[Bug 19913] New: 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

Summary: 2weistein: UseGLSL=disabled dark character with Shader
                    Model 2.0
           Product: Wine
           Version: 1.1.25
          Platform: PC
        OS/Version: Mac OS X 10.5
            Status: UNCONFIRMED
          Severity: trivial
          Priority: P2
         Component: directx-d3d
        AssignedTo: [e-mail]
        ReportedBy: [e-mail]
                CC: [e-mail]

Stefan,

with UseGLSL=disabled, the switch to Shader Model 2.0 shortly after wine-1.1.24
causes the animated character on the left of the main menu to become very dark
-- but not black.

commit e211a0b4a432013b76c1398fc99985951f98dc92
Author: Stefan Dösinger 
Date:   Thu Jun 18 18:27:04 2009 +0200

wined3d: Enable Shader Model 2.0 in ARB.

If the needed constants are available, we can support all vs_2_0 and ps_2_0
    requirements with the plain ARB extensions. We cannot however, run SM 2.0a
or
    SM 2.0b.

This goes along with log output not seen before:
fixme:d3d_shader:shader_arb_generate_vshader HW VertexShader Error at position
0: "Error on line 28: missing comma (hint: '[')"
 repeated 4 times and
fixme:d3d_draw:drawStridedFast >>>>>>>>>>>>>>>>> GL_INVALID_OPERATION (0x502)
from glDrawElements @ drawprim.c / 50
 repeated a thousand times

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

Jörg Höhle  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |download, regression
                URL|                            |http://www.2weistein.de/dem
                   |                            |o-download.html
            Version|1.1.25                      |1.1.28

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #1 from Jörg Höhle   2009-09-02 06:21:35 ---
I forgot to mention that I'm using OSX 10.5.7 (not 5.8) with NVidia 9400
graphics and XQuartz 2.3.3.2 (not 2.4). Will probably switch to 10.6 soon.

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #2 from Stefan Dösinger   2009-09-02 07:43:51 ---
Can you attach a +d3d_shader log?

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #3 from Jörg Höhle   2009-09-03 04:10:22 ---
Created an attachment (id=23398)
 --> (http://bugs.winehq.org/attachment.cgi?id=23398)
d3d_shader log from SM 2.0 commit (bzipped 33KB)

The attached +d3d_shader log may be somewhat redundant, since I had to press
Esc multiple times to skip the cdv animation, brainmonster logo, intro video
and then logo to eventually reach the main menu with the dark character, where
I aborted using ^C.
> Can you attach a +d3d_shader log?

Incredible!  In wine-1.1.28, WINEDEBUG=+d3d_shader or even
-all,trace+d3d_shader causes a completely black screen right from the start. 
This does not happen with the mentioned post-1.1.24 commit, so I used that one
to produce the attached log.

I reviewed all TRACE[d3d_shader] and TRACE_ON(d3d_shader) in wined3d/ but found
nothing obvious (those non-\n-terminated trace() snippets maybe?).
Can you reproduce that other bug? Do you need help with it?
Shall I start another round of regression testing to identify when the screen
turns black?

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #4 from Henri Verbeet   2009-09-03 04:21:36 ---
Created an attachment (id=23399)
 --> (http://bugs.winehq.org/attachment.cgi?id=23399)
+d3d_shader patch

(In reply to comment #3)
> Incredible!  In wine-1.1.28, WINEDEBUG=+d3d_shader or even
> -all,trace+d3d_shader causes a completely black screen right from the start.

Maybe this patch helps there...

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #5 from Jörg Höhle   2009-09-07 03:59:33 ---
Your patch fixes the black screen (so the dark character becomes visible in
1.1.29).
Please make it fix as well the inconsistent usage of TRACE_ON(d3d_shader) with
TRACE_(d3d_constants) in between the 2 changed lines.

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #6 from Jörg Höhle   2010-01-19 14:12:50 ---
Created an attachment (id=25797)
 --> (http://bugs.winehq.org/attachment.cgi?id=25797)
d3d_shader+d3d_caps log from Geheimakte2, wine.git at SM2.0 commit

With i915 Intel graphics on Ubuntu Intrepid, the same SM2.0 commit causes
Geheimakte2 Puritas Cordis to freeze screen and mouse immediately after the
intro sequence (or when loading a scene), while audio finishes playing
normally.  The work around (also valid for 1.1.36) is to disable the pixel
shader in winecfg.

"UseGLSL"="disabled" does not help here.
The normal console log shows no error. The attached log was created without
disabling the pixel shader (i.e. the app hung and was aborted with ^C).

Should I open a distinct issue for this bug as the MacOS/NVidia issue is
certainly different from this Intel i915 on Linux one about Geheimakte2 (and
from bug #19155)?

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

Jörg Höhle  changed:

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

--- Comment #7 from Jörg Höhle   2010-02-22 06:41:26 ---
As of wine-1.1.39, UseGLSL or not makes not difference in 2weistein (demo and
retail) and the animated characters look good (no Moire pattern either).
Closing this issue. I'll open a distinct one for Geheimakte2.

As a side note, wine-1.1.24 still rocks for 3D games on MacOS.  Beside the 2
regressions I've filed since, I still need to bisect 2 others ones
 - 1.1.24>1.1.33 Lego Star Wars 2: graphics freeze 5min within the first level
 - 1.1.38>1.1.39 Lego Star Wars (1 or 2?): NVidia becomes as slow as Intel
graphics

Actually, the performance regression that is extremely strong in Lego Star Wars
perhaps affects 2weistein to a much lesser degree. There's a slight "long
persistence" effect of the character in 2weistein in the main menu when s/he
jumps, that reminds me of the times when those "reminiscence" mouse pointer
hacks where fashion, e.g. as you move the mouse, you still see a lot of mouse
pointers at the former positions, which then slowly fade away.
The more load on the graphics engine, the more visible the reminiscence. Yet
the Mac is still fast enough to bear 2 versions of 2weistein (e.g. demo+retail)
opened side by side in 2 different versions of Wine (e.g. 1.1.24+.39) without
other trouble.

Although I've yet to find a 3D game that I use and that displays FPS rates as a
rough estimate of performance, the general feeling is that wine-1.1.24 is more
responsive than any release afterwards.

Well, I'll try to find time and bisect the cases I know and see if that helps
improve the situation.

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #8 from Stefan Dösinger   2010-02-22 09:13:47 ---
> - 1.1.38>1.1.39 Lego Star Wars (1 or 2?): NVidia becomes as slow as Intel graphics

I suspect the slowdown is caused by ce42470138a4805dce509a1fa806c8880530f324,
but I'd appreciate an independent bisect to confirm this.

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #9 from Henri Verbeet   2010-02-22 10:48:45 ---
(In reply to comment #8)
> > - 1.1.38>1.1.39 Lego Star Wars (1 or 2?): NVidia becomes as slow as Intel graphics
>
> I suspect the slowdown is caused by ce42470138a4805dce509a1fa806c8880530f324,
> but I'd appreciate an independent bisect to confirm this.

That patch has some (expected) performance impact, but it's already in 1.1.38.

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

Alexandre Julliard  changed:

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

--- Comment #10 from Alexandre Julliard   2010-03-05 12:43:08 ---
Closing bugs fixed in 1.1.40.

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #11 from Jörg Höhle   2010-03-24 17:31:00 ---
FWIW, the mentioned Geheimakte 2 freeze with shader model 2.0 is bug #21838.
Lego Star Wars 2: graphics freeze 5min within start is bug #22029.
I'm sorry I've not yet tested the impact of commit ce42470... Wine-1.1.40 feels
quite ok (there are some hick-ups in the intro animation but I need time to
assess whether one version of Wine is truly better than others there).

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

--- Comment #12 from Jörg Höhle   2010-06-29 16:12:46 ---
Stefan,
> I suspect the slowdown is caused by ce42470138a4805dce509a1fa806c8880530f324,
> but I'd appreciate an independent bisect to confirm this.

sorry to come back that late on this wish. Now I've performed some measurements
using WINEDEBUG=+fps with Lego Star Wars 2
commit  cantina    desert    intro    scene1
after       50            30           22-40    38-41
before    33-44    18-21    13-35    40 fps
1.1.39    33-         19            13-80     22
That one commit indeed causes a significant slow down in numbers, even though
the visual difference is hardly noticeable.
But there must be some other slowdown between that commit and wine-1.1.39 that
curiously affects only "scene1", i.e. when starting to play the first level,
past the intro.

[Bug 19913] 2weistein: UseGLSL=disabled dark character with Shader Model 2.0

http://bugs.winehq.org/show_bug.cgi?id=19913

Austin English  changed:

What    |Removed                     |Added
----------------------------------------------------------------------------
         OS/Version|Mac OS X 10.5               |Mac OS X