Minimize moves by introducing WeakVec

[teoxoy]

This should improve the situation in bevyengine/bevy#15893.

Follow-up to #5874.

@DGriffin91 could you give this PR a try?

[ErichDonGubler]

Conditionally approving based on user confirmation that this does, indeed, solve a problem for them. I'm open to merging in the absence of that feedback, but I think waiting a day or so to confirm would be good.

[DGriffin91]

Unfortunately it seems like bevy still has the issue with this PR. It might be slightly faster though. I have the bevy branch with this pr up if you wanted to play with it.

First frame time with cargo run --example many_cubes --release -- --vary-material-data-per-instance

119619ms: bevy w/ minimize-moves
129969ms: bevy main

vtune filtered in on just the part that is hanging w/ minimize-moves:

[tychedelia]

The problem here as far as I can tell is that we are creating 160k unique bind groups that all share the same texture in a loop. Even without the call to retain, WeakVec.push(..) iterates over self.inner referencing all the bind groups we've created so far, which ends up looking exponential:

[teoxoy]

The problem here as far as I can tell is that we are creating 160k unique bind groups that all share the same texture in a loop.

This is worst case scenario for this code path. We can alleviate this issue further by running the code every X invocations of push rather than on every invocation - I will give it a try.

[teoxoy]

I pushed another commit that should be a good balance between perf and memory use.

@DGriffin91 & @tychedelia could you give it a try?

[tychedelia]

@teoxoy Yay! Validated on Metal and Vulkan That seems to do it. 🎉✨