bc7416aa22
Less data accessed and compared gives better batching performance. # Objective - Use a smaller id to represent the lightmap in batch data to enable a faster implementation of draw streams. - Improve batching performance for 3D sorted render phases. ## Solution - 3D batching can use `LightmapSlabIndex` (a `NonMaxU32` which is 4 bytes) instead of the lightmap `AssetId<Image>` (an enum where the largest variant is a 16-byte UUID) to discern the ability to batch. ## Testing Tested main (yellow) vs this PR (red) on an M4 Max using the `many_cubes` example with `WGPU_SETTINGS_PRIO=webgl2` to avoid GPU-preprocessing, and modifying the materials in `many_cubes` to have `AlphaMode::Blend` so that they would rely on the less efficient sorted render phase batching. <img width="1500" alt="Screenshot_2025-03-15_at_12 17 21" src="https://github.com/user-attachments/assets/14709bd3-6d06-40fb-aa51-e1d2d606ebe3" /> A 44.75us or 7.5% reduction in median execution time of the batch and prepare sorted render phase system for the `Transparent3d` phase (handling 160k cubes). --- ## Migration Guide - Changed: `RenderLightmap::new()` no longer takes an `AssetId<Image>` argument for the asset id of the lightmap image. |
||
|---|---|---|
| .. | ||
| src | ||
| Cargo.toml | ||
| LICENSE-APACHE | ||
| LICENSE-MIT | ||
| README.md | ||
Bevy PBR
