teoxoy/encase

nalgebra support: ShaderSize is not implemented for Matrix<f32, Const<3>...

JMLX42 opened this issue · 10 comments

JMLX42 commented

Considering the following WGSL code:

struct Material {
    color: vec4<f32>
}

@group(0)
@binding(0)
var<uniform> material: Material;

This is what wgsl_to_wgpu generates:

#[repr(C)]
#[derive(Debug, Copy, Clone, PartialEq, encase::ShaderType)]
pub struct Material {
    pub color: nalgebra::SVector<f32, 4>,
}

Which causes the following build error:

error[E0277]: the trait bound `Matrix<f32, Const<4>, Const<1>, ArrayStorage<f32, 4, 1>>: ShaderSize` is not satisfied
 --> src\shaders\shader.rs:4:41
  |
4 | #[derive(Debug, Copy, Clone, PartialEq, encase::ShaderType)]
  |                                         ^^^^^^^^^^^^^^^^^^ the trait `ShaderSize` is not implemented for `Matrix<f32, Const<4>, Const<1>, ArrayStorage<f32, 4, 1>>`
  |
  = help: the following other types implement trait `ShaderSize`:
            &T
            &mut T
            Arc<T>
            ArrayLength
            AtomicI32
            AtomicU32
            Box<T>
            Cow<'_, T>
          and 47 others
  = help: see issue #48214
  = note: this error originates in the derive macro `encase::ShaderType` (in Nightly builds, run with -Z macro-backtrace for more info)

Which is unexpected.

Please advise.

teoxoy commented

I think this should work. Do you have the nalgebra feature flag of encase switched on?

JMLX42 commented

@teoxoy yes:

encase = { version = "0.4.1", features = ["nalgebra"] }
ScanMountGoat commented

I'm seeing code for Vector4<T> and vector slices in the nalgebra implementations but not SVector. Is this just a missing implementation?
https://github.com/teoxoy/encase/blob/main/src/impls/nalgebra.rs

JMLX42 commented

I'm seeing code for Vector4 and vector slices in the nalgebra implementations but not SVector. Is this just a missing implementation?

@ScanMountGoat AFAIK Vector is an alias for SVector:

https://nalgebra.org/docs/user_guide/vectors_and_matrices

  • SVector<T, D> is a statically-sized column vector with D rows, e.g., SVector<f32, 10>.

[...]

Aliases exist for statically-sized vectors and matrices with small dimensions:

  • Vector1 .. Vector6: are column vectors of dimension 1 to 6.

That's why I would expect it to work.

teoxoy commented

I'm seeing code for Vector4 and vector slices in the nalgebra implementations but not SVector. Is this just a missing implementation?

@ScanMountGoat AFAIK Vector is an alias for SVector:

Looking at nalgebra's source:

pub type SVector<T, const D: usize> = Matrix<T, Const<D>, U1, ArrayStorage<T, D, 1>>;
pub type Vector4<T> = Matrix<T, U4, U1, ArrayStorage<T, 4, 1>>;

You are using SVector<f32, 4> and encase uses Vector4<T> - they should point to the same definition.

Do you have a minimum reproducible example?

JMLX42 commented

Do you have a minimum reproducible example?

@teoxoy here you go:

https://github.com/JMLX42/wgsl_to_wgpu/tree/bug/encase-22/example

teoxoy commented

Ah, I think I figured it out. I haven't updated encase to support nalgebra v0.32, but you request it in Cargo.toml.
You essentially have both versions in your project now.

JMLX42 commented

Ah, I think I figured it out. I haven't updated encase to support nalgebra v0.32, but you request it in Cargo.toml.

@teoxoy here you go: #23

Not sure if fixes the problem though...

teoxoy commented

The next major release (0.5) will contain the PR. Do you need it released or could you use a git dependency for now?

JMLX42 commented

Do you need it released or could you use a git dependency for now?

@teoxoy nope. I added the following to Cargo.toml and the changes in #23 do resolve the problem:

[patch.crates-io]
encase = { git = "https://github.com/teoxoy/encase.git", rev = "e35c7793a6126f5677b31bcfeec4ea3701cc2348" }
  • 👍1