Global Invocation ID node

[DougLilliequist]

Description

Not sure if I found a bug or not using it correctly, but the globalID node doesn't seem to work.

Solution

If it's behaviour is not what it suggets, I would suggest adding a node for getting the global invocation id.

Alternatives

For now it seems that manually calculating a 3D index via instanceIndex is the way to go.

Additional context

No response

[cmhhelgeson]

Could you provide a reproducible example or link demonstrating how globalId is not functioning properly? I've tried applying it to webgpu_compute_water in lieu of instanceIndex and it appears to be working as expected.

[DougLilliequist]

Sure! This is what I initially did (pardon the formatting...):

wgslFn(computeDivergence(this._width, this._height), [insideBounds, getGlobalId])({
						velocityXPing: isEven
							? storageTexture(this._velocityX.ping).toReadOnly()
							: storageTexture(this._velocityX.pong).toReadOnly(),
						velocityYPing: isEven
							? storageTexture(this._velocityY.ping).toReadOnly()
							: storageTexture(this._velocityY.pong).toReadOnly(),
						velocityZPing: isEven
							? storageTexture(this._velocityZ.ping).toReadOnly()
							: storageTexture(this._velocityZ.pong).toReadOnly(),
						divergence: storageTexture(this._divergence).toReadWrite(),
						index: globalId
					}).compute(this._threadCount, [4, 4, 4])

Notice how I pass globalId which is imported from three/tsl

I ended up using instanceIndex and manually calculate the global id. It works for now as a duct tape solution, but would be ideal to just have instant access to it and use the index math for something else.

[Mugen87]

It would be great if you could actually share a live example like a fiddle so the issue is reproducible. Use one of the links from the issue template.

[Mugen87]

For providing the live example you might want to use the code from #31941 where @cmhhelgeson update an example so it uses globalId.

[Mugen87]

Closing until a fiddle has been provided.