Previews, as seen when this build job started (0b580c7):
WebGPU _{webgpu.idl | Explainer | Correspondence Reference}
WGSL _{grammar.js | wgsl.lalr.txt}

Thanks all! Just need WG approval at this point (I don't think we finalized that previously.)

Let's try to get approval offline: @teoxoy @jimblandy this is basically the same as WGPU's SHADER_PRIMITIVE_INDEX feature, would you be ok merging this in?

LGTM but should it be primitive_index instead of primitive_id? same as vertex/instance/sample ending in _index. We do use ID for some of the builtins and it seemed to me that it was for those that don't have a scalar type (with the only exception being the recent-ish subgroup_invocation_id).

LGTM but should it be primitive_index instead of primitive_id? same as vertex/instance/sample ending in _index. We do use ID for some of the builtins and it seemed to me that it was for those that don't have a scalar type (with the only exception being the recent-ish subgroup_invocation_id).

Oh good observation, reading the relevant WGSL specification I am confused regarding when _id vs _index is used as it doesn't seem consistent. If we renamed subgroup_invocation_id to subgroup_invocation_index and also named this primitive_index then it would seem consistent as pointed out.

That's a really good point. I went with primitive_id because that's how it had initially been discussed and GLSL/HLSL/MSL all refer to it as "Primitive ID". They also use Vertex ID/Instance ID however, and we decided on the "Index" verbiage anyway, so I don't see any reason to break that pattern here.

I'm in favor of renaming to primitive_index, even if it means some minor refactoring Dawn/Chrome. :)

I'm not a WGSL designer but in my mind, that canonical thing is called id, and then IF we need a linearized version of the id, then that is called index. So primitive_id and subgroup_invocation_id are consistent.

I'm not sure I understand the difference, in that case, between vertex_index, instance_index, and primitive_id. All of them are "linearized" ids, and canonically the vertex and instance versions are referred to as ID by the native APIs. (SPIR-V appears to be the outlier?) If we still choose to call them vertex_index and instance_index in WGSL then I'm not sure why primitive would be different?

Ultimately I'll ensure the spec change follows whatever the WGSL group decides, but it does seem like "index" would be the more consistent verbiage.

I agree with @toji 's analysis. vertex_index, instance_index, and sample_index have no corresponding *_id equivalent.

The outlier then becomes subgroup_invocation_id - I'm not sure how much adoption of subgroups exists so far which would allow us to make a breaking change there for consistency

Ah fair point, I missed that about vertex/instance.

I'm still a little unsure if this should be called an "index" since it's not an index into any actual array, but I don't feel strongly about that.

Assuming index suffix means 'index into an array', then yes I think we should leave this as-is (primitive_id). Either naming is fine with me :)