Uploaded October 2024 | Updated November 2026, 14 hours ago
With James Mitchell.
If your game runs with multiple threads, then the chances are you are using locks/mutexes. In this talk I will show how locks are typically used and why different lock implementations can negatively impact the performance of your games, this includes looking at locks typically included with most programming languages and the common spin lock that many game developers still reach for. I will introduce a newer technique to building locks which is inspired by Linux futexes and Webkit’s parking lot, this technique allows you to build extremely efficient single byte locks without the concern of large operating system overheads or the issues that arise when using spin locks. This powerfulness won’t sacrifice performance or debuggability, with a single byte lock we can get better performance then a spin lock and a more powerful debugging experience then either operating system locks or spin locks could achieve. The technique demonstrated also enables the building of purpose specific thread synchronization primitives in straightforward ways without requiring you to be a level 381 multi-threading wizard. This talk will be primarily language agnostic, some examples might be in C++ but should be relative to other programming languages.
Video courtesy of GCAP and IGEA
www.gcap.com.au
With James Mitchell.
If your game runs with multiple threads, then the chances are you are using locks/mutexes. In this talk I will show how locks are typically used and why different lock implementations can negatively impact the performance of your games, this includes looking at locks typically included with most programming languages and the common spin lock that many game developers still reach for. I will introduce a newer technique to building locks which is inspired by Linux futexes and Webkit’s parking lot, this technique allows you to build extremely efficient single byte locks without the concern of large operating system overheads or the issues that arise when using spin locks. This powerfulness won’t sacrifice performance or debuggability, with a single byte lock we can get better performance then a spin lock and a more powerful debugging experience then either operating system locks or spin locks could achieve. The technique demonstrated also enables the building of purpose specific thread synchronization primitives in straightforward ways without requiring you to be a level 381 multi-threading wizard. This talk will be primarily language agnostic, some examples might be in C++ but should be relative to other programming languages.
Video courtesy of GCAP and IGEA
www.gcap.com.au










