Page 1 of 1

conflict libraries

PostPosted: Wed Jun 05, 2013 11:36 am
by samarvafai1392
Hi,

I call the culaZcooBicgstab routine to solve the linear system of equations Ax=b in my code. I use cmake for making the executable file. I worked on Linux Debian operating system. The cuda version on the machine is 4.2. And the GPU model is Tesla M2070. As I don't have a root access to the system I put the cusparse library in my home directory and I couldn't adjust the /etc/ld.so.conf file to include CULASPARSE_LIB_PATH_64 and I just set LD_LIBRARY_PATH environment variable. After running the cmake I got the following warnings that I couldn't fix them:

"CMake Warning at /home/sv/HyENA_010311/examples
CMakeLists.txt:80 (ADD_EXECUTABLE):
Cannot generate a safe runtime search path for target CQM because files in
some directories may conflict with libraries in implicit directories:

runtime library [libcudart.so.4] in /usr/lib/x86_64-linux-gnu may be hidden by files in:
/home/sv/culasparse/lib64
runtime library [libcublas.so.4] in /usr/lib/x86_64-linux-gnu may be hidden by files in:
/home/sv/culasparse/lib64

Some of these libraries may not be found correctly."

The configuration is done and I have the correct result, but I am just curious about these warning messages?

Thank you in adavance.
Samar

Re: conflict libraries

PostPosted: Thu Jun 06, 2013 6:06 am
by john
What you're seeing is a message saying that Cmake doesn't really know how to rectify that CULA has included a copy of libcudart, while there's another one present on the system. It's just telling you it doesn't know which to use, so you'll want to adjust your CMake scripts around to help it out. In theory, either of the two are fine. In practice, though, you'll probably want to use whichever one has the newest minor version number, but unfortunately in the CUDA 4.x series they didn't include the minor version number in the filename, so it can be difficult to tell. Basically - if it works, I wouldn't worry about it. If it fails, you'll have a better time if you move up to CUDA 5.