Myles C. Maxfield

The way it works in Metal is kind of complicated. There are 2 relevant limits: 1. Each pixel has to lie within a maximum number of bytes. This limit is...

Looking at this now.

I have a proposal for this, but I'm not sure it's correct on older versions of macOS/iOS. I don't want to post it until I'm a bit more confident in...

In the [above post](https://github.com/gpuweb/gpuweb/issues/2965#issuecomment-1158100115), the second limit only kicks in for render passes using tile shaders (which are only available in Apple4 or later). The way we know whether the...

Sounds like we should at least start with the Metal ones and expand from there once new data presents itself.

``` var total = 0 foreach renderTarget in renderTargets: let format = renderTarget.pixelFormat var formatWidth = widthOfFormatInBytes(format) // e.g. 8 for RG32Float var componentWidth = widthOfComponentFormatInBytes(format) // e.g. 4 for...

Maybe an extension?

If we were going to expose this in WebGPU, the API would probably have to include calls like Metal's `useResource` and D3D12's `makeResident()` calls.

Another piece that this investigation doesn't include is affordances for the implicit barriers that WebGPU implementations will emit internally. If we don't know which resources are used, it will be...

> WebGPU should go with "descriptor indexing" approach whereas pointer types sit in large already pre-validated arrays Yes, I understand what you mean now. You're right, this would significantly speed...