From: Ondrej Marsalek (
Date: Mon May 18 2009 - 11:30:55 CDT

On Mon, May 18, 2009 at 16:27, Axel Kohlmeyer
<> wrote:
> in any case, you should compile from the cvs code for this and use
> the SHARED compiler option. mind you, there are some restrictions
> (e.g. no CUDA support, at least not on x86_64 as you cannot link
> -fPIC and -fno-PIC code and nvcc does not (yet) support -fpic
> compilation).

Thanks for the instructions.

Some details for others interested in this. It has worked for me after
changing the link command for the plugins in Make-arch from gcc to g++.
With gcc there were unresolved symbols at import in python.

The steps would roughly be the following:

1) configure and build as usual, install
2) add the SHARED option to configure.options, reconfigure and rebuild
3) set the VMDDIR env variable to point to the dir with the installation
4) in python: import vmd (with appropriate path settings so that
can be found, of course)