I have a unique situation in my application where rarely I will issue drawing commands in my Metal app (on both iOS and macOS), and I end up not needing to actually execute them. Is there a way to safely discard one of the render pipeline objects without submitting it to the GPU?
I have tried discarding my MTLRenderCommandEncoder, but that trips an assertion when the object gets deallocated (because I haven't called endEncoding). If I just abandon the MTLCommandBuffer, it remains referenced by the MTLCommandQueue, so it just ends up sitting there (and using up one of my available command buffers). I haven't tried abandoning the entire MTLCommandQueue yet.
Is there some option I have missed? I am guessing the answer is no, but I figure it's worth asking.
NOTE: In case you're wondering what possible reason I could have to issue rendering calls and then 'change my mind' in runtime, the answer is that the calls are coming from another library which is used by 3rd party code. Unfortunately that library supports scenarios where rendering calls will be made on an object that doesn't end up being used. I must issue the calls to Metal, but I'd like to avoid actually executing them if possible.
I could just use my own data structure to queue up the drawing calls, and to only call Metal once I KNOW the stuff needs to get rendered, but that would significantly increase the complexity of my code.