ATI+linux bug hunt

[iluxa]

There is a bug, and it's even mentioned on forums, where EL

client crashes on ATI Radeon cards, when it rains. This is actually ATI

bug, but it'd be nice to work around it, since I don't see a fix coming out

of them. I confirmed that commenting out call to function that renders

rain works around it. Now I was trying to get to C2, and when I got to the

place that is supposed to render the ship (or so I was told), I have 100%

reproduceable crash (I die just as soon as I log in, since client tries to

render that model).

 

I tracked it down specifically to

3d_objects.c line 309:

glDrawArrays(GL_TRIANGLES,array_order.start,array_order.count);

when object_id->file_name is ./3dobjects/structures/ship2.e3d

(this is in latest CVS)

 

My username in the game is iluxa. You can check my coordinates/camera

view direction

to try to reproduce this.

Oh, and I have Radeon R250 [Mobility FireGL 9000] (rev 02) graphics card,

using X.Org version 7.1.1 and here is ATI driver info:

ilya@iluxalaptop ~/games/elc $ fglrxinfo

display: :0.0 screen: 0

OpenGL vendor string: ATI Technologies Inc.

OpenGL renderer string: MOBILITY/RADEON 9000 DDR Generic

OpenGL version string: 1.3.1091 (X4.3.0-8.27.6)

[iluxa]

Some things that might be related:

http://www.gamedev.net/community/forums/to...hichpage=1&

(describes a bug in driver causing a crash when array is larger then 4000 elements)

 

An interesting info on similar problem that Xgl sees when running on top of fglrx can also be

found here: https://launchpad.net/distros/ubuntu/+sourc...-xgl/+bug/63830

 

I put together following patch based on info above, and it seems to work so far

 

Index: 3d_objects.c

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

RCS file: /cvsroot/elc/elc/3d_objects.c,v

retrieving revision 1.105

diff -u -r1.105 3d_objects.c

--- 3d_objects.c 14 Sep 2006 16:43:35 -0000 1.105

+++ 3d_objects.c 27 Dec 2006 02:25:36 -0000

@@ -296,6 +296,7 @@

 

materials_no=object_id->e3d_data->materials_no;

for(i=0;i<materials_no;i++) {

+ int idx, max;

get_and_set_texture_id(array_order.texture_id);

#ifdef DEBUG

// a quick check for errors

@@ -306,7 +307,13 @@

array_order.start, array_order.count);

}

#endif // DEBUG

- glDrawArrays(GL_TRIANGLES,array_order.start,array_order.count);

+ idx = array_order.start;

+ max = array_order.start+array_order.count;

+ while(idx < max) {

+ int num = max-idx < 1000 ? max-idx : 1000;

+ glDrawArrays(GL_TRIANGLES,idx,num);

+ idx += num;

+ }

}

 

glPopMatrix();//restore the scene

[Learner]

Some things that might be related:

http://www.gamedev.net/community/forums/to...hichpage=1&

(describes a bug in driver causing a crash when array is larger then 4000 elements)

 

An interesting info on similar problem that Xgl sees when running on top of fglrx can also be

found here: https://launchpad.net/distros/ubuntu/+sourc...-xgl/+bug/63830

 

I put together following patch based on info above, and it seems to work so far

 

Index: 3d_objects.c

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

RCS file: /cvsroot/elc/elc/3d_objects.c,v

retrieving revision 1.105

diff -u -r1.105 3d_objects.c

--- 3d_objects.c 14 Sep 2006 16:43:35 -0000 1.105

+++ 3d_objects.c 27 Dec 2006 02:25:36 -0000

@@ -296,6 +296,7 @@

 

materials_no=object_id->e3d_data->materials_no;

for(i=0;i<materials_no;i++) {

+ int idx, max;

get_and_set_texture_id(array_order.texture_id);

#ifdef DEBUG

// a quick check for errors

@@ -306,7 +307,13 @@

array_order.start, array_order.count);

}

#endif // DEBUG

- glDrawArrays(GL_TRIANGLES,array_order.start,array_order.count);

+ idx = array_order.start;

+ max = array_order.start+array_order.count;

+ while(idx < max) {

+ int num = max-idx < 1000 ? max-idx : 1000;

+ glDrawArrays(GL_TRIANGLES,idx,num);

+ idx += num;

+ }

}

 

glPopMatrix();//restore the scene

Several patches based on this have gone into e3d_objects.c, shadows.c, reflection.c and weather.c

 

The weather specific patch might even stop the crashing when it rains for some ATI users!

[iluxa]

I'll give it a try this weekend

[iluxa]

OK. I can confirm that rain rendering works for me just fine now.