I had meant to send this to the list, instead of just Randall.
-- Forwarded message --
From: j s
Date: Thu, Jul 30, 2009 at 9:03 AM
Subject: Re: [Paraview] Baffled by "floating point exceptions" in ParaView
3.6.1
To: Randall Hand
If you are running it in gdb, you might be able
Randall - we saw some execution-time failures of 3.6.1 that was related
to an older version of the nVidia driver. We didn't see floating
point exceptions, though - we saw socket errors related to the
client-server connection. Moving to a newer version of the nVidia
driver on the client sid
Ok, more information.
If I compile ParaView without Python and without MPI, it works beautifully.
Enabling MPI, however, causes the problem.. a GDB trace:
(gdb) run
Starting program: /viz/work/rhand/pvezviz/install-MPI/bin/paraview
Program received signal SIGFPE, Arithmetic exception.
0x00
Looking online, that is a nasty FPE, since it is a result of an integer
operation, not a floating point operation.
This is occuring at address NULL, which should not be possible.
#8 0x in ?? ()
It looks like a function at address 0 is being called. Do you have more of
the trace,
A google search for "0x in ?? ()" and open mpi gets a lot of
hits.
Juan
On Thu, Aug 6, 2009 at 3:55 PM, j s wrote:
> Looking online, that is a nasty FPE, since it is a result of an integer
> operation, not a floating point operation.
>
> This is occuring at address NULL, which s
That came from simply "gdb ./paraview", then "run" and "bt" once it died.
--
Randall Hand
Visualization Scientist
http://www.yeraze.com
http://www.vizworld.com
On Thu, Aug 6, 2009 at 3:55 PM, j s wrote:
> Looking online, that is a nasty FPE, since it is a result of an integer
> operation, not
I'm not sure if a function is being called through a null pointer, or gdb is
reporting it for another reason.
Something bad is happening in free(), which has nothing to do with numerics,
but may have something to do with how your libraries are treating memory.
Apparently bt is supposed to give yo
You may be able to try setting the environment variable:
MALLOC_CHECK_
to
2
and the program should abort on the first instance of heap corruption.
Regards,
Juan
On Thu, Aug 6, 2009 at 4:40 PM, j s wrote:
> I'm not sure if a function is being called through a null pointer, or gdb
> is reportin
Randall - try compiling against Openmpi-1.3 and see if you get different
results.We saw issues with compiling Paraview 3.x against
Openmpi-1.2.6 that went away when we switch to Openmpi-1.3.
j s wrote:
Looking online, that is a nasty FPE, since it is a result of an
integer operation, not
My guess is that there is a symbol conflict between something in
ParaView and OpenMPI 1.2.6. If switching to 1.3 does not fix the
problem, we'll have to resort to debugging this some other way.
-berk
On Fri, Aug 7, 2009 at 8:59 AM, Rick Angelini wrote:
> Randall - try compiling against Openmpi-1.
I'm currently waiting for the admins to Install OpenMPI1.3, but in the
meantime I recompiled and linked against Mesa (rather than the nvidia
drivers).
Tada, all works. So there's definately some problem with NVidia Drivers +
OpenMPI1.2.x
--
Randall Hand
Visualization Scientist
http://www.yeraze.c
Randall - that's consistent with what we saw in our environment as well.
Randall Hand wrote:
I'm currently waiting for the admins to Install OpenMPI1.3, but in the
meantime I recompiled and linked against Mesa (rather than the nvidia
drivers).
Tada, all works. So there's definately some prob
12 matches
Mail list logo
