Add BindGroupLayout caching via descriptors by Zeophlite · Pull Request #21205 · bevyengine/bevy

Zeophlite · 2025-09-25T06:32:20Z

Objective

Defer creating BindGroupLayout by using a BindGroupLayoutDescriptor and cache the results
Unblocks bevy_material (render-less material definitions)
Blocked by lift limits up a bit for bevy_material #21533

Solution

Reviewers, look at first commit for mechanism, and following for usage

Testing

CI

alice-i-cecile · 2025-09-26T03:56:03Z

Can you add more motivation to the PR description please? I'm having trouble following why this is a good idea.

tychedelia

Awesome work. Thanks for doing all this. I'd like to see some benchmarking as I'm a little worried about performance. However, my intuition is that we are extremely read heavy and so it feels likely that we can work around any performance issues if they come up. I can help with benchmarking if necessary. But overall the code looks great!

tychedelia · 2025-10-16T19:06:02Z

+#[derive(Clone, Debug, PartialEq, Eq, Hash, Default)]
+pub struct BindGroupLayoutDescriptor {
+    /// Debug label of the bind group layout descriptor. This will show up in graphics debuggers for easy identification.
+    pub label: Option<Cow<'static, str>>,


as an aside, I kinda feel like we should make labels non-optional in all our code but that's for another PR.

Yeah this would be a good follow-up.

We should be using the same DebugName strategy that ECS uses for system names.

I can help with benchmarking if necessary

@tychedelia I would definitely appreciate that (especially if you can explain how)

tychedelia · 2025-10-16T19:09:19Z

 #[derive(Resource)]
 pub struct PipelineCache {
    layout_cache: Arc<Mutex<LayoutCache>>,
+    bindgroup_layout_cache: Arc<Mutex<BindGroupLayoutCache>>,


I'd like to see benchmarking before any changes but this might be a rare scenario RwLock could win. Or maybe even something like dashmap. Anyway, just curious how much contention there actually is on this.

We might get wins doing this on the layout_cache too. I thinkt he shader_cache probably wouldnt see any significant wins from it but might as well be consistent if we switch to RwLock

atlv24

I have made a migration guide Zeophlite#4

atlv24 · 2025-10-16T17:01:27Z

+            #[expect(clippy::option_map_or_none, reason = "TODO")]
+            label: Self::label().map_or(None, |f| Some(f.into())),


Suggested change

#[expect(clippy::option_map_or_none, reason = "TODO")]

label: Self::label().map_or(None, |f| Some(f.into())),

label: Self::label().map(|f| f.into()),

jasmine-nominal

Lgtm once the clippy lint is addressed.

This will be very helpful for an upcoming PR I'm working on to simplify render boilerplate. I'd also like if we could extend this to cache the actual pipelines.

atlv24 · 2025-10-16T19:44:31Z

+                    #[expect(clippy::redundant_closure_for_method_calls, reason = "TODO")]
+                    descriptor.label.as_ref().map(|s| s.as_ref()),


Suggested change

#[expect(clippy::redundant_closure_for_method_calls, reason = "TODO")]

descriptor.label.as_ref().map(|s| s.as_ref()),

descriptor.label.as_ref().map(Cow::as_ref),

atlv24

Thanks for putting in the work for this!

Approved, contingent on my two suggestions and the migration guide being merged.

I will make a follow-up PR for AsBindGroup cleanups to remove the RenderDevice dependency from it.

# Objective - Defer creating `BindGroupLayout` by using a `BindGroupLayoutDescriptor` and cache the results - Unblocks `bevy_material` (render-less material definitions) - Blocked by bevyengine#21533 ## Solution - Reviewers, look at first commit for mechanism, and following for usage ## Testing - CI --------- Co-authored-by: atlas <email@atlasdostal.com>

* Fix compile error when compiling with DLSS enabled after #21205 * Use permutation sampling for ReSTIR DI temporal reuse to fix artifacts under DLSS-RR * For both DI and GI, removed the spatial raytrace, and moved it to the final reservoir before shading. * Reduced DI initial samples 32 -> 8 for better performance at the cost of quality * Various specular GI improvements and bugfixes (still kinda terrible overall, I need to do some research on how people usually do this kind of thing) * Made the world cache adapt faster / be less stable * Switched spatial hashing collisions from to linear probing --------- Co-authored-by: Jasmine Schweitzer <jasmine.schweitzer@nominal.io>

djeedai · 2026-01-31T19:07:33Z

Has anyone looked at perf in the end on that PR? Or memory consumption? @alice-i-cecile or @tychedelia maybe?

I'm upgrading Hanabi, and where I had before a bunch of trivially cloneable BindGroupLayout handles in various caches (hash maps) and other data structures, now everywhere is a .clone() of an entire BindGroupLayoutDescriptor including cloning its Vec<> of entries (so at least 2 allocations per .clone()). Even in BindGroupLayoutCache::get(), the layout descriptor is passed by value, so needs to be cloned by the caller, even if the bind group layout already exists in the cache. Basically unless I'm missing something, since you need to query the cache each frame to rebuild the bind groups (as I understand Bevy rebuilds them all the time, as last time I asked I was told to do the same because creating bind groups was cheap), this means we .clone() every single BindGroupLayoutDescriptor every frame, likely multiple times.

mockersf · 2026-01-31T21:30:57Z

It had no impact on the performance of Bevy's stress tests

atlv24 and others added 12 commits September 25, 2025 11:51

wip

adae66e

blit layout descriptor

9699ab5

Build bevy_core_pipeline

c5635cd

bevy_anti_alias

e1f8db8

bevy_post_process

d212d56

bevy_sprite_render WIP

e8556e5

bevy_ui_render

31d0931

bevy_gizmos

b0ee042

bevy_pbr

302e34c

bevy_solari

66212d0

bevy_anti_alias fmt

77c2407

bevy_gizmos

8e1ff54

Zeophlite commented Sep 25, 2025

View reviewed changes

Comment thread crates/bevy_anti_alias/src/contrast_adaptive_sharpening/node.rs Outdated

Zeophlite commented Sep 25, 2025

View reviewed changes

Comment thread crates/bevy_pbr/src/material.rs

Zeophlite commented Sep 25, 2025

View reviewed changes

Comment thread crates/bevy_render/src/render_resource/bind_group.rs Outdated

Zeophlite added 7 commits September 25, 2025 16:01

Less clone

bb4c7b7

bevy_solari clone-less

3768a46

More fixes

15f7e36

Meshlet fixes

0df6e82

fmt

7b61ec2

Solari binder new

1f0b279

Examples fixes

ff8bc25

Zeophlite force-pushed the ad/bgld branch from 180d8de to ff8bc25 Compare September 25, 2025 10:56

alice-i-cecile added C-Bug An unexpected or incorrect behavior A-Rendering Drawing game state to the screen S-Needs-Review Needs reviewer attention (from anyone!) to move forward labels Sep 26, 2025

github-project-automation Bot added this to Rendering (Old) Sep 26, 2025

Zeophlite added this to the 0.18 milestone Sep 26, 2025

Zeophlite removed the C-Bug An unexpected or incorrect behavior label Sep 26, 2025

Zeophlite mentioned this pull request Oct 14, 2025

Move MaterialProperties from bevy_pbr to bevy_material #21543

Closed

Zeophlite added 4 commits October 15, 2025 03:09

Merge remote-tracking branch 'origin/main' into ad/bgld

435f3e2

fmt

17ca4d5

Merge remote-tracking branch 'bevy/main' into ad/bgld

e2968a1

fmt

95278d6

Zeophlite marked this pull request as ready for review October 14, 2025 23:22

tychedelia approved these changes Oct 16, 2025

View reviewed changes

atlv24 reviewed Oct 16, 2025

View reviewed changes

jasmine-nominal approved these changes Oct 16, 2025

View reviewed changes

atlv24 reviewed Oct 16, 2025

View reviewed changes

atlv24 approved these changes Oct 16, 2025

View reviewed changes

atlv24 and others added 2 commits October 17, 2025 06:55

Add migration guide

08c84a1

clippy

79d4239

atlv24 added S-Ready-For-Final-Review This PR has been approved by the community. It's ready for a maintainer to consider merging it and removed S-Needs-Review Needs reviewer attention (from anyone!) to move forward labels Oct 16, 2025

Zeophlite added 2 commits October 17, 2025 08:20

pls clippy

1a34c68

fmt

f8a6b53

alice-i-cecile added the C-Code-Quality A section of code that is hard to understand or change label Oct 17, 2025

alice-i-cecile enabled auto-merge October 17, 2025 00:34

alice-i-cecile added this pull request to the merge queue Oct 17, 2025

Merged via the queue into bevyengine:main with commit adf8211 Oct 17, 2025
38 checks passed

github-project-automation Bot moved this to Done in Rendering (Old) Oct 17, 2025

JMS55 mentioned this pull request Oct 25, 2025

Various Solari improvements #21649

Merged

brianreavis mentioned this pull request Nov 5, 2025

Optimized BindGroupLayoutCache to reduce unnecessary cloning #21757

Open

ChristopherBiscardi mentioned this pull request Dec 27, 2025

bevy 0.18 StarArawn/bevy_ecs_tilemap#628

Merged

		#[expect(clippy::option_map_or_none, reason = "TODO")]
		label: Self::label().map_or(None, \|f\| Some(f.into())),

	#[expect(clippy::option_map_or_none, reason = "TODO")]
	label: Self::label().map_or(None, \|f\| Some(f.into())),
	label: Self::label().map(\|f\| f.into()),

		#[expect(clippy::redundant_closure_for_method_calls, reason = "TODO")]
		descriptor.label.as_ref().map(\|s\| s.as_ref()),

	#[expect(clippy::redundant_closure_for_method_calls, reason = "TODO")]
	descriptor.label.as_ref().map(\|s\| s.as_ref()),
	descriptor.label.as_ref().map(Cow::as_ref),

Uh oh!

Uh oh!

Conversation

Zeophlite commented Sep 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Objective

Solution

Testing

Uh oh!

Uh oh!

Uh oh!

Uh oh!

alice-i-cecile commented Sep 26, 2025

Uh oh!

tychedelia left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

atlv24 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jasmine-nominal left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

atlv24 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

djeedai commented Jan 31, 2026

Uh oh!

mockersf commented Jan 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

Zeophlite commented Sep 25, 2025 •

edited

Loading

atlv24 left a comment •

edited

Loading