[Bug 35032] New: starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

Bug #: 35032
           Summary: starcraft unbearably slow in fullscreen mode
           Product: Wine
           Version: unspecified
          Platform: x86
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: -unknown
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
    Classification: Unclassified
   Regression SHA1: 94ae743ea668e49d40ae4e2dc5fe1f5d9be018cb

Created attachment 46698
  --> http://bugs.winehq.org/attachment.cgi?id=46698
glxinfo

Hi,

while Starcraft (specifically Broodwar) worked perfectly before, I now observed
that whenever I would let it run in fullscreen mode, it would be horribly slow
(about 2-3 fps) so that even moving the mouse cursor would become a frustrating
task. This issue does not show when starcraft is not drawing fullscreen, for
example when it is in a virtual desktop or when my window manager (awesome)
constricts it to a window.

This might be a problem with my graphics card/driver but since it worked fine
in wine a few years ago, I ran a regression test and found out that the problem
was introduced with commit 94ae743 which changed the DefaultSurfaceType to
opengl.

On my host I'm running Debian jessie (testing) amd64 with mesa 9.1 and intel
2.19 drivers. I did all the regression testing in a chroot environment with
Debian squeeze i386 because the specific wine versions around commit 94ae743
would not compile when libxml2 was too new.

I only have access to Intel graphics hardware, so I can't test on a machine
with a different graphics card. The laptop I was testing on was a Dell E6510
with Ironlake graphics as you can see in the attached glxinfo output.

I was told in IRC that this problem is likely a driver issue but nevertheless I
would like to know whether there is a patch that I can apply to my specific
wine version to restore the functionality I had before commit 94ae743 was
applied?

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #1 from josch  2013-12-02 04:33:24 CST ---
Hi,

I was being made aware of bug#34547. The reporter claims that for him
wine-1.5.28 worked flawlessly which is not the case for me.

Nevertheless, I also got the warn:d3d_perf:wined3d_surface_blt messages but
only when running starcraft in fullscreen mode. Once my window manager
constricts it to something that is not the fullscreen the messages go away and
the game is faster.

I also tried the patch attached to bug#34547 with the current git HEAD. While I
can confirm that the warn:d3d_perf:wined3d_surface_blt messages go away in
fullscreen mode with this patch, I do not see any difference in performance on
my system. Starcraft is still unplayably slow (no visible difference in speed)
during fullscreen mode and faster when not fullscreen.

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |regression

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #2 from josch  2013-12-03 12:44:49 CST ---
I compiled the most recent mesa version from git as of now:
5b331f6fcbf226f18e0c517ffdce30a39bb92982 from 2013-11-23 which was released
after mesa 10.0.

It did not make any difference in the effect I observed.

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #3 from josch  2013-12-03 13:39:28 CST ---
I now recompiled mesa with -DDEBUG to get some output with INTEL_DEBUG=perf and
here it is: intel_debug_perf.log

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #4 from josch  2013-12-03 13:40:02 CST ---
Created attachment 46718
  --> http://bugs.winehq.org/attachment.cgi?id=46718
intel_debug_perf.log

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #5 from josch  2013-12-05 12:41:25 CST ---
Following some suggestions from
http://dri.freedesktop.org/wiki/IntelPerformanceTuning/ yielded:

When the game is windowed (virtual desktop or otherwise) and running smoothly
then CPU is used 80% and GPU around 80%.
When the game is fullscreen and running slow and choppy then CPU is used 40%
and GPU is 100%.

This suggests that the bottleneck during fullscreen mode is the GPU.

I'm unable to execute "perf record -g -e i915:i915_gem_request_wait_begin" for
a more detailed analysis because it returns:
    invalid or unsupported event: 'i915:i915_gem_request_wait_begin

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #6 from josch  2013-12-06 17:25:26 CST ---
Created attachment 46764
  --> http://bugs.winehq.org/attachment.cgi?id=46764
gdi fix

This patch does not fix the issue I have with starcraft and the opengl renderer
(this still persists) but with this patch I can now play starcraft with the gdi
renderer without problems (since commit 6325f3dd the window was black only).

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #7 from Bruno Jesus  2013-12-06 19:51:39 CST ---
Thank you very much for the patch, I'm finally able to play StarCraft in my
intel graphics board now.

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #8 from josch  2013-12-07 01:42:11 CST ---
Bruno: sorry, forgot to mention: the patch is by Henri Verbeet so all thanks
goes to him for fixing this that quickly :)

For further reproducibility of this bug with the opengl renderer on intel
graphics cards I also just tried the starcraft demo from
http://ftp.blizzard.com/pub/starcraft/demos/SCDemo.exe (great times when demos
were 29M).

I can confirm that this bug does not only affect the full version of the game
but also the demo. At least on my intel system and with the opengl renderer.
With Henri Verbeets fix, it works perfectly with gdi.

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

Henri Verbeet  changed:

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

--- Comment #9 from Henri Verbeet  ---
The issue with the GDI ddraw renderer should be fixed by commits
38495706b79bc77759aca918ba60d9d226b072ee and
544e52bff40dd51a3f4f9954a6a32a75ff70ae07. As far as I can tell the issue with
the OpenGL renderer is in the Intel driver, so I'm resolving this UPSTREAM. You
should probably file a bug for that at bugs.freedesktop.org if you haven't
already, and if you like you can CC me on that.

[Bug 35032] starcraft unbearably slow in fullscreen mode

http://bugs.winehq.org/show_bug.cgi?id=35032

[e-mail] changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |programmer11180@programist.
                   |                            |ru

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #10 from Bruno Jesus  ---
(In reply to Henri Verbeet from comment #9)
> The issue with the GDI ddraw renderer should be fixed by commits
> 38495706b79bc77759aca918ba60d9d226b072ee and
> 544e52bff40dd51a3f4f9954a6a32a75ff70ae07. As far as I can tell the issue
> with the OpenGL renderer is in the Intel driver, so I'm resolving this
> UPSTREAM. You should probably file a bug for that at bugs.freedesktop.org if
> you haven't already, and if you like you can CC me on that.

Henri, can you give hints about what to write in the freedesktop report? I just
tested this and it's still an issue.

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #11 from Henri Verbeet  ---
(In reply to Bruno Jesus from comment #10)
> Henri, can you give hints about what to write in the freedesktop report? I
> just tested this and it's still an issue.

At this point, mostly just the usual things about how to reproduce it, like
hardware and software versions, etc., and perhaps a link to this bug report. It
would of course also be helpful if you could already do some profiling on your
own like e.g. the earlier mentioned
http://dri.freedesktop.org/wiki/IntelPerformanceTuning/, but what tends to make
profiling hard is that it's not always obvious how to interpret the results.

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #12 from Bruno Jesus  ---
Created attachment 50976
  --> https://bugs.winehq.org/attachment.cgi?id=50976
intel_gpu_top while running StarCraft

The attached screenshot shows what happens when running StarCraft in
intel_gpu_top. The screenshot was taken using a second screen while the first
screen was fullscreen inside the game.

The screenshot shows that all the time is spent in CS, which I guess is related
to "critical section" or "command stream"? I'm trying to understand the
results...

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #13 from Bruno Jesus  ---
Created attachment 50977
  --> https://bugs.winehq.org/attachment.cgi?id=50977
intel_gpu_top while running StarCraft in Virtual Desktop

This is just for comparison, when playing in virtual desktop the games runs
fine.

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

--- Comment #14 from Henri Verbeet  ---
(In reply to Bruno Jesus from comment #12)
> The screenshot shows that all the time is spent in CS, which I guess is
> related to "critical section" or "command stream"? I'm trying to understand
> the results...

My guess would be the command stream unit as in section 3.2.2 of
http://www.x.org/docs/intel/965/VOL_1_graphics_core.pdf, but that doesn't
necessarily make it more obvious what's going on.

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

Ken Sharp  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |[e-mail]

--- Comment #15 from Ken Sharp  ---
*** Bug 36831 has been marked as a duplicate of this bug. ***

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

Austin English  changed:

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

--- Comment #16 from Austin English  ---
Closing.

[Bug 35032] starcraft unbearably slow in fullscreen mode

https://bugs.winehq.org/show_bug.cgi?id=35032

Austin English  changed:

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

--- Comment #17 from Austin English  ---
This was inadvertently caught up in my unclosed bugs filter. NOTOURBUG should
only be closed when fixed upstream.

Setting back to RESOLVED NOTOURBUG.

Sorry for the spam.