David Neto

And just for completeness sake, approximating 1/sqrt(x) has some fast approximations. https://en.wikipedia.org/wiki/Fast_inverse_square_root

The other thing that probably needs addressing is that now the '[callable](https://gpuweb.github.io/gpuweb/wgsl/#syntax-callable)' grammar element is ambiguous: callable : | [ident](https://gpuweb.github.io/gpuweb/wgsl/#syntax-ident) | [type_decl_without_ident](https://gpuweb.github.io/gpuweb/wgsl/#syntax-type_decl_without_ident) | [vec_prefix](https://gpuweb.github.io/gpuweb/wgsl/#syntax-vec_prefix) | [mat_prefix](https://gpuweb.github.io/gpuweb/wgsl/#syntax-mat_prefix) | [array](https://gpuweb.github.io/gpuweb/wgsl/#syntax-array)

I'm considering defining a "pretype" as a kind of object which, together with a parameterization, names a type. And predeclared identifiers can resolve to those pretypes. Example, `vec2` is a...

Thanks for your patience waiting for this delayed reply. Yes, I'm interested in integrating with OSS-Fuzz. Great project! My team uses Effcee in googletest-based tests in our other projects, so...

Hardware imposes requirements, and you've called them "required alignment" above. Carrying the constraint out to containing aggregates is how we make sure all the math works out: If there's shader...

Sorry, let me take another crack at this. I think there the levels of support breaks down into these slices: | *Num* | *Constraint* | *Enforcement* | *Consequence* | |----|-----|----|----|...

Discussed in [2022-06-07 WGSL meeting.](https://docs.google.com/document/d/1EnjkYkp3ce4A78npEV8-iyAOiXNN8YzvVyt2AHHm4oI/edit#heading=h.us5u6wdocxi2) * Confirmed we should use option 3 from above. * Terminology distinction that may help clarify: "packing constraint" (spacing out data within shader-declared things) vs....

Editorially, it was much simpler to keep concrete vs. abstract for each of shift left and shift right. The key thing this PR does is decouples the type of the...

Group agreed to land this in the 2022-10-11 meeting.

From the discussion at the F2F, the implementation can often ensure it's done better than if the compiler does it. It has proven tricky in the past. Also applies to...