[Bug 105496] Using a single-plane imageview from a multi-plane image is broken

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=105496

--- Comment #24 from atomnuker at gmail.com ---
(In reply to Lionel Landwerlin from comment #23)
> Digging through the khronos gitlab, it seems the extended flags was added
> for another purpose :
> https://gitlab.khronos.org/vulkan/vulkan/merge_requests/1896
> 
> I think the usage you describe looks like it would fix atomnuker's problem,
> but it feels like an unintended consequence :)
Well, the EXTENDED usage flag is for when your imageview's format is supported
by the GPU but the format the image was created with is not. It only applies to
formats, rather than usage flags, tiling, etc.
The flag to be able to create single-plane representations of multi-plane
images is VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT.


> That said, the interaction here between VK_KHR_maintenance2 and
> VK_KHR_sampler_ycbcr_conversion is a bit subtle so I'm not that
> surprised that both of you (along with the CTS authors) missed it.
There should be no interaction between either, maintenance2 isn't required by
the YCBCR extension, but maintenance1 is.


> Yes, just creating a view seems to be nicer than creating an aliasing image.
> However, I don't think that's as ugly as you think it is.  If you create the
> image with the disjoint bit, it is required to work as per section 11.8 of
> the Vulkan 1.1.71 spec.
Well, it sort of is - having the DISJOINT bit set on an image you created means
having to allocate each plane separately, contributing to memory fragmentation,
allocation time, flushing and invalidating, and, depending on implementation,
image accessing speed.


Anyway, I've submitted a proposal to clarify that the STORAGE usage flag on
multiplanar formats only applies to single-plane VKImageViews. Knowing that
another implementation (nvidia's) also doesn't flag the storage bit for
multiplanar images makes me thing I should word the proposal more strongly
(e.g. if all compatible singleplane representations' formats supports the
STORAGE bit, then it must be set as well for the multiplanar format).

-- 
You are receiving this mail because:
You are the assignee for the bug.
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20180318/d118d5dc/attachment.html>