[SciPy-user] Problems with ACML

[Ryan May]

Hi,

Does anyone here use the AMD Core Math Libraries (ACML) as their
underlying libraries for BLAS/LAPACK/etc. in SciPy? I have problems
with (at least) scipy.linalg.eigvals (Fernando discovered this at SciPy
on my laptop). For instance, the following crashes reliably with ACML,
but works fine with ATLAS versions of BLAS/LAPACK:

>>>from scipy.linalg import eigvals
>>>from numpy.random import rand
>>>a = rand(100,100)
>>>eigvals(a)

Anyone else have this problem? Is ACML known to be a bad thing to use
with SciPy?

Thanks,

Ryan

--
Ryan May
Graduate Research Assistant
School of Meteorology
University of Oklahoma

_______________________________________________
SciPy-user mailing list
SciPy...@scipy.org
http://projects.scipy.org/mailman/listinfo/scipy-user

[Robert Kern]

Ryan May wrote:
> Hi,
>
> Does anyone here use the AMD Core Math Libraries (ACML) as their
> underlying libraries for BLAS/LAPACK/etc. in SciPy? I have problems
> with (at least) scipy.linalg.eigvals (Fernando discovered this at SciPy
> on my laptop). For instance, the following crashes reliably with ACML,
> but works fine with ATLAS versions of BLAS/LAPACK:
>
>>> >from scipy.linalg import eigvals
>>> >from numpy.random import rand
>>>> a = rand(100,100)
>>>> eigvals(a)
>
> Anyone else have this problem? Is ACML known to be a bad thing to use
> with SciPy?

No, but it's not an option that's as thoroughly tested as ATLAS, either. If you
could supply a gdb backtrace from the crash, that would help locate the problem.

--
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless enigma
that is made terrible by our own mad attempt to interpret it as though it had
an underlying truth."
-- Umberto Eco

[George Nurser]

On 28/08/07, Ryan May <rm...@ou.edu> wrote:
> Hi,
>
> Does anyone here use the AMD Core Math Libraries (ACML) as their
> underlying libraries for BLAS/LAPACK/etc. in SciPy? I have problems
> with (at least) scipy.linalg.eigvals (Fernando discovered this at SciPy
> on my laptop). For instance, the following crashes reliably with ACML,
> but works fine with ATLAS versions of BLAS/LAPACK:
>
> >>>from scipy.linalg import eigvals
> >>>from numpy.random import rand
> >>>a = rand(100,100)
> >>>eigvals(a)
>
> Anyone else have this problem? Is ACML known to be a bad thing to use
> with SciPy?

Works for me, no problems.
I use the gnu 64 bit version of acml; I think it's v 3.6.0
If you wish i can give a description of how I installed it.

-George.

[Ryan May]

Robert Kern wrote:
> Ryan May wrote:
>> Hi,
>>
>> Does anyone here use the AMD Core Math Libraries (ACML) as their
>> underlying libraries for BLAS/LAPACK/etc. in SciPy? I have problems
>> with (at least) scipy.linalg.eigvals (Fernando discovered this at SciPy
>> on my laptop). For instance, the following crashes reliably with ACML,
>> but works fine with ATLAS versions of BLAS/LAPACK:
>>
>>>> >from scipy.linalg import eigvals
>>>> >from numpy.random import rand
>>>>> a = rand(100,100)
>>>>> eigvals(a)
>> Anyone else have this problem? Is ACML known to be a bad thing to use
>> with SciPy?
>
> No, but it's not an option that's as thoroughly tested as ATLAS, either. If you
> could supply a gdb backtrace from the crash, that would help locate the problem.
>

Well, right now this is what I get (having updated scipy):

Python 2.5.1 (r251:54863, Jul 27 2007, 10:40:24)
[GCC 4.1.2 (Gentoo 4.1.2)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> from scipy.linalg import eigvals
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/lib64/python2.5/site-packages/scipy/linalg/__init__.py",
line 8, in <module>
from basic import *
File "/usr/lib64/python2.5/site-packages/scipy/linalg/basic.py", line
17, in <module>
from lapack import get_lapack_funcs
File "/usr/lib64/python2.5/site-packages/scipy/linalg/lapack.py", line
18, in <module>
from scipy.linalg import clapack
ImportError: /usr/lib64/python2.5/site-packages/scipy/linalg/clapack.so:
undefined symbol: clapack_sgesv
>>> scipy.__version__
'0.5.2.1'

Any ACML users out there have any idea? This is on AMD64 w/
gfortran/int64 edition of ACML 3.6.1.

Ryan

--
Ryan May
Graduate Research Assistant
School of Meteorology
University of Oklahoma

[Ryan May]

> Well, right now this is what I get (having updated scipy):
>
> Python 2.5.1 (r251:54863, Jul 27 2007, 10:40:24)
> [GCC 4.1.2 (Gentoo 4.1.2)] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
>>>> from scipy.linalg import eigvals
> Traceback (most recent call last):
> File "<stdin>", line 1, in <module>
> File "/usr/lib64/python2.5/site-packages/scipy/linalg/__init__.py",
> line 8, in <module>
> from basic import *
> File "/usr/lib64/python2.5/site-packages/scipy/linalg/basic.py", line
> 17, in <module>
> from lapack import get_lapack_funcs
> File "/usr/lib64/python2.5/site-packages/scipy/linalg/lapack.py", line
> 18, in <module>
> from scipy.linalg import clapack
> ImportError: /usr/lib64/python2.5/site-packages/scipy/linalg/clapack.so:
> undefined symbol: clapack_sgesv
>>>> scipy.__version__
> '0.5.2.1'
>
> Any ACML users out there have any idea? This is on AMD64 w/
> gfortran/int64 edition of ACML 3.6.1.
>
> Ryan
>

Well, it appears that the missing symbol was due to some bad combination
of ACML _and_ ATLAS. Removing atlas and recompiling scipy/numpy
resolved it.

As far as the original crash goes, my problem was that I had the int64
version of ACML installed. So for you ACML users out there, don't use
the int64 version unless the code you're using actually uses INTERGER*8.