GPUBindGroup

GPUBindGroup 用 device.createBindGroup(descriptor) 创建。它本身没有方法。布局决定每个绑定号是什么资源，绑定组填入具体对象。

interface GPUBindGroupDescriptor {
  label?: string;
  layout: GPUBindGroupLayout;
  entries: GPUBindGroupEntry[];
}

interface GPUBindGroupEntry {
  binding: GPUIndex32;
  resource: GPUBindingResource;
}

type GPUBindingResource =
  | GPUSampler
  | GPUTexture
  | GPUTextureView
  | GPUBuffer
  | GPUBufferBinding
  | GPUExternalTexture;

interface GPUBufferBinding {
  buffer: GPUBuffer;
  offset?: GPUSize64;
  size?: GPUSize64;
}

直接传 GPUBuffer 等价于 { buffer }。直接传 GPUTexture 等价于它的默认视图。

布局

interface GPUBindGroupLayoutEntry {
  binding: GPUIndex32;
  visibility: GPUShaderStageFlags;
  buffer?: GPUBufferBindingLayout;
  sampler?: GPUSamplerBindingLayout;
  texture?: GPUTextureBindingLayout;
  storageTexture?: GPUStorageTextureBindingLayout;
  externalTexture?: GPUExternalTextureBindingLayout;
}

每一项只能填写一种资源成员。visibility 是 GPUShaderStage.VERTEX（0x1）、FRAGMENT（0x2）、COMPUTE（0x4）的组合。

interface GPUBufferBindingLayout {
  type?: "uniform" | "storage" | "read-only-storage"; // 默认 "uniform"
  hasDynamicOffset?: boolean; // 默认 false
  minBindingSize?: GPUSize64; // 默认 0
}

interface GPUTextureBindingLayout {
  sampleType?: "float" | "unfilterable-float" | "depth" | "sint" | "uint"; // 默认 "float"
  viewDimension?: GPUTextureViewDimension; // 默认 "2d"
  multisampled?: boolean; // 默认 false
}

interface GPUStorageTextureBindingLayout {
  access?: "write-only" | "read-only" | "read-write"; // 默认 "write-only"
  format: GPUTextureFormat;
  viewDimension?: GPUTextureViewDimension; // 默认 "2d"
}

interface GPUSamplerBindingLayout {
  type?: "filtering" | "non-filtering" | "comparison"; // 默认 "filtering"
}

minBindingSize 为 0 时，管线创建不检查大小，绘制或 dispatch 时再检查缓冲是否盖得住着色器变量。Uniform 的 offset 必须是 minUniformBufferOffsetAlignment 的倍数，默认限制是 256。

createPipelineLayout({ bindGroupLayouts }) 的数组下标就是 @group 编号。GPURenderPipeline 和 GPUComputePipeline 都有 getBindGroupLayout(index)；管线使用 layout: "auto" 时，用它取出推断布局。

相关页面

• 教程：绑定组
• 规范：GPUBindGroup