[Bug 91232] New: Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[Bug 91232] New: Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 1812 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

            Bug ID: 91232
           Summary: Torchlight 1&2 (in Wine): glitchy models when hardware
                    skinning enabled
           Product: Mesa
           Version: git
          Hardware: Other
                OS: All
            Status: NEW
          Severity: normal
          Priority: medium
         Component: Drivers/DRI/nouveau
          Assignee: nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org
          Reporter: gyebro69-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org
        QA Contact: nouveau-PD4FTy7X32lNgt0PjOBp9y5qC8QIuHrW@public.gmane.org

Created attachment 116947
  --> https://bugs.freedesktop.org/attachment.cgi?id=116947&action=edit
screenshot, comparison (nouveau vs. nvidia blob)

I believe this is different from bug #90887, I tried the patch from that bug
and it didn't help.
Models in Torchlight 1 and 2 are rendered incorrectly, they have lots of
flashing, white rectangles on them.Disabling hardware skinning in the options
menu makes the problem go away.
I see similar issues in some other games as well so this bug may affect
multiple games.

I tried and reproduced the problem in Mesa 10.1 and 10.4 too.
The software renderer doesn't have the problem.
Disabling shader optimization doesn't help.
The trace replays correctly on Nvidia blob 340.76.

Trace (uncompressed 43 MB):
https://drive.google.com/open?id=0B-tTbLKBl-tORHV2eGdleGk1dzQ

Fedora 22 32-bit
Kernel 4.0.6-300.fc22.i686+PAE
VGA compatible controller: NVIDIA Corporation G92 [GeForce GTS 250] (rev a2)
(prog-if 00 [VGA controller])
Mesa 10.6-branchpoint-826-gff0a41b
libdrm-2.4.62-1-g676c806
xf86-video-nouveau-1.0.11-23-gdfd827c

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 3465 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 409 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #1 from Ilia Mirkin <imirkin-FrUbXkNCsVf2fBVCVOL8/A@public.gmane.org> ---
Reproduced on nvc0 as well. Works with llvmpipe.

Switching to opt=0 makes the flickering purple instead of white. Disabling the
watchdog makes no difference.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 1218 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 525 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #2 from Ilia Mirkin <imirkin-FrUbXkNCsVf2fBVCVOL8/A@public.gmane.org> ---
Appears that the draw @30166 is the one that introduces a white polygon. The
shader program in question is a little sketchy since it sets gl_FrontColor in
vertex shader, but uses gl_SecondaryColor in fragment shader. Investigating
whether this is legal and how to handle it.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 1338 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 2633 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #3 from Ilia Mirkin <imirkin-FrUbXkNCsVf2fBVCVOL8/A@public.gmane.org> ---
Created attachment 116959
  --> https://bugs.freedesktop.org/attachment.cgi?id=116959&action=edit
shader_test that demonstrates the (?) issue

Looks like they use gl_SecondaryColor in the fragment shader, and the GLSL
linker, seeing no writes to gl_SecondaryFrontColor in the vertex shader,
creates this IR:

(declare (location=4 shader_in ) vec4 gl_in_TexCoord0)
(declare (location=5 shader_in ) vec4 gl_in_TexCoord1)
(declare (location=1 shader_in ) vec4 gl_Color)
(declare (temporary ) vec4 gl_in_FrontColor1_dummy)
(declare (location=3 shader_in ) float gl_FogFragCoord)
(declare (location=4 shader_out ) vec4 gl_out_TexCoord0)
(declare (temporary ) vec4 gl_out_TexCoord0)
(declare () vec4 ret)
(declare (location=0 uniform ) sampler2D ps_sampler0)
(declare (location=1 uniform ) samplerCube ps_sampler1)
(declare (location=2 uniform ) vec4 specular_enable)
(declare (location=3 uniform ) #anon_struct_0001@0x24cc5d0 ffp_fog)
(function main
  (signature void
    (parameters
    )
    (
      (declare (temporary ) vec4 textureCube_retval)
      (assign  (xyzw) (var_ref textureCube_retval)  (tex vec4 (var_ref
ps_sampler1)  (swiz xyz (var_ref gl_in_TexCoord1) ) 0 1 () )) 
      (assign  (xyzw) (var_ref ret)  (expression vec4 * (tex vec4 (var_ref
ps_sampler0)  (swiz xy (var_ref gl_in_TexCoord0) ) 0 1 () )(var_ref gl_Color) )
) 
      (assign  (xyz) (var_ref ret)  (expression vec3 sat (expression vec3 +
(swiz xyz (var_ref textureCube_retval) )(swiz xyz (var_ref ret) )) ) ) 
      (assign  (w) (var_ref ret)  (expression float * (swiz w (var_ref
textureCube_retval) )(swiz w (var_ref ret) )) ) 
      (assign  (xyzw) (var_ref gl_out_TexCoord0)  (expression vec4 +
(expression vec4 * (var_ref gl_in_FrontColor1_dummy) (var_ref specular_enable)
) (var_ref ret) ) ) 
      (assign  (xyz) (var_ref gl_out_TexCoord0)  (expression vec3 lrp (swiz xyz
(record_ref (var_ref ffp_fog)  color) )(swiz xyz (var_ref gl_out_TexCoord0)
)(expression float sat (expression float * (expression float + (record_ref
(var_ref ffp_fog)  end) (expression float neg (var_ref gl_FogFragCoord) ) )
(record_ref (var_ref ffp_fog)  scale) ) ) ) ) 
      (assign  (xyzw) (var_ref gl_out_TexCoord0@64)  (var_ref gl_out_TexCoord0)
) 
    ))

)

This in turn becomes an uninitialized read from TEMP[3] in the TGSI, and
nouveau doesn't 0-initialize such vars. specular_enable happens to be (1, 1, 1,
0) here.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 3629 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 509 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #4 from Ilia Mirkin <imirkin-FrUbXkNCsVf2fBVCVOL8/A@public.gmane.org> ---
Created attachment 116960
  --> https://bugs.freedesktop.org/attachment.cgi?id=116960&action=edit
init unknown values to 0

As I suspected, getting nouveau to init these to 0 fixes the issue. But IMO the
glsl compiler is the one doing something dodgy here...

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 1544 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 788 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #5 from Béla Gyebrószki <gyebro69-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> ---
(In reply to Ilia Mirkin from comment #4)
> Created attachment 116960 [details] [review]
> init unknown values to 0
> 
> As I suspected, getting nouveau to init these to 0 fixes the issue. But IMO
> the glsl compiler is the one doing something dodgy here...

The patch works and it fixes the issue not just in Torchlight but also in some
other titles like Silent Hunter III, Evil Genius or the Ankh series.
Some games with similar looking problem still have the the problem, but that
might be a different issue.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 1988 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 1040 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #6 from Béla Gyebrószki <gyebro69-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> ---
Traces from games where the problem is fixed with your patch:

Evil Genius (character model corrupted after loading a saved game):
https://drive.google.com/open?id=0B-tTbLKBl-tOMWtxWXdGVm13LXc

Ankh 2 (menu screen heavily corrupted):
https://drive.google.com/open?id=0B-tTbLKBl-tOQ0kydGFFM2xsbXM

===
Maybe unrelated, here are the traces from 2 games where the problem isn't fixed
by the patch:

Bloodrayne 2 (objects like the tables in front of the model have those
flickering boxes):
https://drive.google.com/open?id=0B-tTbLKBl-tOWGZEZ1dPYUVwUzA
With your patch applied those objects become invisible.

Tomb Raider:Underworld (character model corrupted):
https://drive.google.com/open?id=0B-tTbLKBl-tOOWhES3ZJaGlHeXc
The problem remains with the patch.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 2174 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 2042 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #7 from Ilia Mirkin <imirkin-FrUbXkNCsVf2fBVCVOL8/A@public.gmane.org> ---
(In reply to Béla Gyebrószki from comment #6)
> Traces from games where the problem is fixed with your patch:
> 
> Evil Genius (character model corrupted after loading a saved game):
> https://drive.google.com/open?id=0B-tTbLKBl-tOMWtxWXdGVm13LXc
> 
> Ankh 2 (menu screen heavily corrupted):
> https://drive.google.com/open?id=0B-tTbLKBl-tOQ0kydGFFM2xsbXM

i965 on hsw also has issues with these. And on the original trace in comment 1.
Looking at least at the Ankh 2 trace, it appears to also be the
gl_SecondaryColor issue -- I assume they're using slightly diff versions of the
same engine. All the var names are the same, but the frag shader structure is a
tad different.

> 
> ===
> Maybe unrelated, here are the traces from 2 games where the problem isn't
> fixed by the patch:
> 
> Bloodrayne 2 (objects like the tables in front of the model have those
> flickering boxes):
> https://drive.google.com/open?id=0B-tTbLKBl-tOWGZEZ1dPYUVwUzA
> With your patch applied those objects become invisible.

i965 also has funkiness rendering on the table.

This genius shader actually only sets gl_FrontSecondaryColor and then uses
gl_Color.w (in addition to gl_SecondaryColor). I bet that has to be 1.0 and not
0.0, hence the transparency fail. I guess it's all from the same engine. And it
expects the colors to be (0, 0, 0, 1) when not explicitly set.

> 
> Tomb Raider:Underworld (character model corrupted):
> https://drive.google.com/open?id=0B-tTbLKBl-tOOWhES3ZJaGlHeXc
> The problem remains with the patch.

i965 seems fine with this one. Probably an unrelated bug.

Well, I don't feel so bad. Intel has a QA team (not to mention a dev team). You
have probably done 10x more QA on nouveau than the sum of what had ever been
done prior.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 3397 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 482 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #8 from Ilia Mirkin <imirkin-FrUbXkNCsVf2fBVCVOL8/A@public.gmane.org> ---
Created attachment 116966
  --> https://bugs.freedesktop.org/attachment.cgi?id=116966&action=edit
init unknown values to (0, 0, 0, 1)

This fixes the table in Bloodrayne 2, I think. (Having some unrelated issues,
so not 100% sure.)

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 1539 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 537 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #9 from Ilia Mirkin <imirkin-FrUbXkNCsVf2fBVCVOL8/A@public.gmane.org> ---
I spoke to ChrisF about this, and it appears that the behaviour is undefined at
least in GLSL. As the GLSL is being generated by wine, I mentioned it on
#winehackers, but there was no response, people probably just weren't around.

Perhaps worth filing a bug on bugs.winehq.org as well?

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 1346 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 346 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

--- Comment #10 from Béla Gyebrószki <gyebro69-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> ---
Bug reported in Wine bugtracker: https://bugs.winehq.org/show_bug.cgi?id=38869

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 1237 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 801 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

Henri Verbeet <hverbeet-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> changed:

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

--- Comment #11 from Henri Verbeet <hverbeet-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> ---
Yeah, this is our (Wine's) bug. We should be intializing varyings read by the
fragment shader but not written by the vertex shader, pretty much like the
patch on this bug report does, but in wined3d instead of the driver.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 2314 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau

[Bug 91232] Torchlight 1&2 (in Wine): glitchy models when hardware skinning enabled

[bugzilla-daemon-CC+yJ3UmIYqDUpFQwHEjaQ]

[-- Attachment #1.1: Type: text/plain, Size: 1309 bytes --]

https://bugs.freedesktop.org/show_bug.cgi?id=91232

Béla Gyebrószki <gyebro69-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |CLOSED
                 CC|                            |gyebro69-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

--- Comment #12 from Béla Gyebrószki <gyebro69-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> ---
Fixed in Wine git by the following series of commits:

http://source.winehq.org/git/wine.git/commit/97ee9451edae82154b289cb57b46ef13aedd58b8
http://source.winehq.org/git/wine.git/commit/09eda02b6896f5fe4ca7db5e1c13cbc67f8118e8
http://source.winehq.org/git/wine.git/commit/08b21528ed51d7ece92ea67fe0afe1d2a998c989
http://source.winehq.org/git/wine.git/commit/f511787423f468a679837761ef4ea841ba7a3fda
http://source.winehq.org/git/wine.git/commit/93db8e97da2690c62a5d4586d6214632a98df344

wine-1.7.52-190-gd548639
OpenGL vendor string: nouveau
OpenGL renderer string: Gallium 0.4 on NV92
OpenGL core profile version string: 3.3 (Core Profile) Mesa 11.1.0-devel

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.

[-- Attachment #1.2: Type: text/html, Size: 3350 bytes --]

[-- Attachment #2: Type: text/plain, Size: 153 bytes --]

_______________________________________________
Nouveau mailing list
Nouveau@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/nouveau