[Bug 33318] New: 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

Bug #: 33318
           Summary: 3DMark2001 SE: upgrading to Wine 1.5.27 reduces
                    performance score by 20%
           Product: Wine
           Version: 1.5.27
          Platform: x86-64
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: -unknown
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
    Classification: Unclassified

Created attachment 44068
  --> http://bugs.winehq.org/attachment.cgi?id=44068
Screenshots with example scores for Wine 1.5.27 and Wine 1.4.1

I'm running Linux Mint 13 x86_64 with kernel 3.5.0-26-generic and X.Org Server
1.13.0 on my laptop (ThinkPad X61s with Intel X3100 graphics, see [1]). Before
upgrading Wine from 1.4.1 version to version 1.5.27 using the Ubuntu Wine PPA
[2] I decided to run 3DMark to see if the upgrade affected performance. I then
ran 3DMark after upgrading Wine. The result was an unexpected, consistent 20%
drop in the score.

The scores (averaged from three trials) were as follows:
Wine 1.4.1: 2614
Wine 1.5.27: 2080

I'll see if this can be reproduced on a machine with a different graphics chip.

[1] http://www.thinkwiki.org/wiki/Category:X61s
[2] http://www.ubuntuupdates.org/ppa/wine?dist=precise

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

Danyil Bohdan  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #44068|0                           |1
        is obsolete|                            |

--- Comment #1 from Danyil Bohdan  2013-04-02 13:15:41 CDT ---
Created attachment 44069
  --> http://bugs.winehq.org/attachment.cgi?id=44069
Screenshots with (correct) example scores for Wine 1.5.27 and Wine 1.4.1

Oh snap, I swapped 1.5.27 and 1.4.1 windows in the previous image. Again,
1.5.27 actually gets a _lower_ score. Sorry.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |download, performance,
                   |                            |regression
          Component|-unknown                    |directx-d3d

--- Comment #2 from Austin English  2013-04-02 16:52:48 CDT ---
I suppose you could run a regression test, though I doubt one single patch
affected things that badly..

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

André H.  changed:

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

--- Comment #3 from André H.  2013-04-02 19:32:49 CDT ---
Adding Stefan as he does gaming benchmarks on Wine

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #4 from Stefan Dösinger  2013-04-02 23:58:57 CDT ---
There have been some ups and downs over the years, but I think the major
performance regressions should be fixed.

I'm afraid I'll have to ask you to run a regression test over all those Wine
versions.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #5 from Danyil Bohdan  2013-04-03 11:44:38 CDT ---
I ran 3DMark2001 SE on a work machine today, a desktop PC with an Nvidia
GeForce GTS 250 graphics card running Linux Mint 13 LTS with kernel
3.2.0-39-generic, X.Org Server 1.11.3 and nouveau drivers (reporting as
"Gallium 0.4 on NVCF"). The difference in performance is considerably smaller
but it is still there. Due to the lack of time I was only able to perform a
single trial for each version.

The scores were as follows:

Wine 1.4.1: 10664
Wine 1.5.27: 9375

I'm not sure when and if I'll be able to test each patch version for this
regression, though.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #6 from Danyil Bohdan  2013-04-03 11:47:26 CDT ---
Created attachment 44077
  --> http://bugs.winehq.org/attachment.cgi?id=44077
GeForce GTS 250/nouveau scores for Wine 1.5.27 and Wine 1.4.1

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #7 from Austin English  2013-04-03 19:52:04 CDT ---
With nvidia binary drivers and a GTX 480, the difference between 1.4 and 1.527
was 1%.

Seems to be mesa.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #8 from Stefan Dösinger  2013-04-04 01:17:18 CDT ---
I have 3 systems performing automated performance tracking on a Radeon X1600
(r300g), Radeon HD 5770 (r600g) and Geforce 7600 (nvidia binary). Data is
available here:

http://openbenchmarking.org/result/1304031-STEF-WINE15222&track_view=400
(r300g)
http://openbenchmarking.org/result/1304037-STEF-WINE15299&track_view=400
(r600g)
http://openbenchmarking.org/result/1304031-FO-WINE1524915&track_view=400
(nvidia, new)
http://openbenchmarking.org/result/1302087-FO-WINE1523835&track_view=400

You have to read the results right to left, and the graphs are a bit of a mess.
r300g spotted a small performance drop in 3DMark2k1 around Wine 1.4. r600g was
switched from fglrx to r600g at some point, screwing up performance
comparisons. Furthermore the r600g results are unreadable because the website
is confused - download the raw XML and put it into Libreoffice or gnuplot or
something. The nvidia box was upgraded from some ancient ubuntu to a modern
arch a few test revisions back, which broke the link between the old and new
results and caused a ~5% performance drop. Unfortunately the old results are
unreadable as well because of one misguided test result :-( . I really have to
nag Michael Larabel about that...

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #9 from Danyil Bohdan  2013-04-04 04:17:23 CDT ---
Created attachment 44086
  --> http://bugs.winehq.org/attachment.cgi?id=44086
Drop in Radeon X1600 (r300g) score in version 1.5.4

Stefan, I found a big drop in your benchmark results for r300g between Wine
1.5.3 and Wine 1.5.4. The 3DMark2001 score goes from 5445 down to 4654. Unlike
the score in 3DMark2000, which drops from 6903 to 2247 in the same release but
then rises to 8225 in 1.5.19, it never seems to recover. See the attached
image.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #10 from Stefan Dösinger  2013-04-04 05:30:20 CDT ---
Oh right, this just looks so small because of the big one-time spike. I really
wish this website had a way to remove broken results...

Once upon a time I bisected the 3Dmark2000 regression and ended up with
a8a201cf61b39f79be980c4f193905bf8c0276c4. This patch shouldn't have an effect
on 3DMark 2001 because the drop in 2000 was a bug in ddraw.dll, which was fixed
by a61f68fc5ab67e9c08665bd1861d6811be42d30f.

Nevertheless, you could try to disable GL_ARB_map_buffer_range in directx.c
(just remove it from the extension table at the top). If that helps, we have a
pointer in the right direction.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #11 from Rico  2013-04-04 08:38:21 CDT ---
I got the following with r600g (HD 4350 - RV710):
29893d81621b50e3acca1f2c34767867a9256ecb^ (wine 1.5.3 fast): 5400
29893d81621b50e3acca1f2c34767867a9256ecb (wine 1.5.4 slow): 4300
1.5.27: 4000

29893d81621b50e3acca1f2c34767867a9256ecb is the first bad commit
commit 29893d81621b50e3acca1f2c34767867a9256ecb
Author: Henri Verbeet 
Date:   Wed May 2 21:47:58 2012 +0200

wined3d: Determine index buffer location based on vertex buffer locations.

:040000 040000 fc5dd836116dec3047a41fe7b28c22d4ef4261c8
72a5445b837c0fa479e611f985543f319d528c0a M    dlls

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #12 from Henri Verbeet  2013-04-04 09:39:19 CDT ---
Is it drawing pre-transformed vertices a lot, or e.g. hitting drawStridedSlow()
for some other reason? The buffer_get_sysmem() in there is potentially slower
if the index buffer was moved to VRAM. In general I don't think 3DMark 2000 or
2001 or particularly interesting benchmarks for current hardware though.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

http://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #13 from Danyil Bohdan  2013-04-04 10:13:53 CDT ---
Commenting out GL_ARB_map_buffer_range in dlls/wined3d/directx.c improves the
average score on my X61s from 2080 to 2218. This is in the latest version (Wine
1.5.27-127-ga62b767, git commit a62b7679356abde12a0d4c86926d010099d1393b).

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
    Regression SHA1|                            |29893d81621b50e3acca1f2c347
                   |                            |67867a9256ecb

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #14 from Ken Sharp  ---
Any improvement in Wine 1.7.45 or later?

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

Stefan Dösinger  changed:

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

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

[e-mail] changed:

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

--- Comment #15 from [e-mail] ---
(In reply to Danyil Bohdan from comment #13)
> Commenting out GL_ARB_map_buffer_range in dlls/wined3d/directx.c improves
> the average score on my X61s from 2080 to 2218. This is in the latest
> version (Wine 1.5.27-127-ga62b767, git commit
> a62b7679356abde12a0d4c86926d010099d1393b).

The commit doesn't seem to revert anymore.

What do you actually comment out?

I found this from source

/* GL_ARB_map_buffer_range and GL_APPLE_flush_buffer_range provide the same
* functionality. Prefer the ARB extension */
gl_info->supported[APPLE_FLUSH_BUFFER_RANGE] = FALSE;

Is this actually fixed already?

I got a result of 25997

Got the 3dmark from here

https://www.techpowerup.com/downloads/285/futuremark-3dmark-2001-free-versio...

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #16 from Henri Verbeet  ---
I have patches that bring the score up to slightly above the one for 1.4.1 on
my hardware (AMD CYPRESS, Mesa 13.0.2). Unfortunately those won't make it into
the 2.0 release.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #17 from Henri Verbeet  ---
Is this still an issue with current git?

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

--- Comment #18 from [e-mail] ---
(In reply to Henri Verbeet from comment #17)
> Is this still an issue with current git?

I got a lot higher number now. 29478 compared to 25997.

Run the test only once now and then.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

Stefan Dösinger  changed:

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

--- Comment #19 from Stefan Dösinger  ---
I've benchmarked this app on my MacBookPro10,1 running Linux:

92671487050dfe0d377eb0248b9d458e7156e3ba (the commit before the regression):
2880x1800: 19109 3dmarks
640x480: 42464

Wine 2.0:
2880x1800: 17025 3dmarks
640x480: 37168

Wine 2.0-348 (sorry, forgot to write down the exact SHA1):
2880x1800: 19421 3dmarks
640x480: 40877

I'll call it fixed. I suspect afb2aa2512d4c5ffa7256ce5137bc2c142a5d065 fixed
it. I will confirm this and set the fix SHA1 once I have done so.

There is still a difference in the CPU limited case (42464 -> 40877 3dmarks).

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

Stefan Dösinger  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
      Fixed by SHA1|                            |afb2aa2512d4c5ffa7256ce5137
                   |                            |bc2c142a5d065

--- Comment #20 from Stefan Dösinger  ---
I can confirm that afb2aa2512d4c5ffa7256ce5137bc2c142a5d065 fixed this.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

Alexandre Julliard  changed:

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

--- Comment #21 from Alexandre Julliard  ---
Closing bugs fixed in 2.2.

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

Michael Stefaniuc  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |2.0.x

[Bug 33318] 3DMark2001 SE: upgrading to Wine 1.5.27 reduces performance score by 20%

https://bugs.winehq.org/show_bug.cgi?id=33318

Michael Stefaniuc  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|2.0.x                       |---

--- Comment #22 from Michael Stefaniuc  ---
Removing the 2.0.x milestone from bugs included in 2.0.5.