Not sure if this was discussed yet or not. But I wanted to kick-off a discussion about inclusion of WorkGraph API into WebGPU

What are work graphs?

Very briefly, a it's an API for creating GPU compute pipelines with dynamic resource allocation. For example, if you want to build a mesh on the GPU, you could currently use a storage buffer likes so:

struct Elements{
  end: atomic<u32>,
  elements: array< Triangle >
}

The problem is that you have to pre-allocate such a buffer for worst-case, as you don't know ahead of time how many triangles you'd be generating.

There are other benefits such as better scheduling semantics to improve utilization.

References:

DirectX spec: https://github.com/microsoft/DirectX-Specs/blob/master/d3d/WorkGraphs.md
DirectX blog: https://devblogs.microsoft.com/directx/d3d12-work-graphs/

Here's a nice talk from DirectX guys:
https://www.youtube.com/watch?v=0S8qymwcHIc

AMD talk:
https://www.youtube.com/watch?v=QQP6-JF64DQ

Why is this relevant for the web?

• Resources are more scarce in general, especially on mobile. This limits what we can do quite severely, this feature would help alleviate this issue.
• We have a fairly sequential programming model which makes efficient GPU utilization difficult. Work Graphs would help here as well, by side-stepping the issue more or less.
• Free up main thread a little. Less work to be done on the CPU, this is especially relevant for allocations, which can take a while in my experience and cause hick-ups.