[Bug 13122] New: Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Summary: Graphical regression in Team Fortress 2.
           Product: Wine
           Version: 0.9.61.
          Platform: PC-x86-64
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: trivial
          Priority: P5
         Component: directx-d3d
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
                CC: [e-mail]

When being killed, the camera zooms to the killer allowing the victim to press
F5 to save an image of the killer. In Wine-61 (and rc1) when the camera reaches
the killer, the entire screen goes black, although the HUD is still visible.

628e4eece3b9bc724f8873aa3c078f545da887cc is first bad commit
commit 628e4eece3b9bc724f8873aa3c078f545da887cc
Author: Roderick Colenbrander 
Date:   Mon Apr 28 21:44:21 2008 +0000

wined3d: Add multisampling support.

:040000 040000 7864ad7f4c0aeaa2235255e8d9065edb796f70d2
0b014ba64337d75aba8298596be87e6c9d3066b3 M      dlls
:040000 040000 d5eaaeeaa515f68aa03ecf10c3be9f2fbfa6c182
5e8b0f59abb7bec2c8af2c010b673bfb4b1b9801 M      include

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #1 from Vyacheslav Goltser   2008-05-10 21:29:57 ---
Created an attachment (id=12906)
 --> (http://bugs.winehq.org/attachment.cgi?id=12906)
screenshot showing the glitch in rc1

The screenshot is what the glitch looks like in RC1 (0.9.61 is the same).

wine-0.9.60 does not experience this.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #2 from Vyacheslav Goltser   2008-05-13 00:43:18 ---
The app maintainer has said that TF2 works fine on his system, is it possible
that me being on a 64bit system could affect this?

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #3 from Roderick Colenbrander   2008-05-13 01:13:05 ---
Most likely not. The videocard + drivers are more important. Which are you
using?

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #4 from Vyacheslav Goltser   2008-05-13 01:34:26 ---
nvidia 8800GT 640MB, driver is 169.12 (same as app maintainer).

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #5 from Vyacheslav Goltser   2008-05-13 01:37:16 ---
I do have something interesting to report (just realized it now).

If I go into the menu to select load outs for characters, when I die, the black
screen is displayed, but in the upper left corner, an image of the last viewed
class which I looked at in the load out appears. (it's as if there is a mixup
on which image to display or something of the sort).

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Paul Johnson  changed:

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

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #6 from Paul Johnson   2008-05-14 02:10:17 ---
Created an attachment (id=13043)
 --> (http://bugs.winehq.org/attachment.cgi?id=13043)
Example deathcam shot exhibiting the bug.

The screen goes black, and shows random, oversize text.  Sometimes the text
changes size, though the text is different every time.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #7 from Paul Johnson   2008-05-14 02:10:28 ---
Created an attachment (id=13044)
 --> (http://bugs.winehq.org/attachment.cgi?id=13044)
Example deathcam shot exhibiting the bug.

The screen goes black, and shows random, oversize text.  Sometimes the text
changes size, though the text is different every time.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #8 from Paul Johnson   2008-05-14 02:10:29 ---
Created an attachment (id=13045)
 --> (http://bugs.winehq.org/attachment.cgi?id=13045)
Example deathcam shot exhibiting the bug.

The screen goes black, and shows random, oversize text.  Sometimes the text
changes size, though the text is different every time.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #9 from Paul Johnson   2008-05-14 02:18:55 ---
I'm getting it, too, though in a slightly different form on the deathcam shots.
 This bug presents a severe disadvantage to wine users in Team Fortress 2 as it
prevents people from getting a good look at how they were killed.  I have
attached a few screenshots showing example output of this bug.

I noticed this in 0.9.61 and RC1, prior to that, I did not play TF2.

Currently using RC1, using an nVidia FX5500 256MB, driver is legacy 96.43.05.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Paul Johnson  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1

--- Comment #10 from Paul Johnson   2008-05-19 22:23:23 ---
*** This bug has been confirmed by popular vote. ***

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #11 from Roderick Colenbrander   2008-05-20 11:02:56 ---
Check if it works in the latest GIT version as I have and others have added
some fixes to it which could affect this bug.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Vitaliy Margolen  changed:

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

--- Comment #12 from Vitaliy Margolen   2008-05-21 08:34:21 ---
*** Bug 13334 has been marked as a duplicate of this bug. ***

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #13 from Robert M. Muncrief   2008-05-21 08:49:49 ---
(In reply to comment #12)
> *** Bug 13334 has been marked as a duplicate of this bug. ***

I just filed bug 13334 but it was immediately marked as a duplicate of this
one. Bisect did identify the same commit, but I wanted to be sure the developer
was aware that my bug was about the full Steam Half-Life 2 and HL2 Episode One
games, and the Demo call of Duty 2, and the bug presents itself a little
differently.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Stefan Dösinger  changed:

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

--- Comment #14 from Stefan Dösinger   2008-05-21 09:48:35 ---
It's possible that the framebuffer readback fails with multisampled
framebuffers. Did you enable multisampling(antialiasing) in the game's
settings? Can you try to disable it? Are you using fbos, or the default
offscreen rendering method? I guess we're not enabling multisampling for fbos,
so glBlitFramebuffer might fail.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #15 from Robert M. Muncrief   2008-05-21 10:05:18 ---
(In reply to comment #14)
> It's possible that the framebuffer readback fails with multisampled
> framebuffers. Did you enable multisampling(antialiasing) in the game's
> settings? Can you try to disable it? Are you using fbos, or the default
> offscreen rendering method? I guess we're not enabling multisampling for fbos,
> so glBlitFramebuffer might fail.

Yes, the default settings for the games use antialiasing, and I'm using fbo's.
They have to be used because not using them causes games run at a crawl (2 to
13 FPS instead of 20 to 180 FPS for Half Life games). I should have specified
my video settings, but I've always thought opengl with fbo was the default for
all games. That what I've always used, and until this regression error it
always worked great.

I've been messing around with a lot of configurations and trying to get all my
regular suite of apps and games to work so I'm too tired to do another
recompile now. But I'll pull the latest git tomorrow and see if turning off all
anti aliasing works around the problem, but I would still think its something
that needs to be fixed because it can dramatically improve graphics quality,
and like I said always worked before this regression error.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #16 from Paul Johnson   2008-05-21 14:03:48 ---
I've had antialiasing turned off the entire time (since I know enabling them
destroys framerates like nothing else on an FX5500).  I've tried both FBO (due
to another game) and the default (whatever that is) with no discernable change
whatsoever.  What's a GIT version, and what would it's package name be in
Debian Sid?

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #17 from Austin English   2008-05-21 14:33:00 ---
(In reply to comment #16)
> I've had antialiasing turned off the entire time (since I know enabling them
> destroys framerates like nothing else on an FX5500).  I've tried both FBO (due
> to another game) and the default (whatever that is) with no discernable change
> whatsoever.  What's a GIT version, and what would it's package name be in
> Debian Sid?

You'll have to compile it youself from git (sort of like CVS). See:
http://wiki.winehq.org/GitWine

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #18 from Robert M. Muncrief   2008-05-21 22:46:26 ---
OK, I did a reinstall of git as of about 3:00PM today and the problem with
blank videos in the games I reported is indeed the lack of anti-aliasing
support when using fbo.

If I turned anti-aliasing completely off in Half-Life 2, Episode One, and Call
of Duty 2 all the previously blank video portions played fine.

I was then curious to see the actual effect of anti-aliasing on the game suite
I'm running so I booted up my standby XP installation and ran them. The lack of
anti-aliasing in Wine was then quite apparent. Without anti-aliasing games run
under Wine have a pronounced "shimmering" effect when moving, and when
stationary the edges of objects are noticeably jagged.

I hope that anti-aliasing can be implemented when using fbo, because without
fbo games simply run to slow to be usable.

In the least case however, something should be done so that if a game requests
the use of anti-aliasing that it be ignored. I noticed that some games, like
Far Cry 2, seem to detect the lack of anti-aliasing support and don't try to
use it. Most games seem to think there is some level of anti-aliasing support
though, and this can certainly cause problems for users that expect a game to
run with automatically detected settings.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #19 from Vyacheslav Goltser   2008-05-22 01:23:44 ---
just updated my git tree (git fetch; git rebase origin)

wine-1.0-rc1-182-g9f76085

bug still present. tried turning MSAA off, but that did not help.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #20 from Roderick Colenbrander   2008-05-22 01:27:35 ---
It doesn't proof it has to do with FBOs yet. You should set the
OffscreenRenderingMethod to backbuffer and if it works in that mode with
antialiasing then it is a FBO problem for sure.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #21 from H. Verbeet   2008-05-22 04:03:06 ---
As a random guess, I'd say stretch_rect_fbo() doesn't handle blitting to/from
multisampled buffers.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #22 from Robert M. Muncrief   2008-05-22 05:08:20 ---
(In reply to comment #20)
> It doesn't proof it has to do with FBOs yet. You should set the
> OffscreenRenderingMethod to backbuffer and if it works in that mode with
> antialiasing then it is a FBO problem for sure.

I just spent awhile running the Half Life and Call of Duty 2 games and verified
that anti-aliasing works with OffscreenRenderingMode set to pbuffer or
backbuffer.  I just let wine pick whatever RenderTargetLockMode it wanted.

OpenGL games actually only seem to suffer around a 20% to 40% drop in FPS with
backbuffer and anti-aliasing set to 2x, so Half Life and COD are playable, and
the quality of the AA looks quite good. However, with AA any higher the FPS go
down very quickly. DirectX games were not playable in backbuffer mode at all
though. I literally almost ate a whole bag of microwave popcorn waiting for
Need for Speed Pro Street to get to the first selection screen.

And neither OpenGL or DirectX games run very well with pbuffer, although they
do work and you can see the beneficial effects of AA.

So there definitely appears to be a problem with AA and FOB, at least with the
games I'm using.

I just want to say again though, the quality of the AA was very good, just as
good as under Windows to these untrained eyes. If it worked with FOB it would
be awesome!

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #23 from H. Verbeet   2008-05-22 05:57:32 ---
Actually... the issue could also be doing eg. ReadPixels or CopyTexImage on a
multisampled FBO. In principle these should always be redundant when FBOs are
used, but I'm not completely sure they never happen. Are there any GL errors in
your console output?

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #24 from Stefan Dösinger   2008-05-22 06:01:14 ---
Commenting out GL_EXT_framebuffer_blit from the extension table in direct.c
could show if it's a problem with fb_blit or not.

I guess the glCopyTexImage2D with offscreenrenderingmode != fbo causes the slow
performance with multisampling.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #25 from Robert M. Muncrief   2008-05-22 19:41:32 ---
Created an attachment (id=13267)
 --> (http://bugs.winehq.org/attachment.cgi?id=13267)
Half Life 2/Wine Bad Video Test

OK, I ran a semi-formal test with Half-Life 2 using both "fbo" and "backbuffer"
and attached the test setup and console output.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #26 from Robert M. Muncrief   2008-05-22 19:52:27 ---
(In reply to comment #24)
> Commenting out GL_EXT_framebuffer_blit from the extension table in direct.c
> could show if it's a problem with fb_blit or not.
>
> I guess the glCopyTexImage2D with offscreenrenderingmode != fbo causes the slow
> performance with multisampling.

I'm not familiar with Wine architecture, but I'd be happy to run the test if
you give me a little more instruction. After commenting out
GL_EXT_framebuffer_blit what should I expect to see/not see? What tests would
you like to be run?

I'm also not familiar with glCopyTexImage2D, but if there's a problem with data
transfer or manipulation and there's at least some semi-defined I/O and
architectural documentation I could try fixing it. I have a few decades of
experience as a digital hardware/firmware/software designer, but as I said I
know nothing about the internals of Wine.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #27 from H. Verbeet   2008-05-23 06:26:43 ---
> phalynx@entropod:~$ env WINEPREFIX="/home/phalynx/.wine" WINEDEBUG="fixme-all" wine "C:\Program Files\Steam\Steam.exe" -applaunch 220 -heapsize 1024000

You'll have to remove the WINEDEBUG="fixme-all" part, or GL errors won't show.
If EXT_framebuffer_blit is the problem, commenting out the extension should fix
the blank video.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #28 from Robert M. Muncrief   2008-05-23 22:02:59 ---
Created an attachment (id=13291)
 --> (http://bugs.winehq.org/attachment.cgi?id=13291)
Half-Life 2 Blank Video GL_EXT_framebuffer_blit test

Nice call gentlemen. GL_EXT_framebuffer_blit is indeed the problem. I've
attached a test report with the details and console output. Please let me know
if you need anything else.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Robert M. Muncrief  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #13291|0                           |1
        is obsolete|                            |

--- Comment #29 from Robert M. Muncrief   2008-05-23 22:08:13 ---
Created an attachment (id=13292)
 --> (http://bugs.winehq.org/attachment.cgi?id=13292)
Half-Life 2 Blank Video GL_EXT_framebuffer_blit test

Aaarg, sorry people. I keep forgetting to put hard CR/LF's in my reports.
Here's the same report with CR/LF's.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Phil Costin  changed:

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

--- Comment #30 from Phil Costin   2008-05-24 13:59:09 ---
Garry's Mod is not visible after launching it and on launching Counter-Strike:
Source, the background is visible but not the main menu

Used bisect to get to this.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #31 from Stefan Dösinger   2008-05-24 14:33:03 ---
philcostin: Try to disable HDR - that may get you around the absolutely not
visible problem.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #32 from Vyacheslav Goltser   2008-06-20 21:40:24 ---
Using Wine 1.0 (compiled it myself) in Ubuntu Hardy 8.04 (amd64) I am not
longer experiencing this regression (everything works). As far as I am
concerned, this bug can be closed.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Dmitry Timoshkov  changed:

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

--- Comment #33 from Dmitry Timoshkov   2008-06-20 22:46:23 ---
Reported fixed.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

Alexandre Julliard  changed:

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

--- Comment #34 from Alexandre Julliard   2008-06-27 10:15:33 ---
Closing bugs fixed in 1.1.0.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #35 from Paul Johnson   2008-07-20 18:11:27 ---
I still see this in 1.1.1, please reopen.

[Bug 13122] Graphical regression in Team Fortress 2.

http://bugs.winehq.org/show_bug.cgi?id=13122

--- Comment #36 from Stefan Dösinger   2008-07-20 19:01:20 ---
Paul Johnson, there's a different problem with up to dx9 radeon cards(X1x00 and
earlier), as well as GeForce FX and earlier cards that should be fixed in
current git