`no_std` for `naga` #7249

bushrat011899 · 2025-02-28T05:52:03Z

Connections

Contributes to Support no_std where possible (specifically for wasm32v1-none) #6826

Description

Added std, libm, and codespan-reporting features.
Added std to features which are incompatible with no_std to document compatibility
Switched to using core and alloc instead of std where required
Added naga to the existing no_std CI check

Testing

cargo check -p naga --no-default-features --target=wasm32v1-none --features libm,dot-out,glsl-out,msl-out,msl-out-if-target-apple,wgsl-out,hlsl-out,hlsl-out-if-target-windows,compact

Squash or Rebase?

Squash before merging.

Checklist

Run cargo fmt.
Run taplo format.
Run cargo clippy. If applicable, add:
- --target wasm32-unknown-unknown
Run cargo xtask test to run tests.
Add change to CHANGELOG.md. See simple instructions inside file.

Notes for Reviewers

The diff looks large, but the vast majority of it is a shim file, naga/src/math.rs which just handles bringing in maths operations from either std or libm depending on what features are enabled. The rest of the changes are mostly changing import names and sprinkling a few #[cfg(feature = "...")] statements.

Feature Compatibility

➖ denotes a feature which works and is not required on that target
❌ denotes a feature which is incompatible with that target
✔️ denotes a feature which is required for that target

Feature	`std` (e.g., `x86_64-pc-windows-msvc` for Windows)	`no_std` (e.g., `wasm32v1-none`)
`default`	➖	❌
`std`	➖	❌
`libm`	➖	✔️
`dot-out`	➖	➖
`glsl-out`	➖	➖
`msl-out`	➖	➖
`msl-out-if-target-apple`	➖	➖
`wgsl-out`	➖	➖
`hlsl-out`	➖	➖
`hlsl-out-if-target-windows`	➖	➖
`compact`	➖	➖
`glsl-in`	➖	❌
`serialize`	➖	❌
`deserialize`	➖	❌
`arbitrary`	➖	❌
`spv-in`	➖	❌
`spv-out`	➖	❌
`wgsl-in`	➖	❌
`codespan-reporting`	➖	❌

Note that many of these features could be made to be compatible with no_std, but since they are optional I am leaving them for a follow-up to reduce the size of this PR as much as possible.

--no-default-features compiles on wasm32v1-none

bushrat011899 · 2025-02-28T06:10:47Z

This is an alternative to #6940 (special thanks to @brody4hire for working on this while I was busy doing other things!) I believe this PR is in a more polished state as-is and ready to merge. There are no outstanding todo items, and the only dependency added is libm, and even then only when the libm feature is enabled.

One possible downside to consider is --no-default-features will now fail to compile, since a floating point math backend isn't included as a baseline. This is in-line with how crates like glam work so it isn't unreasonable, but it may be worth considering making libm an unconditional dependency.

It was also raised by @kpreid that the use of an std feature isn't desirable, and to instead include std for any feature which needs it. I've implemented this PR with a simple std feature but I am happy to investigate that approach instead if that is broadly desirable.

cwfitzgerald

Have some high level concerns and some notes

The libm and codespan-reporting changes should be split into 2 separate PRs, both from a reviewability perspective, and ensuring we can pinpoint any regressions.

cwfitzgerald · 2025-02-28T19:06:20Z

naga/src/back/dot/mod.rs

+use alloc::borrow::Cow;
+use alloc::format;
+use alloc::string::String;
+use alloc::string::ToString;
+use alloc::vec::Vec;
+use core::fmt::{Error as FmtError, Write as _};


These should remain as compacted as possible

cwfitzgerald · 2025-03-01T00:39:17Z

naga/src/math.rs

Instead of using traits, which require this very kludgy fully-qualified syntax on calling. Instead I think the better plan would be making two inline modules with wrapper methods, then conditionally compiling each module and re-exporting as appropriate. Overloads can be dealt with suffixes. That way we can call math::sin_f32 and such directly.

As noted above though, this change needs to be in a separate PR.

Why not use a trait with methods? It even already exists, and this is how I handle the problem:

#[cfg(not(feature = "std"))] use num_traits::float::Float as _;

With the libm feature enabled, Float is a nearly perfect polyfill for std float methods.

That would work!

cwfitzgerald · 2025-03-01T00:46:24Z

naga/src/valid/handles.rs

+use super::ValidationError;
+use crate::non_max_u32::NonMaxU32;
 use crate::{
    arena::{BadHandle, BadRangeError},
    diagnostic_filter::DiagnosticFilterNode,
    Handle,
 };
-
-use crate::non_max_u32::NonMaxU32;
 use crate::{Arena, UniqueArena};
+use core::{convert::TryInto, hash::Hash};

-use super::ValidationError;
-
-use std::{convert::TryInto, hash::Hash};
+#[cfg(test)]
+use alloc::string::ToString;


I know we don't follow this super strictly, but we should try to follow rustfmt's StdExternalCrate ordering for imports. You don't need to fix things you didn't touch, but when changing them, we should try to standardize them.

cwfitzgerald · 2025-03-01T00:51:08Z

.github/workflows/ci.yml


          # Check with all features except "std".
          cargo clippy --target ${{ matrix.target }} ${{ matrix.extra-flags }} -p wgpu-types --no-default-features --features strict_asserts,fragile-send-sync-non-atomic-wasm,serde,counters
+          cargo clippy --target ${{ matrix.target }} ${{ matrix.extra-flags }} -p naga --no-default-features --features libm,dot-out,glsl-out,msl-out,msl-out-if-target-apple,wgsl-out,hlsl-out,hlsl-out-if-target-windows,compact


I have concerns about this list atrophying, I'm not sure what should be done about that though.

cwfitzgerald · 2025-03-01T00:52:40Z

naga/Cargo.toml

+## Provide rich error messages using codespan-reporting
+codespan-reporting = ["std", "dep:codespan-reporting", "dep:termcolor"]


I worry that cordoning codespan-reporting off completely will mean that wgpu-core won't be able to format errors like it expects to, as it definitely pretty prints the errors there.

bushrat011899 · 2025-03-01T03:57:47Z

I think these are really good points and very actionable! I'm going to close this PR in favour of the first and least controversial part, which is simply importing from core and alloc instead of std.

bushrat011899 added 5 commits February 28, 2025 14:50

MVP

a8ce252

--no-default-features compiles on wasm32v1-none

Fix existing features

68674ad

Fix features for no_std

f52ec69

Add to CI

dda0aa6

taplo fmt

8704780

bushrat011899 requested a review from a team as a code owner February 28, 2025 05:52

bushrat011899 added 2 commits February 28, 2025 16:52

Update CHANGELOG.md

e7d7cc2

Update math.rs

c569909

cwfitzgerald self-assigned this Feb 28, 2025

cwfitzgerald requested changes Mar 1, 2025

View reviewed changes

bushrat011899 closed this Mar 1, 2025

bushrat011899 mentioned this pull request Mar 1, 2025

Use core and alloc instead of std in naga #7256

Open

6 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

`no_std` for `naga` #7249

`no_std` for `naga` #7249

bushrat011899 commented Feb 28, 2025 •

edited

Loading

bushrat011899 commented Feb 28, 2025

cwfitzgerald left a comment

cwfitzgerald Feb 28, 2025

cwfitzgerald Mar 1, 2025

kpreid Mar 1, 2025

cwfitzgerald Mar 1, 2025

cwfitzgerald Mar 1, 2025

cwfitzgerald Mar 1, 2025

cwfitzgerald Mar 1, 2025

bushrat011899 commented Mar 1, 2025

		## Provide rich error messages using codespan-reporting
		codespan-reporting = ["std", "dep:codespan-reporting", "dep:termcolor"]

no_std for naga #7249

no_std for naga #7249

Conversation

bushrat011899 commented Feb 28, 2025 • edited Loading

bushrat011899 commented Feb 28, 2025

cwfitzgerald left a comment

Choose a reason for hiding this comment

cwfitzgerald Feb 28, 2025

Choose a reason for hiding this comment

cwfitzgerald Mar 1, 2025

Choose a reason for hiding this comment

kpreid Mar 1, 2025

Choose a reason for hiding this comment

cwfitzgerald Mar 1, 2025

Choose a reason for hiding this comment

cwfitzgerald Mar 1, 2025

Choose a reason for hiding this comment

cwfitzgerald Mar 1, 2025

Choose a reason for hiding this comment

cwfitzgerald Mar 1, 2025

Choose a reason for hiding this comment

bushrat011899 commented Mar 1, 2025

`no_std` for `naga` #7249

`no_std` for `naga` #7249

bushrat011899 commented Feb 28, 2025 •

edited

Loading