Experiment with thread local pool of audio render quanta #511
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
| /bench |
|
orottier force-pushed the feature/alloc-thread-local branch from a1f0270 to d915944 Compare | /bench |
|
| Interesting regression on the delay nodes between version one and two.. 🤔 |
| I can't reproduce the regression locally (not with cargo bench, and also the example/benchmarks.rs shows improvement for the delay test..) Let's try again /bench |
|
| /bench |
|
| Hey @b-ma , the idea for this change was growing for a long time on me. Happy to see some benchmark improvements again. Could you have a look? I'm certainly interested what |
| Hey, I'm a bit out of networks this week-end, I'll have a look on Monday |
| Hey, so on my side:
do we agree that I have run |
b-ma reviewed May 20, 2024
src/render/quantum.rs Outdated
| } | ||
| thread_local! { | ||
| /// Thread local pool of pre-allocated sample slice that can be reused after being dropped | ||
| static POOL: RefCell<Vec<Rc<[f32; RENDER_QUANTUM_SIZE]>>> = RefCell::new(Vec::with_capacity(32)); |
There was a problem hiding this comment.
I wonder if we shouldn't have a larger initial pool? As a stereo delay of 1 second (default max_delay_time) uses 375 * 2 AudioRenderQuantumChannel, having something like 1024 preallocated buffers would guarantees we don't allocate at least when creating one delay
There was a problem hiding this comment.
Good point. We could definitely do it but we should realize this could happen on any thread and the memory is never reclaimed, so to be a bit conservative would not be bad.
I'm having a bit more thoughts about this setup though, when using async rendering the OfflineAudioContext can move to any thread of the executor and we may end up leaking a lot of memory.
I could specialize it to be of different size for offline and online rendering. Starting with a small pool for offline rendering is not bad because we can safely allocate more during rendering.
Yup that's how it works. Weird that your sines seem to be affected because none of this PR should matter there.. |
Yup I agree... this is quite weird |
It is still a bit magical to me and I want to see when a new pool is initialized.
| /bench |
|
With the new design of 'silent quantum' we can also instantiate silent buffers on the control thread.
| /bench |
|
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
No description provided.