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
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
| //===- HexagonMachineScheduler.cpp - MI Scheduler for Hexagon -------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// MachineScheduler schedules machine instructions after phi elimination. It
// preserves LiveIntervals so it can be invoked before register allocation.
//
//===----------------------------------------------------------------------===//
#include "HexagonMachineScheduler.h"
#include "HexagonInstrInfo.h"
#include "HexagonSubtarget.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/CodeGen/DFAPacketizer.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineLoopInfo.h"
#include "llvm/CodeGen/RegisterClassInfo.h"
#include "llvm/CodeGen/RegisterPressure.h"
#include "llvm/CodeGen/ScheduleDAG.h"
#include "llvm/CodeGen/ScheduleHazardRecognizer.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSchedule.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/IR/Function.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm>
#include <cassert>
#include <iomanip>
#include <limits>
#include <memory>
#include <sstream>
using namespace llvm;
#define DEBUG_TYPE "machine-scheduler"
static cl::opt<bool> IgnoreBBRegPressure("ignore-bb-reg-pressure",
cl::Hidden, cl::ZeroOrMore, cl::init(false));
static cl::opt<bool> UseNewerCandidate("use-newer-candidate",
cl::Hidden, cl::ZeroOrMore, cl::init(true));
static cl::opt<unsigned> SchedDebugVerboseLevel("misched-verbose-level",
cl::Hidden, cl::ZeroOrMore, cl::init(1));
// Check if the scheduler should penalize instructions that are available to
// early due to a zero-latency dependence.
static cl::opt<bool> CheckEarlyAvail("check-early-avail", cl::Hidden,
cl::ZeroOrMore, cl::init(true));
// This value is used to determine if a register class is a high pressure set.
// We compute the maximum number of registers needed and divided by the total
// available. Then, we compare the result to this value.
static cl::opt<float> RPThreshold("hexagon-reg-pressure", cl::Hidden,
cl::init(0.75f), cl::desc("High register pressure threhold."));
/// Return true if there is a dependence between SUd and SUu.
static bool hasDependence(const SUnit *SUd, const SUnit *SUu,
const HexagonInstrInfo &QII) {
if (SUd->Succs.size() == 0)
return false;
// Enable .cur formation.
if (QII.mayBeCurLoad(*SUd->getInstr()))
return false;
if (QII.canExecuteInBundle(*SUd->getInstr(), *SUu->getInstr()))
return false;
for (const auto &S : SUd->Succs) {
// Since we do not add pseudos to packets, might as well
// ignore order dependencies.
if (S.isCtrl())
continue;
if (S.getSUnit() == SUu && S.getLatency() > 0)
return true;
}
return false;
}
/// Check if scheduling of this SU is possible
/// in the current packet.
/// It is _not_ precise (statefull), it is more like
/// another heuristic. Many corner cases are figured
/// empirically.
bool VLIWResourceModel::isResourceAvailable(SUnit *SU, bool IsTop) {
if (!SU || !SU->getInstr())
return false;
// First see if the pipeline could receive this instruction
// in the current cycle.
switch (SU->getInstr()->getOpcode()) {
default:
if (!ResourcesModel->canReserveResources(*SU->getInstr()))
return false;
break;
case TargetOpcode::EXTRACT_SUBREG:
case TargetOpcode::INSERT_SUBREG:
case TargetOpcode::SUBREG_TO_REG:
case TargetOpcode::REG_SEQUENCE:
case TargetOpcode::IMPLICIT_DEF:
case TargetOpcode::COPY:
case TargetOpcode::INLINEASM:
case TargetOpcode::INLINEASM_BR:
break;
}
MachineBasicBlock *MBB = SU->getInstr()->getParent();
auto &QST = MBB->getParent()->getSubtarget<HexagonSubtarget>();
const auto &QII = *QST.getInstrInfo();
// Now see if there are no other dependencies to instructions already
// in the packet.
if (IsTop) {
for (unsigned i = 0, e = Packet.size(); i != e; ++i)
if (hasDependence(Packet[i], SU, QII))
return false;
} else {
for (unsigned i = 0, e = Packet.size(); i != e; ++i)
if (hasDependence(SU, Packet[i], QII))
return false;
}
return true;
}
/// Keep track of available resources.
bool VLIWResourceModel::reserveResources(SUnit *SU, bool IsTop) {
bool startNewCycle = false;
// Artificially reset state.
if (!SU) {
ResourcesModel->clearResources();
Packet.clear();
TotalPackets++;
return false;
}
// If this SU does not fit in the packet or the packet is now full
// start a new one.
if (!isResourceAvailable(SU, IsTop) ||
Packet.size() >= SchedModel->getIssueWidth()) {
ResourcesModel->clearResources();
Packet.clear();
TotalPackets++;
startNewCycle = true;
}
switch (SU->getInstr()->getOpcode()) {
default:
ResourcesModel->reserveResources(*SU->getInstr());
break;
case TargetOpcode::EXTRACT_SUBREG:
case TargetOpcode::INSERT_SUBREG:
case TargetOpcode::SUBREG_TO_REG:
case TargetOpcode::REG_SEQUENCE:
case TargetOpcode::IMPLICIT_DEF:
case TargetOpcode::KILL:
case TargetOpcode::CFI_INSTRUCTION:
case TargetOpcode::EH_LABEL:
case TargetOpcode::COPY:
case TargetOpcode::INLINEASM:
case TargetOpcode::INLINEASM_BR:
break;
}
Packet.push_back(SU);
#ifndef NDEBUG
LLVM_DEBUG(dbgs() << "Packet[" << TotalPackets << "]:\n");
for (unsigned i = 0, e = Packet.size(); i != e; ++i) {
LLVM_DEBUG(dbgs() << "\t[" << i << "] SU(");
LLVM_DEBUG(dbgs() << Packet[i]->NodeNum << ")\t");
LLVM_DEBUG(Packet[i]->getInstr()->dump());
}
#endif
return startNewCycle;
}
/// schedule - Called back from MachineScheduler::runOnMachineFunction
/// after setting up the current scheduling region. [RegionBegin, RegionEnd)
/// only includes instructions that have DAG nodes, not scheduling boundaries.
void VLIWMachineScheduler::schedule() {
LLVM_DEBUG(dbgs() << "********** MI Converging Scheduling VLIW "
<< printMBBReference(*BB) << " " << BB->getName()
<< " in_func " << BB->getParent()->getName()
<< " at loop depth " << MLI->getLoopDepth(BB) << " \n");
buildDAGWithRegPressure();
Topo.InitDAGTopologicalSorting();
// Postprocess the DAG to add platform-specific artificial dependencies.
postprocessDAG();
SmallVector<SUnit*, 8> TopRoots, BotRoots;
findRootsAndBiasEdges(TopRoots, BotRoots);
// Initialize the strategy before modifying the DAG.
SchedImpl->initialize(this);
LLVM_DEBUG(unsigned maxH = 0;
for (unsigned su = 0, e = SUnits.size(); su != e;
++su) if (SUnits[su].getHeight() > maxH) maxH =
SUnits[su].getHeight();
dbgs() << "Max Height " << maxH << "\n";);
LLVM_DEBUG(unsigned maxD = 0;
for (unsigned su = 0, e = SUnits.size(); su != e;
++su) if (SUnits[su].getDepth() > maxD) maxD =
SUnits[su].getDepth();
dbgs() << "Max Depth " << maxD << "\n";);
LLVM_DEBUG(dump());
initQueues(TopRoots, BotRoots);
bool IsTopNode = false;
while (true) {
LLVM_DEBUG(
dbgs() << "** VLIWMachineScheduler::schedule picking next node\n");
SUnit *SU = SchedImpl->pickNode(IsTopNode);
if (!SU) break;
if (!checkSchedLimit())
break;
scheduleMI(SU, IsTopNode);
// Notify the scheduling strategy after updating the DAG.
SchedImpl->schedNode(SU, IsTopNode);
updateQueues(SU, IsTopNode);
}
assert(CurrentTop == CurrentBottom && "Nonempty unscheduled zone.");
placeDebugValues();
LLVM_DEBUG({
dbgs() << "*** Final schedule for "
<< printMBBReference(*begin()->getParent()) << " ***\n";
dumpSchedule();
dbgs() << '\n';
});
}
void ConvergingVLIWScheduler::initialize(ScheduleDAGMI *dag) {
DAG = static_cast<VLIWMachineScheduler*>(dag);
SchedModel = DAG->getSchedModel();
Top.init(DAG, SchedModel);
Bot.init(DAG, SchedModel);
// Initialize the HazardRecognizers. If itineraries don't exist, are empty, or
// are disabled, then these HazardRecs will be disabled.
const InstrItineraryData *Itin = DAG->getSchedModel()->getInstrItineraries();
const TargetSubtargetInfo &STI = DAG->MF.getSubtarget();
const TargetInstrInfo *TII = STI.getInstrInfo();
delete Top.HazardRec;
delete Bot.HazardRec;
Top.HazardRec = TII->CreateTargetMIHazardRecognizer(Itin, DAG);
Bot.HazardRec = TII->CreateTargetMIHazardRecognizer(Itin, DAG);
delete Top.ResourceModel;
delete Bot.ResourceModel;
Top.ResourceModel = new VLIWResourceModel(STI, DAG->getSchedModel());
Bot.ResourceModel = new VLIWResourceModel(STI, DAG->getSchedModel());
const std::vector<unsigned> &MaxPressure =
DAG->getRegPressure().MaxSetPressure;
HighPressureSets.assign(MaxPressure.size(), 0);
for (unsigned i = 0, e = MaxPressure.size(); i < e; ++i) {
unsigned Limit = DAG->getRegClassInfo()->getRegPressureSetLimit(i);
HighPressureSets[i] =
((float) MaxPressure[i] > ((float) Limit * RPThreshold));
}
assert((!ForceTopDown || !ForceBottomUp) &&
"-misched-topdown incompatible with -misched-bottomup");
}
void ConvergingVLIWScheduler::releaseTopNode(SUnit *SU) {
if (SU->isScheduled)
return;
for (const SDep &PI : SU->Preds) {
unsigned PredReadyCycle = PI.getSUnit()->TopReadyCycle;
unsigned MinLatency = PI.getLatency();
#ifndef NDEBUG
Top.MaxMinLatency = std::max(MinLatency, Top.MaxMinLatency);
#endif
if (SU->TopReadyCycle < PredReadyCycle + MinLatency)
SU->TopReadyCycle = PredReadyCycle + MinLatency;
}
Top.releaseNode(SU, SU->TopReadyCycle);
}
void ConvergingVLIWScheduler::releaseBottomNode(SUnit *SU) {
if (SU->isScheduled)
return;
assert(SU->getInstr() && "Scheduled SUnit must have instr");
for (SUnit::succ_iterator I = SU->Succs.begin(), E = SU->Succs.end();
I != E; ++I) {
unsigned SuccReadyCycle = I->getSUnit()->BotReadyCycle;
unsigned MinLatency = I->getLatency();
#ifndef NDEBUG
Bot.MaxMinLatency = std::max(MinLatency, Bot.MaxMinLatency);
#endif
if (SU->BotReadyCycle < SuccReadyCycle + MinLatency)
SU->BotReadyCycle = SuccReadyCycle + MinLatency;
}
Bot.releaseNode(SU, SU->BotReadyCycle);
}
/// Does this SU have a hazard within the current instruction group.
///
/// The scheduler supports two modes of hazard recognition. The first is the
/// ScheduleHazardRecognizer API. It is a fully general hazard recognizer that
/// supports highly complicated in-order reservation tables
/// (ScoreboardHazardRecognizer) and arbitrary target-specific logic.
///
/// The second is a streamlined mechanism that checks for hazards based on
/// simple counters that the scheduler itself maintains. It explicitly checks
/// for instruction dispatch limitations, including the number of micro-ops that
/// can dispatch per cycle.
///
/// TODO: Also check whether the SU must start a new group.
bool ConvergingVLIWScheduler::VLIWSchedBoundary::checkHazard(SUnit *SU) {
if (HazardRec->isEnabled())
return HazardRec->getHazardType(SU) != ScheduleHazardRecognizer::NoHazard;
unsigned uops = SchedModel->getNumMicroOps(SU->getInstr());
if (IssueCount + uops > SchedModel->getIssueWidth())
return true;
return false;
}
void ConvergingVLIWScheduler::VLIWSchedBoundary::releaseNode(SUnit *SU,
unsigned ReadyCycle) {
if (ReadyCycle < MinReadyCycle)
MinReadyCycle = ReadyCycle;
// Check for interlocks first. For the purpose of other heuristics, an
// instruction that cannot issue appears as if it's not in the ReadyQueue.
if (ReadyCycle > CurrCycle || checkHazard(SU))
Pending.push(SU);
else
Available.push(SU);
}
/// Move the boundary of scheduled code by one cycle.
void ConvergingVLIWScheduler::VLIWSchedBoundary::bumpCycle() {
unsigned Width = SchedModel->getIssueWidth();
IssueCount = (IssueCount <= Width) ? 0 : IssueCount - Width;
assert(MinReadyCycle < std::numeric_limits<unsigned>::max() &&
"MinReadyCycle uninitialized");
unsigned NextCycle = std::max(CurrCycle + 1, MinReadyCycle);
if (!HazardRec->isEnabled()) {
// Bypass HazardRec virtual calls.
CurrCycle = NextCycle;
} else {
// Bypass getHazardType calls in case of long latency.
for (; CurrCycle != NextCycle; ++CurrCycle) {
if (isTop())
HazardRec->AdvanceCycle();
else
HazardRec->RecedeCycle();
}
}
CheckPending = true;
LLVM_DEBUG(dbgs() << "*** Next cycle " << Available.getName() << " cycle "
<< CurrCycle << '\n');
}
/// Move the boundary of scheduled code by one SUnit.
void ConvergingVLIWScheduler::VLIWSchedBoundary::bumpNode(SUnit *SU) {
bool startNewCycle = false;
// Update the reservation table.
if (HazardRec->isEnabled()) {
if (!isTop() && SU->isCall) {
// Calls are scheduled with their preceding instructions. For bottom-up
// scheduling, clear the pipeline state before emitting.
HazardRec->Reset();
}
HazardRec->EmitInstruction(SU);
}
// Update DFA model.
startNewCycle = ResourceModel->reserveResources(SU, isTop());
// Check the instruction group dispatch limit.
// TODO: Check if this SU must end a dispatch group.
IssueCount += SchedModel->getNumMicroOps(SU->getInstr());
if (startNewCycle) {
LLVM_DEBUG(dbgs() << "*** Max instrs at cycle " << CurrCycle << '\n');
bumpCycle();
}
else
LLVM_DEBUG(dbgs() << "*** IssueCount " << IssueCount << " at cycle "
<< CurrCycle << '\n');
}
/// Release pending ready nodes in to the available queue. This makes them
/// visible to heuristics.
void ConvergingVLIWScheduler::VLIWSchedBoundary::releasePending() {
// If the available queue is empty, it is safe to reset MinReadyCycle.
if (Available.empty())
MinReadyCycle = std::numeric_limits<unsigned>::max();
// Check to see if any of the pending instructions are ready to issue. If
// so, add them to the available queue.
for (unsigned i = 0, e = Pending.size(); i != e; ++i) {
SUnit *SU = *(Pending.begin()+i);
unsigned ReadyCycle = isTop() ? SU->TopReadyCycle : SU->BotReadyCycle;
if (ReadyCycle < MinReadyCycle)
MinReadyCycle = ReadyCycle;
if (ReadyCycle > CurrCycle)
continue;
if (checkHazard(SU))
continue;
Available.push(SU);
Pending.remove(Pending.begin()+i);
--i; --e;
}
CheckPending = false;
}
/// Remove SU from the ready set for this boundary.
void ConvergingVLIWScheduler::VLIWSchedBoundary::removeReady(SUnit *SU) {
if (Available.isInQueue(SU))
Available.remove(Available.find(SU));
else {
assert(Pending.isInQueue(SU) && "bad ready count");
Pending.remove(Pending.find(SU));
}
}
/// If this queue only has one ready candidate, return it. As a side effect,
/// advance the cycle until at least one node is ready. If multiple instructions
/// are ready, return NULL.
SUnit *ConvergingVLIWScheduler::VLIWSchedBoundary::pickOnlyChoice() {
if (CheckPending)
releasePending();
auto AdvanceCycle = [this]() {
if (Available.empty())
return true;
if (Available.size() == 1 && Pending.size() > 0)
return !ResourceModel->isResourceAvailable(*Available.begin(), isTop()) ||
getWeakLeft(*Available.begin(), isTop()) != 0;
return false;
};
for (unsigned i = 0; AdvanceCycle(); ++i) {
assert(i <= (HazardRec->getMaxLookAhead() + MaxMinLatency) &&
"permanent hazard"); (void)i;
ResourceModel->reserveResources(nullptr, isTop());
bumpCycle();
releasePending();
}
if (Available.size() == 1)
return *Available.begin();
return nullptr;
}
#ifndef NDEBUG
void ConvergingVLIWScheduler::traceCandidate(const char *Label,
const ReadyQueue &Q, SUnit *SU, int Cost, PressureChange P) {
dbgs() << Label << " " << Q.getName() << " ";
if (P.isValid())
dbgs() << DAG->TRI->getRegPressureSetName(P.getPSet()) << ":"
<< P.getUnitInc() << " ";
else
dbgs() << " ";
dbgs() << "cost(" << Cost << ")\t";
DAG->dumpNode(*SU);
}
// Very detailed queue dump, to be used with higher verbosity levels.
void ConvergingVLIWScheduler::readyQueueVerboseDump(
const RegPressureTracker &RPTracker, SchedCandidate &Candidate,
ReadyQueue &Q) {
RegPressureTracker &TempTracker = const_cast<RegPressureTracker &>(RPTracker);
dbgs() << ">>> " << Q.getName() << "\n";
for (ReadyQueue::iterator I = Q.begin(), E = Q.end(); I != E; ++I) {
RegPressureDelta RPDelta;
TempTracker.getMaxPressureDelta((*I)->getInstr(), RPDelta,
DAG->getRegionCriticalPSets(),
DAG->getRegPressure().MaxSetPressure);
std::stringstream dbgstr;
dbgstr << "SU(" << std::setw(3) << (*I)->NodeNum << ")";
dbgs() << dbgstr.str();
SchedulingCost(Q, *I, Candidate, RPDelta, true);
dbgs() << "\t";
(*I)->getInstr()->dump();
}
dbgs() << "\n";
}
#endif
/// isSingleUnscheduledPred - If SU2 is the only unscheduled predecessor
/// of SU, return true (we may have duplicates)
static inline bool isSingleUnscheduledPred(SUnit *SU, SUnit *SU2) {
if (SU->NumPredsLeft == 0)
return false;
for (auto &Pred : SU->Preds) {
// We found an available, but not scheduled, predecessor.
if (!Pred.getSUnit()->isScheduled && (Pred.getSUnit() != SU2))
return false;
}
return true;
}
/// isSingleUnscheduledSucc - If SU2 is the only unscheduled successor
/// of SU, return true (we may have duplicates)
static inline bool isSingleUnscheduledSucc(SUnit *SU, SUnit *SU2) {
if (SU->NumSuccsLeft == 0)
return false;
for (auto &Succ : SU->Succs) {
// We found an available, but not scheduled, successor.
if (!Succ.getSUnit()->isScheduled && (Succ.getSUnit() != SU2))
return false;
}
return true;
}
/// Check if the instruction changes the register pressure of a register in the
/// high pressure set. The function returns a negative value if the pressure
/// decreases and a positive value is the pressure increases. If the instruction
/// doesn't use a high pressure register or doesn't change the register
/// pressure, then return 0.
int ConvergingVLIWScheduler::pressureChange(const SUnit *SU, bool isBotUp) {
PressureDiff &PD = DAG->getPressureDiff(SU);
for (auto &P : PD) {
if (!P.isValid())
continue;
// The pressure differences are computed bottom-up, so the comparision for
// an increase is positive in the bottom direction, but negative in the
// top-down direction.
if (HighPressureSets[P.getPSet()])
return (isBotUp ? P.getUnitInc() : -P.getUnitInc());
}
return 0;
}
// Constants used to denote relative importance of
// heuristic components for cost computation.
static const unsigned PriorityOne = 200;
static const unsigned PriorityTwo = 50;
static const unsigned PriorityThree = 75;
static const unsigned ScaleTwo = 10;
/// Single point to compute overall scheduling cost.
/// TODO: More heuristics will be used soon.
int ConvergingVLIWScheduler::SchedulingCost(ReadyQueue &Q, SUnit *SU,
SchedCandidate &Candidate,
RegPressureDelta &Delta,
bool verbose) {
// Initial trivial priority.
int ResCount = 1;
// Do not waste time on a node that is already scheduled.
if (!SU || SU->isScheduled)
return ResCount;
LLVM_DEBUG(if (verbose) dbgs()
<< ((Q.getID() == TopQID) ? "(top|" : "(bot|"));
// Forced priority is high.
if (SU->isScheduleHigh) {
ResCount += PriorityOne;
LLVM_DEBUG(dbgs() << "H|");
}
unsigned IsAvailableAmt = 0;
// Critical path first.
if (Q.getID() == TopQID) {
if (Top.isLatencyBound(SU)) {
LLVM_DEBUG(if (verbose) dbgs() << "LB|");
ResCount += (SU->getHeight() * ScaleTwo);
}
LLVM_DEBUG(if (verbose) {
std::stringstream dbgstr;
dbgstr << "h" << std::setw(3) << SU->getHeight() << "|";
dbgs() << dbgstr.str();
});
// If resources are available for it, multiply the
// chance of scheduling.
if (Top.ResourceModel->isResourceAvailable(SU, true)) {
IsAvailableAmt = (PriorityTwo + PriorityThree);
ResCount += IsAvailableAmt;
LLVM_DEBUG(if (verbose) dbgs() << "A|");
} else
LLVM_DEBUG(if (verbose) dbgs() << " |");
} else {
if (Bot.isLatencyBound(SU)) {
LLVM_DEBUG(if (verbose) dbgs() << "LB|");
ResCount += (SU->getDepth() * ScaleTwo);
}
LLVM_DEBUG(if (verbose) {
std::stringstream dbgstr;
dbgstr << "d" << std::setw(3) << SU->getDepth() << "|";
dbgs() << dbgstr.str();
});
// If resources are available for it, multiply the
// chance of scheduling.
if (Bot.ResourceModel->isResourceAvailable(SU, false)) {
IsAvailableAmt = (PriorityTwo + PriorityThree);
ResCount += IsAvailableAmt;
LLVM_DEBUG(if (verbose) dbgs() << "A|");
} else
LLVM_DEBUG(if (verbose) dbgs() << " |");
}
unsigned NumNodesBlocking = 0;
if (Q.getID() == TopQID) {
// How many SUs does it block from scheduling?
// Look at all of the successors of this node.
// Count the number of nodes that
// this node is the sole unscheduled node for.
if (Top.isLatencyBound(SU))
for (const SDep &SI : SU->Succs)
if (isSingleUnscheduledPred(SI.getSUnit(), SU))
++NumNodesBlocking;
} else {
// How many unscheduled predecessors block this node?
if (Bot.isLatencyBound(SU))
for (const SDep &PI : SU->Preds)
if (isSingleUnscheduledSucc(PI.getSUnit(), SU))
++NumNodesBlocking;
}
ResCount += (NumNodesBlocking * ScaleTwo);
LLVM_DEBUG(if (verbose) {
std::stringstream dbgstr;
dbgstr << "blk " << std::setw(2) << NumNodesBlocking << ")|";
dbgs() << dbgstr.str();
});
// Factor in reg pressure as a heuristic.
if (!IgnoreBBRegPressure) {
// Decrease priority by the amount that register pressure exceeds the limit.
ResCount -= (Delta.Excess.getUnitInc()*PriorityOne);
// Decrease priority if register pressure exceeds the limit.
ResCount -= (Delta.CriticalMax.getUnitInc()*PriorityOne);
// Decrease priority slightly if register pressure would increase over the
// current maximum.
ResCount -= (Delta.CurrentMax.getUnitInc()*PriorityTwo);
// If there are register pressure issues, then we remove the value added for
// the instruction being available. The rationale is that we really don't
// want to schedule an instruction that causes a spill.
if (IsAvailableAmt && pressureChange(SU, Q.getID() != TopQID) > 0 &&
(Delta.Excess.getUnitInc() || Delta.CriticalMax.getUnitInc() ||
Delta.CurrentMax.getUnitInc()))
ResCount -= IsAvailableAmt;
LLVM_DEBUG(if (verbose) {
dbgs() << "RP " << Delta.Excess.getUnitInc() << "/"
<< Delta.CriticalMax.getUnitInc() << "/"
<< Delta.CurrentMax.getUnitInc() << ")|";
});
}
// Give a little extra priority to a .cur instruction if there is a resource
// available for it.
auto &QST = DAG->MF.getSubtarget<HexagonSubtarget>();
auto &QII = *QST.getInstrInfo();
if (SU->isInstr() && QII.mayBeCurLoad(*SU->getInstr())) {
if (Q.getID() == TopQID &&
Top.ResourceModel->isResourceAvailable(SU, true)) {
ResCount += PriorityTwo;
LLVM_DEBUG(if (verbose) dbgs() << "C|");
} else if (Q.getID() == BotQID &&
Bot.ResourceModel->isResourceAvailable(SU, false)) {
ResCount += PriorityTwo;
LLVM_DEBUG(if (verbose) dbgs() << "C|");
}
}
// Give preference to a zero latency instruction if the dependent
// instruction is in the current packet.
if (Q.getID() == TopQID && getWeakLeft(SU, true) == 0) {
for (const SDep &PI : SU->Preds) {
if (!PI.getSUnit()->getInstr()->isPseudo() && PI.isAssignedRegDep() &&
PI.getLatency() == 0 &&
Top.ResourceModel->isInPacket(PI.getSUnit())) {
ResCount += PriorityThree;
LLVM_DEBUG(if (verbose) dbgs() << "Z|");
}
}
} else if (Q.getID() == BotQID && getWeakLeft(SU, false) == 0) {
for (const SDep &SI : SU->Succs) {
if (!SI.getSUnit()->getInstr()->isPseudo() && SI.isAssignedRegDep() &&
SI.getLatency() == 0 &&
Bot.ResourceModel->isInPacket(SI.getSUnit())) {
ResCount += PriorityThree;
LLVM_DEBUG(if (verbose) dbgs() << "Z|");
}
}
}
// If the instruction has a non-zero latency dependence with an instruction in
// the current packet, then it should not be scheduled yet. The case occurs
// when the dependent instruction is scheduled in a new packet, so the
// scheduler updates the current cycle and pending instructions become
// available.
if (CheckEarlyAvail) {
if (Q.getID() == TopQID) {
for (const auto &PI : SU->Preds) {
if (PI.getLatency() > 0 &&
Top.ResourceModel->isInPacket(PI.getSUnit())) {
ResCount -= PriorityOne;
LLVM_DEBUG(if (verbose) dbgs() << "D|");
}
}
} else {
for (const auto &SI : SU->Succs) {
if (SI.getLatency() > 0 &&
Bot.ResourceModel->isInPacket(SI.getSUnit())) {
ResCount -= PriorityOne;
LLVM_DEBUG(if (verbose) dbgs() << "D|");
}
}
}
}
LLVM_DEBUG(if (verbose) {
std::stringstream dbgstr;
dbgstr << "Total " << std::setw(4) << ResCount << ")";
dbgs() << dbgstr.str();
});
return ResCount;
}
/// Pick the best candidate from the top queue.
///
/// TODO: getMaxPressureDelta results can be mostly cached for each SUnit during
/// DAG building. To adjust for the current scheduling location we need to
/// maintain the number of vreg uses remaining to be top-scheduled.
ConvergingVLIWScheduler::CandResult ConvergingVLIWScheduler::
pickNodeFromQueue(VLIWSchedBoundary &Zone, const RegPressureTracker &RPTracker,
SchedCandidate &Candidate) {
ReadyQueue &Q = Zone.Available;
LLVM_DEBUG(if (SchedDebugVerboseLevel > 1)
readyQueueVerboseDump(RPTracker, Candidate, Q);
else Q.dump(););
// getMaxPressureDelta temporarily modifies the tracker.
RegPressureTracker &TempTracker = const_cast<RegPressureTracker&>(RPTracker);
// BestSU remains NULL if no top candidates beat the best existing candidate.
CandResult FoundCandidate = NoCand;
for (ReadyQueue::iterator I = Q.begin(), E = Q.end(); I != E; ++I) {
RegPressureDelta RPDelta;
TempTracker.getMaxPressureDelta((*I)->getInstr(), RPDelta,
DAG->getRegionCriticalPSets(),
DAG->getRegPressure().MaxSetPressure);
int CurrentCost = SchedulingCost(Q, *I, Candidate, RPDelta, false);
// Initialize the candidate if needed.
if (!Candidate.SU) {
LLVM_DEBUG(traceCandidate("DCAND", Q, *I, CurrentCost));
Candidate.SU = *I;
Candidate.RPDelta = RPDelta;
Candidate.SCost = CurrentCost;
FoundCandidate = NodeOrder;
continue;
}
// Choose node order for negative cost candidates. There is no good
// candidate in this case.
if (CurrentCost < 0 && Candidate.SCost < 0) {
if ((Q.getID() == TopQID && (*I)->NodeNum < Candidate.SU->NodeNum)
|| (Q.getID() == BotQID && (*I)->NodeNum > Candidate.SU->NodeNum)) {
LLVM_DEBUG(traceCandidate("NCAND", Q, *I, CurrentCost));
Candidate.SU = *I;
Candidate.RPDelta = RPDelta;
Candidate.SCost = CurrentCost;
FoundCandidate = NodeOrder;
}
continue;
}
// Best cost.
if (CurrentCost > Candidate.SCost) {
LLVM_DEBUG(traceCandidate("CCAND", Q, *I, CurrentCost));
Candidate.SU = *I;
Candidate.RPDelta = RPDelta;
Candidate.SCost = CurrentCost;
FoundCandidate = BestCost;
continue;
}
// Choose an instruction that does not depend on an artificial edge.
unsigned CurrWeak = getWeakLeft(*I, (Q.getID() == TopQID));
unsigned CandWeak = getWeakLeft(Candidate.SU, (Q.getID() == TopQID));
if (CurrWeak != CandWeak) {
if (CurrWeak < CandWeak) {
LLVM_DEBUG(traceCandidate("WCAND", Q, *I, CurrentCost));
Candidate.SU = *I;
Candidate.RPDelta = RPDelta;
Candidate.SCost = CurrentCost;
FoundCandidate = Weak;
}
continue;
}
if (CurrentCost == Candidate.SCost && Zone.isLatencyBound(*I)) {
unsigned CurrSize, CandSize;
if (Q.getID() == TopQID) {
CurrSize = (*I)->Succs.size();
CandSize = Candidate.SU->Succs.size();
} else {
CurrSize = (*I)->Preds.size();
CandSize = Candidate.SU->Preds.size();
}
if (CurrSize > CandSize) {
LLVM_DEBUG(traceCandidate("SPCAND", Q, *I, CurrentCost));
Candidate.SU = *I;
Candidate.RPDelta = RPDelta;
Candidate.SCost = CurrentCost;
FoundCandidate = BestCost;
}
// Keep the old candidate if it's a better candidate. That is, don't use
// the subsequent tie breaker.
if (CurrSize != CandSize)
continue;
}
// Tie breaker.
// To avoid scheduling indeterminism, we need a tie breaker
// for the case when cost is identical for two nodes.
if (UseNewerCandidate && CurrentCost == Candidate.SCost) {
if ((Q.getID() == TopQID && (*I)->NodeNum < Candidate.SU->NodeNum)
|| (Q.getID() == BotQID && (*I)->NodeNum > Candidate.SU->NodeNum)) {
LLVM_DEBUG(traceCandidate("TCAND", Q, *I, CurrentCost));
Candidate.SU = *I;
Candidate.RPDelta = RPDelta;
Candidate.SCost = CurrentCost;
FoundCandidate = NodeOrder;
continue;
}
}
// Fall through to original instruction order.
// Only consider node order if Candidate was chosen from this Q.
if (FoundCandidate == NoCand)
continue;
}
return FoundCandidate;
}
/// Pick the best candidate node from either the top or bottom queue.
SUnit *ConvergingVLIWScheduler::pickNodeBidrectional(bool &IsTopNode) {
// Schedule as far as possible in the direction of no choice. This is most
// efficient, but also provides the best heuristics for CriticalPSets.
if (SUnit *SU = Bot.pickOnlyChoice()) {
LLVM_DEBUG(dbgs() << "Picked only Bottom\n");
IsTopNode = false;
return SU;
}
if (SUnit *SU = Top.pickOnlyChoice()) {
LLVM_DEBUG(dbgs() << "Picked only Top\n");
IsTopNode = true;
return SU;
}
SchedCandidate BotCand;
// Prefer bottom scheduling when heuristics are silent.
CandResult BotResult = pickNodeFromQueue(Bot,
DAG->getBotRPTracker(), BotCand);
assert(BotResult != NoCand && "failed to find the first candidate");
// If either Q has a single candidate that provides the least increase in
// Excess pressure, we can immediately schedule from that Q.
//
// RegionCriticalPSets summarizes the pressure within the scheduled region and
// affects picking from either Q. If scheduling in one direction must
// increase pressure for one of the excess PSets, then schedule in that
// direction first to provide more freedom in the other direction.
if (BotResult == SingleExcess || BotResult == SingleCritical) {
LLVM_DEBUG(dbgs() << "Prefered Bottom Node\n");
IsTopNode = false;
return BotCand.SU;
}
// Check if the top Q has a better candidate.
SchedCandidate TopCand;
CandResult TopResult = pickNodeFromQueue(Top,
DAG->getTopRPTracker(), TopCand);
assert(TopResult != NoCand && "failed to find the first candidate");
if (TopResult == SingleExcess || TopResult == SingleCritical) {
LLVM_DEBUG(dbgs() << "Prefered Top Node\n");
IsTopNode = true;
return TopCand.SU;
}
// If either Q has a single candidate that minimizes pressure above the
// original region's pressure pick it.
if (BotResult == SingleMax) {
LLVM_DEBUG(dbgs() << "Prefered Bottom Node SingleMax\n");
IsTopNode = false;
return BotCand.SU;
}
if (TopResult == SingleMax) {
LLVM_DEBUG(dbgs() << "Prefered Top Node SingleMax\n");
IsTopNode = true;
return TopCand.SU;
}
if (TopCand.SCost > BotCand.SCost) {
LLVM_DEBUG(dbgs() << "Prefered Top Node Cost\n");
IsTopNode = true;
return TopCand.SU;
}
// Otherwise prefer the bottom candidate in node order.
LLVM_DEBUG(dbgs() << "Prefered Bottom in Node order\n");
IsTopNode = false;
return BotCand.SU;
}
/// Pick the best node to balance the schedule. Implements MachineSchedStrategy.
SUnit *ConvergingVLIWScheduler::pickNode(bool &IsTopNode) {
if (DAG->top() == DAG->bottom()) {
assert(Top.Available.empty() && Top.Pending.empty() &&
Bot.Available.empty() && Bot.Pending.empty() && "ReadyQ garbage");
return nullptr;
}
SUnit *SU;
if (ForceTopDown) {
SU = Top.pickOnlyChoice();
if (!SU) {
SchedCandidate TopCand;
CandResult TopResult =
pickNodeFromQueue(Top, DAG->getTopRPTracker(), TopCand);
assert(TopResult != NoCand && "failed to find the first candidate");
(void)TopResult;
SU = TopCand.SU;
}
IsTopNode = true;
} else if (ForceBottomUp) {
SU = Bot.pickOnlyChoice();
if (!SU) {
SchedCandidate BotCand;
CandResult BotResult =
pickNodeFromQueue(Bot, DAG->getBotRPTracker(), BotCand);
assert(BotResult != NoCand && "failed to find the first candidate");
(void)BotResult;
SU = BotCand.SU;
}
IsTopNode = false;
} else {
SU = pickNodeBidrectional(IsTopNode);
}
if (SU->isTopReady())
Top.removeReady(SU);
if (SU->isBottomReady())
Bot.removeReady(SU);
LLVM_DEBUG(dbgs() << "*** " << (IsTopNode ? "Top" : "Bottom")
<< " Scheduling instruction in cycle "
<< (IsTopNode ? Top.CurrCycle : Bot.CurrCycle) << " ("
<< reportPackets() << ")\n";
DAG->dumpNode(*SU));
return SU;
}
/// Update the scheduler's state after scheduling a node. This is the same node
/// that was just returned by pickNode(). However, VLIWMachineScheduler needs
/// to update it's state based on the current cycle before MachineSchedStrategy
/// does.
void ConvergingVLIWScheduler::schedNode(SUnit *SU, bool IsTopNode) {
if (IsTopNode) {
Top.bumpNode(SU);
SU->TopReadyCycle = Top.CurrCycle;
} else {
Bot.bumpNode(SU);
SU->BotReadyCycle = Bot.CurrCycle;
}
}
|