Time- and space-efficient flow-sensitive points-to analysis

Abstract

Compilation of real-world programs often requires hours. The term nightly build known to industrial researchers is an artifact of long compilation times. Our goal is to reduce the absolute analysis times for large C codes (of the order of millions of lines). Pointer analysis is one of the key analyses performed during compilation. Its scalability is paramount to achieve the efficiency of the overall compilation process and its precision directly affects that of the client analyses. In this work, we design a time-and space-efficient flow-sensitive pointer analysis and parallelize it on graphics processing units. Our analysis proposes to use an extended bloom filter, called multibloom, to store points-to information in an approximate manner and develops an analysis in terms of the operations over the multibloom. Since bloom filter is a probabilistic data structure, we develop ways to gain back the analysis precision. We achieve effective parallelization by achieving memory coalescing, reducing thread divergence, and improving load balance across GPU warps. Compared to a state-of-the-art sequential solution, our parallel version achieves a 7.8× speedup with less than 5% precision loss on a suite of six large programs. Using two client transformations, we show that this loss in precision only minimally affects a client's precision. ACM Reference Format:Nasre, R. 2013. Time-and space-efficient flow-sensitive points-to analysis. ACM Trans. analysis directly affects the client analyses and transformations [Hind and Pioli 2000].However, industry-strength compilers need to use flow-insensitive pointer analysis because of the high analysis time and memory cost of a flow-sensitive analysis.The benefit of a flow-sensitive pointer analysis over that of a flow-insensitive analysis has not been clear [Hind and Pioli 1998]. However, it has been shown that a precise pointer analysis is helpful to several clients, such as typestate verification [Fink et al. 2008], security analysis [Chang et al. 2008], bug detection [Guyer and Lin 2005], and the analysis of multithreaded programs [Salcianu and Rinard 2001]. As a result, there is a renewed interest in the area of flow-sensitive pointer analysis, and the scalability of such analyses has been greatly improved [Hardekopf and Lin 2011; Li et al. 2011; Yu et al. 2010; Lhoták and Chung 2011; Hardekopf and Lin 2009; Kahlon 2008]. However, despite these efforts, industrial response to the adoption of these analyses has been lukewarm. For instance, widely used compilers like GCC [GCC 2013] and LLVM [Lattner and Adve 2004] rely on flow-insensitive pointer analysis, despite the known advantages of a flow-sensitive analysis. One of the main reasons behind this pessimistic reaction is high absolute running times of several analyses over large-sized codes. As an example, a state-of-the-art flow-sensitive pointer analysis [Hardekopf and Lin 2011] over gs, an open-source postscript viewer, totaling 0.4 million lines of C code, requires more than half an hour to complete! Considering that pointer a...

An Efficient GPU Implementation of Inclusion-Based Pointer Analysis

“…This analysis is the foundation for a number of areas, including program slicing [31], interprocedural SSA analysis [32], precise pointer analysis [10], [11], [33], and bug detection [3], [34].…”

“…Nasre [33] later introduced another GPU implementation of Andersen's analysis, by also considering flowsensitivity. This work explores the use of a bloom filter, trading off between precision and performance.…”

Parallel Pointer Analysis with CFL-Reachability

“…In recent years, there have been a number of attempts on parallelising pointer analysis algorithms for analysing C or Java programs on multi-core CPUs and/or GPUs [3], [7], [8], [9], [10], [14], [20]. As compared in Table II, all these parallel solutions are different forms of Andersen's pointer analysis [2] with varying precision considerations in terms of context-, flow-and field-sensitivity.…”

“…Their analysis is fieldinsensitive but flow-sensitive (only partially since it does not perform strong updates). Recently, both field-and flowsensitivity (with strong updates for singleton objects) are handled for C programs on multi-core CPUs [9] and GPUs [10]. The speedups are up to 2.6X on eight CPU cores and 7.8X (with precision loss) on a Tesla C2070 GPU, respectively.…”

“…There have been a few parallel implementations of Andersen's pointer analysis on multi-core CPUs [3], [8], [9], [14], GPUs [7], [10], and heterogeneous CPU-GPU systems [20], as compared in detail in Table II. As Andersen's analysis is a whole-program pointer analysis, these earlier solutions cannot be applied to parallelise demand-driven pointer analysis.…”