CentOS

long story short, I did something wrong while trying to upgrade CentOS 6.7 to CentOS7.0,
and recovered things(it's now 6.9), vnc etc. now I can do normal things as before.
This morning, I tried to compile a program which I had no problem with,
but couldn't do it with message the linker cannot understand -lcuda option.
So I reinstalled cuda using these commands as shown in CUDA manual.

`sudo rpm -i cuda-repo-rhel6-7-5-local-7.5-18.x86_64.rpm`
`sudo yum clean all`
`sudo yum install cuda`

But the problem still persisted.
I expected libcuda.so.... to have been installed somrewhere (like /usr/local/cuda/lib64)
but it was not there.

rpm -qa shows I have many cuda packages including cuda-7.5-18.x86_64.
So I wanted to see what files have been installed by that package.
repoquery --list cuda-7.5-18.x86_64
shows nothing. rpm -ql cuda-7.5-18.x86_64 shows either nothing.
What can be wrong?

The 'cuda' package is probably a metapackage that supplies no files but Requires: the others. It's an easy way to pull in multiple packages in one go - you Require the others from your main one and yum pulls them in and installs them. Look at the other ones it pulls in.

Trevor,
Yes, the file libcuda.so... was installed by cudacore package not by cuda package.
The libcuda.so file was under /usr/local/cuda/lib64/stubs. So I fixed the Makefile's LDFLAGS line to
LDFLAGS+= -L/usr/local/cuda/lib64 -L/usr/local/cuda/lib64/stubs -lcuda -lcudart -lcublas -lcurand
and it's comiled again! Thanks!
(by the way, this is yolov2 Makefile)
ADD : ups, I see an error during execution. maybe the NVIDIA driver is not installed correctly.