A probabilistic pointer analysis for speculative optimizations

IEEE Trans. Parallel Distrib. Syst.

Hwang

et al. 2012

Probabilistic pointer analysis (PPA) is a compile-time analysis method that estimates the probability that a points-to relationship will hold at a particular program point. The results are useful for optimizing and parallelizing compilers, which need to quantitatively assess the profitability of transformations when performing aggressive optimizations and parallelization. This paper presents a PPA technique using the static single assignment (SSA) form. When computing the probabilistic points-to relationships of a specific pointer, a pointer relation graph (PRG) is first built to represent all of the possible points-to relationships of the pointer. The PRG is transformed by a sequence of reduction operations into a compact graph, from which the probabilistic points-to relationships of the pointer can be determined. In addition, PPA is further extended to interprocedural cases by considering function related statements. We have implemented our proposed scheme including static and profiling versions in the Open64 compiler, and performed experiments to obtain the accuracy and scalability. The static version estimates branch probabilities by assuming that every conditional is equally likely to be true or false, and that every loop executes 10 times before terminating. The profiling version measures branch probabilities dynamically from past program executions using a default workload provided with the benchmark. The average errors for selected benchmarks were 3.80 percent in the profiling version and 9.13 percent in the static version. Finally, SPEC CPU2006 is used to evaluate the scalability, and the result indicates that our scheme is sufficiently efficient in practical use. The average analysis time was 35.59 seconds for an average of 98,696 lines of code.

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Support of Probabilistic Pointer Analysis in the SSA Form

Hung

IEEE Trans. Parallel Distrib. Syst.

Hwang

et al. 2012

“…Silva and Steffan [34] proposed a one-level context-sensitive and flow-sensitive probabilistic pointer analysis algorithm that statically predicts the probability of each pointsto relation at every program point. Their algorithm computes points-to probabilities through the use of linear transfer functions that are efficiently encoded as sparse matrices.…”

Section: Analysis Performancementioning

confidence: 99%

Probabilistic Points-to Analysis for Java

Sun

Zhao

Lecture Notes in Computer Science

2011

Probabilistic points-to analysis is an analysis technique for defining the probabilities on the points-to relations in programs. It provides the compiler with some optimization chances such as speculative dead store elimination, speculative redundancy elimination, and speculative code scheduling. Although several static probabilistic points-to analysis techniques have been developed for C language, they cannot be applied directly to Java because they do not handle the classes, objects, inheritances and invocations of virtual methods. In this paper, we propose a context-insensitive and flow-sensitive probabilistic points-to analysis for Java (JPPA) for statically predicting the probability of points-to relations at all program points (i.e., points before or after statements) of a Java program. JPPA first constructs an interprocedural control flow graph (ICFG) for a Java program, whose edges are labeled with the probabilities calculated by an algorithm based on a static branch prediction approach, and then calculates the probabilistic points-to relations of the program based upon the ICFG. We have also developed a tool called Lukewarm to support JPPA and conducted an experiment to compare JPPA with a traditional context-insensitive and flow-sensitive points-to analysis approach. The experimental results show that JPPA is a precise and effective probabilistic points-to analysis technique for Java.

“…Chen et al [6] developed an algorithm to estimate the probabilities of loop-carried pointer-induced data-dependence relationships. Silva and Steffan [35] introduced linear transfer functions to model the problem of probabilistic alias analysis, whereby the transfer functions are efficiently encoded as sparse matrices. They demonstrated that the approach provides accurate probabilities at compile time.…”

Section: Probabilistic Alias Analysismentioning

confidence: 99%

Probabilistic Alias Analysis of Executable Code

2010

Int J Parallel Prog

In this paper we present a method for flow-sensitive, context-insensitive probabilistic alias analysis at the assembly level. A memory disambiguation algorithm is also developed for revealing the probability of two registers holding the same memory location. The alias analysis and memory disambiguation algorithms are implemented based on the Diablo post-link optimizer. Experimental results show that the technique can estimate the probabilities that registers refer to the same memory address in benchmark programs with an overall average error of about 6.8%. The post-link optimizer can leverage the obtained quantitative information to facilitate aggressive analyses and optimizations.