Kelsey Gilbert

WGSL meeting minutes 2022-05-17 * AB: As stands, it’s an undefined value out of bounds. Any value is possible for that type. Folks have expressed static values should be shader...

WGSL meeting minutes 2022-08-02 * DN: Gives background. * MM: Let’s not solve it ourselves. Let’s defer to another existing well-defined spec. * DN: I looked at C++17, C++11 and...

Specifically, I think 640 would be a fantastic wink here that we're aware that maybe it's low, but also signal that we'd obviously entertain increasing it later. :)

WGSL meeting minutes 2022-08-02 * KG: What do we want to do that is different from what we currently do? * DN: **I want to wait and see where the...

WGSL 2022-09-06 Minutes * DN: DS filed a thing and I was looking at hex floats. Was looking at what JS did it and said 20 digits and think that's...

WGSL 2022-08-09 Minutes * GR: correct we do not have those precision guarantees. Flush denorms. Don’t have explicit atan2 intrinsic. * DN: So, does MSFT have a statement of rules...

TR linked us to the atan2 always-polyfill implementation in DXC: https://github.com/microsoft/DirectXShaderCompiler/blob/b3101a2feb078a9da9c78b44d9b61c8c312b646e/lib/HLSL/HLOperationLower.cpp#L1530 Here's my translation to WGSL-ish: ```rs fn atan2(y: T, x: T) -> T { let tan = y /...

Notably, DXC has e.g. asin and acos as opcodes (like atan). It's possible that DXC's e.g. asin might have better ulp than `atan2(sqrt(1.0 - x * x), x)` that we...

WGSL 2022-08-16 Minutes * KG: Like MM point we shouldn’t care too much and figure out what the bounds are. Folks either us the builtin or roll their own. Can...

WGSL meeting minutes 2022-07-05 * Offline: DN: Google proposal: Background is at: [https://github.com/gpuweb/gpuweb/issues/3081#issuecomment-1170472250](https://github.com/gpuweb/gpuweb/issues/3081#issuecomment-1170472250) but the summary is: * const expressions can be struct. (We already agreed to allow array, in...