A Snapshot of the Performance of Wasm Backends for Managed Languages
Year: 2025
Type: conference-paper
Abstract: WebAssembly (Wasm) has been extended to support features such as garbage collection, references, exceptions, and tail calls that facilitate compilation of managed languages. In this paper, we capture a snapshot of the performance of languages that use these new capabilities from two perspectives. First, we present a language-by-language performance comparison of six managed language implementations on Wasm to the performance to their native implementations. Second, we focus on the implementation of the Bigloo Scheme compiler and explore the impact of different choices for compiling specific aspects of the language. Our findings suggest that Wasm has become a promising compilation target for most managed languages, but that its performance still falls short of that achieved by native code. Our results also show that the quality of the Wasm implementations vary, with the best ones being, on average, about 1.4× slower than the native backend and the worst ones seeing average slowdowns of more than 8× with some tests even failing to execute correctly.
Authors: Manuel Serrano, Robert Bruce Findler
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Funder: National Science Foundation
OpenAlex ID: W4415002667