Refactor GLTF Support

[Schneegans]

In my opinion, the current state of our GLTF support should be improved. Here are a few thoughts:

• We could update our version of tinygltf which would make some code obsolete.
• Environment texture handling should be decoupled from the model. In the settings there could be separate arrays, one for cubemaps and one for models.
• I think currently no graphics memory is freed when models are deleted. This leads to memory leaks when we reload settings or plugins.
• The code in src/cs-graphics/internal/gltfmodel.cpp could make use of some vista classes for OpenGL wrapping in order to reduce the line count significantly.
• Support for more GTLF features (animations?) would be awesome!

What do you think?

[bernstein]

Sounds good

• We could update our version of tinygltf which would make some code obsolete.

we should do that

• Environment texture handling should be decoupled from the model. In the settings there could be separate arrays, one for cubemaps and one for models.

yes, we should definitely do that!

• The code in src/cs-graphics/internal/gltfmodel.cpp could make use of some vista classes for OpenGL wrapping in order to reduce the line count significantly.

yes, thats true

• Support for more GTLF features (animations?) would be awesome!

Animations would be great

• I think currently no graphics memory is freed when models are deleted. This leads to memory leaks when we reload settings or plugins.

Actually gpu resources should be freed when a GltfLoader object and all the nodes created by it are destroyed. We are using shared_ptr with a custom deleter that frees the gpu memory on destruction but there is another design issue here:

• How to handle shared GPU resources in general:
  1. Currently shared resources like textures and bufferdata are managed by the GltfLoader. So multiple Vista nodes might share the same VBO. So as long as a GltfLoader object and/or its created nodes exist, used gpu memory is not freed
  2. Another solution would be that every created VistaGltfNode handles its own resources. So no more shared gpu data. Then removing scene graph nodes and thereby freeing associated gpu memory is simple. A drawback is of cause gpu memory fragmentation because now every node with a mesh has its own vbo.
  3. Or we add a the gpu resource handling to the GraphicsEngine.
  4. Other ideas?
• I just looked into the GraphicsEngine and it already handles some shared resources such as hdrbuffer. I think we should go with solution 3 and use the Engine for resource handling such as the brdf textures and vbos. And maybe we can get rid of the GltfLoader class completely. What do you think?