[PATCH vkd3d 03/12] vkd3d: Do not call vkCmdPushConstants() with a count of zero (KHRONOS_validation).

[Zebediah Figura]

On 4/21/22 11:30, Henri Verbeet wrote:
> On Wed, 20 Apr 2022 at 19:50, Zebediah Figura <zfigura at codeweavers.com> wrote:
>> On 4/20/22 10:40, Henri Verbeet wrote:
>>> Is calling SetGraphicsRoot32BitConstants()/SetComputeRoot32BitConstants()
>>> valid d3d12? If not, we should be fixing the callers instead.
>>>
>>
>> Right, I believe I received the same feedback when submitting this patch
>> before, and never found the time to properly look into it, which is why
>> it has remained in my tree without being resubmitted.
>>
>> Looking into it now, I don't see any indication that a zero count is
>> invalid, either from the "specification" [1] or in the function
>> documentation themselves [2] [3], and in practice it's worked on (AMD)
>> Windows when I've tried it. I haven't tried with the debug layer,
>> though, and I don't know of any other things I could check.
> 
> The debug layer might be interesting to check, but I guess that's good enough.

I was able to install the debug layer, and it turns out it actually does 
complain, sort of:

"ID3D12CommandList::SetGraphicsRoot32BitConstants: The currently set 
root signature declares parameter [0] with type 32bit constants, with 
Num32BitValues 0, so it is invalid to set DestOffsetIn32BitValues of 0."

It doesn't complain about setting a count of zero, though, just about 
offset >= size.

Given that it seems to work in practice, I'm not sure whether it makes 
sense to have this patch or not.