Implement bindless buffers on metal by mate-h · Pull Request #9081 · gfx-rs/wgpu

mate-h · 2026-02-20T08:11:09Z

Connections
Resolves #6741

Description
Problem: Metal on macOS did not support bindless storage buffers.

In wgpu-hal, encode buffer binding arrays using MTLBuffer argument encoder. Metal expects encoded pointers, not raw resource IDs. Creates an argument buffer from an encoder, encode each buffer, and register it in resources_to_use.

In naga MSL writer, emit device T* for buffer binding array elements (which are pointers to device memory) instead of the constant layout used for textures and samplers. For struct members behind binding arrays, emit -> instead of . because the elements are pointers.

In the adapter, expose the BUFFER_BINDING_ARRAY feature on supporting devices, with special handling for MTLCaptureDevice. This is a separate fix for Xcode frame capture, when some functions fail to invoke, in these cases fallback to feature table based limits. I referenced the metal feature table PDF.

Testing

Tested with Bevy Solari on metal / macOS which makes use of bindless storage buffers and raytracing acceleration structures.

Squash

Checklist

Run cargo fmt.
Run taplo format.
Run cargo clippy --tests. If applicable, add:
- --target wasm32-unknown-unknown
Run cargo xtask test to run tests.
If this contains user-facing changes, add a CHANGELOG.md entry.

inner-daemons · 2026-02-20T19:14:25Z

I'm excited to see this and plan on checking out the code. Thanks a ton for working on it!

Vecvec

A couple of things from a read through.

inner-daemons

Ok so lots of thoughts here, mostly questions. Overall looks good. More comments would be great.

Big thing though is snapshot tests. I can't really review this change until I know what it generates and know that what it generates is correct. There might even be existing tests that you can just add METAL as a target to.

Thanks for putting in the work. I look forward to using this myself, and I'd like to get this landed soon!

inner-daemons · 2026-03-11T15:21:39Z

@mate-h This will probably be assigned once you address at least some of the above comments.

msvbg · 2026-03-17T19:31:13Z

Eagerly awaiting this one and will be testing the PR in my own project. I also volunteer to adopt this PR if @mate-h is busy.

msvbg · 2026-03-21T12:52:35Z

The PR appears to work as intended on my end, backported to wgpu v27. Will post back if I notice any issues.

Port of gfx-rs#9081 to the v27 metal crate API. - naga: emit `device` pointer syntax for struct-based binding arrays in MSL, use `->` for member access, skip address space for BindingArray globals - wgpu-hal: add BUFFER_BINDING_ARRAY feature flag on Metal (MSL 3.0 + Tier 2), use MTLArgumentEncoder to encode buffer pointer arrays in bind groups - wgpu-hal: extend map_resource_usage for Buffer types, add pointer_array_argument_descriptor helper

cwfitzgerald · 2026-03-21T18:11:49Z

@msvbg feel free to put up a new PR with things updated, just make sure to keep the authorship on the commits so it's properly attributed

mate-h · 2026-03-21T18:18:11Z

Hi everyone quick update, I am working on addressing the remaining feedback by @inner-daemons and making sure the CI passes. Regarding the backport, for my understanding wondering why it is needed? Let me know @msvbg if you'd like to collaborate on this!

msvbg · 2026-03-21T18:34:18Z

Oh my bad, I did not intend to push to this branch! I'm sorry, that was accidental. I didn't even realize I had push access to this. Can you hard reset/remove my commits?

msvbg · 2026-03-21T18:40:36Z

Okay it seems that I misread the Github interface and I did not push to your branch. Phew! I needed a backport because I'm on an older version of wgpu. It's for my own use. Please disregard the confusion :)

As for collaboration, I have nothing further to add to the PR!

inner-daemons · 2026-03-22T00:31:08Z

@msvbg Just FYI, a back port of this is unlikely to be accepted. We only accept back ports for a few of the most critical bug fixes in every release cycle, and even then we mainly do it for the latest release (27 was 3 releases ago).

Thanks for your interest though!

mate-h · 2026-03-23T08:45:17Z

All right I am finished addressing all the feedback, ready for review again.
for those that want to test out Solari with this PR I posted a branch to bevy: https://github.com/mate-h/bevy/tree/m/solari-metal

checkout this PR locally at wgpu repo. checkout bevy repo at the branch linked above. place this file to bevy/.cargo/config.toml

[patch.crates-io]
wgpu = { path = "../wgpu/wgpu" }
wgpu-core = { path = "../wgpu/wgpu-core" }
wgpu-hal = { path = "../wgpu/wgpu-hal" }
wgpu-types = { path = "../wgpu/wgpu-types" }
wgpu-macros = { path = "../wgpu/wgpu-macros" }
naga = { path = "../wgpu/naga" }
naga_oil = { path = "../naga_oil" }

last line is a local git clone of naga master branch or set to git dep.

…ine layout

inner-daemons

I'm gonna tentatively approve this, since the biggest points have been addressed. I haven't tried incredibly hard yet to reason through all of the writer changes since my last review but what I've noticed seems reasonable.

inner-daemons · 2026-04-14T14:41:38Z

@cwfitzgerald Feel free to take a look

cwfitzgerald

Making solid progress! Some more minor and pedantic issues and a bit more testing

inner-daemons · 2026-05-27T15:27:37Z

This is still waiting on author

msvbg · 2026-06-08T10:31:36Z

Is this still being worked on?

mate-h · 2026-06-11T06:33:40Z

I have not worked on this in a while, I'll come back to it now! I got some more feedback from @cwfitzgerald that I need to address. And also merge in the trunk branch again.

cwfitzgerald

Thanks for the progress, some more comments, mainly around testing

mate-h · 2026-06-29T19:24:10Z

@cwfitzgerald thanks for your continued feedback! I have addressed every outstanding comment on this PR, let me know if there's anything else to resolve.

cwfitzgerald

Approving, looks good with one caveat that we should make an issue for.

cwfitzgerald · 2026-06-29T21:21:00Z

+    int _e104 = uint(0) < 10 && _buffer_sizes.size0[0u] != 0u && uint(non_uniform_index) < 10 && _buffer_sizes.size0[unsigned(non_uniform_index)] != 0u ? storage_array[non_uniform_index].inner->far[0] : DefaultConstructible();
+    u1_ = _e99 + as_type<uint>(_e104);
+    uint _e107 = u1_;
+    int _e112 = uint(0) < 10 && _buffer_sizes.size0[0u] != 0u && uint(7) < 10 && _buffer_sizes.size0[7u] != 0u ? storage_array[7].inner->far[0] : DefaultConstructible();


How does this interact with partially bound binding arrays. We're introducing bounds checks on the binding array itself, but with partially bound arrays the size isn't known to the shader. This is definitely an improvement than before where we had no bounds checks at all, so feel free to push this to an issue/follow up, but something to consider.

# Objective [Ray tracing on Metal](gfx-rs/wgpu#8071) has been available since wgpu 29. I shipped a PR to wgpu for [implementing bindless storage buffers](gfx-rs/wgpu#9081) for metal, which was released with wgpu 30. With the [wgpu 30 PR](bevyengine#24841) in bevy merged, we are fully ready to add support to Solari on metal backends! However because there is a limit of 31 buffer slots per shader stage. Running Solari on main on a raytracing compatible Mac results in a crash. ## Solution I had combine the world cache into a single packed buffer. In theory this should not introduce performance regressions since it’s only reorganizing bindings and doesn’t introduce and copy or other passes. Tried keeping the diff fairly minimal and easy to review. Getting the buffer layout for the world cache on the CPU with `ShaderType` was tricky. To avoid blowing up memory I put it in a private module so you can't construct it. ## Testing - Ran the Solari example on mac ``` cargo run --example solari --features bevy_solari,https,free_camera ``` Currently this PR needs a benchmark against main to ensure there's no perf regression. I could use some help with that! --- ## Showcase <img width="2566" height="1514" alt="shot" src="https://github.com/user-attachments/assets/691b9bd3-129f-47e6-8efe-313cbadd6ff5" /> ## Known limitations Tracing spans incorrectly report 0.0ms , tracking in gfx-rs/wgpu#9414 No denoising support yet. Could theoretically use [MetalFX Ray Reconstruction](https://crates.io/crates/objc2-metal-fx) in place of DLSS for cross-platform denoising support in Solari. In the longer term we are also looking at Open Image Denoise 3.0 that promises temporal denoising support. --------- Co-authored-by: Chris Biscardi <chris@christopherbiscardi.com>

mate-h added 3 commits February 20, 2026 00:01

Implement bindless buffers on metal

a1439aa

Fix CI and changelog

1aafa4a

Update changelog

96ed0d9

inner-daemons self-assigned this Feb 20, 2026

Vecvec reviewed Feb 20, 2026

View reviewed changes

Comment thread wgpu-hal/src/metal/adapter.rs Outdated

Comment thread wgpu-hal/src/metal/adapter.rs Outdated

Address feedback

84c28e6

inner-daemons suggested changes Feb 26, 2026

View reviewed changes

inner-daemons removed their assignment Feb 27, 2026

cwfitzgerald assigned inner-daemons Mar 11, 2026

mate-h added 4 commits March 19, 2026 14:18

Merge branch 'trunk' into m/bindless-buffer-metal

733c4d9

Add snapshot test

a641b92

Update changelog

5f767e1

Fixes to naga writer for snapshot tests

896cc2d

mate-h added 5 commits March 23, 2026 01:09

Address feedback for adapter

51b2282

Restructure bindless argument buffer setup

a4744f8

Merge branch 'trunk' into m/bindless-buffer-metal

f065633

Apply style suggestions while keeping rw and cargo fmt

85a1f8b

Fix CI

3ab9b22

mate-h mentioned this pull request Mar 23, 2026

Enable partial bindless on Metal and reduce bind group overhead (#18149) bevyengine/bevy#23436

Merged

mate-h added 4 commits March 30, 2026 10:09

Test with Xcode capture

75a438d

Stop using tier 1 encoder and simplify

8dd9a39

Merge branch 'trunk' into m/bindless-buffer-metal

222ec14

Element wise storage buffer sizes and binding array counts from pipel…

5400203

…ine layout

inner-daemons self-requested a review April 8, 2026 20:24

mate-h added 2 commits April 11, 2026 22:31

Merge branch 'trunk' into m/bindless-buffer-metal

af8ebe7

CI format

42d7e56

inner-daemons approved these changes Apr 14, 2026

View reviewed changes

Comment thread naga/src/proc/mod.rs

Comment thread naga/tests/in/wgsl/binding-buffer-arrays.toml Outdated

Comment thread wgpu-hal/src/metal/adapter.rs Outdated

Comment thread wgpu-hal/src/metal/adapter.rs

inner-daemons requested a review from cwfitzgerald April 14, 2026 14:41

cwfitzgerald requested changes Apr 15, 2026

View reviewed changes

DGriffin91 mentioned this pull request Jun 20, 2026

Clarify wgpu/metal bindless support bevyengine/bevy-website#2523

Merged

mate-h added 6 commits June 25, 2026 23:24

Merge branch 'trunk' into m/bindless-buffer-metal

1fc93f1

Code style feedback

c18ef12

MSL snapshot for unbounded arrays

a8485e3

Address feedback comments

42fd861

Unbounded binding arrays in SPIRV output

163cf62

Clippy

9abbad3

cwfitzgerald requested changes Jun 26, 2026

View reviewed changes

Comment thread naga-test/tests/binding_array_length_map_toml.rs Outdated

Comment thread naga/tests/in/wgsl/binding-buffer-arrays.wgsl

Comment thread wgpu-hal/src/metal/adapter.rs

mate-h added 2 commits June 26, 2026 18:51

MSL writer for partially bound binding arrays

766c388

Remove Xcode Capture workaround

838c4f0

mate-h mentioned this pull request Jun 27, 2026

Workaround xcode misreporting ray tracing support #9757

Open

8 tasks

cwfitzgerald approved these changes Jun 29, 2026

View reviewed changes

cwfitzgerald merged commit bf5c6cf into gfx-rs:trunk Jun 29, 2026
59 checks passed

mate-h mentioned this pull request Jul 22, 2026

Solari support on metal backends bevyengine/bevy#25123

Merged

Uh oh!

Conversation

mate-h commented Feb 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

inner-daemons commented Feb 20, 2026

Uh oh!

Vecvec left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

inner-daemons 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!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

inner-daemons commented Mar 11, 2026

Uh oh!

msvbg commented Mar 17, 2026

Uh oh!

msvbg commented Mar 21, 2026

Uh oh!

cwfitzgerald commented Mar 21, 2026

Uh oh!

mate-h commented Mar 21, 2026

Uh oh!

msvbg commented Mar 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

msvbg commented Mar 21, 2026

Uh oh!

inner-daemons commented Mar 22, 2026

Uh oh!

mate-h commented Mar 23, 2026

Uh oh!

inner-daemons left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

inner-daemons commented Apr 14, 2026

Uh oh!

cwfitzgerald left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

inner-daemons commented May 27, 2026

Uh oh!

msvbg commented Jun 8, 2026

Uh oh!

mate-h commented Jun 11, 2026

Uh oh!

cwfitzgerald left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mate-h commented Jun 29, 2026

Uh oh!

cwfitzgerald left a comment

mate-h commented Feb 20, 2026 •

edited

Loading

inner-daemons left a comment •

edited

Loading

msvbg commented Mar 21, 2026 •

edited

Loading