vello_gpu: reuse staging belt for strips upload

[HigherOrderLogic]

Clear more TODO.

This remove the queue allocation and use an arena-like staging belt instead.

[LaurenzV]

Ah, sorry, I forgot to mention that there was a reason why we haven't used the wgpu-native staging belt, see this PR: #1532

Or does this problem not exist in this implementation?

[HigherOrderLogic]

Or does this problem not exist in this implementation?

I think it should eliminate the overlapped memory problem, since the old belt is dropped every time a new one is created. However, there would still be overlapped memory spike, for example when frames are submitted too fast, but this should only last for a short duration.

[LaurenzV]

@taj-p What do you think of the current impl? We've changed it now so that we precalculate how many bytes we need, set this as the chunk size and recreate the whole staging belt whenever we have to grow. I've checked the profiler and it results in the same speedup! But not sure if you have better ideas, as it's kind of ugly and brittle to estimate how much space we are going to need. But maybe it's the best option for now.

[HigherOrderLogic]

Cant you keep track of how many bytes allocated through a prop inside the StagingBelt wrapper struct? Should be kinda similar to an arena I think.

[LaurenzV]

I mean yeah, but the problem is we need to know it ahead of time before we start doing the allocations, right?

[HigherOrderLogic]

we need to know it ahead of time before we start doing the allocations, right?

Hmm yeah, that should help with preventing the degenerate case mentioned in the OG PR. However, our StagingBelt is discarded and recreated when size grow so it should be fine I think.

But I havent tested it out yet so let's go with your approach for now.