I am on a Fedora 19, trying to run nvvp from usr/local/cuda-6.0/libnvvp.
I can run the visual profiler, nvvp on CUDA samples that I compiled. Also, my cuda program works fine when I run it from the command line.
But when I try to run my exe with the nvvp, I get this error:
/vhome/username/Project/cudaexe: error while loading shared libraries: libcudart.so.6.0: cannot open shared object file: No such file or directory
A google search for this error shows that no one using nvvp seems to have had this issue before. But in general this error seems to show up when environmental settings are not correct. I did all of that when I first installed (see this post: c++ - Why can't libcudart.so.4 be found when compiling the CUDA samples under Ubuntu? - Stack Overflow), and did it all again anyway to make sure. However, that did not solve my issue, and I did not think it would solve my issue because I can run the profiler with the CUDA samples, and if the environmental variables were not set right, then it wouldn’t have worked for that.
Any insight would be greatly appreciated.
Just a bit more information about my cuda program. The cuda code was compiled into a static library then linked to *.cpp codes using gcc -std=c++11. Could that be causing this error? And if so, is there a way around it? i.e. not using c++11, or not using the cuda code as a library?