Shared Memory & Parallel Programming with JavaScript

on YouTube ↗24m

" summary

Demonstrates real numbers and trade-offs: naive parallel counts over 200 million iterations slow down due to cache and data races, naive Atomics.add serialises and spikes to ~13 seconds, while per-thread local counters plus one Atomic.add restores speed near the single-thread ~400 ms. Shows SharedArrayBuffer and Atomics for coordinated shared memory and a custom compiler TurboScript that emits asm.js-like code, builds a KD-tree for the Stanford dragon and runs an eight-tile global illumination ray tracer in real time in Three.js.

AI-written summary · about

" chapters (14)

1. 0:00Shared memory in JavaScript
2. 1:13Web Workers memory limits
3. 2:24SharedArrayBuffer and Atomics
4. 3:59Parallel slower than single-threaded
5. 5:31Solving data races with Atomics
6. 7:53Naive Atomics performance impact
7. 9:05Optimizing parallel counting
8. 10:33Ray tracing with shared memory
9. 13:35Manual memory serialization challenges
10. 15:17TurboScript introduction
11. 17:38TurboScript and WebAssembly
12. 20:28TurboScript demo
13. 22:44TurboScript debugging and future
14. 23:30Q&A

" related talks

• 33mParallelism experiments in JavaScriptNaveed Ihsanullah · JSConf US 2015
• 25mGetting Over Moore's Law: Parallelization Using JavaScript in the BrowserDamian Schenkelman · JSConf Colombia 2015
• 23mParallel JavaScriptJaswanth Sreeram · JSConf EU 2014
• 41mConcurrency and Parallel Computing in JavaScriptStephan Herhut · Strange Loop 2013
• 32mMulticore JavaScript: Past, Present and FutureUjjwal Sharma · Nordic.js 2022
• 31mJavaScript Compilation Techniques or "Wasn't JavaScript supposed to be slow?"Marijn Haverbeke · JSConf US 2011
• 30mNative Speed on the Web: JavaScript and asm.jsAlon Zakai · Strange Loop 2013
• 46mHow much performance can you get out of JavaScript?Massimiliano Mantione · jsDay 2017