Inferring Vertex Buffers from WGSL to TS

[reczkok]

When inferring vertex buffers form a .wgsl shader we face an ambiguity problem. For example:

@vertex
fn main(
  @location(0) position : vec3f,
  @location(1) normal : vec3f,
  @location(2) uv : vec2f
) -> VertexOutput {
  ...
}

There is not enough information here to properly define what buffers we are dealing with. Is it one vertex buffer with three attributes? Are there three vertex buffers with one attribute each? Furthermore, we don't even know if it's a vertex or instance vertex buffer. Since there is no way to infer this information we need the user to somehow pass it.

[iwoplaza]

(work-in-progress)

// shader.wgsl

struct VertexOutput {
  @builtin(position) position: vec4f,
  uv: vec2f,
}

@vertex
fn main(
  @location(0) position: vec3f,
  @location(1) normal: vec3f,
  @location(2) uv: vec2f,
) -> VertexOutput {
  ...
}

// shader.ts

/* generated via tgpu-cli by TypeGPU */

const VertexOutput = {
  position: builtin.position,
  uv: vec2f,
};

const main = tgpu
  .vertexFn([vec3f, vec3f, vec2f], VertexOutput)
  .does(/* wgsl */`(
      @location(0) position: vec3f,
      @location(1) normal: vec3f,
      @location(2) uv: vec2f,
    ) -> VertexOutput {
      ...
    }
  `)
  .$uses({ get VertexOutput() { return main.Output; } });

// main.ts

import tgpu from 'typegpu';
import { looseArray, looseStruct, vec3f, vec2f, unorm8x2 } from 'typegpu/data';
import { vertexMain } from './shader.ts';

const VertexData = looseStruct({
  position: vec3f,
  normal: vec3f,
  uv: unorm8x2,
});

const vertexLayout = tgpu.vertexLayout((n) => looseArray(VertexData, n), 'vertex');
const { attrib } = vertexLayout;

const vertexBuffer = tgpu
  .createBuffer(vertexLayout.schemaForCount(512))
  .$usage('vertex');

const renderPipeline = tgpu
  .withVertex(vertexMain, [attrib.position, attrib.normal, attrib.uv])
  .withFragment(...)
  .createPipeline();

renderPipeline
  .with(vertexLayout, vertexBuffer)
  .draw(...);

The completeness of the current approach

The following properties have to be known before creating a pipeline:

type GPURenderPipelineDescriptor = {
    vertex: GPUVertexState;
    primitive?: GPUPrimitiveState;
    depthStencil?: GPUDepthStencilState;
    multisample?: GPUMultisampleState;
    fragment?: GPUFragmentState;
};