Model Resource Caching

[Relfos]

This branch provides a modification the internals of the Model class.
Now there is a clear separation between a model and its resources used by the GPU (shaders, textures, vertex arrays, etc).
This means that it is now possible to load multiple instances of the same model without requiring reloading the whole model again, and without wasting gpu resources having multiple copies of the same things.

In order to preserve backwards compatitibility, the previous system will still work as usual. In order to make use of the new caching system, the programmer will have to create a ModelResourceCache and pass it inside the options when creating a new Model.
Currently models resources are cached by URLs (meaning Models loaded from the same URL will reuse the same assets if they are sharing a cache instance).

[pjcozzi]

Thanks for this @Relfos! Can you please merge cesium's master branch into your branch so we'll be able to merge when this is ready:

[pjcozzi]

In order to preserve backwards compatitibility, the previous system will still work as usual. In order to make use of the new caching system, the programmer will have to create a ModelResourceCache and pass it inside the options when creating a new Model.

Do you think we need this? I envisioned this working without any changes to the user's code.

[pjcozzi]

Actually, I'm going to wait until you merge in master so I can look at a clean diff.

[pjcozzi]

@Relfos please add a comment to this pull request when it is ready since github does not automatically send an email when new commits are added to a branch. Thanks!

[Relfos]

Hi Patrick, can you check this pull request soon?
I updated it to Cesium 1.4, if we can finally integrate with it the master would help a lot!

[pjcozzi]

I'll try to look at this on Saturday. Please keep it up-to-date with master.

[Relfos]

Ok, I'll keep it updated until then!

On Mon, Dec 8, 2014 at 3:08 PM, Patrick Cozzi notifications@github.com
wrote:

I'll try to look at this on Saturday. Please keep it up-to-date with
master.

—
Reply to this email directly or view it on GitHub
#2177 (comment)
.

[pjcozzi]

Thanks again for the pull request @Relfos. I think this needs a few changes for robustness. Renderer resources should be cached per context to support the same model in two different Cesium viewers (which can't share resources), and cached items should automatically be deleted from the cache when they are not in use. We would also like to be able to cache using a user-defined key for the cases when the glTF was created procedurally and a URL is not available.

I tried to support all of this in the model-cache branch. Try it and let me know if it works for you. The only other changes I would consider for this branch are:

• Reference counting the animation cache so it is also automatically deleted.
• Ability to invalidate the whole cache or the cache for one url/cache-key
• Avoid potential ping-ponging by delaying actually removing from the cache when the count reaches zero for several frames/seconds. This may also move all the GL delete functions into the render loop, which would be better for requestAnimationFrame timing (however, perhaps not noticeable).

Let me know if you need any of these changes or anything else in that branch.

CC #927

[pjcozzi]

Closing via #2340.