Improving JavaScript performance by deconstructing the type system

Abstract

Increased focus on JavaScript performance has resulted in vast performance improvements for many benchmarks. However, for actual code used in websites, the attained improvements often lag far behind those for popular benchmarks.This paper shows that the main reason behind this shortfall is how the compiler understands types. JavaScript has no concept of types, but the compiler assigns types to objects anyway for ease of code generation. We examine the way that the Chrome V8 compiler defines types, and identify two design decisions that are the main reasons for the lack of improvement: (1) the inherited prototype object is part of the current object's type definition, and (2) method bindings are also part of the type definition. These requirements make types very unpredictable, which hinders type specialization by the compiler. Hence, we modify V8 to remove these requirements, and use it to compile the JavaScript code assembled by JSBench from real websites. On average, we reduce the execution time of JSBench by 36%, and the dynamic instruction count by 49%.

TypeDevil: Dynamic Type Inconsistency Analysis for JavaScript

“…We reported this problem to the developers. 3 The root cause is that two arrays that are supposed to have the same length have a slightly different length. The problem does not crash the benchmark but leads to nonsensical computation.…”

“…Optimizing JIT compilers benefit from inferred types to emit type-specialized code [20], [24], [36]. Ahn et al [3] improve the object representation to enable optimizations even though objects with different prototypes appear at a single code location. Our work avoids this problem by defining types independent of the prototype object.…”

Memento mori: dynamic allocation-site-based optimizations

“…It is sufficient to check a single word of memory directly after the object to determine if the object carries a memento. 1 Why not use an extra word in the object or information in the metadata of the object to store allocation site information? What about a hash map?…”

“…A detailed description of maps is beyond the scope of the paper; other works offer more detail, e.g. [1].…”

Energy and run-time performance practices in software engineering

“…By manually replacing data structures in applications, the authors obtained significant energy savings as illustrated in Table 2.9. Because the authors of the above work report an extensive list of results, we compared the energy consumption of the most and least energy-efficient interface, 10 and we present in Table 2.9 the data structure with the most energy-efficient methods for each interface.…”