[Bug 41394] New: 3DMark2001se 'Car chase' speed regression on mesa/nouveau

https://bugs.winehq.org/show_bug.cgi?id=41394

Bug ID: 41394
           Summary: 3DMark2001se 'Car chase' speed regression on
                    mesa/nouveau
           Product: Wine
           Version: 1.9.16
          Hardware: x86
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: directx-d3d
          Assignee: [e-mail]
          Reporter: [e-mail]
      Distribution: ---

After upgrading wine to ver. 1.9.18 I noticed strange speed regression. After
bisecting wine I found this commit was source of it:

first bad commit: [b62f4139eda58e8e25cbc58864fcea2599a20e97] wined3d: Swizzle
D3DCOLOR attributes in the GLSL FFP replacement when necessary

I'm using nv92 (GeForce 8800 GS) video card with relatively new kernel
(4.8.0-rc7-x64+) and 32-bit Slackware OS with many manual updates, including
git Mesa drivers. Initially I blamed mesa for speed regression, or my specific
build of it, but after testing Mesa 12.0.3 and seeing same very low speed there
I tried wine bisect first. May be there is interplay between some bugs, or
nouveau simply can't handle slightly different kind of load. But most other
tests in this benchmark works fine, and older 3DMark2000 also speedy.

I tried to make apitrace of this issue but under apitrace both good and bad
commits played 'car chase' scene (low/high detail) smoothly. Without apitrace
running bad commit mostly slows down to 0-1 fps even on low detail scene.

Compressed traces are at 
https://cloud.mail.ru/public/75Mf/fFuSprBh9

Bug still present in current git head (post 1.9.19).

[Bug 41394] 3DMark2001se 'Car chase' speed regression on mesa/nouveau

https://bugs.winehq.org/show_bug.cgi?id=41394

--- Comment #1 from Matteo Bruni  ---
That's weird indeed, Mesa on your GPU should support ARB_vertex_array_bgra so
this patch should not make any difference on your setup. It's also pretty
suspicious that it's not slow with apitrace.

I have downloaded the traces but I haven't looked at them yet. I also have a
box with a similar GPU, I'll try to reproduce there (if I can manage to start
that box again, at least...)

I guess you could also try to attach (or link to, if it's too big after
compressing it) a +tid,ddraw,d3d,d3d_shader log with current Wine, maybe that
will give us some clue.

[Bug 41394] 3DMark2001se 'Car chase' speed regression on mesa/nouveau

https://bugs.winehq.org/show_bug.cgi?id=41394

--- Comment #2 from Andrew  ---
Created attachment 55754
  --> https://bugs.winehq.org/attachment.cgi?id=55754
WINEDEBUG=+tid,ddraw,d3d,d3d_shader log (xz compressed)

151 mb uncompressed!

[Bug 41394] 3DMark2001se 'Car chase' speed regression on mesa/nouveau

https://bugs.winehq.org/show_bug.cgi?id=41394

Andrew  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |FIXED

--- Comment #3 from Andrew  ---
I think this one was fixed - I have another montherboard/cpu now, but same
videocard.

Using wine wine-2.2-33-g60c98ca on linux 4.10 git / mesa Mesa 17.1.0-devel
(git-5398d00) I can't observe any serious slowdowns even if I set  cpus to
1.4Ghz speed (current minimum)

Thanks for fixing this!

[Bug 41394] 3DMark2001se 'Car chase' speed regression on mesa/nouveau

https://bugs.winehq.org/show_bug.cgi?id=41394

Alexandre Julliard  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED

--- Comment #4 from Alexandre Julliard  ---
Closing bugs fixed in 2.3.