Improved VK_EXT_descriptor_heap align limits - #2787
Conversation
|
Could you identify the spec language behind your comment "The alignment limits must be less than or equal to the corresponding maximum size limits."? We weren't able to resolve this in real time on the teleconference today and people are not sure about this change. |
|
@mematthias for context, it is allowed for drivers to have a It is already invalid in the spec to have that flipped, where the alignment is less than the size |
|
The fields of
@spencer-lunarg According to the constraints, If the |
Sorry, you are right, but the "precise" size from |
|
@HansKristian-Work I think i messed this up from what you said on the call, the spec does say |
|
Given the constraints, it could even be shortened further; so that it is no longer |
|
Right, it's the individual descriptor types as queried by GetDescriptorSizeEXT that can return "odd" sizes ... Missed the requirement that bufferDescriptorSize and friends had to be POT. The PR LGTM after more spec review. |
I think the min makes sense since it makes the spec constraint clearer. |
The alignment limits must be less than or equal to the corresponding maximum size limits.
This is now represented by
min.