What rules exist around BathRendererGroup, Compute/Graphics Buffers, and instance data?

I know there's a new version of BatchRendererGroup coming, and maybe that will change the answer to this question, or maybe it is orthogonal.

So I want to run a Compute Shader only on instances that pass culling. However, it seems currently HR V2 uploads all instance data and compute buffers (skinning) prior to culling. I recognize that culling happens in multiple stages for multiple cameras. That's fine if I have to incrementally update the compute buffer and run additional compute shader passes with a different range of the buffer. So I guess my questions are:

• Is updating the instance data buffers safe during culling?
• Is updating a Compute Buffer safe during culling?
• Is dispatching a Compute Shader safe during culling?
• Is there some other SRP hook where I could be doing this uploading and/or Compute Shader dispatch instead after knowing the culling results?

 

It has been a while, but I wanted to follow up and state that so far this is working surprisingly well!

I have been dispatching a compute shader from the culling callback for a while now. But this weekend I got the entire material property updates migrated over. I keep track of which entities have been rendered in a frame and only change values for new entities which are passing a culling callback for the first time. This gives me a last-moment opportunity to set an offset into a buffer populated by a compute shader on an entity. I also have to preserve old buffers for a few frames when resizing to a new one, but overall, VRAM usage is way down!