[Bug 17437] ATI and EVE Premium: no models display / crashes on startup

http://bugs.winehq.org/show_bug.cgi?id=17437

Florian Sievert  changed:

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

Willem Brosz  changed:

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

rig0r  changed:

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

Dan Merillat  changed:

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

Vitaliy Margolen  changed:

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

hpestilence  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #20746|0                           |1
        is obsolete|                            |

Mads  changed:

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

Joachim Rousseau  changed:

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

hpestilence  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #20974|0                           |1
        is obsolete|                            |
  Attachment #20998|0                           |1
        is obsolete|                            |

--- Comment #57 from Matthieu Bertin   2009-04-09 03:54:14 ---
how do you remove the compiled version of wine then ? before i compile it
again...

--- Comment #58 from Eddy Y.   2009-04-09 04:19:17 ---
(In reply to comment #57)
> how do you remove the compiled version of wine then ? before i compile it
> again...

You just overwrite the current version with the one you will be compiling.

First, patch your source code, making sure that you have the corresponding code
version and patch (ie. do not apply a patch meant for wine 1.1.7 to wine
1.1.6).

Second, check where your existing wine is with 'whereis wine'.

If it's in /usr/bin, then use

$ ./configure --prefix=/usr

Otherwise, do

$ ./configure

Then,

$ make
$ sudo make install

--- Comment #59 from Matthieu Bertin   2009-04-09 17:47:25 ---
Here it is. I did what you said, and it still doesn't work.
I downloaded 1.1.18 from sourceforge and i am using latest proprietary drivers
available on synaptic.

here is a short version:

brzhk@gauss:~/winecomp/wine-1.1.18$ patch -p1 < apocrypha.shaders.1.1.17.patch 
patching file dlls/wined3d/baseshader.c
patching file dlls/wined3d/directx.c
patching file dlls/wined3d/wined3d_main.c
patching file dlls/wined3d/wined3d_private.h
Hunk #1 succeeded at 290 (offset -8 lines).
brzhk@gauss:~/winecomp/wine-1.1.18$ ./configure 
(conf. trace)
brzhk@gauss:~/winecomp/wine-1.1.18$ make depend && make
(traces)
brzhk@gauss:~/winecomp/wine-1.1.18$ sudo make install

i also tried with --prefix=/usr/local as it was the path shown by whereis wine.

--- Comment #60 from Eddy Y.   2009-04-10 03:13:49 ---
(In reply to comment #59)
> Here it is. I did what you said, and it still doesn't work.
> I downloaded 1.1.18 from sourceforge and i am using latest proprietary drivers
> available on synaptic.

I use the latest drivers from ATI. Could you try with that please? Also, do you
have the user.reg settings specified earlier in the thread?

See http://bugs.winehq.org/show_bug.cgi?id=17437#c23
> brzhk@gauss:~/winecomp/wine-1.1.18$ sudo make install
>
> i also tried with --prefix=/usr/local as it was the path shown by whereis wine.

I'm out of ideas.

What error are you getting? Does EVE not run at all? Or, does it run with
graphics glitches?

--- Comment #61 from Matthieu Bertin   2009-04-10 03:58:51 ---
(In reply to comment #60)
> (In reply to comment #59)
> > Here it is. I did what you said, and it still doesn't work.
> > I downloaded 1.1.18 from sourceforge and i am using latest proprietary drivers
> > available on synaptic.
>
> I use the latest drivers from ATI. Could you try with that please? Also, do you
> have the user.reg settings specified earlier in the thread?
>
> See http://bugs.winehq.org/show_bug.cgi?id=17437#c23

IT JUST DID THE TRICK ! Thanks a lot.

everything works fine now !

--- Comment #62 from Florian Sievert   2009-04-10 15:27:01 ---
> IT JUST DID THE TRICK ! Thanks a lot.

I can confirm this. I tried for some days now to get the patch for my ati card.
However, everything seems to fail. Adding the options to the user.reg and the
patch seems to work fine for ati users. Thanks!

--- Comment #63 from Connor   2009-04-12 18:11:29 ---
We need a new patch for wine 1.1.19.

Tried to use the 1.1.17 patch on wine-1.1.19 with ATI HD3850 and Catalyst 9.3.

Patched wine compiles, but too much has changed and models are gone again.

Cmd-line shows:

fixme:d3d_shader:shader_glsl_select >>>>>>>>>>>>>>>>> GL_INVALID_OPERATION
(0x502) from glUseProgramObjectARB @ glsl_shader.c / 3757
fixme:d3d_shader:shader_glsl_load_constants >>>>>>>>>>>>>>>>>
GL_INVALID_OPERATION (0x502) from glUniform4fvARB @ glsl_shader.c / 573

--- Comment #64 from Mathieu Belanger   2009-04-12 22:12:13 ---
Created an attachment (id=20410)
 --> (http://bugs.winehq.org/attachment.cgi?id=20410)
Wine 1.1.19 patch for Apocrypha

Here the new patch for Wine 1.1.19!

Enjoy!

--- Comment #65 from Mathieu Belanger   2009-04-12 22:14:14 ---
Created an attachment (id=20411)
 --> (http://bugs.winehq.org/attachment.cgi?id=20411)
Wine 1.1.19 patch for Apocrypha

Here the new patch for Wine 1.1.19!

Enjoy!

--- Comment #66 from Mathieu Belanger   2009-04-12 22:16:22 ---
Created an attachment (id=20412)
 --> (http://bugs.winehq.org/attachment.cgi?id=20412)
Wine 1.1.19 patch for Apocrypha

Here the new patch for Wine 1.1.19!

Enjoy!

--- Comment #67 from Mathieu Belanger   2009-04-12 22:17:41 ---
Sorry for the triple post, got a Time out error from winehq....

--- Comment #68 from A.T.   2009-04-13 14:06:02 ---
author    Stefan Dösinger

Sun, 22 Mar 2009 11:24:28 +0000 (12:24 +0100)

committer    Alexandre Julliard

Fri, 10 Apr 2009 09:22:55 +0000 (11:22 +0200)

commit    754b5cf2adc73a917d7d60c35d5cad5897a5e92d

tree    fe1490bead9bf88cc15e0b6d58d17b9c1f35fafe    tree | snapshot

parent    0af0bafd6ae40f9832d9c630c24a6696ac284631    commit | diff

wined3d: Deal with reserved shader constants in the backend.

This moves the GLSL and ARB specific reserved constants out of directx.c into

the get_caps methods of the shader backends. That way the number of reserved

constants remains in the backend.

GL_LIMITS({v/p}shader_constantsF) now contains the real number of constants as

advertised by GL instead of some mixture of GL info and backend implementation

specifics. This makes it easier for backends to decide how many constants to

use.

===========================================================

from what i understand this is the (partial) patch already integrated by Stefan
in 1.1.19 (should work for nvidia 7xxx series ? ...i will test asap and post
results for 7600GT);

the only things missing seems to be the registry keys to force correct values
for buggy drivers ( ati 9.3 or older ???.... some people who tested 9.4 beta
reported on phoronix forum that 9.4 solved this issue, lets hope is true....9.4
for M$ WIN was released few days ago, so linux 9.4 final seems to be just
around the corner)

--- Comment #69 from H. Verbeet   2009-04-13 14:25:18 ---
(In reply to comment #68)
> from what i understand this is the (partial) patch already integrated by Stefan
> in 1.1.19 (should work for nvidia 7xxx series ? ...i will test asap and post
> results for 7600GT);

No, it's related, but doesn't solve this issue yet. You need
http://www.winehq.org/pipermail/wine-patches/2009-April/071698.html and (for
fglrx) the following patches for that. Those still need some work though.

--- Comment #70 from Mathieu Belanger   2009-04-13 15:03:06 ---
This patch (apocrypha.shaders.1.1.19.patch) is still needed for EVE on ATI card
and Nvidia 7xxx-

- The ATI Drivers are still buggy and still need overide via registery key for
uniform and varying.

- The max_constantsF = GL_LIMITS(vshader_constantsF) stuff moved but the patch
is still needed for that because the formula are the same : max_constantsF =
GL_LIMITS(vshader_constantsF) - (MAX_CONST_B / 4) - MAX_CONST_I - 1 and
MAX_CONST_B & MAX_CONST_I are still static constantes (16 and 16).... So
without the patch, ATI and Nvidia 7xxx don't have enough vec4 for EVE Shader.

--- Comment #71 from hpestilence   2009-04-14 23:10:35 ---
I tested out the patches on the mailing list and the wine from git with my
HD4850 and here are the results:

Everything seems to render fine on medium and low shader settings, including
shadows and hdr.

On high shaders the title screen was black with hdr and bloom on, turning those
off gave me a normal title screen. When I logged on in the game the station /
models didn't render regardless if hdr or bloom was on or off.

Looking through my logs it seems it's related to the original ati varyings bug
which is easily worked around by hardcoding a lower varying number. Only a
problem if you want to run in high settings though. :)

If your a nvidia card owner everything should work for you guys "out of the
box" if these patches get accepted.

--- Comment #72 from hpestilence   2009-04-22 01:53:28 ---
Created an attachment (id=20605)
 --> (http://bugs.winehq.org/attachment.cgi?id=20605)
shader error from wine git

I found a few models that didn't render specifically the covetor (possibly
other mining ships) and the dreadnoughts. I found the error and it seems a
shader uses 246 uniforms and only 245 is defined.

This is with the current wine in the git and just a little patch to lower my
varyings to 52.

--- Comment #73 from Vitaliy Margolen   2009-04-26 02:00:26 ---
*** Bug 18217 has been marked as a duplicate of this bug. ***

--- Comment #74 from hpestilence   2009-04-27 02:58:03 ---
Created an attachment (id=20746)
 --> (http://bugs.winehq.org/attachment.cgi?id=20746)
wine-1.1.20 patch

This patch is a mix of Stefan's "official" fix for the uniforms issue and the
registry option to change the max varyings for ATI users.

This should allow all models to render on medium or low shader settings with
all other settings on high, but on high shaders the mining ships and
dreadnoughts don't render and maybe some other stuff.

--- Comment #75 from hpestilence   2009-04-27 03:03:23 ---
oops looks like it copied a whole file into the patch, I'll fix it tomorrow :)

--- Comment #76 from H. Verbeet   2009-04-27 03:07:17 ---
For what it's worth, "git apply" is better at applying patches than "patch".

--- Comment #77 from hpestilence   2009-04-27 17:46:11 ---
Created an attachment (id=20762)
 --> (http://bugs.winehq.org/attachment.cgi?id=20762)
fixed wine-1.1.20 patch

--- Comment #78 from Mads   2009-04-28 12:25:10 ---
still same problem here.. 7900GS - Ubuntu 9.04 - Wine 1.1.20 - tried the patch
above with no luck so far

--- Comment #79 from Connor   2009-04-28 14:53:20 ---
(In reply to comment #78)
> still same problem here.. 7900GS - Ubuntu 9.04 - Wine 1.1.20 - tried the patch
> above with no luck so far

Patch from post #70 for wine-1.1.19 work for wine-1.1.20 on ATI 3850/3870 with
Catalyst 9.4.

The patch from post #77 gives the usual errors and shows no models at all:

fixme:d3d_shader:shader_glsl_load_constants >>>>>>>>>>>>>>>>>
GL_INVALID_OPERATION (0x502) from glUniform4fvARB @ glsl_shader.c / 574
fixme:d3d_shader:shader_glsl_select >>>>>>>>>>>>>>>>> GL_INVALID_OPERATION
(0x502) from glUseProgramObjectARB @ glsl_shader.c / 3757

--- Comment #80 from Joachim Rousseau   2009-04-28 16:40:37 ---
(In reply to comment #79)

All patches until 1.1.19 solved the issue with my NVidia 6800GT. With last
patch for wine 1.1.20 I had a crash at first Eve run (between the charachter
choice end the station loading), but the second launch was fine. And this crash
may be unrelated, or due to some odd cache effect.
Not tested for a long time.

--- Comment #81 from Bozhan Boiadzhiev   2009-04-30 15:08:26 ---
why patch isn't accepted yet?
i hate to compile things... :((((

--- Comment #82 from Joachim Rousseau   2009-04-30 17:59:58 ---
(In reply to comment #80)
> (In reply to comment #79)

Only one problem : salvaging effect is very slow (less than 0.5fps) though I
set shader quality to low. I may discover other issues like that I suppose.

--- Comment #83 from hpestilence   2009-05-08 17:39:36 ---
Created an attachment (id=20974)
 --> (http://bugs.winehq.org/attachment.cgi?id=20974)
wine-1.1.21 ati varying patch

This is just the part of the previous patch that makes it optional to set the
varyings in the registry for dx10 ATI cards that have problems displaying
models when they're used in a shader.

[Software\Wine\Direct3D]
"MaxVaryings" = "52"  (multiple of 4, example of a 68 varying card)

This is an example of a card supporting 68 varyings, use glxinfo -l to find out
what fglrx is reporting. A lower value than 52 might be needed for some cards.

--- Comment #84 from Stefan Dösinger   2009-05-08 18:31:32 ---
I recommend to define a driver quirk in the quirk table in directx.c for this
card/driver. I cannot do this because I do not have a dx10 ATI card to test
with.

Besides, I think this should be fixed in fglrx 9.4, or am I mistaken?

--- Comment #85 from hpestilence   2009-05-08 20:04:38 ---
Nothing involving uniforms or varyings is fixed in the 9.4 drivers for my card
at least glxinfo -l seems to report the same as the previous drivers. One of
the people who beta tests the driver says 9.5 has more wine fixes, not sure
which ones specifically though.

--- Comment #86 from Connor   2009-05-09 08:36:33 ---
(In reply to comment #83)
> Created an attachment (id=20974)

--> (http://bugs.winehq.org/attachment.cgi?id=20974) [details]
> wine-1.1.21 ati varying patch
>
> This is just the part of the previous patch that makes it optional to set the
> varyings in the registry for dx10 ATI cards that have problems displaying
> models when they're used in a shader.
>
> [Software\Wine\Direct3D]
> "MaxVaryings" = "52"  (multiple of 4, example of a 68 varying card)
>
> This is an example of a card supporting 68 varyings, use glxinfo -l to find out
> what fglrx is reporting. A lower value than 52 might be needed for some cards.

wine-1.1.21 with the patch does not work for me on ATI 3870.

wine compiles fine but hangs on login.
Console says:

err:ntdll:RtlpWaitForCriticalSection section 0x110054 "heap.c: main process
heap section" wait timed out in thread 0024, blocked by 0021, retrying (60 sec)
err:ntdll:RtlpWaitForCriticalSection section 0x110054 "heap.c: main process
heap section" wait timed out in thread 0028, blocked by 0021, retrying (60 sec)
err:ntdll:RtlpWaitForCriticalSection section 0x110054 "heap.c: main process
heap section" wait timed out in thread 0025, blocked by 0021, retrying (60 sec)
err:ntdll:RtlpWaitForCriticalSection section 0x110054 "heap.c: main process
heap section" wait timed out in thread 0027, blocked by 0021, retrying (60 sec)
err:ntdll:RtlpWaitForCriticalSection section 0x110054 "heap.c: main process
heap section" wait timed out in thread 0026, blocked by 0021, retrying (60 sec)

--- Comment #87 from Connor   2009-05-09 08:40:01 ---
(In reply to comment #86)
> (In reply to comment #83)

The above was wine-1.1.21 at log in into a station.

When i use another account to log into space wine crashes and closes:

wine: Unhandled page fault on write access to 0x6c666556 at address 0x7ef93ca5
(thread 0019), starting debugger...
err:seh:raise_exception Unhandled exception code c0000005 flags 0 addr
0x7ef93ca5

wine-1.1.19 and wine-1.1.20 work fine

--- Comment #88 from hpestilence   2009-05-10 00:17:34 ---
Created an attachment (id=20998)
 --> (http://bugs.winehq.org/attachment.cgi?id=20998)
ati dx10 varying float fix patch

This patch should work for those with dx10 ATI cards and using fglrx, it
changes the max varying floats from 68 to 48 through the quirk table. I chose
48 to cover all hd2000 to hd4000 cards since it's been reported that 52 is too
high for some.

--- Comment #89 from Stefan Dösinger   2009-05-10 04:47:40 ---
This patch looks good to me. If it works, go ahead and send it to wine-patches.

--- Comment #90 from Connor   2009-05-11 17:09:41 ---
(In reply to comment #87)
> (In reply to comment #86)
> > (In reply to comment #83)
>
> The above was wine-1.1.21 at log in into a station.
>
> When i use another account to log into space wine crashes and closes:
>
> wine: Unhandled page fault on write access to 0x6c666556 at address 0x7ef93ca5
> (thread 0019), starting debugger...
> err:seh:raise_exception Unhandled exception code c0000005 flags 0 addr
> 0x7ef93ca5
>
> wine-1.1.19 and wine-1.1.20 work fine

I did a git regression test and received this as the patch causing the crash:

f9276a64295e5fdff6e8b68958ed9ef0e9a29c31 is first bad commit
commit f9276a64295e5fdff6e8b68958ed9ef0e9a29c31
Author: Stefan Dösinger 
Date:   Thu May 7 17:31:20 2009 +0200

wined3d: Keep track of used float constants.

:040000 040000 ae0082c6c1dfce84694ea90dcab94cdf2d67d107
3d6199e8e08684e6c71188ee40a8b9a41ad3d82a M      dlls

--- Comment #91 from Stefan Dösinger   2009-05-12 04:46:09 ---
Maybe the crash is related to the incorrectly reported constants? Ie, if the
driver says 128 constants and the fixup code doesn't catch it, the code will
allocate a bitmap for 128 constants. If the shader uses 256 it will blow up
memory.

Maybe we need to change something there to prevent a driver bug from causing a
memory corruption in wined3d. The game will still not work, but at least we
don't corrupt memory.

--- Comment #92 from Stefan Dösinger   2009-05-12 04:48:18 ---
    {
        match_ati_r300_to_500,
        quirk_ati_dx9,
        "ATI GLSL constant and normalized texrect quirk"
    },

Ah yeah, I limited the uniform fixup to <= r500 cards because (a) I expected a
fixed r600+ driver soon, and (b) I don't have an r600 card to test. So someone
may look into adding the uniform quirk for r600 cards too. (but don't do that
normalized texrect stuff on r600. It has real NP2 support)

--- Comment #93 from Connor   2009-05-12 09:02:17 ---
(In reply to comment #91)
> Maybe the crash is related to the incorrectly reported constants? Ie, if the
> driver says 128 constants and the fixup code doesn't catch it, the code will
> allocate a bitmap for 128 constants. If the shader uses 256 it will blow up
> memory.

What i did notice is that my driver does not report any maxvalues when "glxinfo
-l" is issued.

I use opensuse 11.1 64bit, ATI 3850 and 3870 with Catalyst 9.4 and a "glxinfo
-l" reports no maxvaryings or uniforms.

--- Comment #94 from hpestilence   2009-05-14 20:53:53 ---
Created an attachment (id=21105)
 --> (http://bugs.winehq.org/attachment.cgi?id=21105)
atidx10quirks.patch

Sorry, my card was getting the ati dx9 quirks applied so I didn't include the
arb constants quirk.

This patch should apply to all ATI DX10 hardware.

I'm currently running EVE in high shader mode with everything rendering
including exhumers and dreadnoughts :)

--- Comment #95 from hpestilence   2009-05-14 23:53:38 ---
Oh yeah, I did leave a bug report at the unofficial ATI bugzilla about the
varying vec4 array > 13 is used.

Other ATI DX10 card users should report there too if you can't use 64
max_glsl_varyings.

http://ati.cchtml.com/show_bug.cgi?id=1555

--- Comment #96 from Connor   2009-05-15 13:58:21 ---
(In reply to comment #94)
> Created an attachment (id=21105)

--> (http://bugs.winehq.org/attachment.cgi?id=21105) [details]
> atidx10quirks.patch
>
> Sorry, my card was getting the ati dx9 quirks applied so I didn't include the
> arb constants quirk.
>
> This patch should apply to all ATI DX10 hardware.
>
> I'm currently running EVE in high shader mode with everything rendering
> including exhumers and dreadnoughts :)

Confirming patch working on opensuse 11.1 64bit, wine-1.1.21, catalyst 9.4, ATI
3850.

No crash on login anymore.

--- Comment #97 from hpestilence   2009-05-15 16:50:33 ---
grr, catalyst 9.5 still reports the wrong opengl capabilities.