petermcneeleychromium

**Adding some data from existing code usages** [Bevy link to change from storage to texture](https://jms55.github.io/posts/2024-11-14-virtual-geometry-bevy-0-15/) ``` let visibility = (u64(depth) R64Uint texture atomics would both be faster than using buffers,...

I have split my original proposal into two independent proposals. One for min/max on storage and one for the other ops (storage and workgroup). The first one is realistically something...

I was talking about storage buffers not texture atomics. Is the visbuffer resolve a different pass?

As discussed this can likely be worked around via bitcast and a uint32 CAS with possible exception of NaNs https://www.w3.org/TR/WGSL/#bitcast-builtin

You can probably avoid parallel for each vertex in mesh: normalize(vertex normal) If you just normalize in the CAS. I wonder if the is a ABA problem with this?

You are correct. You would need to store additional information (a magnitude) if you wanted to avoid this secondary pass

I have seen this usage in other code (bug submission crbug.com/396384013) I wonder if we should just add the function and internally pollyfill.

Pull request for the spec https://github.com/gpuweb/gpuweb/pull/5199 Pull request for the cts https://github.com/gpuweb/cts/pull/4392 Public data from device https://docs.google.com/spreadsheets/d/1i5n8sFlkHi0QrFwa00oXVcY8iBo6VftXzxOsziWzrsg/edit?gid=176471445#gid=176471445

The polyfill (if possible) would be useful for platform consistency. I can see someone writing a feature that uses sample mask in such a way only to find out on...

Assuming this pr gets approved this pull is waiting in the wings https://github.com/gpuweb/cts/pull/4392