[Bug 12923] New: GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Summary: GTR2 graphics are seen through or invisible
           Product: Wine
           Version: 0.9.61.
          Platform: PC-x86-64
        OS/Version: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: directx-d3d
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]

4428cc32de1f99463ebaa990726c8bd7a3eef6c4 is first bad commit
commit 4428cc32de1f99463ebaa990726c8bd7a3eef6c4
Author: Roderick Colenbrander 
Date:   Mon Apr 28 21:13:12 2008 +0000

wined3d: Fix a few small bugs in WineD3D_ChoosePixelFormat.

:040000 040000 6e6e5f451c4a9961328c12a57c3042b6b622067c
b5090b2050987a2273e1e20d55c91be1117c1c92 M      dlls

Ingame most of the models and textures are missing, this making the game
unplayable.
Probably also an issue in BMW M3 Challenge.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #1 from Gregor Münch   2008-05-03 08:21:45 ---
Created an attachment (id=12664)
 --> (http://bugs.winehq.org/attachment.cgi?id=12664)
image describing the bug

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Gregor Münch  changed:

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

--- Comment #2 from Gregor Münch   2008-05-03 08:23:25 ---
adding author of the patch...

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #3 from Gregor Münch   2008-05-03 08:28:54 ---
Created an attachment (id=12665)
 --> (http://bugs.winehq.org/attachment.cgi?id=12665)
+wgl

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #4 from Roderick Colenbrander   2008-05-03 09:47:04 ---
Create a log of WINEDEBUG=+d3d further what directx version does this game
require?

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #5 from Gregor Münch   2008-05-03 11:12:36 ---
The game is a directx9 one.
Interesting is, if I select AA (2x or 4x) in games option program before
gamestart everything renders correctly.
http://sharebase.de/files/gShQSmhT2t.html

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #6 from Roderick Colenbrander   2008-05-03 12:57:45 ---
You can just attach the log as an attachment to here. The other download site
doesn't work for me.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Richie  changed:

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

--- Comment #7 from Richie   2008-05-09 20:51:04 ---
This is for call of duty 4.  I was going to log a new bug but done a search on
this commit.   I hope it is ok to post here. I first noticed corrupt graphics
in COD4 at wine-0.9.61 and days later I tried the latest git as of May 8,
8:00pm EST and problem was still there.  I then did a full regression from .60
to .61 and came up with the following commit:

$ git bisect good
4428cc32de1f99463ebaa990726c8bd7a3eef6c4 is first bad commit
commit 4428cc32de1f99463ebaa990726c8bd7a3eef6c4
Author: Roderick Colenbrander 
Date:   Mon Apr 28 21:13:12 2008 +0000

wined3d: Fix a few small bugs in WineD3D_ChoosePixelFormat.

:040000 040000 6e6e5f451c4a9961328c12a57c3042b6b622067c
b5090b2050987a2273e1e20d55c91be1117c1c92 M      dlls

Notes: 
- both Nvida drivers versions 169.12 and 173.08 beta were tested, however for
the duration of the regression testing I remained on 169.12.  This not simply
confirms nothing changes between versions.
- nvidia 8800 GTS 512
- Additionaly, I tested today's wine-1.0-rc1 and confirmed that it still
exhibited this issue.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #8 from Richie   2008-05-09 20:52:47 ---
Created an attachment (id=12860)
 --> (http://bugs.winehq.org/attachment.cgi?id=12860)
demonstration of graphical corruption in COD4

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #9 from Lei Zhang   2008-05-09 21:04:40 ---
same problem as bug 13025?

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #10 from Richie   2008-05-10 00:39:01 ---
Created an attachment (id=12866)
 --> (http://bugs.winehq.org/attachment.cgi?id=12866)
patch for cached opengl pixel format table code loop

I have COD4 consistently displaying correct graphics on wine 1.0-rc1 now. No
other tests done.

Regarding the initial "(cached) opengl pixel format table code loop" within
dlls/wined3d/context.c... Without knowing the content/structure of the cfgs
data, among other things, my first thoughts was that if it is expected that we
use the *first* iPixelFormat found -when our depth requirements aren't met- (as
per the code comments) it seems logical that we would need to break at that
point.

However, it could simply be that by having added this loop break, the use of a
*higher* iPixelFormat value is now being avoided (which for whatever reason
could be causing the corruption issue).  Perhaps within that there is the
underlying issue?

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #11 from Roderick Colenbrander   2008-05-10 02:41:02 ---
The way I wrote the code is that we try to find a pixel format which exactly
matches our requirements. The pixel format table is sorted by the display
drivers and there is no real pattern in it and it is different for each driver.
Further down to the end of the list there tend to be 'extended' pixel formats.
Those formats have the same basic properties as the ones at the top of the list
but different extended features like multisampling and others.

Due to the sorting of the list we store the first format which reaches the end
of the loop (because that format matches out basis requirements). We then
continue to look for a format for which depth also matches exactly. If that one
is found we break else we use the format which first met our requirements.

It would be useful to see what the requirements specified to
WineD3D_ChoosePixelFormat are (what parameters are passed to it). Likely
depth/stencil isn't requested. If that is the case we need to know why it
wasn't requested.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #12 from Richie   2008-05-10 11:14:30 ---
Ok I understand now that it would only define it, *if its undefined*, which
would imply the first match where the depth requirements aren't met. So it
seems more likey that I'm now avoiding a particular *depth matching* format by
having added that break.
--
Roderick, please give specifics on how you want me to execute tests in wine to
gather the information that you need.  If you want to provide a customised file
for this case then I will rebuild against that.  Let me know.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #13 from Roderick Colenbrander   2008-05-10 11:25:09 ---
Actually I think that bug 12923 is a duplicate of this one. Let me quickly
explain the problem.

At some stage CreateContext is called to create an OpenGL context. It uses info
passed to it at D3D device creation time to WineD3D_ChoosePixelFormat to choose
a pixel format. WineD3D_ChoosePixelFormat works fine.

The issue is I think that the program manually attaches a 'depth stencil
surface' using SetDepthStencilSurface. Check if that is called. The call should
recreate an OpenGL context using the proper pixel format. Right now
SetDepthStencilSurface is not doing this at all. The reason is just a leftover
of the old pixel format limitation. When it was around the only format we had
always had depth and stencil.

So check if SetDepthStencilSurface is used to manually add a depth/stencil
buffer.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #14 from Richie   2008-05-12 21:05:34 ---
Roderick, I can't get the game to load up completely ( screen goes black and
first logo video doesn't start) when trying to run cod4 with WINEDEBUG=+d3d.  I
hope that someone else who reads this and has the same issue can upload the
trace data.  I'd really like to see this get fixed!

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #15 from Roderick Colenbrander   2008-05-17 11:55:52 ---
Created an attachment (id=13124)
 --> (http://bugs.winehq.org/attachment.cgi?id=13124)
Hacky patch

This is a first attempt to a patch to fix the issue. It is far from clean but
give it a try.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Roderick Colenbrander  changed:

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

--- Comment #16 from Roderick Colenbrander   2008-05-17 11:56:45 ---
*** Bug 13025 has been marked as a duplicate of this bug. ***

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

haarp  changed:

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

--- Comment #17 from haarp   2008-05-17 12:29:58 ---
For Supreme Commander, this patch results in a huge performance hit and severe
graphical corruption, even in the intro videos and main menu. Console spits
stuff like this out (Offscreenrenderer=fbo):

err:d3d:IWineD3DDeviceImpl_SetDepthStencilSurface depthstencil change:
WINED3DFMT_D24S8
err:d3d:IWineD3DDeviceImpl_SetDepthStencilSurface depthstencil change:
WINED3DFMT_D24S8
err:d3d_surface:IWineD3DSurfaceImpl_LoadLocation Surface does not have any up
to date location
fixme:d3d:check_fbo_status FBO status GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT_EXT
(0x8cd6)
fixme:d3d:check_fbo_status      Color attachment 0: (0x1c4ebe88)
WINED3DFMT_A8R8G8B8 1920x1200
fixme:d3d:check_fbo_status      Depth attachment: (0x1c4ebfd0) WINED3DFMT_D24S8
1920x1200
fixme:d3d:IWineD3DDeviceImpl_ClearSurface >>>>>>>>>>>>>>>>>
GL_INVALID_FRAMEBUFFER_OPERATION_EXT (0x506) from glClear @ device.c / 5080
fixme:d3d:ActivateContext >>>>>>>>>>>>>>>>> GL_INVALID_OPERATION (0x502) from
glDrawBuffer @ context.c / 1040

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #18 from Roderick Colenbrander   2008-05-17 14:54:03 ---
Likely a lot of unneeded context recreation is done now. Further I'm not
restoring the contents of the backbuffer.

In short SetDepthStencilSurface is something which we can't easily do in
opengl. In OpenGL you select a pixel format for your drawable (a window, a
pixmap or a pbuffer) and this pixel format describes the RGB colors but it also
fixes the depth and stencil buffer. In d3d RGB colors aren't tied to depth
stencil, you can just attach a depth buffer after you create your 'drawable'.
What we must do in case of opengl is to recreate the window and opengl context.
This is very tricky stuff and I'm now doing it quickly using a reset command.
(but I'm not restoring the backbuffer yet)

Let me create a version which at least won't do a reset when it isn't needed.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #19 from Roderick Colenbrander   2008-05-17 14:58:35 ---
Created an attachment (id=13130)
 --> (http://bugs.winehq.org/attachment.cgi?id=13130)
Attempt 2

This is a quickly updated patch. It only resets the device when no depth
stencil format was previously set. (I still need to check for format 'upgrades'
though)

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #20 from Richie   2008-05-17 15:28:34 ---
Created an attachment (id=13131)
 --> (http://bugs.winehq.org/attachment.cgi?id=13131)
hacky_patch test results

With your "Hacky patch", COD4 did not show any graphical corruption or
noticable performance problems.  My initial tests show that it is working,
however this is with Shadows OFF.  The game hard locks when turning Shadows ON
(during game or after launch from inital options screen).  While we can't
expect a performance hit when turning Shadows ON, we shouldn't expect a hard
lock so I will upload output.
--
I will follup up with results on the Attempt 2 patch.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #21 from Richie   2008-05-17 15:57:30 ---
excerpt from my previous post should have read "While we *could*
expect a performance hit when turning Shadows ON, we shouldn't expect a hard
lock so I will upload output."
--
--
COD4: The Attempt 2 patch worked with both Shadows ON or OFF and I received the
typical performance hit that I have in the past when turning Shadows ON.  Other
settings were (have remained) Specular Map YES, Depth of Field NO, Glow NO,
Number of Dynamic lights OFF, Soften Smoke Edges NO
--
I didn't do tests for these other options being ON, but near basic settings
leads to a working game.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #22 from haarp   2008-05-17 16:46:48 ---
Created an attachment (id=13133)
 --> (http://bugs.winehq.org/attachment.cgi?id=13133)
Pic describing problems

Ok, the 2nd patch makes it actually possible to use Supreme Commander. Although
it appears it introduces more problems than it solves.
> From the pic (left good, right with patch):

Top: Lightning seems to break
Middle: Ground textures become "blocky" (ignore the black bar, this was a bug
in my graphics software)
Bottom: More texture problems.

Also, using this patch makes the game slower than not using it. Additionally,
all fonts disappear in the main menu. :/

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #23 from Roderick Colenbrander   2008-05-18 02:51:43 ---
What sort of errrors/messages are you seeing in the console? As I mentioned we
are recreating everything and that's why this patch is so tricky.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #24 from haarp   2008-05-18 04:36:06 ---
fixme:d3d:check_fbo_status FBO status GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT_EXT
(0x8cd6)
fixme:d3d:check_fbo_status      Color attachment 0: (0x1c4eb830)
WINED3DFMT_A8R8G8B8 1920x1200
fixme:d3d:check_fbo_status      Depth attachment: (0x1c4ebfd0) WINED3DFMT_D24S8
1920x1200
fixme:d3d_draw:drawStridedFast >>>>>>>>>>>>>>>>>
GL_INVALID_FRAMEBUFFER_OPERATION_EXT (0x506) from glDrawArrays @ drawprim.c /
279
err:d3d_surface:IWineD3DSurfaceImpl_LoadLocation Surface does not have any up
to date location

whereas the fixmes get spammed while the last err only pops up "occasionally"

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #25 from Roderick Colenbrander   2008-05-18 11:03:45 ---
I'm about to submit a workaround for the problem to wine-patches. The real fix
is a lot of work and way to risky during the 1.0 code freeze.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #26 from haarp   2008-05-18 11:10:41 ---
Created an attachment (id=13158)
 --> (http://bugs.winehq.org/attachment.cgi?id=13158)
Stalker broken

Hopefully it's not one of the patches in this bug. Because I just noticed
another game (Stalker) whose graphics completely screw up when the patch is
used ;) again with lightning issues and some other stuff

Also, every game seems to slow down to a crawl...

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #27 from Richie   2008-05-18 12:35:20 ---
haarp, ooc does Supreme Commander and Stalker work correctly on 0.9.60 ?  The
core commit listed here at the beginning of the bug occured about midway
between .60 and .61.  I'm using .60 as an initial reference point as to what to
expect verses what the current head +/- any patches here produce.

Roderick, with the WineD3D_ChoosePixelFormat changes that you've recently made
and with this new exposed bug, is it actually more benefical to revert those
commits and have the pre-existing functionality restored as opposed to now
leading to a work around to avoid breaking 1.0 release canidates?  I don't know
what things were fixed by those WineD3D_ChoosePixelFormat updates, but I'm
curious if you've weighed the options in this respect.  I respect that bugs
need to be fixed as those commits begin to address, but perhaps this is
something that could be re-introduced, and with more fixes, and after 1.0?  
Just a thought.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #28 from haarp   2008-05-18 12:43:17 ---
(In reply to comment #27)
> haarp, ooc does Supreme Commander and Stalker work correctly on 0.9.60 ?  The
> core commit listed here at the beginning of the bug occured about midway
> between .60 and .61.  I'm using .60 as an initial reference point as to what to
> expect verses what the current head +/- any patches here produce.

I am using 1.0-rc1 as base to apply Roderick's patches. In any way, this bug
got introduced between 0.9.60 and 61. SupCom works fine in 0.9.60, Stalker
exhibits bug 12794. Luckily, I had a Stalker savegame that is not affected by
this bug with which I tested it.
(Interestingly, Stalker's bug 12974 exhibits similiar terminal messages as this
bug here, but it was a different commit that broke it)

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #29 from haarp   2008-05-18 12:45:23 ---
Oops, meant to say bug 12794 in the second link..

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #30 from Roderick Colenbrander   2008-05-20 06:39:01 ---
The workaround is in GIT.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #31 from haarp   2008-05-20 09:29:31 ---
By workaround, do you mean reverting the patch that caused the problem or a
proper patch?
Because as far as I'm concerned, this problem is fixed in SupCom.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #32 from Roderick Colenbrander   2008-05-20 09:45:59 ---
As I explained the issue is basically that D3D programs have to decide either
when they initialize D3D or add a later stage whether they want to use a depth
and stencil buffer. In case of OpenGL you need to request depth and stencil
support roughly at the moment you initialize OpenGL. In order to support the
'delayed' depth / stencil usage which D3D allows we need to 'reinitialize'
OpenGL and that is tricky (we need to recreate all textures, shaders, ..). This
is too risky to do during the 1.0 codefreeze. Stefan and I thought it would be
easy but we missed some things.

The workaround just hard codes the use of a depth stencil format at OpenGL
initialization time and that way works around the issue.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Gregor Münch  changed:

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

--- Comment #33 from Gregor Münch   2008-05-20 14:10:02 ---
GTR2 works again. Thx.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #34 from Jonathan   2008-05-21 06:47:42 ---
Robot Wars is also fixed. Thanks.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #35 from haarp   2008-05-21 08:44:17 ---
I wouldn't close this bug yet as the proper fix isn't implemented yet.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Roderick Colenbrander  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|FIXED                       |REMIND

--- Comment #36 from Roderick Colenbrander   2008-05-21 09:44:46 ---
I'm leaving it at REMIND.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

--- Comment #37 from Dmitry Timoshkov   2008-05-21 09:47:55 ---
REMIND is a bad choice. I'd argue that we shouldn't use this resolution
for Wine bugs as all, it's meaningless.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Dmitry Timoshkov  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|REMIND                      |ABANDONED

--- Comment #38 from Dmitry Timoshkov   2008-06-09 01:57:51 ---
Changing the resolution to abandoned. If the problem persists please open
a separate bug report.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Dmitry Timoshkov  changed:

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

--- Comment #39 from Dmitry Timoshkov   2008-06-09 01:58:01 ---
Closing.

[Bug 12923] GTR2 graphics are seen through or invisible

http://bugs.winehq.org/show_bug.cgi?id=12923

Gregor Münch  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|ABANDONED                   |FIXED

--- Comment #40 from Gregor Münch   2008-06-09 09:14:18 ---
The issue was simply fixed. Even if its not the _fully_ correct approach. There
are no impacts for the user. Its now just basically like it was before.