Does instance_index work on Metal ?

[olofwalker]

I have rewritten my rendering pipeline to use DrawIndexedIndirect and rely on instance_index being set to base_instance to correctly index arrays with transforms and other stuff in the shader.

However, I can't get it to work, it seems like the builtin instance_index is always zero.

This is how the vertex shader entrypoint looks like

fn main_vs(
    @builtin(instance_index) object_id: u32,
    input: VertexInput,
) -> VertexOutput {

Here is how I create the indirect commands

        for (i, (vertex_count, indices_count)) in lengths.iter().enumerate() {
            indirect_data.push(DrawIndexedIndirect {
                vertex_count: *indices_count as u32, // Will be renamed in 0.19
                instance_count: 1,
                base_index: index_offset,
                vertex_offset,
                base_instance: i as u32,
            });
            vertex_offset += *vertex_count as i32;
            index_offset += *indices_count as u32;
        }

And yes, I have enabled the non-zero based instance index feature

        let (device, queue) = adapter
            .request_device(
                &DeviceDescriptor {
                    label: None,
                    features: Features::TEXTURE_BINDING_ARRAY
                        | Features::SAMPLED_TEXTURE_AND_STORAGE_BUFFER_ARRAY_NON_UNIFORM_INDEXING
                        | Features::PUSH_CONSTANTS
                        | Features::POLYGON_MODE_LINE
                        | Features::INDIRECT_FIRST_INSTANCE
                        | Features::VERTEX_WRITABLE_STORAGE,
                    limits: Limits {
                        max_push_constant_size: 4,
                        ..Limits::default()
                    },
                },
                None, // Trace path
            )
            .await
            .unwrap();

Is anyone else seeing this?

Specs: Macos Sonoma, Rust 1.72, WGPU 0.18

[cwfitzgerald]

It should, could you locally run our test for this thing.

cargo install cargo-nextest
// in wgpu checkout
cargo xtask test vertex_indices

[olofwalker]

@cwfitzgerald Thanks, this works, and I quickly discovered the problem when running my code.

My buffer holding DrawIndexedIndirect contains more than one entry, and I just figured that draw_indexed_indirect would draw all of them, but I should have used multi_draw_indexed_indirect.

The question is, why is there a need for two different functions doing in essences the same thing?

[cwfitzgerald]

It's a hardware limitation fundamentally. DX11 and GL4.0-4.2 only have single indirect. Multi draw indirect came in dx12/GL4.3. We emulate multi-draw on platforms that don't support it through repeated calls to draw_indirect.