Title :
Augmenting JavaScript JIT with ahead-of-time compilation
Author :
Roman Zhuykov;Vahagn Vardanyan;Dmitry Melnik;Ruben Buchatskiy;Eugeniy Sharygin
Author_Institution :
Institute for System Programming of Russian Academy of Sciences, Alexander Solzhenitsyn st., 25, Moscow, Russia
Abstract :
Modern JavaScript engines use just-in-time (JIT) compilation to produce a binary code. JIT compilers are limited in a complexity of optimizations they can perform at a runtime without delaying an execution. On the contrary, ahead-of-time (AOT) compilers don´t have such limitations, but they are not well suited for compiling dynamic languages such as JavaScript. In the paper we discuss methods for augmenting multi-tiered JavaScript JIT with a capability for AOT compilation, so to reduce program startup time and to move complex optimizations to AOT phase. We have implemented saving of JavaScript programs as a binary package containing bytecode and native code in open-source WebKit library. Our implementation allows shipping of JavaScript programs not only as a source code, but also as application binary packages with a precompiled code. In addition, our approach does not require any language feature restrictions. This has resulted in performance gain for popular JavaScript benchmarks such as SunSpider and Kraken on ARM platform, however, at a cost of increased package size.
Keywords :
"Optimization","Engines","Benchmark testing","Loading","Binary codes","Dynamic compiler","Libraries"
Conference_Titel :
Computer Science and Information Technologies (CSIT), 2015
DOI :
10.1109/CSITechnol.2015.7358262