Proceedings of the 2014 International Conference on Principles and Practices of Programming on the Java Platform: Virtual Machi 2014
DOI: 10.1145/2647508.2647524
|View full text |Cite
|
Sign up to set email alerts
|

Comparing points-to static analysis with runtime recorded profiling data

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
6
0

Year Published

2015
2015
2024
2024

Publication Types

Select...
3
2
1

Relationship

0
6

Authors

Journals

citations
Cited by 8 publications
(6 citation statements)
references
References 26 publications
0
6
0
Order By: Relevance
“…Any unsoundness is, in principle, bad, so a quantification is treacherous. However, following the example of other work Lhoták 2012, 2013;Li et al 2014;Stancu et al 2014], we quantify the unsoundness of the static analysis in terms of missing callgraph edges, compared to a dynamic call-graph. We consider only call-graph edges originating from application code, since library classes contain a fair amount of non-analyzable native methods.…”
Section: Discussionmentioning
confidence: 99%
See 1 more Smart Citation
“…Any unsoundness is, in principle, bad, so a quantification is treacherous. However, following the example of other work Lhoták 2012, 2013;Li et al 2014;Stancu et al 2014], we quantify the unsoundness of the static analysis in terms of missing callgraph edges, compared to a dynamic call-graph. We consider only call-graph edges originating from application code, since library classes contain a fair amount of non-analyzable native methods.…”
Section: Discussionmentioning
confidence: 99%
“…Our experiments are substantially more representative in regards to the actual empirical soundness of a joint reflection and pointer analysis. Stancu et al [2014] present an empirical study that compares profiling data with a pointsto static analysis, but do not support reflection. They target only the most reflection-light benchmarks of the DaCapo 9.12-Bach suite (avrora, luindex, and lusearch), and patch the code to avoid reflection entirely.…”
Section: Related Workmentioning
confidence: 95%
“…The LLVMbased backend works in several phases. In the first phase, it performs points-to analysis [30] to discover all the code it will possibly have to compile. Then, an ahead-of-time inlining phase is performed.…”
Section: Proposed Approachmentioning
confidence: 99%
“…Call-graph construction is one of the best-known clients of points-to analysis [2,3,80] and has the added benefit of quantifying how much code the analysis truly reaches. We compare the call-graph edges found by our static analysis to a dynamic call-graph-a comparison also found in other recent work [126]. For a sound static analysis, no edge should occur dynamically but not predicted statically.…”
Section: Rq2: Do the Presented Techniques Have Reasonable Running Times?mentioning
confidence: 60%
“…Reflection is one of their examples but they only apply it to small benchmarks. Stancu et al [126] empirically compare profiling data with a points-to static analysis. However, they target only the most reflection-light benchmarks of the DaCapo 9.12-Bach suite (avrora, luindex, and lusearch), and patch the code to avoid reflection entirely.…”
Section: Static Analysis and Reflectionmentioning
confidence: 99%