Exploring off-main-thread computation: possible fit with ComputeKit

[tapava]

Hi everyone,

I am the author of an open-source toolkit called ComputeKit, which focuses on running expensive computations via a worker pool, with optional React and WASM support. I'm not suggesting it's a solution by default. I am just curious whether an approach like this could make sense for parts of deck.gl, or if the project has intentionally avoided this trade-off.

I'd really appreciate hearing maintainers' or contributors' thoughts on:

• Whether off-main-thread computation is relevant for this project
• If there are architectural reasons ComputeKit solution wouldn’t fit
• Or if similar approaches have already been evaluated

Thanks for the great work on the project, and happy to learn either way 😄

[ibgreen]

@tapava Thanks for the links, and some off-the-cuff feedback.

• This looks like a nicely designed library that could be a good complement for deck.gl app writers that want to use WASM / off-thread computation.

• So far we do make extensive use of workers threads and WASM modules mainly in our loaders.gl framework, where most loader modules come with pre-built workers.

• That said, we haven't really started writing code in WASM, we mostly use WASM to integrate code written in other languages and we mainly use the GPU to accelerate things.

• I see that you are supporting AssemblyScript, this is interesting and looks like it could make refactoring smaller pieces of code into WASM a lot easier. We might start to experiment with that, perhaps in our math.gl library.

• To set expectations, we do have a high bar for adding external dependencies.

[tapava]

@ibgreen Thanks for your feedback!

Yeah, I took a look at loaders.gl, you've clearly got the worker story covered, no need to add anything there.

On the AssemblyScript side, you're already getting the real wins from GPU acceleration anyway.

Appreciate you taking the time to look at it. If there's ever a use case where off-thread CPU work makes sense for you, I'm around.

Thanks again.