textureLoad does not allows texture_multisampled_2d

[IcanDivideBy0]

no overload of `textureLoad` found for given arguments. Found (texture_2d_multisampled<f32>, vec2<i32>, i32), expected one of:
fn(texture_1d<T>, i32, i32) -> vec4<T>
fn(texture_2d<T>, vec2<i32>, i32) -> vec4<T>
fn(texture_2d_array<T>, vec2<i32>, i32, i32) -> vec4<T>
fn(texture_3d<T>, vec3<i32>, i32) -> vec4<T>
fn(texture_2d_multisampled_array<T>, vec2<i32>, i32) -> vec4<T>
fn(texture_depth_2d, vec2<i32>, i32) -> f32
fn(texture_depth_2d_array, vec2<i32>, i32, i32) -> f32
fn(texture_depth_2d_multisampled, vec2<i32>, i32) -> f32
fn(texture_external, vec2<i32>) -> vec4<f32>
fn(texture_storage_1d<F,read>, i32) -> F::StorageType
fn(texture_storage_2d<F,read>, vec2<i32>) -> F::StorageType
fn(texture_storage_2d_array<F,read>, vec2<i32>, i32) -> F::StorageType
fn(texture_storage_3d<F,read>, vec3<i32>) -> F::StorageType

[IcanDivideBy0]

Just installed the extension from market place,
It looks like the analyzer should suport it tho ... https://github.com/wgsl-analyzer/wgsl-analyzer/blob/main/crates/hir_ty/builtins.wgsl#L183

[IcanDivideBy0]

What's strange is that the error reports a texture_2d_multisampled but i'm using a texture_multisampled_2d

[IcanDivideBy0]

Looks like the order is incorrect here https://github.com/wgsl-analyzer/wgsl-analyzer/blob/main/crates/hir_def/src/type_ref.rs#L179-L190

[jakobhellermann]

The order is incorrect as you pointed out, and the issue with textureLoad is that the builtins get generated from a file, but the multisampled_2d texture was recognized as arrayed: true which is false.

I pushed fixes for the formatting and the incorrect array texture, a new version 0.1.4 should be released once the CI finishes running.

[jakobhellermann]

The new version should be released now, can you check if the issue is fixed?

Edit: nevermind, I think it's still incorrect
Edit 2: Now it should work.

[IcanDivideBy0]

thanks, the issue seems to be fixed for texture_multisampled_2d, now i'm having a similar error with texture_depth_multisampled_2d:

no overload of `textureLoad` found for given arguments. Found (texture_depth_multisampled_1d, vec2<i32>, i32), expected one of:
fn(texture_1d<T>, i32, i32) -> vec4<T>
fn(texture_2d<T>, vec2<i32>, i32) -> vec4<T>
fn(texture_2d_array<T>, vec2<i32>, i32, i32) -> vec4<T>
fn(texture_3d<T>, vec3<i32>, i32) -> vec4<T>
fn(texture_multisampled_2d<T>, vec2<i32>, i32) -> vec4<T>
fn(texture_depth_2d, vec2<i32>, i32) -> f32
fn(texture_depth_2d_array, vec2<i32>, i32, i32) -> f32
fn(texture_depth_multisampled_2d, vec2<i32>, i32) -> f32
fn(texture_external, vec2<i32>) -> vec4<f32>
fn(texture_storage_1d<F,read>, i32) -> F::StorageType
fn(texture_storage_2d<F,read>, vec2<i32>) -> F::StorageType
fn(texture_storage_2d_array<F,read>, vec2<i32>, i32) -> F::StorageType
fn(texture_storage_3d<F,read>, vec3<i32>) -> F::StorageType

[jakobhellermann]

Fixed it: e3cd69a.

Is the shader you are using open source so I can try it locally or are are there other errors I should fix before releasing 0.1.6?

[IcanDivideBy0]

yes, please, take a look here: https://github.com/IcanDivideBy0/calva/blob/master/calva-renderer/src/shaders/light.shadow.wgsl
there's many other errors like:

no overload of `dot` found for given arguments. Found (vec3<T>, vec3<f32>), expected one of:
fn(vecN<f32>, vecN<f32>) -> f32

[jakobhellermann]

I found the issue, vec3<f32> and vec3<T> didn't get unified correctly when type checking the built-ins. Pushed the fix and released 0.1.6, everything should be working correctly now.

[IcanDivideBy0]

thanks, this shader is fine now, i still have other issues tho, you can check other shaders in the same directory: https://github.com/IcanDivideBy0/calva/blob/master/calva-renderer/src/shaders/

most notabely:

let frag_pos_view = camera.inv_proj * vec4<f32>(in.ndc, z, 1.0);
let frag_pos_view = frag_pos_view.xyz / frag_pos_view.w;

let L = normalize(in.l_position - frag_pos_view); // frag_pos_view is considered a vec4 by wgsl-analyzer

and:

fn get_skinning_matrix(frame: u32, in: VertexInput) -> mat4x4<f32> {
    let joints = vec4<u32>(
        in.joints >>  0u & 0xFFu,
        in.joints >>  8u & 0xFFu,
        in.joints >> 16u & 0xFFu,
        in.joints >> 24u & 0xFFu,
    );

    let m1 = get_joint_matrix(frame, joints.x) * in.weights.x; // multiply matrix by a scalar is reported as an error
    let m2 = get_joint_matrix(frame, joints.y) * in.weights.y;
    let m3 = get_joint_matrix(frame, joints.z) * in.weights.z;
    let m4 = get_joint_matrix(frame, joints.w) * in.weights.w;

    return mat4x4<f32>(
        m1.x + m2.x + m3.x + m4.x, // accessing matrix rows with .x, .y, etc is reported as an error
        m1.y + m2.y + m3.y + m4.y,
        m1.z + m2.z + m3.z + m4.z,
        m1.w + m2.w + m3.w + m4.w,
    );
}

[jakobhellermann]

The first example isn't valid wgsl, you can't shadow variables in the same scope. It seems to be accepted in naga right now but it shouldn't be: https://gpuweb.github.io/gpuweb/wgsl/#declaration-and-scope

[IcanDivideBy0]

Oh, ok, thanks for pointing it out 👍

[jakobhellermann]

Accessing matrix rows by x, y, z also isn't valid in mesh.simple.wgsl: https://gpuweb.github.io/gpuweb/wgsl/#matrix-access-expr

    let view3 = mat3x3<f32>(
        camera.view.x.xyz,
        camera.view.y.xyz,
        camera.view.z.xyz,
    );

You should use [0] based indices.

[IcanDivideBy0]

Ok, those have been fixed, only thing left is the matrix multiplication by a scalar which is allowed by the spec

[jakobhellermann]

Those were missing but are added now, 0.2.0 is on its way to be released.

[IcanDivideBy0]

Works like a charm 💯 thanks for your responsiveness!