We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
Hacker News·Jun 20·5 min read
Bun has an open PR adding shared-memory threads to JavaScriptCore
TL;DR
JavaScriptCore lacked efficient parallel execution capabilities, limiting performance in multi-threaded environments. A new feature, shared-memory threads, allows functions to run on separate threads while accessing the same memory heap without the need for complex data transfer methods.
✦ Why It Matters
Engineers can leverage shared-memory threads to enhance performance in JavaScript applications requiring parallel processing.
Key Takeaways
- Explore shared-memory threads for improved performance.
- Understand the implications of direct memory access.
- Monitor ongoing developments in JavaScriptCore.
- Participate in discussions around the pull request.
Full Summary
JavaScriptCore, the engine behind JavaScript execution in various environments, previously struggled with efficient multi-threading due to the need for structured data transfer methods like message passing. The introduction of shared-memory threads allows developers to create new threads that can directly access shared objects in memory, simplifying the process.
This implementation runs functions in parallel across all four Just-In-Time (JIT) compilation tiers without requiring a global lock, which is a significant improvement. Initial tests indicate that the thread test suite passes under this new model, although there are still outstanding tasks such as thread-sanitizer cleanup and additional fuzz testing.
The existing fallback mode and configurations that disable threads remain intact and verified. The development log transparently details challenges faced during implementation, emphasizing the experimental nature of this pull request.
While promising, the feature may not be merged into the main codebase without further validation.