Moreover, are you building Amber on a machine that actually HAS a GPU? If
you're inadvertently logged into the wrong box you'll get weird results.
CUDA_HOME can be set in that case and appear fine but for one or two
libraries that may show up as missing until you're on the right hardware.
On Apr 4, 2017 9:34 PM, "David Case" <david.case.rutgers.edu> wrote:
On Tue, Apr 04, 2017, Daniel Mermelstein wrote:
>
> Has anyone tried compiling the current master branch for serial cuda? I
> keep getting "/usr/bin/ld: cannot find -lcuda". My cuda library is not in
> /usr/bin/, but I've got the library location in my LD_LIBRARY_PATH.
Was CUDA_HOME set correctly when you ran configure? It would be good to
post
your config.h, plus the actual location of the cuda library.
LD_LIBRARY_PATH is used to find shared libraries at run time, not compile
time
(I think). To make the executable, your link command has to have
something like "-L/usr/local/cuda/lib64". If configure is not creating
the correct arguments, changing LD_LIBRARY_PATH won't help.
...dac
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Apr 04 2017 - 19:00:04 PDT
