TODO.txt

-Shader should have Vertex type baked in 
-Textures should be able to be created empty. Then they don't need a staging buffer etc
-Copy buffer should have the option to use graphics or copy queue
-Viewport needs to be set as dynamic state (instead of having to create PSOs when that changes) (Should this be done?)
-Make sure the shader compiler is included in the project
-Add ImGUI and delete from file dependency
-When storing uniform buffers per swapchain buffer. We should only allocate one buffer and use disjoint ranges that way we do not have multiple discrete buffers (note: you only need duplicate data for types that are mutable and CPU accessible like uniform buffers. That way if multiple frames are in-flight you aren't modifying uniform memory for a previous frame. GBuffers only need one copy because it is only ever used by the GPU and only one thing will be reading and writing to it at the same time and they will be synced thanks to semaphores and other GPU sync objects already)
-Look into push constants in order to push UBOs to the GPU more efficiently
-When creating an image by creating a buffer and an image and changing the layout and copying to it and then changing the layout to an optimal read-only format. All of these steps should be done in a single command buffer instead of each of them synchronously making their own command buffer and flushing (vkQueueWaitIdle calls)
-TODO Doesn't have to fail if a device doesn't support ANISO 16x, instead could just not use that filtering on samplers
-For the textures I think it makes the most sense to have a descriptor set per material that includes the texture references (and re-uses one sampler for each) and includes other material specific values. For UBOs I would similarly have a descriptor set per mesh that includes all the mesh specific values. That way you can efficiently swap out descriptors based on groups of resources.
-https://developer.nvidia.com/vulkan-shader-resource-binding

Long term:
-Use the VulkanMemoryAllocator instead of calling vkAllocateMemory for each buffer/resource
That way you will have a memory pool that you can use memory from, thus lowering the driver overhead of those frequent allocation calls
-Once a sample is up and running, test current resource implementation of VK_SHARING_MODE_CONCURRENT and switch it to VK_SHARING_MODE_EXCLUSIVE and transfer the resources after creation to the graphics queue. Any performance difference?