llvm.org GIT mirror llvm / release_35@215010 lib / Target / AArch64 / AArch64SchedCyclone.td
release_35@215010

Tree @release_35@215010 (Download .tar.gz)

AArch64SchedCyclone.td @release_35@215010raw · history · blame

  1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
//=- ARMSchedCyclone.td - AArch64 Cyclone Scheduling Defs ----*- tablegen -*-=//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the machine model for AArch64 Cyclone to support
// instruction scheduling and other instruction cost heuristics.
//
//===----------------------------------------------------------------------===//

def CycloneModel : SchedMachineModel {
  let IssueWidth = 6; // 6 micro-ops are dispatched per cycle.
  let MicroOpBufferSize = 192; // Based on the reorder buffer.
  let LoadLatency = 4; // Optimistic load latency.
  let MispredictPenalty = 16; // 14-19 cycles are typical.
}

//===----------------------------------------------------------------------===//
// Define each kind of processor resource and number available on Cyclone.

// 4 integer pipes
def CyUnitI : ProcResource<4> {
  let BufferSize = 48;
}

// 2 branch units: I[0..1]
def CyUnitB : ProcResource<2> {
  let Super  = CyUnitI;
  let BufferSize = 24;
}

// 1 indirect-branch unit: I[0]
def CyUnitBR : ProcResource<1> {
  let Super  = CyUnitB;
}

// 2 shifter pipes: I[2..3]
// When an instruction consumes a CyUnitIS, it also consumes a CyUnitI
def CyUnitIS : ProcResource<2> {
  let Super = CyUnitI;
  let BufferSize = 24;
}

// 1 mul pipe: I[0]
def CyUnitIM : ProcResource<1> {
  let Super = CyUnitBR;
  let BufferSize = 32;
}

// 1 div pipe: I[1]
def CyUnitID : ProcResource<1> {
  let Super = CyUnitB;
  let BufferSize = 16;
}

// 1 integer division unit. This is driven by the ID pipe, but only
// consumes the pipe for one cycle at issue and another cycle at writeback.
def CyUnitIntDiv : ProcResource<1>;

// 2 ld/st pipes.
def CyUnitLS : ProcResource<2> {
  let BufferSize = 28;
}

// 3 fp/vector pipes.
def CyUnitV : ProcResource<3> {
  let BufferSize = 48;
}
// 2 fp/vector arithmetic and multiply pipes: V[0-1]
def CyUnitVM : ProcResource<2> {
  let Super = CyUnitV;
  let BufferSize = 32;
}
// 1 fp/vector division/sqrt pipe: V[2]
def CyUnitVD : ProcResource<1> {
  let Super = CyUnitV;
  let BufferSize = 16;
}
// 1 fp compare pipe: V[0]
def CyUnitVC : ProcResource<1> {
  let Super = CyUnitVM;
  let BufferSize = 16;
}

// 2 fp division/square-root units.  These are driven by the VD pipe,
// but only consume the pipe for one cycle at issue and a cycle at writeback.
def CyUnitFloatDiv : ProcResource<2>;

//===----------------------------------------------------------------------===//
// Define scheduler read/write resources and latency on Cyclone.
// This mirrors sections 7.7-7.9 of the Tuning Guide v1.0.1.

let SchedModel = CycloneModel in {

//---
// 7.8.1. Moves
//---

// A single nop micro-op (uX).
def WriteX : SchedWriteRes<[]> { let Latency = 0; }

// Move zero is a register rename (to machine register zero).
// The move is replaced by a single nop micro-op.
// MOVZ Rd, #0
// AND Rd, Rzr, #imm
def WriteZPred : SchedPredicate<[{TII->isGPRZero(MI)}]>;
def WriteImmZ  : SchedWriteVariant<[
                   SchedVar<WriteZPred, [WriteX]>,
                   SchedVar<NoSchedPred, [WriteImm]>]>;
def : InstRW<[WriteImmZ], (instrs MOVZWi,MOVZXi,ANDWri,ANDXri)>;

// Move GPR is a register rename and single nop micro-op.
// ORR Xd, XZR, Xm
// ADD Xd, Xn, #0
def WriteIMovPred : SchedPredicate<[{TII->isGPRCopy(MI)}]>;
def WriteVMovPred : SchedPredicate<[{TII->isFPRCopy(MI)}]>;
def WriteMov      : SchedWriteVariant<[
                      SchedVar<WriteIMovPred, [WriteX]>,
                      SchedVar<WriteVMovPred, [WriteX]>,
                      SchedVar<NoSchedPred,   [WriteI]>]>;
def : InstRW<[WriteMov], (instrs COPY,ORRXrr,ADDXrr)>;

// Move non-zero immediate is an integer ALU op.
// MOVN,MOVZ,MOVK
def : WriteRes<WriteImm, [CyUnitI]>;

//---
// 7.8.2-7.8.5. Arithmetic and Logical, Comparison, Conditional,
//              Shifts and Bitfield Operations
//---

// ADR,ADRP
// ADD(S)ri,SUB(S)ri,AND(S)ri,EORri,ORRri
// ADD(S)rr,SUB(S)rr,AND(S)rr,BIC(S)rr,EONrr,EORrr,ORNrr,ORRrr
// ADC(S),SBC(S)
// Aliases: CMN, CMP, TST
//
// Conditional operations.
// CCMNi,CCMPi,CCMNr,CCMPr,
// CSEL,CSINC,CSINV,CSNEG
//
// Bit counting and reversal operations.
// CLS,CLZ,RBIT,REV,REV16,REV32
def : WriteRes<WriteI, [CyUnitI]>;

// ADD with shifted register operand is a single micro-op that
// consumes a shift pipeline for two cycles.
// ADD(S)rs,SUB(S)rs,AND(S)rs,BIC(S)rs,EONrs,EORrs,ORNrs,ORRrs
// EXAMPLE: ADDrs Xn, Xm LSL #imm
def : WriteRes<WriteISReg, [CyUnitIS]> {
  let Latency = 2;
  let ResourceCycles = [2];
}

// ADD with extended register operand is the same as shifted reg operand.
// ADD(S)re,SUB(S)re
// EXAMPLE: ADDXre Xn, Xm, UXTB #1
def : WriteRes<WriteIEReg, [CyUnitIS]> {
  let Latency = 2;
  let ResourceCycles = [2];
}

// Variable shift and bitfield operations.
// ASRV,LSLV,LSRV,RORV,BFM,SBFM,UBFM
def : WriteRes<WriteIS, [CyUnitIS]>;

// EXTR Shifts a pair of registers and requires two micro-ops.
// The second micro-op is delayed, as modeled by ReadExtrHi.
// EXTR Xn, Xm, #imm
def : WriteRes<WriteExtr, [CyUnitIS, CyUnitIS]> {
  let Latency = 2;
  let NumMicroOps = 2;
}

// EXTR's first register read is delayed by one cycle, effectively
// shortening its writer's latency.
// EXTR Xn, Xm, #imm
def : ReadAdvance<ReadExtrHi, 1>;

//---
// 7.8.6. Multiplies
//---

// MUL/MNEG are aliases for MADD/MSUB.
// MADDW,MSUBW,SMADDL,SMSUBL,UMADDL,UMSUBL
def : WriteRes<WriteIM32, [CyUnitIM]> {
  let Latency = 4;
}
// MADDX,MSUBX,SMULH,UMULH
def : WriteRes<WriteIM64, [CyUnitIM]> {
  let Latency = 5;
}

//---
// 7.8.7. Divide
//---

// 32-bit divide takes 7-13 cycles. 10 cycles covers a 20-bit quotient.
// The ID pipe is consumed for 2 cycles: issue and writeback.
// SDIVW,UDIVW
def : WriteRes<WriteID32, [CyUnitID, CyUnitIntDiv]> {
  let Latency = 10;
  let ResourceCycles = [2, 10];
}
// 64-bit divide takes 7-21 cycles. 13 cycles covers a 32-bit quotient.
// The ID pipe is consumed for 2 cycles: issue and writeback.
// SDIVX,UDIVX
def : WriteRes<WriteID64, [CyUnitID, CyUnitIntDiv]> {
  let Latency = 13;
  let ResourceCycles = [2, 13];
}

//---
// 7.8.8,7.8.10. Load/Store, single element
//---

// Integer loads take 4 cycles and use one LS unit for one cycle.
def : WriteRes<WriteLD, [CyUnitLS]> {
  let Latency = 4;
}

// Store-load forwarding is 4 cycles.
//
// Note: The store-exclusive sequence incorporates this
// latency. However, general heuristics should not model the
// dependence between a store and subsequent may-alias load because
// hardware speculation works.
def : WriteRes<WriteST, [CyUnitLS]> {
  let Latency = 4;
}

// Load from base address plus an optionally scaled register offset.
// Rt latency is latency WriteIS + WriteLD.
// EXAMPLE: LDR Xn, Xm [, lsl 3]
def CyWriteLDIdx : SchedWriteVariant<[
  SchedVar<ScaledIdxPred, [WriteIS, WriteLD]>, // Load from scaled register.
  SchedVar<NoSchedPred,   [WriteLD]>]>;        // Load from register offset.
def : SchedAlias<WriteLDIdx, CyWriteLDIdx>;    // Map AArch64->Cyclone type.

// EXAMPLE: STR Xn, Xm [, lsl 3]
def CyWriteSTIdx : SchedWriteVariant<[
  SchedVar<ScaledIdxPred, [WriteIS, WriteST]>, // Store to scaled register.
  SchedVar<NoSchedPred,   [WriteST]>]>;        // Store to register offset.
def : SchedAlias<WriteSTIdx, CyWriteSTIdx>;    // Map AArch64->Cyclone type.

// Read the (unshifted) base register Xn in the second micro-op one cycle later.
// EXAMPLE: LDR Xn, Xm [, lsl 3]
def ReadBaseRS : SchedReadAdvance<1>;
def CyReadAdrBase : SchedReadVariant<[
  SchedVar<ScaledIdxPred, [ReadBaseRS]>, // Read base reg after shifting offset.
  SchedVar<NoSchedPred,   [ReadDefault]>]>;   // Read base reg with no shift.
def : SchedAlias<ReadAdrBase, CyReadAdrBase>; // Map AArch64->Cyclone type.

//---
// 7.8.9,7.8.11. Load/Store, paired
//---

// Address pre/post increment is a simple ALU op with one cycle latency.
def : WriteRes<WriteAdr, [CyUnitI]>;

// LDP high register write is fused with the load, but a nop micro-op remains.
def : WriteRes<WriteLDHi, []> {
  let Latency = 4;
}

// STP is a vector op and store, except for QQ, which is just two stores.
def : SchedAlias<WriteSTP, WriteVSTShuffle>;
def : InstRW<[WriteST, WriteST], (instrs STPQi)>;

//---
// 7.8.13. Branches
//---

// Branches take a single micro-op.
// The misprediction penalty is defined as a SchedMachineModel property.
def : WriteRes<WriteBr,    [CyUnitB]>  {let Latency = 0;}
def : WriteRes<WriteBrReg, [CyUnitBR]> {let Latency = 0;}

//---
// 7.8.14. Never-issued Instructions, Barrier and Hint Operations
//---

// NOP,SEV,SEVL,WFE,WFI,YIELD
def : WriteRes<WriteHint, []> {let Latency = 0;}
// ISB
def : InstRW<[WriteI], (instrs ISB)>;
// SLREX,DMB,DSB
def : WriteRes<WriteBarrier, [CyUnitLS]>;

// System instructions get an invalid latency because the latency of
// other operations across them is meaningless.
def : WriteRes<WriteSys, []> {let Latency = -1;}

//===----------------------------------------------------------------------===//
// 7.9 Vector Unit Instructions

// Simple vector operations take 2 cycles.
def : WriteRes<WriteV, [CyUnitV]> {let Latency = 2;}

// Define some longer latency vector op types for Cyclone.
def CyWriteV3 : SchedWriteRes<[CyUnitV]> {let Latency = 3;}
def CyWriteV4 : SchedWriteRes<[CyUnitV]> {let Latency = 4;}
def CyWriteV5 : SchedWriteRes<[CyUnitV]> {let Latency = 5;}
def CyWriteV6 : SchedWriteRes<[CyUnitV]> {let Latency = 6;}

// Simple floating-point operations take 2 cycles.
def : WriteRes<WriteF, [CyUnitV]> {let Latency = 2;}

//---
// 7.9.1 Vector Moves
//---

// TODO: Add Cyclone-specific zero-cycle zeros. LLVM currently
// generates expensive int-float conversion instead:
// FMOVDi Dd, #0.0
// FMOVv2f64ns Vd.2d, #0.0

// FMOVSi,FMOVDi
def : WriteRes<WriteFImm, [CyUnitV]> {let Latency = 2;}

// MOVI,MVNI are WriteV
// FMOVv2f32ns,FMOVv2f64ns,FMOVv4f32ns are WriteV

// Move FPR is a register rename and single nop micro-op.
// ORR.16b Vd,Vn,Vn
// COPY is handled above in the WriteMov Variant.
def WriteVMov    : SchedWriteVariant<[
                     SchedVar<WriteVMovPred, [WriteX]>,
                     SchedVar<NoSchedPred,   [WriteV]>]>;
def : InstRW<[WriteVMov], (instrs ORRv16i8)>;

// FMOVSr,FMOVDr are WriteF.

// MOV V,V is a WriteV.

// CPY D,V[x] is a WriteV

// INS V[x],V[y] is a WriteV.

// FMOVWSr,FMOVXDr,FMOVXDHighr
def : WriteRes<WriteFCopy, [CyUnitLS]> {
  let Latency = 5;
}

// FMOVSWr,FMOVDXr
def : InstRW<[WriteLD], (instrs FMOVSWr,FMOVDXr,FMOVDXHighr)>;

// INS V[x],R
def CyWriteCopyToFPR : WriteSequence<[WriteVLD, WriteV]>;
def : InstRW<[CyWriteCopyToFPR], (instregex "INSv")>;

// SMOV,UMOV R,V[x]
def CyWriteCopyToGPR : WriteSequence<[WriteLD, WriteI]>;
def : InstRW<[CyWriteCopyToGPR], (instregex "SMOVv","UMOVv")>;

// DUP V,R
def : InstRW<[CyWriteCopyToFPR], (instregex "DUPv")>;

// DUP V,V[x] is a WriteV.

//---
// 7.9.2 Integer Arithmetic, Logical, and Comparisons
//---

// BIC,ORR V,#imm are WriteV

def : InstRW<[CyWriteV3], (instregex "ABSv")>;

// MVN,NEG,NOT are WriteV

def : InstRW<[CyWriteV3], (instregex "SQABSv","SQNEGv")>;

// ADDP is a WriteV.
def CyWriteVADDLP : SchedWriteRes<[CyUnitV]> {let Latency = 2;}
def : InstRW<[CyWriteVADDLP], (instregex "SADDLPv","UADDLPv")>;

def : InstRW<[CyWriteV3],
             (instregex "ADDVv","SMAXVv","UMAXVv","SMINVv","UMINVv")>;

def : InstRW<[CyWriteV3], (instregex "SADDLV","UADDLV")>;

// ADD,SUB are WriteV

// Forward declare.
def CyWriteVABD : SchedWriteRes<[CyUnitV]> {let Latency = 3;}

// Add/Diff and accumulate uses the vector multiply unit.
def CyWriteVAccum : SchedWriteRes<[CyUnitVM]> {let Latency = 3;}
def CyReadVAccum  : SchedReadAdvance<1,
                    [CyWriteVAccum, CyWriteVADDLP, CyWriteVABD]>;

def : InstRW<[CyWriteVAccum, CyReadVAccum],
             (instregex "SADALP","UADALP")>;

def : InstRW<[CyWriteVAccum, CyReadVAccum],
             (instregex "SABAv","UABAv","SABALv","UABALv")>;

def : InstRW<[CyWriteV3], (instregex "SQADDv","SQSUBv","UQADDv","UQSUBv")>;

def : InstRW<[CyWriteV3], (instregex "SUQADDv","USQADDv")>;

def : InstRW<[CyWriteV4], (instregex "ADDHNv","RADDHNv", "RSUBHNv", "SUBHNv")>;

// WriteV includes:
// AND,BIC,CMTST,EOR,ORN,ORR
// ADDP
// SHADD,SHSUB,SRHADD,UHADD,UHSUB,URHADD
// SADDL,SSUBL,UADDL,USUBL
// SADDW,SSUBW,UADDW,USUBW

def : InstRW<[CyWriteV3], (instregex "CMEQv","CMGEv","CMGTv",
                                     "CMLEv","CMLTv",
                                     "CMHIv","CMHSv")>;

def : InstRW<[CyWriteV3], (instregex "SMAXv","SMINv","UMAXv","UMINv",
                                     "SMAXPv","SMINPv","UMAXPv","UMINPv")>;

def : InstRW<[CyWriteVABD], (instregex "SABDv","UABDv",
                                       "SABDLv","UABDLv")>;

//---
// 7.9.3 Floating Point Arithmetic and Comparisons
//---

// FABS,FNEG are WriteF

def : InstRW<[CyWriteV4], (instrs FADDPv2i32p)>;
def : InstRW<[CyWriteV5], (instrs FADDPv2i64p)>;

def : InstRW<[CyWriteV3], (instregex "FMAXPv2i","FMAXNMPv2i",
                                     "FMINPv2i","FMINNMPv2i")>;

def : InstRW<[CyWriteV4], (instregex "FMAXVv","FMAXNMVv","FMINVv","FMINNMVv")>;

def : InstRW<[CyWriteV4], (instrs FADDSrr,FADDv2f32,FADDv4f32,
                                  FSUBSrr,FSUBv2f32,FSUBv4f32,
                                  FADDPv2f32,FADDPv4f32,
                                  FABD32,FABDv2f32,FABDv4f32)>;
def : InstRW<[CyWriteV5], (instrs FADDDrr,FADDv2f64,
                                  FSUBDrr,FSUBv2f64,
                                  FADDPv2f64,
                                  FABD64,FABDv2f64)>;

def : InstRW<[CyWriteV3], (instregex "FCMEQ","FCMGT","FCMLE","FCMLT")>;

def : InstRW<[CyWriteV3], (instregex "FACGE","FACGT",
                                     "FMAXS","FMAXD","FMAXv",
                                     "FMINS","FMIND","FMINv",
                                     "FMAXNMS","FMAXNMD","FMAXNMv",
                                     "FMINNMS","FMINNMD","FMINNMv",
                                     "FMAXPv2f","FMAXPv4f",
                                     "FMINPv2f","FMINPv4f",
                                     "FMAXNMPv2f","FMAXNMPv4f",
                                     "FMINNMPv2f","FMINNMPv4f")>;

// FCMP,FCMPE,FCCMP,FCCMPE
def : WriteRes<WriteFCmp, [CyUnitVC]> {let Latency = 4;}

// FCSEL is a WriteF.

//---
// 7.9.4 Shifts and Bitfield Operations
//---

// SHL is a WriteV

def CyWriteVSHR : SchedWriteRes<[CyUnitV]> {let Latency = 2;}
def : InstRW<[CyWriteVSHR], (instregex "SSHRv","USHRv")>;

def CyWriteVSRSHR : SchedWriteRes<[CyUnitV]> {let Latency = 3;}
def : InstRW<[CyWriteVSRSHR], (instregex "SRSHRv","URSHRv")>;

// Shift and accumulate uses the vector multiply unit.
def CyWriteVShiftAcc : SchedWriteRes<[CyUnitVM]> {let Latency = 3;}
def CyReadVShiftAcc  : SchedReadAdvance<1,
                        [CyWriteVShiftAcc, CyWriteVSHR, CyWriteVSRSHR]>;
def : InstRW<[CyWriteVShiftAcc, CyReadVShiftAcc],
             (instregex "SRSRAv","SSRAv","URSRAv","USRAv")>;

// SSHL,USHL are WriteV.

def : InstRW<[CyWriteV3], (instregex "SRSHLv","URSHLv")>;

// SQSHL,SQSHLU,UQSHL are WriteV.

def : InstRW<[CyWriteV3], (instregex "SQRSHLv","UQRSHLv")>;

// WriteV includes:
// SHLL,SSHLL,USHLL
// SLI,SRI
// BIF,BIT,BSL
// EXT
// CLS,CLZ,CNT,RBIT,REV16,REV32,REV64,XTN
// XTN2

def : InstRW<[CyWriteV4],
             (instregex "RSHRNv","SHRNv",
                        "SQRSHRNv","SQRSHRUNv","SQSHRNv","SQSHRUNv",
                        "UQRSHRNv","UQSHRNv","SQXTNv","SQXTUNv","UQXTNv")>;

//---
// 7.9.5 Multiplication
//---

def CyWriteVMul : SchedWriteRes<[CyUnitVM]> { let Latency = 4;}
def : InstRW<[CyWriteVMul], (instregex "MULv","SMULLv","UMULLv",
                             "SQDMULLv","SQDMULHv","SQRDMULHv")>;

// FMUL,FMULX,FNMUL default to WriteFMul.
def : WriteRes<WriteFMul, [CyUnitVM]> { let Latency = 4;}

def CyWriteV64Mul : SchedWriteRes<[CyUnitVM]> { let Latency = 5;}
def : InstRW<[CyWriteV64Mul], (instrs FMULDrr,FMULv2f64,FMULv2i64_indexed,
                               FNMULDrr,FMULX64,FMULXv2f64,FMULXv2i64_indexed)>;

def CyReadVMulAcc : SchedReadAdvance<1, [CyWriteVMul, CyWriteV64Mul]>;
def : InstRW<[CyWriteVMul, CyReadVMulAcc],
             (instregex "MLA","MLS","SMLAL","SMLSL","UMLAL","UMLSL",
              "SQDMLAL","SQDMLSL")>;

def CyWriteSMul : SchedWriteRes<[CyUnitVM]> { let Latency = 8;}
def CyWriteDMul : SchedWriteRes<[CyUnitVM]> { let Latency = 10;}
def CyReadSMul : SchedReadAdvance<4, [CyWriteSMul]>;
def CyReadDMul : SchedReadAdvance<5, [CyWriteDMul]>;

def : InstRW<[CyWriteSMul, CyReadSMul],
             (instrs FMADDSrrr,FMSUBSrrr,FNMADDSrrr,FNMSUBSrrr,
              FMLAv2f32,FMLAv4f32,
              FMLAv1i32_indexed,FMLAv1i64_indexed,FMLAv2i32_indexed)>;
def : InstRW<[CyWriteDMul, CyReadDMul],
             (instrs FMADDDrrr,FMSUBDrrr,FNMADDDrrr,FNMSUBDrrr,
              FMLAv2f64,FMLAv2i64_indexed,
              FMLSv2f64,FMLSv2i64_indexed)>;

def CyWritePMUL : SchedWriteRes<[CyUnitVD]> { let Latency = 3; }
def : InstRW<[CyWritePMUL], (instregex "PMULv", "PMULLv")>;

//---
// 7.9.6 Divide and Square Root
//---

// FDIV,FSQRT
// TODO: Add 64-bit variant with 19 cycle latency.
// TODO: Specialize FSQRT for longer latency.
def : WriteRes<WriteFDiv, [CyUnitVD, CyUnitFloatDiv]> {
  let Latency = 17;
  let ResourceCycles = [2, 17];
}

def : InstRW<[CyWriteV4], (instregex "FRECPEv","FRECPXv","URECPEv","URSQRTEv")>;

def WriteFRSQRTE : SchedWriteRes<[CyUnitVM]> { let Latency = 4; }
def : InstRW<[WriteFRSQRTE], (instregex "FRSQRTEv")>;

def WriteFRECPS : SchedWriteRes<[CyUnitVM]> { let Latency = 8; }
def WriteFRSQRTS : SchedWriteRes<[CyUnitVM]> { let Latency = 10; }
def : InstRW<[WriteFRECPS],  (instregex "FRECPSv")>;
def : InstRW<[WriteFRSQRTS], (instregex "FRSQRTSv")>;

//---
// 7.9.7 Integer-FP Conversions
//---

// FCVT lengthen f16/s32
def : InstRW<[WriteV], (instrs FCVTSHr,FCVTDHr,FCVTDSr)>;

// FCVT,FCVTN,FCVTXN
// SCVTF,UCVTF V,V
// FRINT(AIMNPXZ) V,V
def : WriteRes<WriteFCvt, [CyUnitV]> {let Latency = 4;}

// SCVT/UCVT S/D, Rd = VLD5+V4: 9 cycles.
def CyWriteCvtToFPR : WriteSequence<[WriteVLD, CyWriteV4]>;
def : InstRW<[CyWriteCopyToFPR], (instregex "FCVT[AMNPZ][SU][SU][WX][SD]r")>;

// FCVT Rd, S/D = V6+LD4: 10 cycles
def CyWriteCvtToGPR : WriteSequence<[CyWriteV6, WriteLD]>;
def : InstRW<[CyWriteCvtToGPR], (instregex "[SU]CVTF[SU][WX][SD]r")>;

// FCVTL is a WriteV

//---
// 7.9.8-7.9.10 Cryptography, Data Transposition, Table Lookup
//---

def CyWriteCrypto2 : SchedWriteRes<[CyUnitVD]> {let Latency = 2;}
def : InstRW<[CyWriteCrypto2], (instrs AESIMCrr, AESMCrr, SHA1Hrr,
                                       AESDrr, AESErr, SHA1SU1rr, SHA256SU0rr,
                                       SHA1SU0rrr)>;

def CyWriteCrypto3 : SchedWriteRes<[CyUnitVD]> {let Latency = 3;}
def : InstRW<[CyWriteCrypto3], (instrs SHA256SU1rrr)>;

def CyWriteCrypto6 : SchedWriteRes<[CyUnitVD]> {let Latency = 6;}
def : InstRW<[CyWriteCrypto6], (instrs SHA1Crrr, SHA1Mrrr, SHA1Prrr,
                                       SHA256Hrrr,SHA256H2rrr)>;

// TRN,UZP,ZUP are WriteV.

// TBL,TBX are WriteV.

//---
// 7.9.11-7.9.14 Load/Store, single element and paired
//---

// Loading into the vector unit takes 5 cycles vs 4 for integer loads.
def : WriteRes<WriteVLD, [CyUnitLS]> {
  let Latency = 5;
}

// Store-load forwarding is 4 cycles.
def : WriteRes<WriteVST, [CyUnitLS]> {
  let Latency = 4;
}

// WriteVLDPair/VSTPair sequences are expanded by the target description.

//---
// 7.9.15 Load, element operations
//---

// Only the first WriteVLD and WriteAdr for writeback matches def operands.
// Subsequent WriteVLDs consume resources. Since all loaded values have the
// same latency, this is acceptable.

// Vd is read 5 cycles after issuing the vector load.
def : ReadAdvance<ReadVLD, 5>;

def : InstRW<[WriteVLD],
             (instregex "LD1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLD, WriteAdr],
             (instregex "LD1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST")>;

// Register writes from the load's high half are fused micro-ops.
def : InstRW<[WriteVLD],
             (instregex "LD1Twov(8b|4h|2s|1d)$")>;
def : InstRW<[WriteVLD, WriteAdr],
             (instregex "LD1Twov(8b|4h|2s|1d)_POST")>;
def : InstRW<[WriteVLD, WriteVLD],
             (instregex "LD1Twov(16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLD, WriteAdr, WriteVLD],
             (instregex "LD1Twov(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVLD, WriteVLD],
             (instregex "LD1Threev(8b|4h|2s|1d)$")>;
def : InstRW<[WriteVLD, WriteAdr, WriteVLD],
             (instregex "LD1Threev(8b|4h|2s|1d)_POST")>;
def : InstRW<[WriteVLD, WriteVLD, WriteVLD],
             (instregex "LD1Threev(16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLD, WriteAdr, WriteVLD, WriteVLD],
             (instregex "LD1Threev(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVLD, WriteVLD],
             (instregex "LD1Fourv(8b|4h|2s|1d)$")>;
def : InstRW<[WriteVLD, WriteAdr, WriteVLD],
             (instregex "LD1Fourv(8b|4h|2s|1d)_POST")>;
def : InstRW<[WriteVLD, WriteVLD, WriteVLD, WriteVLD],
             (instregex "LD1Fourv(16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLD, WriteAdr, WriteVLD, WriteVLD, WriteVLD],
             (instregex "LD1Fourv(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVLDShuffle, ReadVLD],
             (instregex "LD1i(8|16|32)$")>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr],
             (instregex "LD1i(8|16|32)_POST")>;

def : InstRW<[WriteVLDShuffle, ReadVLD],          (instrs LD1i64)>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr],(instrs LD1i64_POST)>;

def : InstRW<[WriteVLDShuffle],
             (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr],
             (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;

def : InstRW<[WriteVLDShuffle, WriteV],
             (instregex "LD2Twov(8b|4h|2s)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteV],
             (instregex "LD2Twov(8b|4h|2s)_POST$")>;
def : InstRW<[WriteVLDShuffle, WriteVLDShuffle],
             (instregex "LD2Twov(16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteVLDShuffle],
             (instregex "LD2Twov(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVLDShuffle, ReadVLD, WriteV],
             (instregex "LD2i(8|16|32)$")>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr, WriteV],
             (instregex "LD2i(8|16|32)_POST")>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteV],
             (instregex "LD2i64$")>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr, WriteV],
             (instregex "LD2i64_POST")>;

def : InstRW<[WriteVLDShuffle, WriteV],
             (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteV],
             (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVLDShuffle, WriteVLDShuffle, WriteV],
             (instregex "LD3Threev(8b|4h|2s)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteVLDShuffle, WriteV],
             (instregex "LD3Threev(8b|4h|2s)_POST")>;
def : InstRW<[WriteVLDShuffle, WriteVLDShuffle, WriteVLDShuffle],
             (instregex "LD3Threev(16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteVLDShuffle, WriteVLDShuffle],
             (instregex "LD3Threev(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVLDShuffle, ReadVLD, WriteV, WriteV],
             (instregex "LD3i(8|16|32)$")>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr, WriteV, WriteV],
             (instregex "LD3i(8|16|32)_POST")>;

def : InstRW<[WriteVLDShuffle, ReadVLD, WriteVLDShuffle, WriteV],
             (instregex "LD3i64$")>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr, WriteVLDShuffle, WriteV],
             (instregex "LD3i64_POST")>;

def : InstRW<[WriteVLDShuffle, WriteV, WriteV],
             (instregex "LD3Rv(8b|4h|2s|16b|8h|4s)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteV, WriteV],
             (instregex "LD3Rv(8b|4h|2s|16b|8h|4s)_POST")>;

def : InstRW<[WriteVLDShuffle, WriteVLDShuffle, WriteV],
             (instrs LD3Rv1d,LD3Rv2d)>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteVLDShuffle, WriteV],
             (instrs LD3Rv2d_POST,LD3Rv2d_POST)>;

def : InstRW<[WriteVLDShuffle, WriteVLDShuffle, WriteV, WriteV],
             (instregex "LD4Fourv(8b|4h|2s)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteVLDShuffle, WriteV, WriteV],
             (instregex "LD4Fourv(8b|4h|2s)_POST")>;
def : InstRW<[WriteVLDPairShuffle, WriteVLDPairShuffle,
              WriteVLDPairShuffle, WriteVLDPairShuffle],
             (instregex "LD4Fourv(16b|8h|4s|2d)$")>;
def : InstRW<[WriteVLDPairShuffle, WriteAdr, WriteVLDPairShuffle,
              WriteVLDPairShuffle, WriteVLDPairShuffle],
             (instregex "LD4Fourv(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVLDShuffle, ReadVLD, WriteV, WriteV, WriteV],
             (instregex "LD4i(8|16|32)$")>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr, WriteV, WriteV, WriteV],
             (instregex "LD4i(8|16|32)_POST")>;


def : InstRW<[WriteVLDShuffle, ReadVLD, WriteVLDShuffle, WriteV, WriteV],
             (instrs LD4i64)>;
def : InstRW<[WriteVLDShuffle, ReadVLD, WriteAdr, WriteVLDShuffle, WriteV],
             (instrs LD4i64_POST)>;

def : InstRW<[WriteVLDShuffle, WriteV, WriteV, WriteV],
             (instregex "LD4Rv(8b|4h|2s|16b|8h|4s)$")>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteV, WriteV, WriteV],
             (instregex "LD4Rv(8b|4h|2s|16b|8h|4s)_POST")>;

def : InstRW<[WriteVLDShuffle, WriteVLDShuffle, WriteV, WriteV],
             (instrs LD4Rv1d,LD4Rv2d)>;
def : InstRW<[WriteVLDShuffle, WriteAdr, WriteVLDShuffle, WriteV, WriteV],
             (instrs LD4Rv1d_POST,LD4Rv2d_POST)>;

//---
// 7.9.16 Store, element operations
//---

// Only the WriteAdr for writeback matches a def operands.
// Subsequent WriteVLDs only consume resources.

def : InstRW<[WriteVST],
             (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
def : InstRW<[WriteAdr, WriteVST],
             (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVSTShuffle],
             (instregex "ST1Twov(8b|4h|2s|1d)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle],
             (instregex "ST1Twov(8b|4h|2s|1d)_POST")>;
def : InstRW<[WriteVST, WriteVST],
             (instregex "ST1Twov(16b|8h|4s|2d)$")>;
def : InstRW<[WriteAdr, WriteVST, WriteVST],
             (instregex "ST1Twov(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVSTShuffle, WriteVST],
             (instregex "ST1Threev(8b|4h|2s|1d)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle, WriteVST],
             (instregex "ST1Threev(8b|4h|2s|1d)_POST")>;
def : InstRW<[WriteVST, WriteVST, WriteVST],
             (instregex "ST1Threev(16b|8h|4s|2d)$")>;
def : InstRW<[WriteAdr, WriteVST, WriteVST, WriteVST],
             (instregex "ST1Threev(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST1Fourv(8b|4h|2s|1d)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST1Fourv(8b|4h|2s|1d)_POST")>;
def : InstRW<[WriteVST, WriteVST, WriteVST, WriteVST],
             (instregex "ST1Fourv(16b|8h|4s|2d)$")>;
def : InstRW<[WriteAdr, WriteVST, WriteVST, WriteVST, WriteVST],
             (instregex "ST1Fourv(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVSTShuffle],           (instregex "ST1i(8|16|32)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle], (instregex "ST1i(8|16|32)_POST")>;

def : InstRW<[WriteVSTShuffle],           (instrs ST1i64)>;
def : InstRW<[WriteAdr, WriteVSTShuffle], (instrs ST1i64_POST)>;

def : InstRW<[WriteVSTShuffle],
             (instregex "ST2Twov(8b|4h|2s)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle],
             (instregex "ST2Twov(8b|4h|2s)_POST")>;
def : InstRW<[WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST2Twov(16b|8h|4s|2d)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST2Twov(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVSTShuffle],           (instregex "ST2i(8|16|32)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle], (instregex "ST2i(8|16|32)_POST")>;
def : InstRW<[WriteVSTShuffle],           (instrs ST2i64)>;
def : InstRW<[WriteAdr, WriteVSTShuffle], (instrs ST2i64_POST)>;

def : InstRW<[WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST3Threev(8b|4h|2s)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST3Threev(8b|4h|2s)_POST")>;
def : InstRW<[WriteVSTShuffle, WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST3Threev(16b|8h|4s|2d)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle, WriteVSTShuffle, WriteVSTShuffle],
             (instregex "ST3Threev(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVSTShuffle],           (instregex "ST3i(8|16|32)$")>;
def : InstRW<[WriteAdr, WriteVSTShuffle], (instregex "ST3i(8|16|32)_POST")>;

def :InstRW<[WriteVSTShuffle, WriteVSTShuffle],           (instrs ST3i64)>;
def :InstRW<[WriteAdr, WriteVSTShuffle, WriteVSTShuffle], (instrs ST3i64_POST)>;

def : InstRW<[WriteVSTPairShuffle, WriteVSTPairShuffle],
            (instregex "ST4Fourv(8b|4h|2s|1d)$")>;
def : InstRW<[WriteAdr, WriteVSTPairShuffle, WriteVSTPairShuffle],
            (instregex "ST4Fourv(8b|4h|2s|1d)_POST")>;
def : InstRW<[WriteVSTPairShuffle, WriteVSTPairShuffle,
              WriteVSTPairShuffle, WriteVSTPairShuffle],
             (instregex "ST4Fourv(16b|8h|4s|2d)$")>;
def : InstRW<[WriteAdr, WriteVSTPairShuffle, WriteVSTPairShuffle,
              WriteVSTPairShuffle, WriteVSTPairShuffle],
             (instregex "ST4Fourv(16b|8h|4s|2d)_POST")>;

def : InstRW<[WriteVSTPairShuffle],           (instregex "ST4i(8|16|32)$")>;
def : InstRW<[WriteAdr, WriteVSTPairShuffle], (instregex "ST4i(8|16|32)_POST")>;

def : InstRW<[WriteVSTShuffle, WriteVSTShuffle],          (instrs ST4i64)>;
def : InstRW<[WriteAdr, WriteVSTShuffle, WriteVSTShuffle],(instrs ST4i64_POST)>;

//---
// Unused SchedRead types
//---

def : ReadAdvance<ReadI, 0>;
def : ReadAdvance<ReadISReg, 0>;
def : ReadAdvance<ReadIEReg, 0>;
def : ReadAdvance<ReadIM, 0>;
def : ReadAdvance<ReadIMA, 0>;
def : ReadAdvance<ReadID, 0>;

} // SchedModel = CycloneModel