[Python-modules-team] Bug#489726: Fw: should numpy be built with atlas?

[Tiziano Zito]

Hi Ondrej, 

> $ ./test_atlas.py
> Using ATLAS:
> 0.53141283989
> $ wajig remove atlas3-base libatlas3gf-base
> $ ./test_atlas.py
> Using ATLAS:
> 1.64572000504
> 
> So it seems to work, even though the difference is not so big.

the difference is not so big because the package contains a
_dotblas.so file. When you remove the ATLAS libs, it simply uses the
available lapack+blas routines, which are slower then ATLAS but
still not so bad. If you build the package without the patch and
without using ATLAS, not _dotblas.so file is present, and the slow
down is much higher:

$ aptitude search ~i~natlas
i   libatlas3gf-sse2                - Automatically Tuned Linear Algebra Softwar
$ ldd /usr/lib/python2.5/site-packages/numpy/core/_dotblas.so 
        linux-gate.so.1 =>  (0xffffe000)
        libblas.so.3gf => /usr/lib/sse2/atlas/libblas.so.3gf (0xb79b6000)
        libgfortran.so.3 => /usr/lib/libgfortran.so.3 (0xb7904000)
        libm.so.6 => /lib/i686/cmov/libm.so.6 (0xb78dd000)
        libgcc_s.so.1 => /lib/libgcc_s.so.1 (0xb78d0000)
        libc.so.6 => /lib/i686/cmov/libc.so.6 (0xb7775000)
        /lib/ld-linux.so.2 (0x80000000)
$ python test_atlas.py 
Using ATLAS:
0.539413928986
$ aptitude purge libatlas3gf-sse2
$ python test_atlas.py 
Using ATLAS:
1.90855002403
$ ldd /usr/lib/python2.5/site-packages/numpy/core/_dotblas.so
        linux-gate.so.1 =>  (0xffffe000)
        libblas.so.3gf => /usr/lib/libblas.so.3gf (0xb7f60000)
        libgfortran.so.3 => /usr/lib/libgfortran.so.3 (0xb7eae000)
        libm.so.6 => /lib/i686/cmov/libm.so.6 (0xb7e87000)
        libgcc_s.so.1 => /lib/libgcc_s.so.1 (0xb7e7a000)
        libc.so.6 => /lib/i686/cmov/libc.so.6 (0xb7d1f000)
        /lib/ld-linux.so.2 (0x80000000)
$ mv /usr/lib/python2.5/site-packages/numpy/core/_dotblas.so /usr/lib/python2.5/site-packages/numpy/core/_dotblas.so.bak
$ python test_atlas.py 
No ATLAS:
7.52080416679

bye, 
tiziano