Description

I'm going through some compute shader work and get confused with instanceIndex from instances and geometry and instanceIndex with compute shaders.

From my (limited) research it's more common to use something like "Invocation" or "thread" etc. But the most common is "Invocation" Thats What Vulkan Calls it

Thats also what compute in the webgpu spec is called.

I think while TSL is still in early beta we should consider matching normal semantics for this.

If not, be explicit in the documentation on WHY we are different.

Solution

Use Industry terms like "Invocation"

Alternatives

"Threads" "Passes" etc.

The regular term is invocation_id with global_invocation_id or local_invocation_id so invocationId makes sense, but if you must have index invocationIndex would also work.

This would give us good clear split from geometry instances.