Profile-guided receiver class prediction

Abstract

The use of dynamically-dispatched procedure calls is a key mechanism for writing extensible and flexible code in object-oriented languages. Unfortunately, dynamic dispatching imposes a runtime performance penalty. Some recent implementations of pure object-oriented languages have utilized profile-guided receiver class prediction to reduce this performance penalty, and some researchers have argued for applying receiver class prediction in hybrid languages like C++. We performed a detailed examination of the dynamic profiles of eight large object-oriented applications written in C++ and Cecil, determining that the receiver class distributions are strongly peaked and stable across both inputs and program versions through time. We describe techniques for gathering and manipulating profile information at varying degrees of precision, particularly in the presence of optimizations such as inlining. Our implementation of profile-guided receiver class prediction improves the performance of large Cecil applications by more than a factor of two over solely static optimizations.

Adaptive online context-sensitive inlining

“…Previous work on offline optimization of object-oriented languages demonstrated that using context-sensitive trace profiling can enable significant improvements in the effectiveness of profile directed inlining [13]. The goal of our work is to achieve these benefits online in a virtual machine.…”

“…Unlike Jikes RVM and the IBM DK, HotSpot only gathers call edge data during initial interpretation of a method and thus is vulnerable to mispredictions if the application's calling patterns are different during initialization and steady state execution. The Vortex optimizing compiler used offline contextsensitive profile data to guide its inlining decisions [13]. Vortex demonstrated that making inlining decisions using context-sensitive profile data can be quite valuable in optimizing object-oriented programs: context-sensitive profile data enabled speedups of up to 24% over context-insensitive profile data for large Cecil programs.…”

“…One common technique to improve the quality of inlining decisions is to utilize profile information to identify frequently executed call edges [8,15,13]. Focusing inlining effort on the most frequently executed call sites increases the odds of maximizing benefit and minimizing cost.…”

“…Focusing inlining effort on the most frequently executed call sites increases the odds of maximizing benefit and minimizing cost. An additional benefit of profile-directed inlining is that it can enable guarded inlining of the dominant target(s) of a virtual dispatch when the distribution of receiver classes is skewed [6,15,13].…”

“…Previous work in offline profile-directed inlining has demonstrated that context-sensitivity can increase the benefits and reduce the costs of profile directed inlining [13]. In this paper, we study the effectiveness of adaptive, online, contextsensitive profile information to guide inlining decisions in the Jikes TM Research Virtual Machine (Jikes RVM).…”