[SciPy-dev] newcore, icc compiled ATLAS problem
Arnd Baecker
arnd.baecker at web.de
Thu Dec 22 09:09:24 EST 2005
Hi,
I am again trying to install scipy core on an Itanium with ATLAS.
FOUND:
libraries = ['lapack', 'ptf77blas', 'ptcblas', 'atlas']
library_dirs = ['/home/baecker/python2/lib/atlas']
language = f77
define_macros = [('ATLAS_INFO', '"\\"3.7.11\\""')]
include_dirs = ['/usr/include']
The problem is that ATLAS is compiled with the intel compiler.
Why is this so:
According to
http://math-atlas.sourceforge.net/errata.html#WhatComp
on Itanium 2 icc 8.0 is MUCH (sic) faster than gcc 3.3!
======================================================================
ERROR: test_basic (scipy.base.matrix.test_matrix.test_algebra)
----------------------------------------------------------------------
Traceback (most recent call last):
File
"/home/baecker/python2/scipy_lintst_atlas/lib/python2.4/site-packages/scipy/base/tests/test_matrix.py",
line 99, in test_basic
from scipy.basic import linalg
File
"/home/baecker/python2//scipy_lintst_atlas/lib/python2.4/site-packages/scipy/basic/__init__.py",
line 6, in ?
import linalg
File
"/home/baecker/python2//scipy_lintst_atlas/lib/python2.4/site-packages/scipy/basic/linalg.py",
line 9, in ?
import lapack_lite
ImportError:
/home/baecker/python2/scipy_lintst_atlas/lib/python2.4/site-packages/scipy/basic/lapack_lite.so:
undefined symbol: for_cpstr
The library `libifcore.so` provides the needed routine
nm /opt/intel/fc_90/lib/libifcore.so.6 | grep for_cpstr
00000000000d6900 T for_cpstr
(+libcxa is needed by this one ...)
So the solution is to get the following compile command
/work/home/baecker/python2/bin/g77 -shared build/temp.linux-ia64-2.4/scipy/basic/lapack_lite/lapack_litemodule.o
-L/home/baecker/python2/lib/atlas -llapack -lptf77blas -lptcblas -latlas -lg2c -L /opt/intel/fc_90/lib -lifcore
-lcxa -o build/lib.linux-ia64-2.4/scipy/basic/lapack_lite.so
I.e.:
-L/opt/intel/fc_90/lib -lifcore -lcxa
has to be added
and then all tests of scipy core work fine.
How can one achieve this?
Best, Arnd
More information about the SciPy-Dev
mailing list