alan-baker
alan-baker
SSBO initialized values are tested (0 and non-zero) via the atomic execution tests (e.g. `add` uses 0 and `and` uses 0xfffffff).
https://github.com/gpuweb/gpuweb/pull/4801 resolves the spec issue.
The current spec just tries to mirror the Vulkan restrictions since D3D restrictions aren't well documented. Vulkan's restrictions are: >The following instructions must not flush denormalized values: OpConstant, OpConstantComposite, OpSpecConstant,...
WGSL WG meeting 2024-07-30: we revisited this issue and decided it was more appropriate to make compat errors happen at pipeline creation
I like the way it is right now. I don't see the harm in being able to set overrides in the module that aren't used by the shader. The next...
My interpretation of the https://github.com/gpuweb/cts/issues/1991 is that the title doesn't match the description very well. To me, the description says make sure to cover more indirect uses of overrides (e.g....
> I'm not sure if this is the same issue but does this produce an error if I only reference `bar`? It feels like it does given the recent decision...
We discussed this internally today. We think that pipeline-creation should be restricted to the stage being compiled generally. There are a few interesting cases that ought to lead to spec...
Have you tried this on D3D? From my searches it seems like this isn't actually implemented in drivers, just reference implementations and maybe not DXC at all (see https://github.com/microsoft/hlsl-specs/issues/245). Not...
All formats are now covered with read and write tests via #3841 and #3849.