Kevin Day wrote:
>
> I've got a small program that will cause a segfault:
>
> Program received signal SIGFPE, Arithmetic exception.
> 0x281824d8 in gl_x86_transform_points3_3d_no_rot_raw ()
> from /t/video/libGL.so.3
> (gdb) bt
> #0 0x281824d8 in gl_x86_transform_points3_3d_no_rot_raw ()
> from /t/video/libGL.so.3
> #1 0x805c2f0 in ?? ()
> #2 0x8062ba8 in ?? ()
> #3 0x280aa34c in gl_Accum (ctx=0x0, op=GL_FALSE, value=0)
> from /t/video/libGL.so.3
> #4 0x80c6000 in ?? ()
>
> Yet this doesn't crash when run with a non-mesa OpenGL implementation, so
> I'm leaning towards "Mesa Bug" here. This uses OSMesa, and I really don't
> have any Mesa system with X installed to see if it's specific to OSMesa or
> not. Here's the program given to me by a coworker. It contains a bit of our
> specific glue to draw it using our display system, so ignore that bit. :)
> This is using FreeBSD, which traps a lot of FP exceptions by default, should
> I be turning these off?
I can't compile your program because I don't have gamedef.h
First, I suggest trying Mesa 3.2 beta 1.
I recently found an FP exception problem in another one of Mesa's
X86 assembly functions so it's probably you've found another one.
Mesa manipulates the x86 FP control word in order try to speed-up
some FP operations, but we shouldn't be masking FP exceptions. It's
best to catch them and fix them. In Mesa 3.3 I added a debugging
option to enable aborts on FP exceptions. It's exposed a couple
bugs so far.
-Brian
_______________________________________________
Mesa-bug maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-bug
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
