Correlating Effectiveness of Pointer Analysis Techniques with Patterns in Embedded System Code

TL;DR

This study evaluates the effectiveness of different pointer analysis techniques on embedded system code, revealing that most cases show similar precision, with slight improvements from context-sensitive analysis, and correlates these gains with coding patterns.

Contribution

It is the first to correlate pointer analysis precision gains with coding patterns in embedded systems, comparing three analysis techniques across industry-used code sets.

Findings

90% of cases had identical pointer information across all analysis types

Context-sensitive analysis was 2.6% more precise than flow-sensitive

Flow-sensitive analysis was 6.8% more precise than flow-insensitive

Abstract

A pointer analysis maps the pointers in a program to the memory locations they point to. In this work, we study the effectiveness of the three flavors of pointer analysis namely flow sensitive, flow insensitive, and context sensitive analysis on seven embedded code sets used in the industry. We compare precision gain i.e., the reduction in the number of spurious memory locations pointed by a pointer in each of these settings. We found that in 90% of cases the pointer information was same in all three settings. In other cases, context sensitive analysis was 2.6% more precise than flow sensitive analysis which was 6.8% more precise than flow insensitive analysis on average. We correlate precision gain with coding patterns in the embedded systems-which we believe to be first of its kind activity.