Zero Copy Mesh

[atlv24]

Objective

• Another step towards Reduce number of copies in the Mesh API #15558

Solution

• Instead of allocating a Vec and then having wgpu copy it into a staging buffer, write directly into the staging buffer.
• gets rid of another hidden copy, in pad_to_alignment.

future work:

• why is there a gcd implementation in here (and its subpar, use binary_gcd. its in the hot path, run twice for every mesh, every frame i think?) make it better and put it in bevy_math
• zero-copy custom mesh api to avoid having to write out a Mesh from a custom rep

Testing

• lighting and many_cubes run fine (and slightly faster. havent benchmarked though)

---

Showcase

• look ma... no copies

at least when RenderAssetUsage is GPU only :3

[IceSentry]

I haven't tested it yet, but the code LGTM