Fix examples on nvidia linux

[rukai]

The examples currently crash almost immediately on nvidia linux.
This is due to an OutOfHostMemory error as described in this issue: #923, #627
This PR changes the examples to wait on the previous frame before submitting the next frame.

I think this is the best solution but other solutions I thought of are:

• call wait method immediately after submitting frame
  • Only need the Box<GpuFuture> variable (dont need the Option<Arc<vulkano::sync::FenceSignalFuture<_>>>)
  • Wastes time blocking
• Add a wait method to GpuFuture trait that fails silently/panics/returns error enum if called on any future other than FenceSignalFuture
  • Only need the Box<GpuFuture> variable
  • Doesnt waste time blocking
  • More potential runtime errors

[rukai]

As mentioned in #627 this issue is caused by this unresolved bug in nvidia drivers (apparently windows as well as linux) https://vulkan.lunarg.com/app/issues/56ca3a477ef24d0001787448
Because its a driver bug, this PR should not be merged. However it should be kept open until the issue is resolved as it documents a work around.

[jonathansty]

@rukai just as a heads up that driver bug seems to be closed. The validation message definitely sounds like it's not a bug but just wrong usage in our case?

22:44:41 [ERROR] "Fence 0x12 is in use. The Vulkan spec states: Each element of pFences must not be currently associated with any queue command that has not yet completed execution on that queue (https://www.khronos.org/registry/vulkan/specs/1.1-extensions/html/vkspec.html#VUID-vkResetFences-pFences-01123)"

[rukai]

oh, good to know, we'd better figure out whats going wrong then.

[jonathansty]

I don't have enough knowledge on how vulkano handles the GPU futures.

However I did give a try to using some raw functions and using fences instead of the GpuFutures to wait. In my application this commit fixed the issue.

Although probably not the most optimal fix as I'm waiting before doing the submit and present and then also immediately waiting after.

[knappador]

Motion to close; expiration date seems lapsed (the examples run on Nvidia hardware)

I do suspect we are missing a lot of behavior on synchronization given the validation, but mainly I'm just having difficulty getting Intel chips to run like Nvidia ones right now.

[AustinJ235]

I know at least the image example still has the issue haven't tested the others. It is probably a good idea to resolve this pull request or open another one to allow the examples to run on nvidia cards.

[AustinJ235]

closed in favor of #1246