Experimenting with wine-pba

[Matteo Bruni]

On Fri, Jun 7, 2019 at 5:59 PM Gabriel Ivăncescu
<gabrielopcode at gmail.com> wrote:
>
> On 6/7/19 6:15 PM, Henri Verbeet wrote:
> > On Fri, 7 Jun 2019 at 19:24, Józef Kucia <joseph.kucia at gmail.com> wrote:
> >> On Fri, Jun 7, 2019 at 4:24 PM Gabriel Ivăncescu
> >> <gabrielopcode at gmail.com> wrote:
> >>> I'm interested in comparing the performance between it and DXVK on some
> >>> games I'm testing, especially those where DXVK seems to provide almost
> >>> no benefit (or sometimes even worse performance!) over wined3d, which is
> >>> a bit odd. I've seen numerous claims that wine-pba provides even greater
> >>> performance than DXVK on certain titles, so I'm curious.
> >>
> >> The PBA patchset is a wined3d patchset. It doesn't have any influence on DXVK.
> >>
> > If you're interested in doing performance work though, I'd argue it
> > would be more interesting to try to close the gap for those cases
> > where DXVK is currently faster than wined3d. It's great that DXVK is
> > working so well for some people, but it's also ultimately a dead end.
> >
> Yeah, in most case DXVK seems to be indeed faster, sometimes it's slower
> though (I don't know if it's my setup, or the game),

What is your setup / what games, for reference?

> but I see a lot of
> people who test both DXVK and wined3d with pba claim that wine-pba is
> actually faster than DXVK by a nice margin (10%-20% or more) in many
> cases, which is why I want to test it out for myself. :-)

Do you recall which GPU and drivers? FWIW wine-pba is mostly a
proof-of-concept (if a VERY interesting one :), a number of things are
simply ignored / not handled, which means applications might crash or
otherwise not work correctly. Performance might also be affected,
along the lines of "it's fast, but it's also wrong".

> Since DXVK already uses persistent buffers, it's a much fairer
> comparison between an OpenGL backend and Vulkan if we also use pba for
> wined3d.

I don't know about fair but yes, we have room for improvement in wined3d.

> > On the subject of PBA, I'm not aware of a more recent rebase, although
> > that doesn't necessarily mean there doesn't exist one. Matteo has a
> > set of patches using a somewhat similar approach that you may find
> > interesting though. Those aren't publicly available, but he may be
> > willing to let you have a look if you ask nicely.
> >
>
> That would be great. FWIW I don't mind if they're hacks at all or in
> crude condition, since it's just for testing purposes.
>
> Matteo would you please let me try out your patchset?

Sure, as Henri mentioned I don't have them in any public place at the
moment and yeah, they are not exactly clean in places (the hope is to
start upstreaming them in the near future though). I'll send you a
separate email pointing to some place you can find them.