Dnia 2010-10-31, nie o godzinie 16:33 -0400, Andreas Kloeckner pisze:
On Sun, 31 Oct 2010 16:07:29 -0400, Andreas Kloeckner
> On Sat, 30 Oct 2010 13:04:03 +0200, Tomasz Rybak <bogomips(a)post.pl> wrote:
> > Hello.
> > I have been trying to create some OpenGL examples for PyCUDA
> > and found program that segfaults.
> > I cannot debug it fully as the last 8 functions are from
> > NVIDIA libraries.
> > The last known function in the trace is cuda::context::detach.
> > Error is present on GTX460 (Fermi) and ION (9400M),
> > on PyCUDA with my patch and without it.
> > System information:
> > Debian 64-bit, kernel 2.6.32, drivers 256.53, CUDA 3.1.
> > PyCUDA from git.
> > Can someone look at program and stack trace and tell whether
> > there is some obvious mistake I am missing?
> > If not, I think I will report upstream bug to NVIDIA.
> Doesn't appear to crash with 260.24. (3.2RC available from Nvidia if you
> sign up for their developer program) Haven't tried 3.2rc2.
I lied. It does crash when you close the window.
(2.6.35, amd64, 260.24, 3.2rc)
So is this error in my code or CUDA?
BTW - I do not want to install 3.2rc for now,
as I am still trying to convince Debian Release team
to accept PyOpenCL 0.92 into Squeeze and need to be
able to build it with stable versions of libraries.
Tomasz Rybak <bogomips(a)post.pl> GPG/PGP key ID: 2AD5 9860
Fingerprint A481 824E 7DD3 9C0E C40A 488E C654 FB33 2AD5 9860