forked from llvm-mirror/llvm
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathLazyCallGraph.cpp
1950 lines (1679 loc) · 71.7 KB
/
LazyCallGraph.cpp
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
//===- LazyCallGraph.cpp - Analysis of a Module's call graph --------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
#include "llvm/Analysis/LazyCallGraph.h"
#include "llvm/ADT/ScopeExit.h"
#include "llvm/ADT/Sequence.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/ScopeExit.h"
#include "llvm/IR/CallSite.h"
#include "llvm/IR/InstVisitor.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/PassManager.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/GraphWriter.h"
#include <utility>
using namespace llvm;
#define DEBUG_TYPE "lcg"
void LazyCallGraph::EdgeSequence::insertEdgeInternal(Node &TargetN,
Edge::Kind EK) {
EdgeIndexMap.insert({&TargetN, Edges.size()});
Edges.emplace_back(TargetN, EK);
}
void LazyCallGraph::EdgeSequence::setEdgeKind(Node &TargetN, Edge::Kind EK) {
Edges[EdgeIndexMap.find(&TargetN)->second].setKind(EK);
}
bool LazyCallGraph::EdgeSequence::removeEdgeInternal(Node &TargetN) {
auto IndexMapI = EdgeIndexMap.find(&TargetN);
if (IndexMapI == EdgeIndexMap.end())
return false;
Edges[IndexMapI->second] = Edge();
EdgeIndexMap.erase(IndexMapI);
return true;
}
static void addEdge(SmallVectorImpl<LazyCallGraph::Edge> &Edges,
DenseMap<LazyCallGraph::Node *, int> &EdgeIndexMap,
LazyCallGraph::Node &N, LazyCallGraph::Edge::Kind EK) {
if (!EdgeIndexMap.insert({&N, Edges.size()}).second)
return;
DEBUG(dbgs() << " Added callable function: " << N.getName() << "\n");
Edges.emplace_back(LazyCallGraph::Edge(N, EK));
}
LazyCallGraph::EdgeSequence &LazyCallGraph::Node::populateSlow() {
assert(!Edges && "Must not have already populated the edges for this node!");
DEBUG(dbgs() << " Adding functions called by '" << getName()
<< "' to the graph.\n");
Edges = EdgeSequence();
SmallVector<Constant *, 16> Worklist;
SmallPtrSet<Function *, 4> Callees;
SmallPtrSet<Constant *, 16> Visited;
// Find all the potential call graph edges in this function. We track both
// actual call edges and indirect references to functions. The direct calls
// are trivially added, but to accumulate the latter we walk the instructions
// and add every operand which is a constant to the worklist to process
// afterward.
//
// Note that we consider *any* function with a definition to be a viable
// edge. Even if the function's definition is subject to replacement by
// some other module (say, a weak definition) there may still be
// optimizations which essentially speculate based on the definition and
// a way to check that the specific definition is in fact the one being
// used. For example, this could be done by moving the weak definition to
// a strong (internal) definition and making the weak definition be an
// alias. Then a test of the address of the weak function against the new
// strong definition's address would be an effective way to determine the
// safety of optimizing a direct call edge.
for (BasicBlock &BB : *F)
for (Instruction &I : BB) {
if (auto CS = CallSite(&I))
if (Function *Callee = CS.getCalledFunction())
if (!Callee->isDeclaration())
if (Callees.insert(Callee).second) {
Visited.insert(Callee);
addEdge(Edges->Edges, Edges->EdgeIndexMap, G->get(*Callee),
LazyCallGraph::Edge::Call);
}
for (Value *Op : I.operand_values())
if (Constant *C = dyn_cast<Constant>(Op))
if (Visited.insert(C).second)
Worklist.push_back(C);
}
// We've collected all the constant (and thus potentially function or
// function containing) operands to all of the instructions in the function.
// Process them (recursively) collecting every function found.
visitReferences(Worklist, Visited, [&](Function &F) {
addEdge(Edges->Edges, Edges->EdgeIndexMap, G->get(F),
LazyCallGraph::Edge::Ref);
});
return *Edges;
}
void LazyCallGraph::Node::replaceFunction(Function &NewF) {
assert(F != &NewF && "Must not replace a function with itself!");
F = &NewF;
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
LLVM_DUMP_METHOD void LazyCallGraph::Node::dump() const {
dbgs() << *this << '\n';
}
#endif
LazyCallGraph::LazyCallGraph(Module &M) {
DEBUG(dbgs() << "Building CG for module: " << M.getModuleIdentifier()
<< "\n");
for (Function &F : M)
if (!F.isDeclaration() && !F.hasLocalLinkage()) {
DEBUG(dbgs() << " Adding '" << F.getName()
<< "' to entry set of the graph.\n");
addEdge(EntryEdges.Edges, EntryEdges.EdgeIndexMap, get(F), Edge::Ref);
}
// Now add entry nodes for functions reachable via initializers to globals.
SmallVector<Constant *, 16> Worklist;
SmallPtrSet<Constant *, 16> Visited;
for (GlobalVariable &GV : M.globals())
if (GV.hasInitializer())
if (Visited.insert(GV.getInitializer()).second)
Worklist.push_back(GV.getInitializer());
DEBUG(dbgs() << " Adding functions referenced by global initializers to the "
"entry set.\n");
visitReferences(Worklist, Visited, [&](Function &F) {
addEdge(EntryEdges.Edges, EntryEdges.EdgeIndexMap, get(F),
LazyCallGraph::Edge::Ref);
});
}
LazyCallGraph::LazyCallGraph(LazyCallGraph &&G)
: BPA(std::move(G.BPA)), NodeMap(std::move(G.NodeMap)),
EntryEdges(std::move(G.EntryEdges)), SCCBPA(std::move(G.SCCBPA)),
SCCMap(std::move(G.SCCMap)), LeafRefSCCs(std::move(G.LeafRefSCCs)) {
updateGraphPtrs();
}
LazyCallGraph &LazyCallGraph::operator=(LazyCallGraph &&G) {
BPA = std::move(G.BPA);
NodeMap = std::move(G.NodeMap);
EntryEdges = std::move(G.EntryEdges);
SCCBPA = std::move(G.SCCBPA);
SCCMap = std::move(G.SCCMap);
LeafRefSCCs = std::move(G.LeafRefSCCs);
updateGraphPtrs();
return *this;
}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
LLVM_DUMP_METHOD void LazyCallGraph::SCC::dump() const {
dbgs() << *this << '\n';
}
#endif
#ifndef NDEBUG
void LazyCallGraph::SCC::verify() {
assert(OuterRefSCC && "Can't have a null RefSCC!");
assert(!Nodes.empty() && "Can't have an empty SCC!");
for (Node *N : Nodes) {
assert(N && "Can't have a null node!");
assert(OuterRefSCC->G->lookupSCC(*N) == this &&
"Node does not map to this SCC!");
assert(N->DFSNumber == -1 &&
"Must set DFS numbers to -1 when adding a node to an SCC!");
assert(N->LowLink == -1 &&
"Must set low link to -1 when adding a node to an SCC!");
for (Edge &E : **N)
assert(E.getNode() && "Can't have an unpopulated node!");
}
}
#endif
bool LazyCallGraph::SCC::isParentOf(const SCC &C) const {
if (this == &C)
return false;
for (Node &N : *this)
for (Edge &E : N->calls())
if (OuterRefSCC->G->lookupSCC(E.getNode()) == &C)
return true;
// No edges found.
return false;
}
bool LazyCallGraph::SCC::isAncestorOf(const SCC &TargetC) const {
if (this == &TargetC)
return false;
LazyCallGraph &G = *OuterRefSCC->G;
// Start with this SCC.
SmallPtrSet<const SCC *, 16> Visited = {this};
SmallVector<const SCC *, 16> Worklist = {this};
// Walk down the graph until we run out of edges or find a path to TargetC.
do {
const SCC &C = *Worklist.pop_back_val();
for (Node &N : C)
for (Edge &E : N->calls()) {
SCC *CalleeC = G.lookupSCC(E.getNode());
if (!CalleeC)
continue;
// If the callee's SCC is the TargetC, we're done.
if (CalleeC == &TargetC)
return true;
// If this is the first time we've reached this SCC, put it on the
// worklist to recurse through.
if (Visited.insert(CalleeC).second)
Worklist.push_back(CalleeC);
}
} while (!Worklist.empty());
// No paths found.
return false;
}
LazyCallGraph::RefSCC::RefSCC(LazyCallGraph &G) : G(&G) {}
#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
LLVM_DUMP_METHOD void LazyCallGraph::RefSCC::dump() const {
dbgs() << *this << '\n';
}
#endif
#ifndef NDEBUG
void LazyCallGraph::RefSCC::verify() {
assert(G && "Can't have a null graph!");
assert(!SCCs.empty() && "Can't have an empty SCC!");
// Verify basic properties of the SCCs.
SmallPtrSet<SCC *, 4> SCCSet;
for (SCC *C : SCCs) {
assert(C && "Can't have a null SCC!");
C->verify();
assert(&C->getOuterRefSCC() == this &&
"SCC doesn't think it is inside this RefSCC!");
bool Inserted = SCCSet.insert(C).second;
assert(Inserted && "Found a duplicate SCC!");
auto IndexIt = SCCIndices.find(C);
assert(IndexIt != SCCIndices.end() &&
"Found an SCC that doesn't have an index!");
}
// Check that our indices map correctly.
for (auto &SCCIndexPair : SCCIndices) {
SCC *C = SCCIndexPair.first;
int i = SCCIndexPair.second;
assert(C && "Can't have a null SCC in the indices!");
assert(SCCSet.count(C) && "Found an index for an SCC not in the RefSCC!");
assert(SCCs[i] == C && "Index doesn't point to SCC!");
}
// Check that the SCCs are in fact in post-order.
for (int i = 0, Size = SCCs.size(); i < Size; ++i) {
SCC &SourceSCC = *SCCs[i];
for (Node &N : SourceSCC)
for (Edge &E : *N) {
if (!E.isCall())
continue;
SCC &TargetSCC = *G->lookupSCC(E.getNode());
if (&TargetSCC.getOuterRefSCC() == this) {
assert(SCCIndices.find(&TargetSCC)->second <= i &&
"Edge between SCCs violates post-order relationship.");
continue;
}
assert(TargetSCC.getOuterRefSCC().Parents.count(this) &&
"Edge to a RefSCC missing us in its parent set.");
}
}
// Check that our parents are actually parents.
for (RefSCC *ParentRC : Parents) {
assert(ParentRC != this && "Cannot be our own parent!");
auto HasConnectingEdge = [&] {
for (SCC &C : *ParentRC)
for (Node &N : C)
for (Edge &E : *N)
if (G->lookupRefSCC(E.getNode()) == this)
return true;
return false;
};
assert(HasConnectingEdge() && "No edge connects the parent to us!");
}
}
#endif
bool LazyCallGraph::RefSCC::isDescendantOf(const RefSCC &C) const {
// Walk up the parents of this SCC and verify that we eventually find C.
SmallVector<const RefSCC *, 4> AncestorWorklist;
AncestorWorklist.push_back(this);
do {
const RefSCC *AncestorC = AncestorWorklist.pop_back_val();
if (AncestorC->isChildOf(C))
return true;
for (const RefSCC *ParentC : AncestorC->Parents)
AncestorWorklist.push_back(ParentC);
} while (!AncestorWorklist.empty());
return false;
}
/// Generic helper that updates a postorder sequence of SCCs for a potentially
/// cycle-introducing edge insertion.
///
/// A postorder sequence of SCCs of a directed graph has one fundamental
/// property: all deges in the DAG of SCCs point "up" the sequence. That is,
/// all edges in the SCC DAG point to prior SCCs in the sequence.
///
/// This routine both updates a postorder sequence and uses that sequence to
/// compute the set of SCCs connected into a cycle. It should only be called to
/// insert a "downward" edge which will require changing the sequence to
/// restore it to a postorder.
///
/// When inserting an edge from an earlier SCC to a later SCC in some postorder
/// sequence, all of the SCCs which may be impacted are in the closed range of
/// those two within the postorder sequence. The algorithm used here to restore
/// the state is as follows:
///
/// 1) Starting from the source SCC, construct a set of SCCs which reach the
/// source SCC consisting of just the source SCC. Then scan toward the
/// target SCC in postorder and for each SCC, if it has an edge to an SCC
/// in the set, add it to the set. Otherwise, the source SCC is not
/// a successor, move it in the postorder sequence to immediately before
/// the source SCC, shifting the source SCC and all SCCs in the set one
/// position toward the target SCC. Stop scanning after processing the
/// target SCC.
/// 2) If the source SCC is now past the target SCC in the postorder sequence,
/// and thus the new edge will flow toward the start, we are done.
/// 3) Otherwise, starting from the target SCC, walk all edges which reach an
/// SCC between the source and the target, and add them to the set of
/// connected SCCs, then recurse through them. Once a complete set of the
/// SCCs the target connects to is known, hoist the remaining SCCs between
/// the source and the target to be above the target. Note that there is no
/// need to process the source SCC, it is already known to connect.
/// 4) At this point, all of the SCCs in the closed range between the source
/// SCC and the target SCC in the postorder sequence are connected,
/// including the target SCC and the source SCC. Inserting the edge from
/// the source SCC to the target SCC will form a cycle out of precisely
/// these SCCs. Thus we can merge all of the SCCs in this closed range into
/// a single SCC.
///
/// This process has various important properties:
/// - Only mutates the SCCs when adding the edge actually changes the SCC
/// structure.
/// - Never mutates SCCs which are unaffected by the change.
/// - Updates the postorder sequence to correctly satisfy the postorder
/// constraint after the edge is inserted.
/// - Only reorders SCCs in the closed postorder sequence from the source to
/// the target, so easy to bound how much has changed even in the ordering.
/// - Big-O is the number of edges in the closed postorder range of SCCs from
/// source to target.
///
/// This helper routine, in addition to updating the postorder sequence itself
/// will also update a map from SCCs to indices within that sequecne.
///
/// The sequence and the map must operate on pointers to the SCC type.
///
/// Two callbacks must be provided. The first computes the subset of SCCs in
/// the postorder closed range from the source to the target which connect to
/// the source SCC via some (transitive) set of edges. The second computes the
/// subset of the same range which the target SCC connects to via some
/// (transitive) set of edges. Both callbacks should populate the set argument
/// provided.
template <typename SCCT, typename PostorderSequenceT, typename SCCIndexMapT,
typename ComputeSourceConnectedSetCallableT,
typename ComputeTargetConnectedSetCallableT>
static iterator_range<typename PostorderSequenceT::iterator>
updatePostorderSequenceForEdgeInsertion(
SCCT &SourceSCC, SCCT &TargetSCC, PostorderSequenceT &SCCs,
SCCIndexMapT &SCCIndices,
ComputeSourceConnectedSetCallableT ComputeSourceConnectedSet,
ComputeTargetConnectedSetCallableT ComputeTargetConnectedSet) {
int SourceIdx = SCCIndices[&SourceSCC];
int TargetIdx = SCCIndices[&TargetSCC];
assert(SourceIdx < TargetIdx && "Cannot have equal indices here!");
SmallPtrSet<SCCT *, 4> ConnectedSet;
// Compute the SCCs which (transitively) reach the source.
ComputeSourceConnectedSet(ConnectedSet);
// Partition the SCCs in this part of the port-order sequence so only SCCs
// connecting to the source remain between it and the target. This is
// a benign partition as it preserves postorder.
auto SourceI = std::stable_partition(
SCCs.begin() + SourceIdx, SCCs.begin() + TargetIdx + 1,
[&ConnectedSet](SCCT *C) { return !ConnectedSet.count(C); });
for (int i = SourceIdx, e = TargetIdx + 1; i < e; ++i)
SCCIndices.find(SCCs[i])->second = i;
// If the target doesn't connect to the source, then we've corrected the
// post-order and there are no cycles formed.
if (!ConnectedSet.count(&TargetSCC)) {
assert(SourceI > (SCCs.begin() + SourceIdx) &&
"Must have moved the source to fix the post-order.");
assert(*std::prev(SourceI) == &TargetSCC &&
"Last SCC to move should have bene the target.");
// Return an empty range at the target SCC indicating there is nothing to
// merge.
return make_range(std::prev(SourceI), std::prev(SourceI));
}
assert(SCCs[TargetIdx] == &TargetSCC &&
"Should not have moved target if connected!");
SourceIdx = SourceI - SCCs.begin();
assert(SCCs[SourceIdx] == &SourceSCC &&
"Bad updated index computation for the source SCC!");
// See whether there are any remaining intervening SCCs between the source
// and target. If so we need to make sure they all are reachable form the
// target.
if (SourceIdx + 1 < TargetIdx) {
ConnectedSet.clear();
ComputeTargetConnectedSet(ConnectedSet);
// Partition SCCs so that only SCCs reached from the target remain between
// the source and the target. This preserves postorder.
auto TargetI = std::stable_partition(
SCCs.begin() + SourceIdx + 1, SCCs.begin() + TargetIdx + 1,
[&ConnectedSet](SCCT *C) { return ConnectedSet.count(C); });
for (int i = SourceIdx + 1, e = TargetIdx + 1; i < e; ++i)
SCCIndices.find(SCCs[i])->second = i;
TargetIdx = std::prev(TargetI) - SCCs.begin();
assert(SCCs[TargetIdx] == &TargetSCC &&
"Should always end with the target!");
}
// At this point, we know that connecting source to target forms a cycle
// because target connects back to source, and we know that all of the SCCs
// between the source and target in the postorder sequence participate in that
// cycle.
return make_range(SCCs.begin() + SourceIdx, SCCs.begin() + TargetIdx);
}
SmallVector<LazyCallGraph::SCC *, 1>
LazyCallGraph::RefSCC::switchInternalEdgeToCall(Node &SourceN, Node &TargetN) {
assert(!(*SourceN)[TargetN].isCall() && "Must start with a ref edge!");
SmallVector<SCC *, 1> DeletedSCCs;
#ifndef NDEBUG
// In a debug build, verify the RefSCC is valid to start with and when this
// routine finishes.
verify();
auto VerifyOnExit = make_scope_exit([&]() { verify(); });
#endif
SCC &SourceSCC = *G->lookupSCC(SourceN);
SCC &TargetSCC = *G->lookupSCC(TargetN);
// If the two nodes are already part of the same SCC, we're also done as
// we've just added more connectivity.
if (&SourceSCC == &TargetSCC) {
SourceN->setEdgeKind(TargetN, Edge::Call);
return DeletedSCCs;
}
// At this point we leverage the postorder list of SCCs to detect when the
// insertion of an edge changes the SCC structure in any way.
//
// First and foremost, we can eliminate the need for any changes when the
// edge is toward the beginning of the postorder sequence because all edges
// flow in that direction already. Thus adding a new one cannot form a cycle.
int SourceIdx = SCCIndices[&SourceSCC];
int TargetIdx = SCCIndices[&TargetSCC];
if (TargetIdx < SourceIdx) {
SourceN->setEdgeKind(TargetN, Edge::Call);
return DeletedSCCs;
}
// Compute the SCCs which (transitively) reach the source.
auto ComputeSourceConnectedSet = [&](SmallPtrSetImpl<SCC *> &ConnectedSet) {
#ifndef NDEBUG
// Check that the RefSCC is still valid before computing this as the
// results will be nonsensical of we've broken its invariants.
verify();
#endif
ConnectedSet.insert(&SourceSCC);
auto IsConnected = [&](SCC &C) {
for (Node &N : C)
for (Edge &E : N->calls())
if (ConnectedSet.count(G->lookupSCC(E.getNode())))
return true;
return false;
};
for (SCC *C :
make_range(SCCs.begin() + SourceIdx + 1, SCCs.begin() + TargetIdx + 1))
if (IsConnected(*C))
ConnectedSet.insert(C);
};
// Use a normal worklist to find which SCCs the target connects to. We still
// bound the search based on the range in the postorder list we care about,
// but because this is forward connectivity we just "recurse" through the
// edges.
auto ComputeTargetConnectedSet = [&](SmallPtrSetImpl<SCC *> &ConnectedSet) {
#ifndef NDEBUG
// Check that the RefSCC is still valid before computing this as the
// results will be nonsensical of we've broken its invariants.
verify();
#endif
ConnectedSet.insert(&TargetSCC);
SmallVector<SCC *, 4> Worklist;
Worklist.push_back(&TargetSCC);
do {
SCC &C = *Worklist.pop_back_val();
for (Node &N : C)
for (Edge &E : *N) {
if (!E.isCall())
continue;
SCC &EdgeC = *G->lookupSCC(E.getNode());
if (&EdgeC.getOuterRefSCC() != this)
// Not in this RefSCC...
continue;
if (SCCIndices.find(&EdgeC)->second <= SourceIdx)
// Not in the postorder sequence between source and target.
continue;
if (ConnectedSet.insert(&EdgeC).second)
Worklist.push_back(&EdgeC);
}
} while (!Worklist.empty());
};
// Use a generic helper to update the postorder sequence of SCCs and return
// a range of any SCCs connected into a cycle by inserting this edge. This
// routine will also take care of updating the indices into the postorder
// sequence.
auto MergeRange = updatePostorderSequenceForEdgeInsertion(
SourceSCC, TargetSCC, SCCs, SCCIndices, ComputeSourceConnectedSet,
ComputeTargetConnectedSet);
// If the merge range is empty, then adding the edge didn't actually form any
// new cycles. We're done.
if (MergeRange.begin() == MergeRange.end()) {
// Now that the SCC structure is finalized, flip the kind to call.
SourceN->setEdgeKind(TargetN, Edge::Call);
return DeletedSCCs;
}
#ifndef NDEBUG
// Before merging, check that the RefSCC remains valid after all the
// postorder updates.
verify();
#endif
// Otherwise we need to merge all of the SCCs in the cycle into a single
// result SCC.
//
// NB: We merge into the target because all of these functions were already
// reachable from the target, meaning any SCC-wide properties deduced about it
// other than the set of functions within it will not have changed.
for (SCC *C : MergeRange) {
assert(C != &TargetSCC &&
"We merge *into* the target and shouldn't process it here!");
SCCIndices.erase(C);
TargetSCC.Nodes.append(C->Nodes.begin(), C->Nodes.end());
for (Node *N : C->Nodes)
G->SCCMap[N] = &TargetSCC;
C->clear();
DeletedSCCs.push_back(C);
}
// Erase the merged SCCs from the list and update the indices of the
// remaining SCCs.
int IndexOffset = MergeRange.end() - MergeRange.begin();
auto EraseEnd = SCCs.erase(MergeRange.begin(), MergeRange.end());
for (SCC *C : make_range(EraseEnd, SCCs.end()))
SCCIndices[C] -= IndexOffset;
// Now that the SCC structure is finalized, flip the kind to call.
SourceN->setEdgeKind(TargetN, Edge::Call);
// And we're done!
return DeletedSCCs;
}
void LazyCallGraph::RefSCC::switchTrivialInternalEdgeToRef(Node &SourceN,
Node &TargetN) {
assert((*SourceN)[TargetN].isCall() && "Must start with a call edge!");
#ifndef NDEBUG
// In a debug build, verify the RefSCC is valid to start with and when this
// routine finishes.
verify();
auto VerifyOnExit = make_scope_exit([&]() { verify(); });
#endif
assert(G->lookupRefSCC(SourceN) == this &&
"Source must be in this RefSCC.");
assert(G->lookupRefSCC(TargetN) == this &&
"Target must be in this RefSCC.");
assert(G->lookupSCC(SourceN) != G->lookupSCC(TargetN) &&
"Source and Target must be in separate SCCs for this to be trivial!");
// Set the edge kind.
SourceN->setEdgeKind(TargetN, Edge::Ref);
}
iterator_range<LazyCallGraph::RefSCC::iterator>
LazyCallGraph::RefSCC::switchInternalEdgeToRef(Node &SourceN, Node &TargetN) {
assert((*SourceN)[TargetN].isCall() && "Must start with a call edge!");
#ifndef NDEBUG
// In a debug build, verify the RefSCC is valid to start with and when this
// routine finishes.
verify();
auto VerifyOnExit = make_scope_exit([&]() { verify(); });
#endif
assert(G->lookupRefSCC(SourceN) == this &&
"Source must be in this RefSCC.");
assert(G->lookupRefSCC(TargetN) == this &&
"Target must be in this RefSCC.");
SCC &TargetSCC = *G->lookupSCC(TargetN);
assert(G->lookupSCC(SourceN) == &TargetSCC && "Source and Target must be in "
"the same SCC to require the "
"full CG update.");
// Set the edge kind.
SourceN->setEdgeKind(TargetN, Edge::Ref);
// Otherwise we are removing a call edge from a single SCC. This may break
// the cycle. In order to compute the new set of SCCs, we need to do a small
// DFS over the nodes within the SCC to form any sub-cycles that remain as
// distinct SCCs and compute a postorder over the resulting SCCs.
//
// However, we specially handle the target node. The target node is known to
// reach all other nodes in the original SCC by definition. This means that
// we want the old SCC to be replaced with an SCC contaning that node as it
// will be the root of whatever SCC DAG results from the DFS. Assumptions
// about an SCC such as the set of functions called will continue to hold,
// etc.
SCC &OldSCC = TargetSCC;
SmallVector<std::pair<Node *, EdgeSequence::call_iterator>, 16> DFSStack;
SmallVector<Node *, 16> PendingSCCStack;
SmallVector<SCC *, 4> NewSCCs;
// Prepare the nodes for a fresh DFS.
SmallVector<Node *, 16> Worklist;
Worklist.swap(OldSCC.Nodes);
for (Node *N : Worklist) {
N->DFSNumber = N->LowLink = 0;
G->SCCMap.erase(N);
}
// Force the target node to be in the old SCC. This also enables us to take
// a very significant short-cut in the standard Tarjan walk to re-form SCCs
// below: whenever we build an edge that reaches the target node, we know
// that the target node eventually connects back to all other nodes in our
// walk. As a consequence, we can detect and handle participants in that
// cycle without walking all the edges that form this connection, and instead
// by relying on the fundamental guarantee coming into this operation (all
// nodes are reachable from the target due to previously forming an SCC).
TargetN.DFSNumber = TargetN.LowLink = -1;
OldSCC.Nodes.push_back(&TargetN);
G->SCCMap[&TargetN] = &OldSCC;
// Scan down the stack and DFS across the call edges.
for (Node *RootN : Worklist) {
assert(DFSStack.empty() &&
"Cannot begin a new root with a non-empty DFS stack!");
assert(PendingSCCStack.empty() &&
"Cannot begin a new root with pending nodes for an SCC!");
// Skip any nodes we've already reached in the DFS.
if (RootN->DFSNumber != 0) {
assert(RootN->DFSNumber == -1 &&
"Shouldn't have any mid-DFS root nodes!");
continue;
}
RootN->DFSNumber = RootN->LowLink = 1;
int NextDFSNumber = 2;
DFSStack.push_back({RootN, (*RootN)->call_begin()});
do {
Node *N;
EdgeSequence::call_iterator I;
std::tie(N, I) = DFSStack.pop_back_val();
auto E = (*N)->call_end();
while (I != E) {
Node &ChildN = I->getNode();
if (ChildN.DFSNumber == 0) {
// We haven't yet visited this child, so descend, pushing the current
// node onto the stack.
DFSStack.push_back({N, I});
assert(!G->SCCMap.count(&ChildN) &&
"Found a node with 0 DFS number but already in an SCC!");
ChildN.DFSNumber = ChildN.LowLink = NextDFSNumber++;
N = &ChildN;
I = (*N)->call_begin();
E = (*N)->call_end();
continue;
}
// Check for the child already being part of some component.
if (ChildN.DFSNumber == -1) {
if (G->lookupSCC(ChildN) == &OldSCC) {
// If the child is part of the old SCC, we know that it can reach
// every other node, so we have formed a cycle. Pull the entire DFS
// and pending stacks into it. See the comment above about setting
// up the old SCC for why we do this.
int OldSize = OldSCC.size();
OldSCC.Nodes.push_back(N);
OldSCC.Nodes.append(PendingSCCStack.begin(), PendingSCCStack.end());
PendingSCCStack.clear();
while (!DFSStack.empty())
OldSCC.Nodes.push_back(DFSStack.pop_back_val().first);
for (Node &N : make_range(OldSCC.begin() + OldSize, OldSCC.end())) {
N.DFSNumber = N.LowLink = -1;
G->SCCMap[&N] = &OldSCC;
}
N = nullptr;
break;
}
// If the child has already been added to some child component, it
// couldn't impact the low-link of this parent because it isn't
// connected, and thus its low-link isn't relevant so skip it.
++I;
continue;
}
// Track the lowest linked child as the lowest link for this node.
assert(ChildN.LowLink > 0 && "Must have a positive low-link number!");
if (ChildN.LowLink < N->LowLink)
N->LowLink = ChildN.LowLink;
// Move to the next edge.
++I;
}
if (!N)
// Cleared the DFS early, start another round.
break;
// We've finished processing N and its descendents, put it on our pending
// SCC stack to eventually get merged into an SCC of nodes.
PendingSCCStack.push_back(N);
// If this node is linked to some lower entry, continue walking up the
// stack.
if (N->LowLink != N->DFSNumber)
continue;
// Otherwise, we've completed an SCC. Append it to our post order list of
// SCCs.
int RootDFSNumber = N->DFSNumber;
// Find the range of the node stack by walking down until we pass the
// root DFS number.
auto SCCNodes = make_range(
PendingSCCStack.rbegin(),
find_if(reverse(PendingSCCStack), [RootDFSNumber](const Node *N) {
return N->DFSNumber < RootDFSNumber;
}));
// Form a new SCC out of these nodes and then clear them off our pending
// stack.
NewSCCs.push_back(G->createSCC(*this, SCCNodes));
for (Node &N : *NewSCCs.back()) {
N.DFSNumber = N.LowLink = -1;
G->SCCMap[&N] = NewSCCs.back();
}
PendingSCCStack.erase(SCCNodes.end().base(), PendingSCCStack.end());
} while (!DFSStack.empty());
}
// Insert the remaining SCCs before the old one. The old SCC can reach all
// other SCCs we form because it contains the target node of the removed edge
// of the old SCC. This means that we will have edges into all of the new
// SCCs, which means the old one must come last for postorder.
int OldIdx = SCCIndices[&OldSCC];
SCCs.insert(SCCs.begin() + OldIdx, NewSCCs.begin(), NewSCCs.end());
// Update the mapping from SCC* to index to use the new SCC*s, and remove the
// old SCC from the mapping.
for (int Idx = OldIdx, Size = SCCs.size(); Idx < Size; ++Idx)
SCCIndices[SCCs[Idx]] = Idx;
return make_range(SCCs.begin() + OldIdx,
SCCs.begin() + OldIdx + NewSCCs.size());
}
void LazyCallGraph::RefSCC::switchOutgoingEdgeToCall(Node &SourceN,
Node &TargetN) {
assert(!(*SourceN)[TargetN].isCall() && "Must start with a ref edge!");
assert(G->lookupRefSCC(SourceN) == this && "Source must be in this RefSCC.");
assert(G->lookupRefSCC(TargetN) != this &&
"Target must not be in this RefSCC.");
#ifdef EXPENSIVE_CHECKS
assert(G->lookupRefSCC(TargetN)->isDescendantOf(*this) &&
"Target must be a descendant of the Source.");
#endif
// Edges between RefSCCs are the same regardless of call or ref, so we can
// just flip the edge here.
SourceN->setEdgeKind(TargetN, Edge::Call);
#ifndef NDEBUG
// Check that the RefSCC is still valid.
verify();
#endif
}
void LazyCallGraph::RefSCC::switchOutgoingEdgeToRef(Node &SourceN,
Node &TargetN) {
assert((*SourceN)[TargetN].isCall() && "Must start with a call edge!");
assert(G->lookupRefSCC(SourceN) == this && "Source must be in this RefSCC.");
assert(G->lookupRefSCC(TargetN) != this &&
"Target must not be in this RefSCC.");
#ifdef EXPENSIVE_CHECKS
assert(G->lookupRefSCC(TargetN)->isDescendantOf(*this) &&
"Target must be a descendant of the Source.");
#endif
// Edges between RefSCCs are the same regardless of call or ref, so we can
// just flip the edge here.
SourceN->setEdgeKind(TargetN, Edge::Ref);
#ifndef NDEBUG
// Check that the RefSCC is still valid.
verify();
#endif
}
void LazyCallGraph::RefSCC::insertInternalRefEdge(Node &SourceN,
Node &TargetN) {
assert(G->lookupRefSCC(SourceN) == this && "Source must be in this RefSCC.");
assert(G->lookupRefSCC(TargetN) == this && "Target must be in this RefSCC.");
SourceN->insertEdgeInternal(TargetN, Edge::Ref);
#ifndef NDEBUG
// Check that the RefSCC is still valid.
verify();
#endif
}
void LazyCallGraph::RefSCC::insertOutgoingEdge(Node &SourceN, Node &TargetN,
Edge::Kind EK) {
// First insert it into the caller.
SourceN->insertEdgeInternal(TargetN, EK);
assert(G->lookupRefSCC(SourceN) == this && "Source must be in this RefSCC.");
RefSCC &TargetC = *G->lookupRefSCC(TargetN);
assert(&TargetC != this && "Target must not be in this RefSCC.");
#ifdef EXPENSIVE_CHECKS
assert(TargetC.isDescendantOf(*this) &&
"Target must be a descendant of the Source.");
#endif
// The only change required is to add this SCC to the parent set of the
// callee.
TargetC.Parents.insert(this);
#ifndef NDEBUG
// Check that the RefSCC is still valid.
verify();
#endif
}
SmallVector<LazyCallGraph::RefSCC *, 1>
LazyCallGraph::RefSCC::insertIncomingRefEdge(Node &SourceN, Node &TargetN) {
assert(G->lookupRefSCC(TargetN) == this && "Target must be in this RefSCC.");
RefSCC &SourceC = *G->lookupRefSCC(SourceN);
assert(&SourceC != this && "Source must not be in this RefSCC.");
#ifdef EXPENSIVE_CHECKS
assert(SourceC.isDescendantOf(*this) &&
"Source must be a descendant of the Target.");
#endif
SmallVector<RefSCC *, 1> DeletedRefSCCs;
#ifndef NDEBUG
// In a debug build, verify the RefSCC is valid to start with and when this
// routine finishes.
verify();
auto VerifyOnExit = make_scope_exit([&]() { verify(); });
#endif
int SourceIdx = G->RefSCCIndices[&SourceC];
int TargetIdx = G->RefSCCIndices[this];
assert(SourceIdx < TargetIdx &&
"Postorder list doesn't see edge as incoming!");
// Compute the RefSCCs which (transitively) reach the source. We do this by
// working backwards from the source using the parent set in each RefSCC,
// skipping any RefSCCs that don't fall in the postorder range. This has the
// advantage of walking the sparser parent edge (in high fan-out graphs) but
// more importantly this removes examining all forward edges in all RefSCCs
// within the postorder range which aren't in fact connected. Only connected
// RefSCCs (and their edges) are visited here.
auto ComputeSourceConnectedSet = [&](SmallPtrSetImpl<RefSCC *> &Set) {
Set.insert(&SourceC);
SmallVector<RefSCC *, 4> Worklist;
Worklist.push_back(&SourceC);
do {
RefSCC &RC = *Worklist.pop_back_val();
for (RefSCC &ParentRC : RC.parents()) {
// Skip any RefSCCs outside the range of source to target in the
// postorder sequence.
int ParentIdx = G->getRefSCCIndex(ParentRC);
assert(ParentIdx > SourceIdx && "Parent cannot precede source in postorder!");
if (ParentIdx > TargetIdx)
continue;
if (Set.insert(&ParentRC).second)
// First edge connecting to this parent, add it to our worklist.
Worklist.push_back(&ParentRC);
}
} while (!Worklist.empty());
};
// Use a normal worklist to find which SCCs the target connects to. We still
// bound the search based on the range in the postorder list we care about,
// but because this is forward connectivity we just "recurse" through the
// edges.
auto ComputeTargetConnectedSet = [&](SmallPtrSetImpl<RefSCC *> &Set) {
Set.insert(this);
SmallVector<RefSCC *, 4> Worklist;
Worklist.push_back(this);
do {
RefSCC &RC = *Worklist.pop_back_val();
for (SCC &C : RC)
for (Node &N : C)
for (Edge &E : *N) {
RefSCC &EdgeRC = *G->lookupRefSCC(E.getNode());
if (G->getRefSCCIndex(EdgeRC) <= SourceIdx)
// Not in the postorder sequence between source and target.
continue;
if (Set.insert(&EdgeRC).second)
Worklist.push_back(&EdgeRC);
}
} while (!Worklist.empty());
};
// Use a generic helper to update the postorder sequence of RefSCCs and return
// a range of any RefSCCs connected into a cycle by inserting this edge. This
// routine will also take care of updating the indices into the postorder
// sequence.
iterator_range<SmallVectorImpl<RefSCC *>::iterator> MergeRange =
updatePostorderSequenceForEdgeInsertion(
SourceC, *this, G->PostOrderRefSCCs, G->RefSCCIndices,
ComputeSourceConnectedSet, ComputeTargetConnectedSet);
// Build a set so we can do fast tests for whether a RefSCC will end up as
// part of the merged RefSCC.
SmallPtrSet<RefSCC *, 16> MergeSet(MergeRange.begin(), MergeRange.end());
// This RefSCC will always be part of that set, so just insert it here.
MergeSet.insert(this);
// Now that we have identified all of the SCCs which need to be merged into
// a connected set with the inserted edge, merge all of them into this SCC.
SmallVector<SCC *, 16> MergedSCCs;
int SCCIndex = 0;
for (RefSCC *RC : MergeRange) {
assert(RC != this && "We're merging into the target RefSCC, so it "
"shouldn't be in the range.");
// Merge the parents which aren't part of the merge into the our parents.
for (RefSCC *ParentRC : RC->Parents)
if (!MergeSet.count(ParentRC))
Parents.insert(ParentRC);
RC->Parents.clear();
// Walk the inner SCCs to update their up-pointer and walk all the edges to
// update any parent sets.