Paper Info
Title
Parallel incremental whole-program optimizations for Scala.js.
Abstract
Whole-program optimizations are powerful tools that can dramatically improve performance, size and other aspects of programs. Because they depend on global knowledge, they must typically be reapplied to the whole program when small changes are made, which makes them too slow for the development cycle. This is an issue for some environments that require, or benefit a lot from, whole-program optimizations, such as compilation to JavaScript or to the Dalvik VM, because their development cycle is slowed down either by the lack of optimizations, or by the time spent on applying them. We present a new approach to designing incremental whole-program optimizers for object-oriented and functional languages: when part of a program changes, only the portions affected by the changes are reoptimized. An incremental optimizer using this approach for Scala.js, the Scala to JavaScript compiler, demonstrates speedups from 10x to 100x compared to its batch version. As a result, the optimizer's running time becomes insignificant compared to separate compilation, making it fit for use on every compilation run during the development cycle. We also show how to parallelize the incremental algorithm to take advantage of multicore hardware.
Year
DOI
Venue
2016
10.1145/2983990.2984013
OOPSLA
Field
DocType
Citations 
Programming language,Scala,Functional programming,Computer science,Parallel computing,Compiler,Single Compilation Unit,Multi-core processor,JavaScript
Conference
1
PageRank 
References 
Authors
0.34
15
2
Name
Order
Citations
PageRank
Sébastien Doeraene131.44
Tobias Schlatter2141.85