alan-baker

That gets covered in particular expressions. For example the [array access expression](https://gpuweb.github.io/gpuweb/wgsl/#array-access-expr) section has entries for three cases: * a value (the case this issue is about) * a reference...

I think probably no because I expect to loosen the restriction on `const` in the future and allow it on user-defined functions.

>[Must](https://gpuweb.github.io/gpuweb/wgsl/#shader-creation-error) be a power of 2, and [must](https://gpuweb.github.io/gpuweb/wgsl/#shader-creation-error) satisfy the required-alignment for the member type: >If align(n) is applied to a member of S with type T, and S is...

Yes, you only need the initializer on one of the variables. There are (or will be) CTS tests for that interaction.

Can you describe the overall design succinctly in one of these pull requests? My understanding is the following: 1. clspv creates a descriptor for a storage buffer for printf and...

Is the story of accuracy roughly the same at fp16 as fp32 for Vulkan devices? Vulkan calls out some differences, but I haven't looked too hard at OpenCL's requirements. Do...

Ha turns out the issue I was thinking was fixed previously and I reviewed it. https://reviews.llvm.org/D81999 for reference.

Why not also infinities? The WGSL spec doesn't require infinity support currently so any infinity is effectively an undefined value. They also seem more common than NaNs so I'd expect...

Implementations could optimize adding the coherent qualifiers if shader does not contain any barriers since they would be necessary to perform the required synchronization (since WGSL only has relaxed atomics).

Given that we can't promise image synchronization using barriers currently, I would think there wouldn't be a point yet in marking write only storage images as coherent. There is a...