jiangzhaoming
jiangzhaoming
I thought the spec said `x / 0 = x`. > If T is a signed integral type, the scalar case, evaluates to: > * e1, when e2 is zero.
I see, `e1 / e2` has no clear definition for floating point in this table. However, for `e1 % e2` we do have the result equals to `e1 - e2...
I understand, thanks
I have a problem about this. As for C/C++, we have the operators precedence `&` lower than `+`, and if we write expression `1+2&4` and `4&2+1`, these are ***valid*** and...
Oops, sorry I missed this......
@qjia7, @fs-eire, @guschmue, @gyagp PTAL
> is there any disadvantage of appending "enable subgroups;" in a program that does not use subgroup feature? For now I don't know there would be overhead to enable `subgroups`.
One proposal is to add an attribute, `adapterInfo`, into `GPUDevice`, to access the adapter information of the adapter that created this `GPUDevice`. The detailed proposal can be found in #4809....
Another proposal could be allowing getting a `GPUAdapter` out from a `GPUDevice`, referencing to `GPUDevice.[[device]].[[adapter]]`. According to current WebGPU spec, an `adapter` would fall into "consumed" state if the adapter...