Annotating Java class files with virtual registers for performance

Abstract

SUMMARYThe Java .class file is a compact encoding of programs for a stack-based virtual machine. It is intended for use in a networked environment, which requires machine independence and minimized consumption of network bandwidth. However, as in all interpreted virtual machines, performance does not match that of code generated for the target machine. We propose verifiable, machine-independent annotations to the Java class file to bring the quality of the code generated by a 'just-in-time' compiler closer to that of an optimizing compiler without a significant increase in code generation time. This division of labor has expensive machine-independent analysis performed off-line and inexpensive machine-dependent codegeneration performed on the client. We call this phenomenon 'super-linear analysis and linear exploitation.' These annotations were designed mindful of the concurrency features of the Java language. In this paper we report results from our machine-independent, prioritized register assignment. We also discuss other possible annotations.

Split Register Allocation: Linear Complexity Without the Performance Penalty

“…This can lead to annotation bloat (although linear), with total annotation size potentially larger than the bytecode itself. Jones and Kamin do not address the problem [5]. 2.…”

“…This is several orders of magnitude more effective than state-of-the-art approaches [5], and even comes with a formal guarantee about optimality. The addition compression row reports the benefits of Algorithm 3, and confirm its important role in making the annotation size negligible w.r.t.…”

“…We discard annotations regarding those live ranges when compiling for another instruction-set architecture (ISA). 5 Fortunately, besides representing a small minority, these live ranges also feature a very short temporal locality and a low degree of interference with other live ranges. This reduces the chances of impacting an important allocation decision that would result in a significant performance difference.…”

“…This work demonstrated how to achieve performance competitive with native priority-based graph coloring allocation. Jones and Kamin [5] extended their virtual register allocation approach, dealing with correctness, calling conventions and portability (addressing variations of the number of physical registers only).…”

Coupling on-line and off-line profile information to improve program performance

“…More recently, off-line profile information has been communicated as part of class files via annotation to reduce dynamic compilation overhead [9]. Annotation has also been used to communicate analysis information that is collected off-line to enable optimized execution times [14,8,4]. Our system combines the use of annotation with adaptive optimization to enable optimized execution times with very little on-line overhead.…”