Encountering array stride issues with DynamicUniformBuffer

[aojiaoxiaolinlin]

I’ve run into an issue while working with Bevy’s render pipeline and was hoping to get some help figuring it out!

I created a pipeline resource that includes a BindGroupLayout, defined like this:

let view_bind_group_layout = render_device.create_bind_group_layout(
    "纹理变换矩阵布局",
    &BindGroupLayoutEntries::single(ShaderStages::VERTEX, uniform_buffer::<Mat4>(true)),
);

I also set up a DynamicUniformBuffer resource:

#[derive(Resource)]
pub struct FilterUniformBuffers {
    pub view_uniform_buffer: DynamicUniformBuffer<Mat4>,
    pub color_matrix_uniform_buffer: DynamicUniformBuffer<ColorMatrixUniform>,
    pub blur_uniform_buffer: DynamicUniformBuffer<BlurUniform>,
    pub glow_uniform_buffer: DynamicUniformBuffer<GlowFilterUniform>,
    pub bevel_uniform_buffer: DynamicUniformBuffer<BevelUniform>,
    pub filter_vertex_with_double_blur_buffer: BufferVec<FilterVertexWithDoubleBlur>,
}

Then I added data to the buffer:

let view_matrix = Mat4::from_cols_array_2d(&[
    [2.0 * scale.x / size.x, 0.0, 0.0, 0.0],
    [0.0, 2.0 * scale.y / size.y, 0.0, 0.0],
    [0.0, 0.0, 1.0, 0.0],
    [0.0, 0.0, 0.0, 1.0],
]);
let view_offset = filter_uniform_buffers
    .view_uniform_buffer
    .push(&view_matrix);

But when I tried to create a BindGroup with it:

let view_bind_group = render_device.create_bind_group(
    "offscreen_bind_group",
    &offscreen_mesh2d_pipeline.view_bind_group_layout,
    &BindGroupEntries::single(view_buffer.binding().unwrap()),
);

I got an error: array stride must be a multiple of 16 (current stride: 4).

Has anyone run into a similar issue with DynamicUniformBuffer or bind group stride mismatches? Any ideas on what I might be missing here? Thanks in advance for the help!

[IceSentry]

Instead of using .push() I think you should use DynamicUniformBuffer::get_writer() then use that to write the value to the buffer. This will handle aligment correctly for you. Alternatively, you can wrap the Mat4 in a struct that has the necessary padding.

[aojiaoxiaolinlin]

Thanks, I'll give it another try. I'd also like to confirm something: currently, I'm pushing data using  .push()  first, and then calling  .write_buffer()  at the end. Does this method not handle automatic alignment? You mentioned wrapping the Mat4 in a struct with padding—could you provide an example? I'm not quite sure I understand.

[IceSentry]

Does this method not handle automatic alignment?

No, IIRC, it will simply push each structs next to each other with no padding so alignment won't be correct.

could you provide an example?

I meant something like

struct Mat4Wrapper {
	mat: Mat4,
	pad_0: i32,
	pad_1: i32,
}

And add as many padding fields as needed until it works.

[aojiaoxiaolinlin]

I found that the error shouldn't be in view_buffer.binding(). Through log printing, I discovered that the failure occurs when executing .push(blur_uniform), whereas .push(view_uniform) succeeds.