Check smoketest buffer/memory binding behavior

[karl-lunarg]

Issue by tobine (MIGRATED)
Thursday Nov 17, 2016 at 21:49 GMT
Originally opened as KhronosGroup/Vulkan-LoaderAndValidationLayers#1172

---

Smoketest has a buffer that is bound to memory without calling vkGetBufferMemoryRequirements(). The spec implies that this should not be done and there's a pending validation warning to flag this case. Understand smoketest behavior in this regard and update as appropriate.

[karl-lunarg]

Comment by Tony-LunarG (MIGRATED)
Friday Nov 18, 2016 at 17:16 GMT

---

If I create N buffers, all with identical BufferCreateInfo structs, and call GetBufferMemoryRequirements only for the first buffer, wouldn't the returned values apply to all N buffers? I think that's what's going on here. Should that really be a warning?

[karl-lunarg]

Comment by tobine (MIGRATED)
Friday Nov 18, 2016 at 18:36 GMT

---

@Tony-LunarG in most reasonable cases I would think the returned value for one buffer should apply to all identically created buffers but I don't believe that's required by the driver. There are some guarantees given by the implementation in the spec, but size doesn't appear to ever be guaranteed.
I do think it should be a warning if a user is binding memory to a buffer or image for which the memory requirements have not been explicitly requested for that exact buffer or image.

If a user is comfortable creating a bunch of identical buffers, only retrieving memory requirements for one of them, and then binding them all to memory, they can simply ignore the warning.

I suspect there will be a reasonable number of new Vulkan users who fail to call vkGetBufferMemoryRequirements() at all and make assumptions about memory that will benefit from the warning.

[mark-lunarg]

This issue should be opened against the Vulkan-Tools repo. Closing.

[karl-lunarg]

Remigrated to KhronosGroup/Vulkan-Tools#2