DriftEngineRenderer · function
generateMipChain
Fill every level of a texture from the one above it, which is what WebGPU asks for instead of a call.
Explained in Hello world.
function generateMipChain(device: GPUDevice, pipelines: MipPipelines, texture: GPUTexture, format: GPUTextureFormat, levels: number,import { generateMipChain } from '@driftengine/core';Parameters
| Parameter | Type | Description |
|---|---|---|
device | GPUDevice | |
pipelines | MipPipelines | |
texture | GPUTexture | |
format | GPUTextureFormat | |
levels | number | |
/** How many array layers to fill | | |
each its own chain. One for a plain texture. */ layers?: number | |
In depth
gl.generateMipmap is one line and WebGPU has no equivalent at all, so this is the whole of
the difference between the two backends for any texture that wants a chain. One encoder, one
render pass per level, submitted together, once, at upload.
The source view is a single level, so the sample cannot read the level being written — that would be one texture bound as an attachment and as a resource at once, which is rejected, and rejected quietly enough to matter.
A free function because there are two callers now. A surface texture on a mesh has wanted a chain since it existed; a sprite sheet wanted one the day a consumer baked type into an atlas and drew it at a seventh of its authored size. Writing the blit twice would be two statements of one thing, and the second copy is the one that would quietly disagree about the format.