (OpNode (v2i64 V128:$Rn),
6977 vecshiftR64Narrow:$imm)),
6978 (!cast<Instruction>(NAME # "v4i32_shift")
6979 (INSERT_SUBREG (IMPLICIT_DEF), V64:$Rd, dsub),
6980 V128:$Rn, vecshiftR64Narrow:$imm)>;
6983 multiclass SIMDVectorLShiftBHSD<bit U, bits<5> opc, string asm,
6984 SDPatternOperator OpNode> {
6985 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
6986 V64, V64, vecshiftL8,
6988 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
6989 (i32 vecshiftL8:$imm)))]> {
6991 let Inst{18-16} = imm;
6994 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
6995 V128, V128, vecshiftL8,
6996 asm, ".16b", ".16b",
6997 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
6998 (i32 vecshiftL8:$imm)))]> {
7000 let Inst{18-16} = imm;
7003 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7004 V64, V64, vecshiftL16,
7006 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
7007 (i32 vecshiftL16:$imm)))]> {
7009 let Inst{19-16} = imm;
7012 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7013 V128, V128, vecshiftL16,
7015 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
7016 (i32 vecshiftL16:$imm)))]> {
7018 let Inst{19-16} = imm;
7021 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7022 V64, V64, vecshiftL32,
7024 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
7025 (i32 vecshiftL32:$imm)))]> {
7027 let Inst{20-16} = imm;
7030 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7031 V128, V128, vecshiftL32,
7033 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
7034 (i32 vecshiftL32:$imm)))]> {
7036 let Inst{20-16} = imm;
7039 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
7040 V128, V128, vecshiftL64,
7042 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
7043 (i32 vecshiftL64:$imm)))]> {
7045 let Inst{21-16} = imm;
7049 multiclass SIMDVectorRShiftBHSD<bit U, bits<5> opc, string asm,
7050 SDPatternOperator OpNode> {
7051 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7052 V64, V64, vecshiftR8,
7054 [(set (v8i8 V64:$Rd), (OpNode (v8i8 V64:$Rn),
7055 (i32 vecshiftR8:$imm)))]> {
7057 let Inst{18-16} = imm;
7060 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7061 V128, V128, vecshiftR8,
7062 asm, ".16b", ".16b",
7063 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn),
7064 (i32 vecshiftR8:$imm)))]> {
7066 let Inst{18-16} = imm;
7069 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7070 V64, V64, vecshiftR16,
7072 [(set (v4i16 V64:$Rd), (OpNode (v4i16 V64:$Rn),
7073 (i32 vecshiftR16:$imm)))]> {
7075 let Inst{19-16} = imm;
7078 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7079 V128, V128, vecshiftR16,
7081 [(set (v8i16 V128:$Rd), (OpNode (v8i16 V128:$Rn),
7082 (i32 vecshiftR16:$imm)))]> {
7084 let Inst{19-16} = imm;
7087 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7088 V64, V64, vecshiftR32,
7090 [(set (v2i32 V64:$Rd), (OpNode (v2i32 V64:$Rn),
7091 (i32 vecshiftR32:$imm)))]> {
7093 let Inst{20-16} = imm;
7096 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7097 V128, V128, vecshiftR32,
7099 [(set (v4i32 V128:$Rd), (OpNode (v4i32 V128:$Rn),
7100 (i32 vecshiftR32:$imm)))]> {
7102 let Inst{20-16} = imm;
7105 def v2i64_shift : BaseSIMDVectorShift<1, U, opc, {1,?,?,?,?,?,?},
7106 V128, V128, vecshiftR64,
7108 [(set (v2i64 V128:$Rd), (OpNode (v2i64 V128:$Rn),
7109 (i32 vecshiftR64:$imm)))]> {
7111 let Inst{21-16} = imm;
7115 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
7116 multiclass SIMDVectorRShiftBHSDTied<bit U, bits<5> opc, string asm,
7117 SDPatternOperator OpNode = null_frag> {
7118 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
7119 V64, V64, vecshiftR8, asm, ".8b", ".8b",
7120 [(set (v8i8 V64:$dst),
7121 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
7122 (i32 vecshiftR8:$imm)))]> {
7124 let Inst{18-16} = imm;
7127 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
7128 V128, V128, vecshiftR8, asm, ".16b", ".16b",
7129 [(set (v16i8 V128:$dst),
7130 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
7131 (i32 vecshiftR8:$imm)))]> {
7133 let Inst{18-16} = imm;
7136 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
7137 V64, V64, vecshiftR16, asm, ".4h", ".4h",
7138 [(set (v4i16 V64:$dst),
7139 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
7140 (i32 vecshiftR16:$imm)))]> {
7142 let Inst{19-16} = imm;
7145 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
7146 V128, V128, vecshiftR16, asm, ".8h", ".8h",
7147 [(set (v8i16 V128:$dst),
7148 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7149 (i32 vecshiftR16:$imm)))]> {
7151 let Inst{19-16} = imm;
7154 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
7155 V64, V64, vecshiftR32, asm, ".2s", ".2s",
7156 [(set (v2i32 V64:$dst),
7157 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7158 (i32 vecshiftR32:$imm)))]> {
7160 let Inst{20-16} = imm;
7163 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
7164 V128, V128, vecshiftR32, asm, ".4s", ".4s",
7165 [(set (v4i32 V128:$dst),
7166 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7167 (i32 vecshiftR32:$imm)))]> {
7169 let Inst{20-16} = imm;
7172 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
7173 V128, V128, vecshiftR64,
7174 asm, ".2d", ".2d", [(set (v2i64 V128:$dst),
7175 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
7176 (i32 vecshiftR64:$imm)))]> {
7178 let Inst{21-16} = imm;
7182 multiclass SIMDVectorLShiftBHSDTied<bit U, bits<5> opc, string asm,
7183 SDPatternOperator OpNode = null_frag> {
7184 def v8i8_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,0,1,?,?,?},
7185 V64, V64, vecshiftL8,
7187 [(set (v8i8 V64:$dst),
7188 (OpNode (v8i8 V64:$Rd), (v8i8 V64:$Rn),
7189 (i32 vecshiftL8:$imm)))]> {
7191 let Inst{18-16} = imm;
7194 def v16i8_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,0,1,?,?,?},
7195 V128, V128, vecshiftL8,
7196 asm, ".16b", ".16b",
7197 [(set (v16i8 V128:$dst),
7198 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn),
7199 (i32 vecshiftL8:$imm)))]> {
7201 let Inst{18-16} = imm;
7204 def v4i16_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,0,1,?,?,?,?},
7205 V64, V64, vecshiftL16,
7207 [(set (v4i16 V64:$dst),
7208 (OpNode (v4i16 V64:$Rd), (v4i16 V64:$Rn),
7209 (i32 vecshiftL16:$imm)))]> {
7211 let Inst{19-16} = imm;
7214 def v8i16_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,0,1,?,?,?,?},
7215 V128, V128, vecshiftL16,
7217 [(set (v8i16 V128:$dst),
7218 (OpNode (v8i16 V128:$Rd), (v8i16 V128:$Rn),
7219 (i32 vecshiftL16:$imm)))]> {
7221 let Inst{19-16} = imm;
7224 def v2i32_shift : BaseSIMDVectorShiftTied<0, U, opc, {0,1,?,?,?,?,?},
7225 V64, V64, vecshiftL32,
7227 [(set (v2i32 V64:$dst),
7228 (OpNode (v2i32 V64:$Rd), (v2i32 V64:$Rn),
7229 (i32 vecshiftL32:$imm)))]> {
7231 let Inst{20-16} = imm;
7234 def v4i32_shift : BaseSIMDVectorShiftTied<1, U, opc, {0,1,?,?,?,?,?},
7235 V128, V128, vecshiftL32,
7237 [(set (v4i32 V128:$dst),
7238 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
7239 (i32 vecshiftL32:$imm)))]> {
7241 let Inst{20-16} = imm;
7244 def v2i64_shift : BaseSIMDVectorShiftTied<1, U, opc, {1,?,?,?,?,?,?},
7245 V128, V128, vecshiftL64,
7247 [(set (v2i64 V128:$dst),
7248 (OpNode (v2i64 V128:$Rd), (v2i64 V128:$Rn),
7249 (i32 vecshiftL64:$imm)))]> {
7251 let Inst{21-16} = imm;
7255 multiclass SIMDVectorLShiftLongBHSD<bit U, bits<5> opc, string asm,
7256 SDPatternOperator OpNode> {
7257 def v8i8_shift : BaseSIMDVectorShift<0, U, opc, {0,0,0,1,?,?,?},
7258 V128, V64, vecshiftL8, asm, ".8h", ".8b",
7259 [(set (v8i16 V128:$Rd), (OpNode (v8i8 V64:$Rn), vecshiftL8:$imm))]> {
7261 let Inst{18-16} = imm;
7264 def v16i8_shift : BaseSIMDVectorShift<1, U, opc, {0,0,0,1,?,?,?},
7265 V128, V128, vecshiftL8,
7266 asm#"2", ".8h", ".16b",
7267 [(set (v8i16 V128:$Rd),
7268 (OpNode (extract_high_v16i8 V128:$Rn), vecshiftL8:$imm))]> {
7270 let Inst{18-16} = imm;
7273 def v4i16_shift : BaseSIMDVectorShift<0, U, opc, {0,0,1,?,?,?,?},
7274 V128, V64, vecshiftL16, asm, ".4s", ".4h",
7275 [(set (v4i32 V128:$Rd), (OpNode (v4i16 V64:$Rn), vecshiftL16:$imm))]> {
7277 let Inst{19-16} = imm;
7280 def v8i16_shift : BaseSIMDVectorShift<1, U, opc, {0,0,1,?,?,?,?},
7281 V128, V128, vecshiftL16,
7282 asm#"2", ".4s", ".8h",
7283 [(set (v4i32 V128:$Rd),
7284 (OpNode (extract_high_v8i16 V128:$Rn), vecshiftL16:$imm))]> {
7287 let Inst{19-16} = imm;
7290 def v2i32_shift : BaseSIMDVectorShift<0, U, opc, {0,1,?,?,?,?,?},
7291 V128, V64, vecshiftL32, asm, ".2d", ".2s",
7292 [(set (v2i64 V128:$Rd), (OpNode (v2i32 V64:$Rn), vecshiftL32:$imm))]> {
7294 let Inst{20-16} = imm;
7297 def v4i32_shift : BaseSIMDVectorShift<1, U, opc, {0,1,?,?,?,?,?},
7298 V128, V128, vecshiftL32,
7299 asm#"2", ".2d", ".4s",
7300 [(set (v2i64 V128:$Rd),
7301 (OpNode (extract_high_v4i32 V128:$Rn), vecshiftL32:$imm))]> {
7303 let Inst{20-16} = imm;
7309 // Vector load/store
7311 // SIMD ldX/stX no-index memory references don't allow the optional
7312 // ", #0" constant and handle post-indexing explicitly, so we use
7313 // a more specialized parse method for them. Otherwise, it's the same as
7314 // the general am_noindex handling.
7315 def MemorySIMDNoIndexOperand : AsmOperandClass {
7316 let Name = "MemorySIMDNoIndex";
7317 let ParserMethod = "tryParseNoIndexMemory";
7319 def am_simdnoindex : Operand<i64>,
7320 ComplexPattern<i64, 1, "SelectAddrModeNoIndex", []> {
7321 let PrintMethod = "printAMNoIndex";
7322 let ParserMatchClass = MemorySIMDNoIndexOperand;
7323 let MIOperandInfo = (ops GPR64sp:$base);
7324 let DecoderMethod = "DecodeGPR64spRegisterClass";
7327 class BaseSIMDLdSt<bit Q, bit L, bits<4> opcode, bits<2> size,
7328 string asm, dag oops, dag iops, list<dag> pattern>
7329 : I<oops, iops, asm, "\t$Vt, $vaddr", "", pattern> {
7334 let Inst{29-23} = 0b0011000;
7336 let Inst{21-16} = 0b000000;
7337 let Inst{15-12} = opcode;
7338 let Inst{11-10} = size;
7339 let Inst{9-5} = vaddr;
7343 class BaseSIMDLdStPost<bit Q, bit L, bits<4> opcode, bits<2> size,
7344 string asm, dag oops, dag iops>
7345 : I<oops, iops, asm, "\t$Vt, $vaddr, $Xm", "", []> {
7351 let Inst{29-23} = 0b0011001;
7354 let Inst{20-16} = Xm;
7355 let Inst{15-12} = opcode;
7356 let Inst{11-10} = size;
7357 let Inst{9-5} = vaddr;
7359 let DecoderMethod = "DecodeSIMDLdStPost";
7362 // The immediate form of AdvSIMD post-indexed addressing is encoded with
7363 // register post-index addressing from the zero register.
7364 multiclass SIMDLdStAliases<string asm, string layout, string Count,
7365 int Offset, int Size> {
7366 // E.g. "ld1 { v0.8b, v1.8b }, [x1], #16"
7367 // "ld1\t$Vt, $vaddr, #16"
7368 // may get mapped to
7369 // (LD1Twov8b_POST VecListTwo8b:$Vt, am_simdnoindex:$vaddr, XZR)
7370 def : InstAlias<asm # "\t$Vt, $vaddr, #" # Offset,
7371 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7372 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
7373 am_simdnoindex:$vaddr, XZR), 1>;
7375 // E.g. "ld1.8b { v0, v1 }, [x1], #16"
7376 // "ld1.8b\t$Vt, $vaddr, #16"
7377 // may get mapped to
7378 // (LD1Twov8b_POST VecListTwo64:$Vt, am_simdnoindex:$vaddr, XZR)
7379 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, #" # Offset,
7380 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7381 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7382 am_simdnoindex:$vaddr, XZR), 0>;
7384 // E.g. "ld1.8b { v0, v1 }, [x1]"
7385 // "ld1\t$Vt, $vaddr"
7386 // may get mapped to
7387 // (LD1Twov8b VecListTwo64:$Vt, am_simdnoindex:$vaddr)
7388 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr",
7389 (!cast<Instruction>(NAME # Count # "v" # layout)
7390 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7391 am_simdnoindex:$vaddr), 0>;
7393 // E.g. "ld1.8b { v0, v1 }, [x1], x2"
7394 // "ld1\t$Vt, $vaddr, $Xm"
7395 // may get mapped to
7396 // (LD1Twov8b_POST VecListTwo64:$Vt, am_simdnoindex:$vaddr, GPR64pi8:$Xm)
7397 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, $Xm",
7398 (!cast<Instruction>(NAME # Count # "v" # layout # "_POST")
7399 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7400 am_simdnoindex:$vaddr,
7401 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
7404 multiclass BaseSIMDLdN<string Count, string asm, string veclist, int Offset128,
7405 int Offset64, bits<4> opcode> {
7406 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
7407 def v16b: BaseSIMDLdSt<1, 1, opcode, 0b00, asm,
7408 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
7409 (ins am_simdnoindex:$vaddr), []>;
7410 def v8h : BaseSIMDLdSt<1, 1, opcode, 0b01, asm,
7411 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
7412 (ins am_simdnoindex:$vaddr), []>;
7413 def v4s : BaseSIMDLdSt<1, 1, opcode, 0b10, asm,
7414 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
7415 (ins am_simdnoindex:$vaddr), []>;
7416 def v2d : BaseSIMDLdSt<1, 1, opcode, 0b11, asm,
7417 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
7418 (ins am_simdnoindex:$vaddr), []>;
7419 def v8b : BaseSIMDLdSt<0, 1, opcode, 0b00, asm,
7420 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
7421 (ins am_simdnoindex:$vaddr), []>;
7422 def v4h : BaseSIMDLdSt<0, 1, opcode, 0b01, asm,
7423 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
7424 (ins am_simdnoindex:$vaddr), []>;
7425 def v2s : BaseSIMDLdSt<0, 1, opcode, 0b10, asm,
7426 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
7427 (ins am_simdnoindex:$vaddr), []>;
7430 def v16b_POST: BaseSIMDLdStPost<1, 1, opcode, 0b00, asm,
7431 (outs !cast<RegisterOperand>(veclist # "16b"):$Vt),
7432 (ins am_simdnoindex:$vaddr,
7433 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7434 def v8h_POST : BaseSIMDLdStPost<1, 1, opcode, 0b01, asm,
7435 (outs !cast<RegisterOperand>(veclist # "8h"):$Vt),
7436 (ins am_simdnoindex:$vaddr,
7437 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7438 def v4s_POST : BaseSIMDLdStPost<1, 1, opcode, 0b10, asm,
7439 (outs !cast<RegisterOperand>(veclist # "4s"):$Vt),
7440 (ins am_simdnoindex:$vaddr,
7441 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7442 def v2d_POST : BaseSIMDLdStPost<1, 1, opcode, 0b11, asm,
7443 (outs !cast<RegisterOperand>(veclist # "2d"):$Vt),
7444 (ins am_simdnoindex:$vaddr,
7445 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7446 def v8b_POST : BaseSIMDLdStPost<0, 1, opcode, 0b00, asm,
7447 (outs !cast<RegisterOperand>(veclist # "8b"):$Vt),
7448 (ins am_simdnoindex:$vaddr,
7449 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7450 def v4h_POST : BaseSIMDLdStPost<0, 1, opcode, 0b01, asm,
7451 (outs !cast<RegisterOperand>(veclist # "4h"):$Vt),
7452 (ins am_simdnoindex:$vaddr,
7453 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7454 def v2s_POST : BaseSIMDLdStPost<0, 1, opcode, 0b10, asm,
7455 (outs !cast<RegisterOperand>(veclist # "2s"):$Vt),
7456 (ins am_simdnoindex:$vaddr,
7457 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7460 defm : SIMDLdStAliases<asm, "16b", Count, Offset128, 128>;
7461 defm : SIMDLdStAliases<asm, "8h", Count, Offset128, 128>;
7462 defm : SIMDLdStAliases<asm, "4s", Count, Offset128, 128>;
7463 defm : SIMDLdStAliases<asm, "2d", Count, Offset128, 128>;
7464 defm : SIMDLdStAliases<asm, "8b", Count, Offset64, 64>;
7465 defm : SIMDLdStAliases<asm, "4h", Count, Offset64, 64>;
7466 defm : SIMDLdStAliases<asm, "2s", Count, Offset64, 64>;
7469 // Only ld1/st1 has a v1d version.
7470 multiclass BaseSIMDStN<string Count, string asm, string veclist, int Offset128,
7471 int Offset64, bits<4> opcode> {
7472 let hasSideEffects = 0, mayStore = 1, mayLoad = 0 in {
7473 def v16b : BaseSIMDLdSt<1, 0, opcode, 0b00, asm, (outs),
7474 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
7475 am_simdnoindex:$vaddr), []>;
7476 def v8h : BaseSIMDLdSt<1, 0, opcode, 0b01, asm, (outs),
7477 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
7478 am_simdnoindex:$vaddr), []>;
7479 def v4s : BaseSIMDLdSt<1, 0, opcode, 0b10, asm, (outs),
7480 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
7481 am_simdnoindex:$vaddr), []>;
7482 def v2d : BaseSIMDLdSt<1, 0, opcode, 0b11, asm, (outs),
7483 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
7484 am_simdnoindex:$vaddr), []>;
7485 def v8b : BaseSIMDLdSt<0, 0, opcode, 0b00, asm, (outs),
7486 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
7487 am_simdnoindex:$vaddr), []>;
7488 def v4h : BaseSIMDLdSt<0, 0, opcode, 0b01, asm, (outs),
7489 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
7490 am_simdnoindex:$vaddr), []>;
7491 def v2s : BaseSIMDLdSt<0, 0, opcode, 0b10, asm, (outs),
7492 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
7493 am_simdnoindex:$vaddr), []>;
7495 def v16b_POST : BaseSIMDLdStPost<1, 0, opcode, 0b00, asm, (outs),
7496 (ins !cast<RegisterOperand>(veclist # "16b"):$Vt,
7497 am_simdnoindex:$vaddr,
7498 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7499 def v8h_POST : BaseSIMDLdStPost<1, 0, opcode, 0b01, asm, (outs),
7500 (ins !cast<RegisterOperand>(veclist # "8h"):$Vt,
7501 am_simdnoindex:$vaddr,
7502 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7503 def v4s_POST : BaseSIMDLdStPost<1, 0, opcode, 0b10, asm, (outs),
7504 (ins !cast<RegisterOperand>(veclist # "4s"):$Vt,
7505 am_simdnoindex:$vaddr,
7506 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7507 def v2d_POST : BaseSIMDLdStPost<1, 0, opcode, 0b11, asm, (outs),
7508 (ins !cast<RegisterOperand>(veclist # "2d"):$Vt,
7509 am_simdnoindex:$vaddr,
7510 !cast<RegisterOperand>("GPR64pi" # Offset128):$Xm)>;
7511 def v8b_POST : BaseSIMDLdStPost<0, 0, opcode, 0b00, asm, (outs),
7512 (ins !cast<RegisterOperand>(veclist # "8b"):$Vt,
7513 am_simdnoindex:$vaddr,
7514 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7515 def v4h_POST : BaseSIMDLdStPost<0, 0, opcode, 0b01, asm, (outs),
7516 (ins !cast<RegisterOperand>(veclist # "4h"):$Vt,
7517 am_simdnoindex:$vaddr,
7518 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7519 def v2s_POST : BaseSIMDLdStPost<0, 0, opcode, 0b10, asm, (outs),
7520 (ins !cast<RegisterOperand>(veclist # "2s"):$Vt,
7521 am_simdnoindex:$vaddr,
7522 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7525 defm : SIMDLdStAliases<asm, "16b", Count, Offset128, 128>;
7526 defm : SIMDLdStAliases<asm, "8h", Count, Offset128, 128>;
7527 defm : SIMDLdStAliases<asm, "4s", Count, Offset128, 128>;
7528 defm : SIMDLdStAliases<asm, "2d", Count, Offset128, 128>;
7529 defm : SIMDLdStAliases<asm, "8b", Count, Offset64, 64>;
7530 defm : SIMDLdStAliases<asm, "4h", Count, Offset64, 64>;
7531 defm : SIMDLdStAliases<asm, "2s", Count, Offset64, 64>;
7534 multiclass BaseSIMDLd1<string Count, string asm, string veclist,
7535 int Offset128, int Offset64, bits<4> opcode>
7536 : BaseSIMDLdN<Count, asm, veclist, Offset128, Offset64, opcode> {
7538 // LD1 instructions have extra "1d" variants.
7539 let hasSideEffects = 0, mayLoad = 1, mayStore = 0 in {
7540 def v1d : BaseSIMDLdSt<0, 1, opcode, 0b11, asm,
7541 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
7542 (ins am_simdnoindex:$vaddr), []>;
7544 def v1d_POST : BaseSIMDLdStPost<0, 1, opcode, 0b11, asm,
7545 (outs !cast<RegisterOperand>(veclist # "1d"):$Vt),
7546 (ins am_simdnoindex:$vaddr,
7547 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7550 defm : SIMDLdStAliases<asm, "1d", Count, Offset64, 64>;
7553 multiclass BaseSIMDSt1<string Count, string asm, string veclist,
7554 int Offset128, int Offset64, bits<4> opcode>
7555 : BaseSIMDStN<Count, asm, veclist, Offset128, Offset64, opcode> {
7557 // ST1 instructions have extra "1d" variants.
7558 let hasSideEffects = 0, mayLoad = 0, mayStore = 1 in {
7559 def v1d : BaseSIMDLdSt<0, 0, opcode, 0b11, asm, (outs),
7560 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
7561 am_simdnoindex:$vaddr), []>;
7563 def v1d_POST : BaseSIMDLdStPost<0, 0, opcode, 0b11, asm, (outs),
7564 (ins !cast<RegisterOperand>(veclist # "1d"):$Vt,
7565 am_simdnoindex:$vaddr,
7566 !cast<RegisterOperand>("GPR64pi" # Offset64):$Xm)>;
7569 defm : SIMDLdStAliases<asm, "1d", Count, Offset64, 64>;
7572 multiclass SIMDLd1Multiple<string asm> {
7573 defm One : BaseSIMDLd1<"One", asm, "VecListOne", 16, 8, 0b0111>;
7574 defm Two : BaseSIMDLd1<"Two", asm, "VecListTwo", 32, 16, 0b1010>;
7575 defm Three : BaseSIMDLd1<"Three", asm, "VecListThree", 48, 24, 0b0110>;
7576 defm Four : BaseSIMDLd1<"Four", asm, "VecListFour", 64, 32, 0b0010>;
7579 multiclass SIMDSt1Multiple<string asm> {
7580 defm One : BaseSIMDSt1<"One", asm, "VecListOne", 16, 8, 0b0111>;
7581 defm Two : BaseSIMDSt1<"Two", asm, "VecListTwo", 32, 16, 0b1010>;
7582 defm Three : BaseSIMDSt1<"Three", asm, "VecListThree", 48, 24, 0b0110>;
7583 defm Four : BaseSIMDSt1<"Four", asm, "VecListFour", 64, 32, 0b0010>;
7586 multiclass SIMDLd2Multiple<string asm> {
7587 defm Two : BaseSIMDLdN<"Two", asm, "VecListTwo", 32, 16, 0b1000>;
7590 multiclass SIMDSt2Multiple<string asm> {
7591 defm Two : BaseSIMDStN<"Two", asm, "VecListTwo", 32, 16, 0b1000>;
7594 multiclass SIMDLd3Multiple<string asm> {
7595 defm Three : BaseSIMDLdN<"Three", asm, "VecListThree", 48, 24, 0b0100>;
7598 multiclass SIMDSt3Multiple<string asm> {
7599 defm Three : BaseSIMDStN<"Three", asm, "VecListThree", 48, 24, 0b0100>;
7602 multiclass SIMDLd4Multiple<string asm> {
7603 defm Four : BaseSIMDLdN<"Four", asm, "VecListFour", 64, 32, 0b0000>;
7606 multiclass SIMDSt4Multiple<string asm> {
7607 defm Four : BaseSIMDStN<"Four", asm, "VecListFour", 64, 32, 0b0000>;
7611 // AdvSIMD Load/store single-element
7614 class BaseSIMDLdStSingle<bit L, bit R, bits<3> opcode,
7615 string asm, string operands, dag oops, dag iops,
7617 : I<oops, iops, asm, operands, "", pattern> {
7621 let Inst{29-24} = 0b001101;
7624 let Inst{15-13} = opcode;
7625 let Inst{9-5} = vaddr;
7627 let DecoderMethod = "DecodeSIMDLdStSingle";
7630 class BaseSIMDLdStSingleTied<bit L, bit R, bits<3> opcode,
7631 string asm, string operands, dag oops, dag iops,
7633 : I<oops, iops, asm, operands, "$Vt = $dst", pattern> {
7637 let Inst{29-24} = 0b001101;
7640 let Inst{15-13} = opcode;
7641 let Inst{9-5} = vaddr;
7643 let DecoderMethod = "DecodeSIMDLdStSingleTied";
7647 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7648 class BaseSIMDLdR<bit Q, bit R, bits<3> opcode, bit S, bits<2> size, string asm,
7650 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, $vaddr",
7651 (outs listtype:$Vt), (ins am_simdnoindex:$vaddr), []> {
7654 let Inst{20-16} = 0b00000;
7656 let Inst{11-10} = size;
7658 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7659 class BaseSIMDLdRPost<bit Q, bit R, bits<3> opcode, bit S, bits<2> size,
7660 string asm, Operand listtype, Operand GPR64pi>
7661 : BaseSIMDLdStSingle<1, R, opcode, asm, "\t$Vt, $vaddr, $Xm",
7662 (outs listtype:$Vt),
7663 (ins am_simdnoindex:$vaddr, GPR64pi:$Xm), []> {
7667 let Inst{20-16} = Xm;
7669 let Inst{11-10} = size;
7672 multiclass SIMDLdrAliases<string asm, string layout, string Count,
7673 int Offset, int Size> {
7674 // E.g. "ld1r { v0.8b }, [x1], #1"
7675 // "ld1r.8b\t$Vt, $vaddr, #1"
7676 // may get mapped to
7677 // (LD1Rv8b_POST VecListOne8b:$Vt, am_simdnoindex:$vaddr, XZR)
7678 def : InstAlias<asm # "\t$Vt, $vaddr, #" # Offset,
7679 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7680 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
7681 am_simdnoindex:$vaddr, XZR), 1>;
7683 // E.g. "ld1r.8b { v0 }, [x1], #1"
7684 // "ld1r.8b\t$Vt, $vaddr, #1"
7685 // may get mapped to
7686 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, XZR)
7687 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, #" # Offset,
7688 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7689 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7690 am_simdnoindex:$vaddr, XZR), 0>;
7692 // E.g. "ld1r.8b { v0 }, [x1]"
7693 // "ld1r.8b\t$Vt, $vaddr"
7694 // may get mapped to
7695 // (LD1Rv8b VecListOne64:$Vt, am_simdnoindex:$vaddr)
7696 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr",
7697 (!cast<Instruction>(NAME # "v" # layout)
7698 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7699 am_simdnoindex:$vaddr), 0>;
7701 // E.g. "ld1r.8b { v0 }, [x1], x2"
7702 // "ld1r.8b\t$Vt, $vaddr, $Xm"
7703 // may get mapped to
7704 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, GPR64pi1:$Xm)
7705 def : InstAlias<asm # "." # layout # "\t$Vt, $vaddr, $Xm",
7706 (!cast<Instruction>(NAME # "v" # layout # "_POST")
7707 !cast<RegisterOperand>("VecList" # Count # Size):$Vt,
7708 am_simdnoindex:$vaddr,
7709 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
7712 multiclass SIMDLdR<bit R, bits<3> opcode, bit S, string asm, string Count,
7713 int Offset1, int Offset2, int Offset4, int Offset8> {
7714 def v8b : BaseSIMDLdR<0, R, opcode, S, 0b00, asm,
7715 !cast<Operand>("VecList" # Count # "8b")>;
7716 def v16b: BaseSIMDLdR<1, R, opcode, S, 0b00, asm,
7717 !cast<Operand>("VecList" # Count #"16b")>;
7718 def v4h : BaseSIMDLdR<0, R, opcode, S, 0b01, asm,
7719 !cast<Operand>("VecList" # Count #"4h")>;
7720 def v8h : BaseSIMDLdR<1, R, opcode, S, 0b01, asm,
7721 !cast<Operand>("VecList" # Count #"8h")>;
7722 def v2s : BaseSIMDLdR<0, R, opcode, S, 0b10, asm,
7723 !cast<Operand>("VecList" # Count #"2s")>;
7724 def v4s : BaseSIMDLdR<1, R, opcode, S, 0b10, asm,
7725 !cast<Operand>("VecList" # Count #"4s")>;
7726 def v1d : BaseSIMDLdR<0, R, opcode, S, 0b11, asm,
7727 !cast<Operand>("VecList" # Count #"1d")>;
7728 def v2d : BaseSIMDLdR<1, R, opcode, S, 0b11, asm,
7729 !cast<Operand>("VecList" # Count #"2d")>;
7731 def v8b_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b00, asm,
7732 !cast<Operand>("VecList" # Count # "8b"),
7733 !cast<Operand>("GPR64pi" # Offset1)>;
7734 def v16b_POST: BaseSIMDLdRPost<1, R, opcode, S, 0b00, asm,
7735 !cast<Operand>("VecList" # Count # "16b"),
7736 !cast<Operand>("GPR64pi" # Offset1)>;
7737 def v4h_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b01, asm,
7738 !cast<Operand>("VecList" # Count # "4h"),
7739 !cast<Operand>("GPR64pi" # Offset2)>;
7740 def v8h_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b01, asm,
7741 !cast<Operand>("VecList" # Count # "8h"),
7742 !cast<Operand>("GPR64pi" # Offset2)>;
7743 def v2s_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b10, asm,
7744 !cast<Operand>("VecList" # Count # "2s"),
7745 !cast<Operand>("GPR64pi" # Offset4)>;
7746 def v4s_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b10, asm,
7747 !cast<Operand>("VecList" # Count # "4s"),
7748 !cast<Operand>("GPR64pi" # Offset4)>;
7749 def v1d_POST : BaseSIMDLdRPost<0, R, opcode, S, 0b11, asm,
7750 !cast<Operand>("VecList" # Count # "1d"),
7751 !cast<Operand>("GPR64pi" # Offset8)>;
7752 def v2d_POST : BaseSIMDLdRPost<1, R, opcode, S, 0b11, asm,
7753 !cast<Operand>("VecList" # Count # "2d"),
7754 !cast<Operand>("GPR64pi" # Offset8)>;
7756 defm : SIMDLdrAliases<asm, "8b", Count, Offset1, 64>;
7757 defm : SIMDLdrAliases<asm, "16b", Count, Offset1, 128>;
7758 defm : SIMDLdrAliases<asm, "4h", Count, Offset2, 64>;
7759 defm : SIMDLdrAliases<asm, "8h", Count, Offset2, 128>;
7760 defm : SIMDLdrAliases<asm, "2s", Count, Offset4, 64>;
7761 defm : SIMDLdrAliases<asm, "4s", Count, Offset4, 128>;
7762 defm : SIMDLdrAliases<asm, "1d", Count, Offset8, 64>;
7763 defm : SIMDLdrAliases<asm, "2d", Count, Offset8, 128>;
7766 class SIMDLdStSingleB<bit L, bit R, bits<3> opcode, string asm,
7767 dag oops, dag iops, list<dag> pattern>
7768 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7770 // idx encoded in Q:S:size fields.
7772 let Inst{30} = idx{3};
7774 let Inst{20-16} = 0b00000;
7775 let Inst{12} = idx{2};
7776 let Inst{11-10} = idx{1-0};
7778 class SIMDLdStSingleBTied<bit L, bit R, bits<3> opcode, string asm,
7779 dag oops, dag iops, list<dag> pattern>
7780 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7782 // idx encoded in Q:S:size fields.
7784 let Inst{30} = idx{3};
7786 let Inst{20-16} = 0b00000;
7787 let Inst{12} = idx{2};
7788 let Inst{11-10} = idx{1-0};
7790 class SIMDLdStSingleBPost<bit L, bit R, bits<3> opcode, string asm,
7792 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7794 // idx encoded in Q:S:size fields.
7797 let Inst{30} = idx{3};
7799 let Inst{20-16} = Xm;
7800 let Inst{12} = idx{2};
7801 let Inst{11-10} = idx{1-0};
7803 class SIMDLdStSingleBTiedPost<bit L, bit R, bits<3> opcode, string asm,
7805 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7807 // idx encoded in Q:S:size fields.
7810 let Inst{30} = idx{3};
7812 let Inst{20-16} = Xm;
7813 let Inst{12} = idx{2};
7814 let Inst{11-10} = idx{1-0};
7817 class SIMDLdStSingleH<bit L, bit R, bits<3> opcode, bit size, string asm,
7818 dag oops, dag iops, list<dag> pattern>
7819 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7821 // idx encoded in Q:S:size<1> fields.
7823 let Inst{30} = idx{2};
7825 let Inst{20-16} = 0b00000;
7826 let Inst{12} = idx{1};
7827 let Inst{11} = idx{0};
7828 let Inst{10} = size;
7830 class SIMDLdStSingleHTied<bit L, bit R, bits<3> opcode, bit size, string asm,
7831 dag oops, dag iops, list<dag> pattern>
7832 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7834 // idx encoded in Q:S:size<1> fields.
7836 let Inst{30} = idx{2};
7838 let Inst{20-16} = 0b00000;
7839 let Inst{12} = idx{1};
7840 let Inst{11} = idx{0};
7841 let Inst{10} = size;
7844 class SIMDLdStSingleHPost<bit L, bit R, bits<3> opcode, bit size, string asm,
7846 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7848 // idx encoded in Q:S:size<1> fields.
7851 let Inst{30} = idx{2};
7853 let Inst{20-16} = Xm;
7854 let Inst{12} = idx{1};
7855 let Inst{11} = idx{0};
7856 let Inst{10} = size;
7858 class SIMDLdStSingleHTiedPost<bit L, bit R, bits<3> opcode, bit size, string asm,
7860 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7862 // idx encoded in Q:S:size<1> fields.
7865 let Inst{30} = idx{2};
7867 let Inst{20-16} = Xm;
7868 let Inst{12} = idx{1};
7869 let Inst{11} = idx{0};
7870 let Inst{10} = size;
7872 class SIMDLdStSingleS<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7873 dag oops, dag iops, list<dag> pattern>
7874 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7876 // idx encoded in Q:S fields.
7878 let Inst{30} = idx{1};
7880 let Inst{20-16} = 0b00000;
7881 let Inst{12} = idx{0};
7882 let Inst{11-10} = size;
7884 class SIMDLdStSingleSTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7885 dag oops, dag iops, list<dag> pattern>
7886 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7888 // idx encoded in Q:S fields.
7890 let Inst{30} = idx{1};
7892 let Inst{20-16} = 0b00000;
7893 let Inst{12} = idx{0};
7894 let Inst{11-10} = size;
7896 class SIMDLdStSingleSPost<bit L, bit R, bits<3> opcode, bits<2> size,
7897 string asm, dag oops, dag iops>
7898 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7900 // idx encoded in Q:S fields.
7903 let Inst{30} = idx{1};
7905 let Inst{20-16} = Xm;
7906 let Inst{12} = idx{0};
7907 let Inst{11-10} = size;
7909 class SIMDLdStSingleSTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
7910 string asm, dag oops, dag iops>
7911 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7913 // idx encoded in Q:S fields.
7916 let Inst{30} = idx{1};
7918 let Inst{20-16} = Xm;
7919 let Inst{12} = idx{0};
7920 let Inst{11-10} = size;
7922 class SIMDLdStSingleD<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7923 dag oops, dag iops, list<dag> pattern>
7924 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7926 // idx encoded in Q field.
7930 let Inst{20-16} = 0b00000;
7932 let Inst{11-10} = size;
7934 class SIMDLdStSingleDTied<bit L, bit R, bits<3> opcode, bits<2> size, string asm,
7935 dag oops, dag iops, list<dag> pattern>
7936 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr", oops, iops,
7938 // idx encoded in Q field.
7942 let Inst{20-16} = 0b00000;
7944 let Inst{11-10} = size;
7946 class SIMDLdStSingleDPost<bit L, bit R, bits<3> opcode, bits<2> size,
7947 string asm, dag oops, dag iops>
7948 : BaseSIMDLdStSingle<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7950 // idx encoded in Q field.
7955 let Inst{20-16} = Xm;
7957 let Inst{11-10} = size;
7959 class SIMDLdStSingleDTiedPost<bit L, bit R, bits<3> opcode, bits<2> size,
7960 string asm, dag oops, dag iops>
7961 : BaseSIMDLdStSingleTied<L, R, opcode, asm, "\t$Vt$idx, $vaddr, $Xm",
7963 // idx encoded in Q field.
7968 let Inst{20-16} = Xm;
7970 let Inst{11-10} = size;
7973 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7974 multiclass SIMDLdSingleBTied<bit R, bits<3> opcode, string asm,
7975 RegisterOperand listtype,
7976 RegisterOperand GPR64pi> {
7977 def i8 : SIMDLdStSingleBTied<1, R, opcode, asm,
7978 (outs listtype:$dst),
7979 (ins listtype:$Vt, VectorIndexB:$idx,
7980 am_simdnoindex:$vaddr), []>;
7982 def i8_POST : SIMDLdStSingleBTiedPost<1, R, opcode, asm,
7983 (outs listtype:$dst),
7984 (ins listtype:$Vt, VectorIndexB:$idx,
7985 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
7987 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
7988 multiclass SIMDLdSingleHTied<bit R, bits<3> opcode, bit size, string asm,
7989 RegisterOperand listtype,
7990 RegisterOperand GPR64pi> {
7991 def i16 : SIMDLdStSingleHTied<1, R, opcode, size, asm,
7992 (outs listtype:$dst),
7993 (ins listtype:$Vt, VectorIndexH:$idx,
7994 am_simdnoindex:$vaddr), []>;
7996 def i16_POST : SIMDLdStSingleHTiedPost<1, R, opcode, size, asm,
7997 (outs listtype:$dst),
7998 (ins listtype:$Vt, VectorIndexH:$idx,
7999 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8001 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8002 multiclass SIMDLdSingleSTied<bit R, bits<3> opcode, bits<2> size,string asm,
8003 RegisterOperand listtype,
8004 RegisterOperand GPR64pi> {
8005 def i32 : SIMDLdStSingleSTied<1, R, opcode, size, asm,
8006 (outs listtype:$dst),
8007 (ins listtype:$Vt, VectorIndexS:$idx,
8008 am_simdnoindex:$vaddr), []>;
8010 def i32_POST : SIMDLdStSingleSTiedPost<1, R, opcode, size, asm,
8011 (outs listtype:$dst),
8012 (ins listtype:$Vt, VectorIndexS:$idx,
8013 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8015 let mayLoad = 1, mayStore = 0, hasSideEffects = 0 in
8016 multiclass SIMDLdSingleDTied<bit R, bits<3> opcode, bits<2> size, string asm,
8017 RegisterOperand listtype, RegisterOperand GPR64pi> {
8018 def i64 : SIMDLdStSingleDTied<1, R, opcode, size, asm,
8019 (outs listtype:$dst),
8020 (ins listtype:$Vt, VectorIndexD:$idx,
8021 am_simdnoindex:$vaddr), []>;
8023 def i64_POST : SIMDLdStSingleDTiedPost<1, R, opcode, size, asm,
8024 (outs listtype:$dst),
8025 (ins listtype:$Vt, VectorIndexD:$idx,
8026 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8028 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8029 multiclass SIMDStSingleB<bit R, bits<3> opcode, string asm,
8030 RegisterOperand listtype, RegisterOperand GPR64pi> {
8031 def i8 : SIMDLdStSingleB<0, R, opcode, asm,
8032 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
8033 am_simdnoindex:$vaddr), []>;
8035 def i8_POST : SIMDLdStSingleBPost<0, R, opcode, asm,
8036 (outs), (ins listtype:$Vt, VectorIndexB:$idx,
8037 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8039 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8040 multiclass SIMDStSingleH<bit R, bits<3> opcode, bit size, string asm,
8041 RegisterOperand listtype, RegisterOperand GPR64pi> {
8042 def i16 : SIMDLdStSingleH<0, R, opcode, size, asm,
8043 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
8044 am_simdnoindex:$vaddr), []>;
8046 def i16_POST : SIMDLdStSingleHPost<0, R, opcode, size, asm,
8047 (outs), (ins listtype:$Vt, VectorIndexH:$idx,
8048 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8050 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8051 multiclass SIMDStSingleS<bit R, bits<3> opcode, bits<2> size,string asm,
8052 RegisterOperand listtype, RegisterOperand GPR64pi> {
8053 def i32 : SIMDLdStSingleS<0, R, opcode, size, asm,
8054 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
8055 am_simdnoindex:$vaddr), []>;
8057 def i32_POST : SIMDLdStSingleSPost<0, R, opcode, size, asm,
8058 (outs), (ins listtype:$Vt, VectorIndexS:$idx,
8059 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8061 let mayLoad = 0, mayStore = 1, hasSideEffects = 0 in
8062 multiclass SIMDStSingleD<bit R, bits<3> opcode, bits<2> size, string asm,
8063 RegisterOperand listtype, RegisterOperand GPR64pi> {
8064 def i64 : SIMDLdStSingleD<0, R, opcode, size, asm,
8065 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
8066 am_simdnoindex:$vaddr), []>;
8068 def i64_POST : SIMDLdStSingleDPost<0, R, opcode, size, asm,
8069 (outs), (ins listtype:$Vt, VectorIndexD:$idx,
8070 am_simdnoindex:$vaddr, GPR64pi:$Xm)>;
8073 multiclass SIMDLdStSingleAliases<string asm, string layout, string Type,
8074 string Count, int Offset, Operand idxtype> {
8075 // E.g. "ld1 { v0.8b }[0], [x1], #1"
8076 // "ld1\t$Vt, $vaddr, #1"
8077 // may get mapped to
8078 // (LD1Rv8b_POST VecListOne8b:$Vt, am_simdnoindex:$vaddr, XZR)
8079 def : InstAlias<asm # "\t$Vt$idx, $vaddr, #" # Offset,
8080 (!cast<Instruction>(NAME # Type # "_POST")
8081 !cast<RegisterOperand>("VecList" # Count # layout):$Vt,
8082 idxtype:$idx, am_simdnoindex:$vaddr, XZR), 1>;
8084 // E.g. "ld1.8b { v0 }[0], [x1], #1"
8085 // "ld1.8b\t$Vt, $vaddr, #1"
8086 // may get mapped to
8087 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, XZR)
8088 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr, #" # Offset,
8089 (!cast<Instruction>(NAME # Type # "_POST")
8090 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8091 idxtype:$idx, am_simdnoindex:$vaddr, XZR), 0>;
8093 // E.g. "ld1.8b { v0 }[0], [x1]"
8094 // "ld1.8b\t$Vt, $vaddr"
8095 // may get mapped to
8096 // (LD1Rv8b VecListOne64:$Vt, am_simdnoindex:$vaddr)
8097 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr",
8098 (!cast<Instruction>(NAME # Type)
8099 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8100 idxtype:$idx, am_simdnoindex:$vaddr), 0>;
8102 // E.g. "ld1.8b { v0 }[0], [x1], x2"
8103 // "ld1.8b\t$Vt, $vaddr, $Xm"
8104 // may get mapped to
8105 // (LD1Rv8b_POST VecListOne64:$Vt, am_simdnoindex:$vaddr, GPR64pi1:$Xm)
8106 def : InstAlias<asm # "." # layout # "\t$Vt$idx, $vaddr, $Xm",
8107 (!cast<Instruction>(NAME # Type # "_POST")
8108 !cast<RegisterOperand>("VecList" # Count # "128"):$Vt,
8109 idxtype:$idx, am_simdnoindex:$vaddr,
8110 !cast<RegisterOperand>("GPR64pi" # Offset):$Xm), 0>;
8113 multiclass SIMDLdSt1SingleAliases<string asm> {
8114 defm : SIMDLdStSingleAliases<asm, "b", "i8", "One", 1, VectorIndexB>;
8115 defm : SIMDLdStSingleAliases<asm, "h", "i16", "One", 2, VectorIndexH>;
8116 defm : SIMDLdStSingleAliases<asm, "s", "i32", "One", 4, VectorIndexS>;
8117 defm : SIMDLdStSingleAliases<asm, "d", "i64", "One", 8, VectorIndexD>;
8120 multiclass SIMDLdSt2SingleAliases<string asm> {
8121 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Two", 2, VectorIndexB>;
8122 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Two", 4, VectorIndexH>;
8123 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Two", 8, VectorIndexS>;
8124 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Two", 16, VectorIndexD>;
8127 multiclass SIMDLdSt3SingleAliases<string asm> {
8128 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Three", 3, VectorIndexB>;
8129 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Three", 6, VectorIndexH>;
8130 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Three", 12, VectorIndexS>;
8131 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Three", 24, VectorIndexD>;
8134 multiclass SIMDLdSt4SingleAliases<string asm> {
8135 defm : SIMDLdStSingleAliases<asm, "b", "i8", "Four", 4, VectorIndexB>;
8136 defm : SIMDLdStSingleAliases<asm, "h", "i16", "Four", 8, VectorIndexH>;
8137 defm : SIMDLdStSingleAliases<asm, "s", "i32", "Four", 16, VectorIndexS>;
8138 defm : SIMDLdStSingleAliases<asm, "d", "i64", "Four", 32, VectorIndexD>;
8141 //----------------------------------------------------------------------------
8142 // Crypto extensions
8143 //----------------------------------------------------------------------------
8145 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8146 class AESBase<bits<4> opc, string asm, dag outs, dag ins, string cstr,
8148 : I<outs, ins, asm, "{\t$Rd.16b, $Rn.16b|.16b\t$Rd, $Rn}", cstr, pat>,
8152 let Inst{31-16} = 0b0100111000101000;
8153 let Inst{15-12} = opc;
8154 let Inst{11-10} = 0b10;
8159 class AESInst<bits<4> opc, string asm, Intrinsic OpNode>
8160 : AESBase<opc, asm, (outs V128:$Rd), (ins V128:$Rn), "",
8161 [(set (v16i8 V128:$Rd), (OpNode (v16i8 V128:$Rn)))]>;
8163 class AESTiedInst<bits<4> opc, string asm, Intrinsic OpNode>
8164 : AESBase<opc, asm, (outs V128:$dst), (ins V128:$Rd, V128:$Rn),
8166 [(set (v16i8 V128:$dst),
8167 (OpNode (v16i8 V128:$Rd), (v16i8 V128:$Rn)))]>;
8169 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8170 class SHA3OpTiedInst<bits<3> opc, string asm, string dst_lhs_kind,
8171 dag oops, dag iops, list<dag> pat>
8172 : I<oops, iops, asm,
8173 "{\t$Rd" # dst_lhs_kind # ", $Rn" # dst_lhs_kind # ", $Rm.4s" #
8174 "|.4s\t$Rd, $Rn, $Rm}", "$Rd = $dst", pat>,
8179 let Inst{31-21} = 0b01011110000;
8180 let Inst{20-16} = Rm;
8182 let Inst{14-12} = opc;
8183 let Inst{11-10} = 0b00;
8188 class SHATiedInstQSV<bits<3> opc, string asm, Intrinsic OpNode>
8189 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
8190 (ins FPR128:$Rd, FPR32:$Rn, V128:$Rm),
8191 [(set (v4i32 FPR128:$dst),
8192 (OpNode (v4i32 FPR128:$Rd), (i32 FPR32:$Rn),
8193 (v4i32 V128:$Rm)))]>;
8195 class SHATiedInstVVV<bits<3> opc, string asm, Intrinsic OpNode>
8196 : SHA3OpTiedInst<opc, asm, ".4s", (outs V128:$dst),
8197 (ins V128:$Rd, V128:$Rn, V128:$Rm),
8198 [(set (v4i32 V128:$dst),
8199 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn),
8200 (v4i32 V128:$Rm)))]>;
8202 class SHATiedInstQQV<bits<3> opc, string asm, Intrinsic OpNode>
8203 : SHA3OpTiedInst<opc, asm, "", (outs FPR128:$dst),
8204 (ins FPR128:$Rd, FPR128:$Rn, V128:$Rm),
8205 [(set (v4i32 FPR128:$dst),
8206 (OpNode (v4i32 FPR128:$Rd), (v4i32 FPR128:$Rn),
8207 (v4i32 V128:$Rm)))]>;
8209 let mayLoad = 0, mayStore = 0, hasSideEffects = 0 in
8210 class SHA2OpInst<bits<4> opc, string asm, string kind,
8211 string cstr, dag oops, dag iops,
8213 : I<oops, iops, asm, "{\t$Rd" # kind # ", $Rn" # kind #
8214 "|" # kind # "\t$Rd, $Rn}", cstr, pat>,
8218 let Inst{31-16} = 0b0101111000101000;
8219 let Inst{15-12} = opc;
8220 let Inst{11-10} = 0b10;
8225 class SHATiedInstVV<bits<4> opc, string asm, Intrinsic OpNode>
8226 : SHA2OpInst<opc, asm, ".4s", "$Rd = $dst", (outs V128:$dst),
8227 (ins V128:$Rd, V128:$Rn),
8228 [(set (v4i32 V128:$dst),
8229 (OpNode (v4i32 V128:$Rd), (v4i32 V128:$Rn)))]>;
8231 class SHAInstSS<bits<4> opc, string asm, Intrinsic OpNode>
8232 : SHA2OpInst<opc, asm, "", "", (outs FPR32:$Rd), (ins FPR32:$Rn),
8233 [(set (i32 FPR32:$Rd), (OpNode (i32 FPR32:$Rn)))]>;
8235 // Allow the size specifier tokens to be upper case, not just lower.
8236 def : TokenAlias<".8B", ".8b">;
8237 def : TokenAlias<".4H", ".4h">;
8238 def : TokenAlias<".2S", ".2s">;
8239 def : TokenAlias<".1D", ".1d">;
8240 def : TokenAlias<".16B", ".16b">;
8241 def : TokenAlias<".8H", ".8h">;
8242 def : TokenAlias<".4S", ".4s">;
8243 def : TokenAlias<".2D", ".2d">;
8244 def : TokenAlias<".1Q", ".1q">;
8245 def : TokenAlias<".B", ".b">;
8246 def : TokenAlias<".H", ".h">;
8247 def : TokenAlias<".S", ".s">;
8248 def : TokenAlias<".D", ".d">;
8249 def : TokenAlias<".Q", ".q">;