Non-render-target textures interop

[lien0n]

Hi @jherico, thanks for the helpful example.

I am having a GL app, which uses big amount of 3D textures (e.g., thousands of 48x48x48 texture objects). There is a situation that I need to release all the GL textures to make room for another App. However, the GL texture release is delayed (even with glFinish() calls) by the driver to a time point that's unknown to me, and the other app's GPU RAM allocation would fail from time to time.

Given your example, I am thinking to have Vulkan taking care of the device memory allocation/deallocation. Am I on the right track with this? If I am, given that my 3D textures are not rendering targets and they are read-only to shaders, do I still need the two semaphores and layout-changing command buffer?