[Pkg-scicomp-devel] Bug#578052: Bug#578052: [atlas] providing >20 separate optimized library packages is insane
Sylvestre Ledru
sylvestre at debian.org
Fri Apr 16 12:45:09 UTC 2010
Hello Fabian,
Thanks, it is always interesting to have feedbacks!
Le vendredi 16 avril 2010 à 13:52 +0200, Fabian Greffrath a écrit :
> thanks alot for the effort you put in packaging the several optimized
> versions of the atlas libraries, BUT I consider it completely
> unnecessary to put them all into separate packages and leave it up to
> the user to manually select the one optimized for his system.
Well, some people do not agree on this.
I think it is important to let the user select the appropriate
optimisation for his computer.
> 1) Many users do not even know which features their processor
> supports. What happens if I install the core2sse3-optimized library
> (and it becomes the main alternative due to its higher priority in the
> update-alternatives machanism) if my processor is a simple Pentium 3?
I believe that a user looking for performances on tool based on BLAS
implementation (R, Scilab, Code Saturne, etc) is aware that it is
related to the CPU they have.
> 2) It is possible to let the dynamic linker select special optimized
> libraries based on information the kernel knows about my system (no
> human intervention necessary). You can simply install libraries in
> subdirectories of /usr/lib which are named after the feature that they
> make use of. For example, you put the altivec-optimized library into
> /usr/lib/altivec and the normal one into /usr/lib. The linker will
> select the right one for you and give the optimized one a higher
> priority. It works on all architectures and has already proven in
> other packages that benefit from architecture-dependent optimizations.
I noticed your message on my blog. I didn't have time to investigate it:
I am currently more focused on build issues on some architectures but I
have it in mind.
However, if I understand correctly, this would mean that I would have to
ship all libraries into a single package or to add some Depends on the
base package, isn't it ?
An atlas package is almost 5M. For example, under x86, we have 5
optimized packages. This would increase by 5 time the install of Atlas.
> Please consider reverting the separation into the multitude of library
> packages and make use of this neat feature instead.
The separation has been in Debian for a very long time (more than 10
years) and it is the first strong complain that I see about that (I know
it is not an argument for not doing it).
Sylvestre
More information about the Pkg-scicomp-devel
mailing list