On Wed Nov 30 10:54:56 2022 +0000, Georg Lehmann wrote:
Yes, we can remove it when we stop using direct calls for descriptor set updates. But at the moment I really don't want to have worse performance with descriptor buffers compared to descriptor sets caused by wine. Especially since it's unlikely that Rémi's work will be in 8.0. This is not just some theoretical problem. With the current non direct calls, vkd3d-proton's descriptor buffer path regresses performance for radv users.
I'm more optimistic about 8.0. Anyway, I wouldn't mind committing it if we need to. At the same time, since there is an active work on the final solution, we may as well wait a few days and see how it goes. Let me defer that to Alexandre (I think he took over that optimization work).