libcudart segfaults on close

Accelerated Computing CUDA CUDA Programming and Performance
1

I’m running CUDA 1.0 x86-64 on a fresh install of Ubuntu 7.04.

The program runs fine, as it did under 0.8, but it segfaults when the application terminates. Here’s what I’m getting from GDB:

Program received signal SIGSEGV, Segmentation fault.

[Switching to Thread 47532137146368 (LWP 15029)]

0x00002b3aee3a6702 in ?? () from /usr/local/cuda/lib/libcudart.so

(gdb) bt

#0  0x00002b3aee3a6702 in ?? () from /usr/local/cuda/lib/libcudart.so

#1  0x00002b3af0163c65 in __tlsHookDeleteIndex () from /usr/local/cuda/lib/libtlshook.so

#2  0x00002b3aee3a4df2 in ?? () from /usr/local/cuda/lib/libcudart.so

#3  0x00007fffbcf585d0 in ?? ()

#4  0x00002b3aee3c5071 in ?? () from /usr/local/cuda/lib/libcudart.so

#5  0x0000000000000000 in ?? ()

Just to check, I added a “press enter to quit” thing at the very end of the program, after all the cleanup is done. It’s definitely happening when the main loop terminates, and not a delayed reaction from freeing the resources.

2

I’m afraid that Ubuntu isn’t explicitly supported for the 1.0 release. We’re hoping to add Ubuntu for the next major release.

Which 1.0 package did you install? Does the problem persist if you try a different distribution’s package?

3

I used the RHEL 4 toolkit. I haven’t had time to try a different package, but I’ll let you know how it goes when I do.

If it still doesn’t work, I might try it with OpenSUSE.

4

I’ve seen reports that the SUSE specific packages are known to work with Ubuntu.

5

OK, it appears that this was an issue with the program (a double free) that for some reason didn’t cause any noticeable issues under CUDA 0.8…