On Tue, 20 Jul 2021 at 05:24, Zebediah Figura zfigura@codeweavers.com wrote:
I don't know if this (or the next patch) offers a measurable performance benefit, but it seems like probably a good idea...
With the "right" combination of application and GPU, quite possibly. However:
- It's always good to have some data to back this kind of change, even if only for future reference. - I think it's a little harder than this patch, unfortunately. In particular, we track "bind_mask" per texture, not per sub-resource. So if after this patch we e.g. were to first call wined3d_texture_vk_barrier() for SRV usage of one level/layer of a texture, and then subsequently for a different level/layer, we wouldn't create a Vulkan barrier for the second call.