feat(layers): ScatterplotLayer uniform buffer

[ibgreen]

For #

Background

• Start transition to uniform buffers

Change List

• Collects scatterplot-specific uniforms in a uniform buffer
• Uses loaders.gl UniformStore class to generate the uniform buffer
• Set the uniform buffer using model bindings property
• Types ScatterplotLayer state so that we can benefit from uniformStore.setUniforms() type checking.

[felixpalmer]

As it needs to be identical, could you define in one place? Derive it from scatterplot.uniformTypes?

[ibgreen]

Yes, though there are some tradeoffs to it. So far it seems the shader linker complains loudly if they are different so it does not seem to be a subtle source of errors.

I tried breaking the uniforms into a separate string and injecting it. A minor problem with that is that I use the vscode GLSL syntax highlighter to make these glsl tagged strings readable, and it breaks if I add template string arguments.

I actually added code for generating uniform declarations from uniform types (in the @luma.gl/shadertools module), but I haven't started using that yet.

The high-level question for the TSC is perhaps: how much shader code should be autogenerated? Once one starts down that dark path...

[felixpalmer]

Do all the uniforms need to be typed out in both the vertex and fragment shader? It seems unnecessary to have to include uniforms in the vertex shader that only the fragment shader reads and vice-versa

I do agree there is a tradeoff between saving time and having shaders that are actually readable in the source

[ibgreen]

Do all the uniforms need to be typed out in both the vertex and fragment shader? It seems unnecessary to have to include uniforms in the vertex shader that only the fragment shader reads and vice-versa

Yes I believe so, though more experience may show otherwise. Also we need to see how this works in both GLSL and WGSL.

[chrisgervang]

Agree with you there. Auto-generation does add a layer of complexity, making it all the more important to have solid debugging tools. I'm all for it, just think we really need to keep an eye on the debugging side of things to make sure everything runs smoothly.

[felixpalmer]

OK, in the spirit of keeping it simple & landing v9 I'm for just typing out the uniforms in full at this stage

[felixpalmer]

More restrictive type? 0 | 1 | 2

[Pessimistress]

It's a little confusing to see this constant referenced in the layer code. Maybe name it ScatterplotUniformLayout?