fix incorrect implementation of getting CallGraphSCCRevTopoOrder #1640
Conversation
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #1640 +/- ##
=======================================
Coverage 63.70% 63.70%
=======================================
Files 245 245
Lines 25999 26001 +2
Branches 4509 4508 -1
=======================================
+ Hits 16563 16565 +2
Misses 9436 9436
|
|
Would it be possible to put the implementation in SCC.h so that every graph including calligraphscc can use it? Line 128 in 5e0491c |
| NodeStack revTopoStack; | ||
| NodeStack& topoStack = this->getSCCDetector()->topoNodeStack(); | ||
| while (!topoStack.empty()) | ||
| NodeStack revTopoStack = this->getSCCDetector()->revTopoNodeStack(); |
There was a problem hiding this comment.
Why making changes on SCCDetect ? It was correct, right?
svf/lib/MSSA/MemRegion.cpp
Outdated
| @@ -250,6 +250,7 @@ void MRGenerator::collectModRefForCall() | |||
| { | |||
| PTACallGraphNode* subCallGraphNode = callGraph->getCallGraphNode(*it); | |||
| /// Get mod-ref of all callsites calling callGraphNode | |||
| std::cout << subCallGraphNode->getFunction()->getName() << std::endl; | |||
There was a problem hiding this comment.
remove this debugging line.
svf/lib/WPA/Andersen.cpp
Outdated
| while (!topoOrder.empty()) | ||
| NodeStack revTopoOrder = getSCCDetector()->revTopoNodeStack(); | ||
| /// referenced topoOrder is used for restoring the topological order for later solving. | ||
| NodeStack &topoOrder = getSCCDetector()->topoNodeStack(); |
There was a problem hiding this comment.
topoOrder has been there when you retrieve it and should not be changed later at line 725.
svf/include/Graphs/SCC.h
Outdated
| GNodeStack &topoOrder = topoNodeStack(); | ||
| while(!topoOrder.empty()) | ||
| { | ||
| NodeID callgraphNodeID = topoOrder.top(); |
There was a problem hiding this comment.
callgraphNodeID => nodeID
svf/lib/WPA/Andersen.cpp
Outdated
| NodeID nodeId = revTopoOrder.top(); | ||
| revTopoOrder.pop(); | ||
| topoOrder.push(nodeId); | ||
| /// restore the topological order for later solving. |
There was a problem hiding this comment.
no restore is needed as you never popped any element in topoOrder
svf/lib/WPA/Andersen.cpp
Outdated
| while (!topoOrder.empty()) | ||
| NodeStack revTopoOrder = getSCCDetector()->revTopoNodeStack(); | ||
| /// referenced topoOrder is used for restoring the topological order for later solving. | ||
| NodeStack &topoOrder = getSCCDetector()->topoNodeStack(); |
There was a problem hiding this comment.
Remove this line: NodeStack &topoOrder = getSCCDetector()->topoNodeStack()
svf/lib/WPA/Andersen.cpp
Outdated
| revTopoOrder.pop(); | ||
| topoOrder.push(nodeId); | ||
| /// restore the topological order for later solving. | ||
| topoOrder.push(repNodeId); |
There was a problem hiding this comment.
Remove this line topoOrder.push(repNodeId);
|
Could you run the below bcs and ensure you patch produces the same average points-to size using https://app.filemail.com/d/uayyovelgnuoixz |
|
You could also validate using your local bcs and report back the stats. |
Yeah, I first validate the second url. The first url seems be out-dated. |
Do you have any bc in your local machine? Try to validate another bc and upload your bc and results. |
Yes, I find some bcs to validate. |
|
I choose 3 project bcs with top size in https://github.com/SVF-tools/Test-Suite/tree/master/test_cases_bc/crux-bc I run the In the following bash.bc, source is: https://github.com/SVF-tools/Test-Suite/blob/master/test_cases_bc/crux-bc/bash.bc $ diff -u bash.bc.before.txt bash.bc.after.txt
--- bash.bc.before.txt 2025-01-26 19:15:13
+++ bash.bc.after.txt 2025-01-26 19:27:00
@@ -19,7 +19,7 @@
IndCallSites 72
LoadsNum 59146
MaxStructSize 24
-NonPtrObj 8132
+NonPtrObj 8135
ReturnsNum 5385
StackObjs 9595
StoresNum 27665
@@ -32,9 +32,9 @@
VarArrayObj 157
VarStructObj 462
----------------Time and memory stats--------------------
-LLVMIRTime 5.542
-SVFIRTime 7.674
-SymbolTableTime 2.728
+LLVMIRTime 6.388
+SVFIRTime 7.678
+SymbolTableTime 2.713
#######################################################
*********PTACallGraph Stats (Andersen analysis)***************
@@ -65,7 +65,7 @@
MemoryUsageVmsize 0
SCCDetectTime 0
SCCMergeTime 0
-TotalTime 58.191
+TotalTime 52.389
UpdateCGTime 0
----------------Numbers stats----------------------------
AddrProcessed 15085
@@ -116,21 +116,21 @@
****Persistent Points-To Cache Statistics: Andersen's analysis bitvector****
################ (program : bash.bc)###############
UniquePointsToSets 15689
-TotalUnions 174202
+TotalUnions 174203
PropertyUnions 142918
UniqueUnions 1401
LookupUnions 27249
-PreemptiveUnions 2634
+PreemptiveUnions 2635
TotalComplements 668673
PropertyComplements 644062
UniqueComplements 515
LookupComplements 23581
PreemptiveComplements 515
-TotalIntersections 1067892
-PropertyIntersections 1066301
-UniqueIntersections 5
+TotalIntersections 1067893
+PropertyIntersections 1066300
+UniqueIntersections 6
LookupIntersections 552
-PreemptiveIntersections 1034
+PreemptiveIntersections 1035
#######################################################
Process finished with exit code 0libcurl.so.bc, source is: https://github.com/SVF-tools/Test-Suite/blob/master/test_cases_bc/crux-bc/libcurl.so.bc $ diff -u libcurl.so.bc.before.txt libcurl.so.bc.after.txt
--- libcurl.so.bc.before.txt 2025-01-26 19:17:05
+++ libcurl.so.bc.after.txt 2025-01-26 19:29:09
@@ -32,9 +32,9 @@
VarArrayObj 220
VarStructObj 199
----------------Time and memory stats--------------------
-LLVMIRTime 3.352
-SVFIRTime 5.138
-SymbolTableTime 1.566
+LLVMIRTime 3.648
+SVFIRTime 5.701
+SymbolTableTime 1.737
#######################################################
*********PTACallGraph Stats (Andersen analysis)***************
@@ -65,7 +65,7 @@
MemoryUsageVmsize 0
SCCDetectTime 0
SCCMergeTime 0
-TotalTime 65.245
+TotalTime 61.742
UpdateCGTime 0
----------------Numbers stats----------------------------
AddrProcessed 10897
@@ -115,22 +115,22 @@
****Persistent Points-To Cache Statistics: Andersen's analysis bitvector****
################ (program : libcurl.so.bc)###############
-UniquePointsToSets 17799
+UniquePointsToSets 17800
TotalUnions 271805
-PropertyUnions 82352
-UniqueUnions 5906
-LookupUnions 173276
-PreemptiveUnions 10271
-TotalComplements 1215598
+PropertyUnions 82343
+UniqueUnions 5905
+LookupUnions 173285
+PreemptiveUnions 10272
+TotalComplements 1215597
PropertyComplements 1042920
UniqueComplements 5128
-LookupComplements 162427
+LookupComplements 162426
PreemptiveComplements 5123
-TotalIntersections 507640
-PropertyIntersections 494808
+TotalIntersections 507641
+PropertyIntersections 494809
UniqueIntersections 59
-LookupIntersections 2489
-PreemptiveIntersections 10284
+LookupIntersections 2488
+PreemptiveIntersections 10285
#######################################################
Process finished with exit code 0tmux.bc, source is: https://github.com/SVF-tools/Test-Suite/blob/master/test_cases_bc/crux-bc/tmux.bc $ diff -u tmux.bc.before.txt tmux.bc.after.txt
--- tmux.bc.before.txt 2025-01-26 19:20:19
+++ tmux.bc.after.txt 2025-01-26 19:33:41
@@ -32,9 +32,9 @@
VarArrayObj 127
VarStructObj 385
----------------Time and memory stats--------------------
-LLVMIRTime 4.205
-SVFIRTime 7.519
-SymbolTableTime 2.422
+LLVMIRTime 4.263
+SVFIRTime 8.364
+SymbolTableTime 2.929
#######################################################
*********PTACallGraph Stats (Andersen analysis)***************
@@ -65,7 +65,7 @@
MemoryUsageVmsize 0
SCCDetectTime 0
SCCMergeTime 0
-TotalTime 137.919
+TotalTime 132.443
UpdateCGTime 0
----------------Numbers stats----------------------------
AddrProcessed 14918
@@ -116,21 +116,21 @@
****Persistent Points-To Cache Statistics: Andersen's analysis bitvector****
################ (program : tmux.bc)###############
UniquePointsToSets 20463
-TotalUnions 734596
+TotalUnions 734597
PropertyUnions 137709
UniqueUnions 6728
LookupUnions 577604
-PreemptiveUnions 12555
+PreemptiveUnions 12556
TotalComplements 1868738
PropertyComplements 1330628
UniqueComplements 5273
LookupComplements 527564
PreemptiveComplements 5273
-TotalIntersections 1197616
-PropertyIntersections 1187010
-UniqueIntersections 20
-LookupIntersections 28
-PreemptiveIntersections 10558
+TotalIntersections 1197617
+PropertyIntersections 1187016
+UniqueIntersections 21
+LookupIntersections 21
+PreemptiveIntersections 10559
#######################################################
Process finished with exit code 0
|
|
Could you also run this https://github.com/user-attachments/files/18392297/502.gcc_r-16.bc.zip |
Yeah, I have a try. |
When I run two testcases, all nearly full of memory with my laptop. I don't know why the time cost difference is 1000+ s. I also find that: wpa will not reached the function My laptop is mac os: I run two testcases in release mode, and diff is here below. $ diff -u 502.gcc_r-16.bc.before.txt 502.gcc_r-16.bc.after.txt
--- 502.gcc_r-16.bc.before.txt 2025-01-27 19:45:11
+++ 502.gcc_r-16.bc.after.txt 2025-01-27 14:38:53
@@ -30,9 +30,9 @@
VarArrayObj 1505
VarStructObj 5840
----------------Time and memory stats--------------------
-LLVMIRTime 47.759
-SVFIRTime 47.614
-SymbolTableTime 22.9
+LLVMIRTime 51.387
+SVFIRTime 54.674
+SymbolTableTime 23.792
#######################################################
*********PTACallGraph Stats (Andersen analysis)***************
@@ -63,7 +63,7 @@
MemoryUsageVmsize 0
SCCDetectTime 0
SCCMergeTime 0
-TotalTime 6557.64
+TotalTime 7626.45
UpdateCGTime 0
----------------Numbers stats----------------------------
AddrProcessed 164468
@@ -139,7 +139,7 @@
MemoryUsageVmsize 0
SCCDetectTime 0
SCCMergeTime 0
-TotalTime 6608.77
+TotalTime 7710.31
UpdateCGTime 0
----------------Numbers stats----------------------------
AddrProcessed 164468
@@ -190,20 +190,20 @@
****Persistent Points-To Cache Statistics: Andersen's analysis bitvector****
################ (program : 502.gcc_r-16.bc)###############
UniquePointsToSets 204056
-TotalUnions 5326743
+TotalUnions 5326741
PropertyUnions 3024648
UniqueUnions 30687
LookupUnions 2212694
-PreemptiveUnions 58714
+PreemptiveUnions 58712
TotalComplements 24523125
PropertyComplements 22608331
UniqueComplements 19815
LookupComplements 1875164
PreemptiveComplements 19815
-TotalIntersections 66660467
-PropertyIntersections 66620516
-UniqueIntersections 28
-LookupIntersections 278
-PreemptiveIntersections 39645
+TotalIntersections 66660465
+PropertyIntersections 66620533
+UniqueIntersections 25
+LookupIntersections 264
+PreemptiveIntersections 39643
####################################################### |
|
You should use release SVF for testing. The time differences are quite large. Could you re-try it? It would be good to understand which modifications you made in the patch affected the performance. |
it's release mode. I have a re-try to understand some other problems. |
|
I run two testcases in release mode every time I restart the machine. I think the time cost is not real, it's not memory efficient in my laptop, maybe it spends much time to swap the memory. Maybe when I have a high performance computer (with more memory, e.g. 64GB memory) in hands, I will re-try later again. 
Here below is diff: diff -u before.txt after.txt
--- before.txt 2025-01-28 11:43:03
+++ after.txt 2025-01-28 11:43:21
@@ -17,7 +17,7 @@
IndCallSites 3208
LoadsNum 792650
MaxStructSize 266
-NonPtrObj 68433
+NonPtrObj 68428
ReturnsNum 121494
StackObjs 111699
StoresNum 287262
\ No newline at end of file
@@ -30,15 +30,15 @@
VarArrayObj 1505
VarStructObj 5840
----------------Time and memory stats--------------------
-LLVMIRTime 46.64
-SVFIRTime 43.333
-SymbolTableTime 22.643
+LLVMIRTime 77.089
+SVFIRTime 69.757
+SymbolTableTime 24.676
#######################################################
*********PTACallGraph Stats (Andersen analysis)***************
################ (program : 502.gcc_r-16.bc)###############
----------------Numbers stats----------------------------
-CalRetPairInCycle 138791
+CalRetPairInCycle 138795
MaxNodeInCycle 13798
NodeInCycle 14258
TotalCycle 245
\ No newline at end of file
@@ -63,7 +63,7 @@
MemoryUsageVmsize 0
SCCDetectTime 0
SCCMergeTime 0
-TotalTime 5440.85
+TotalTime 5634.06
UpdateCGTime 0
----------------Numbers stats----------------------------
AddrProcessed 164468
\ No newline at end of file
@@ -114,7 +114,7 @@
*********PTACallGraph Stats (Andersen analysis)***************
################ (program : 502.gcc_r-16.bc)###############
----------------Numbers stats----------------------------
-CalRetPairInCycle 138791
+CalRetPairInCycle 138795
MaxNodeInCycle 13798
NodeInCycle 14258
TotalCycle 245
\ No newline at end of file
@@ -139,7 +139,7 @@
MemoryUsageVmsize 0
SCCDetectTime 0
SCCMergeTime 0
-TotalTime 5480.38
+TotalTime 5673.53
UpdateCGTime 0
----------------Numbers stats----------------------------
AddrProcessed 164468
\ No newline at end of file
@@ -190,19 +190,19 @@
****Persistent Points-To Cache Statistics: Andersen's analysis bitvector****
################ (program : 502.gcc_r-16.bc)###############
UniquePointsToSets 204113
-TotalUnions 5327937
+TotalUnions 5327933
PropertyUnions 3024978
UniqueUnions 30760
LookupUnions 2213359
-PreemptiveUnions 58840
+PreemptiveUnions 58836
TotalComplements 24522206
PropertyComplements 22607432
UniqueComplements 19856
LookupComplements 1875062
PreemptiveComplements 19856
-TotalIntersections 66662738
-PropertyIntersections 66622702
-UniqueIntersections 30
-LookupIntersections 277
-PreemptiveIntersections 39729
+TotalIntersections 66662744
+PropertyIntersections 66622724
+UniqueIntersections 25
+LookupIntersections 270
+PreemptiveIntersections 39725
####################################################### |
|
@bjjwwang could you test the time differences on our machine? |
an example of test.c:
before fix,
MRGenerator::getCallGraphSCCRevTopoOrderwill get:after fix, it will get: