| 1 |
On Sunday 21 August 2005 16:33, Darren Dale wrote: |
| 2 |
> On Sunday 21 August 2005 10:25 am, Peter Bienstman wrote: |
| 3 |
> > On Sunday 21 August 2005 15:22, Darren Dale wrote: |
| 4 |
> > > I use numeric and scipy, and have been trying for a long time to |
| 5 |
> > > improve the ebuilds to use blas/lapack/atlas. I submitted an ebuild for |
| 6 |
> > > numeric six months ago (http://bugs.gentoo.org/show_bug.cgi?id=81520), |
| 7 |
> > > but it has languished in buzilla because it is not clear how to deal |
| 8 |
> > > with ATLAS (my atlas USE flag was rejected). How should ATLAS |
| 9 |
> > > dependencies be handled in the new infrastructure? |
| 10 |
> > |
| 11 |
> > In theory it's as simple as depending on virtual/blas and virtual/lapack. |
| 12 |
> > The user can then switch between different lapack and blas |
| 13 |
> > implementations at runtime with blas-config and lapack-config. |
| 14 |
> |
| 15 |
> It is not that simple. If numeric and scipy are to make use of atlas, atlas |
| 16 |
> has to be in their external library list at compile time. If atlas is in |
| 17 |
> the list, but not installed on the system, compilation will fail. |
| 18 |
|
| 19 |
That's why I said 'in theory' ;-) |
| 20 |
|
| 21 |
I'm currently working on a patch which just ignores any explicit atlas stuff |
| 22 |
at build time of SciPy. The idea is to build a library which simply links |
| 23 |
to /usr/lib/liblapack.so. SciPy doesn't need to know this is a symlink |
| 24 |
to /usr/lib/lapack/atlas/liblapack.so or any other implementation. |
| 25 |
|
| 26 |
Peter |
Archives