Bug#913567: python-numpy: Why depend on libatlas3-base

2019-01-05 Thread Michael Banck
Hi,

also, one of the PSI4 testcases fails with blas/lapack set to ATLAS, see
https://github.com/psi4/psi4/issues/391 and 
https://github.com/psi4/psi4/issues/1461

So having that removed for buster would be very must appreciated.


Michael



Bug#913567: python-numpy: Why depend on libatlas3-base?

2019-01-02 Thread Thorsten Glaser
Hi Sandro,

>provided by ATLAS; is this causing any problem?

yes: ATLAS does not work on all architectures (see #918076
for an example) and is generally harder to port.

I’ve seen you will upload a new version that works without,
but I wanted to still contribute this data point, so you
see that porters will also appreciate it ;-)

bye,
//mirabilos

PS: When adding pypy, please remember that also it is not
portable and will need an architecture whitelist. Thanks!
-- 
> emacs als auch vi zum Kotzen finde (joe rules) und pine für den einzig
> bedienbaren textmode-mailclient halte (und ich hab sie alle ausprobiert). ;)
Hallo, ich bin der Holger ("Hallo Holger!"), und ich bin ebenfalls
... pine-User, und das auch noch gewohnheitsmäßig ("Oooohhh").  [aus dasr]



Bug#913567: python-numpy: Why depend on libatlas3-base?

2019-01-02 Thread Jörg-Volker Peetz
Hi Sandro,

I've discussed this matter on the debian science e-mail list asking for an cblas
abstraction. I got a reply from Mo Zhou
(https://lists.debian.org/debian-science/2019/01/msg2.html ).
In short he suggested to file a bug against numpy (severity: important).

I repeat the main part of the e-mail here:
"""
Debian's default (generic, no optimization) Atlas is really SLOW, such
that I can write a faster BLAS by simply following this guide:

  https://github.com/flame/how-to-optimize-gemm

Debian's reference BLAS (netlib) contains both BLAS and CBLAS API/ABI,
which should be able to be linked against by numpy, and then we can make
numpy use a high-performance BLAS backend by simply switching
alternatives.

Any user of a modern x84_64 CPU can easily find the giant performance
gap between the default Atlas (generic code) and the default OpenBLAS
(generic+optimized, runtime-dispatch). MKL would be even faster if
the user doesn't mind installing non-free.

I guess "cblas" is supposed to mean a backend with "cblas_*" API/ABI,
and "blas" for "*_" (Fortran) API/ABI. If this is true, then at least
MKL and openblas are expected to work well.

As for numpy's linkage to libcblas.so ... I guess Sandro isn't aware of
the fact that BLAS implementations such as OpenBLAS squashed both C and
Fortran ABI into to a single shared object. And AFAIK only Atlas split
the C ABI into an individual library libcblas.so .
"""

Regards,
Jörg.



Bug#913567: python-numpy: Why depend on libatlas3-base?

2019-01-02 Thread Jörg-Volker Peetz
reopen 913567 python-numpy 1:1.16.0~rc1-3



Bug#913567: python-numpy: Why depend on libatlas3-base?

2018-12-08 Thread Jörg-Volker Peetz
Just wanted to add that openblas also provides a cblas interface in package
libopenblas-dev via the header file /usr/include/x86_64-linux-gnu/cblas.h.

Regards,
Jörg.



Bug#913567: python-numpy: Why depend on libatlas3-base?

2018-11-12 Thread Jörg-Volker Peetz
Sandro Tosi wrote on 12/11/2018 14:55:
>> the package now depends on libatlas3-base besides the dependencies on
>> BLAS/LAPACK. Couldn't this be avoided?
> 
> libcblas was required to build numpy properly and, AFAICS, it's only
> provided by ATLAS; is this causing any problem?
> 

Hi,

no it's not causing any problem at the moment.
I was just concerned about the additional blas/lapack packages, since I have
already installed libopenblas-dev and libopenblas-base.

I was just wondering why some libraries now depend on libcblas as well as
libblas but maybe this is a question better asked upstream.

Regards,
Jörg.



Bug#913567: python-numpy: Why depend on libatlas3-base?

2018-11-12 Thread Sandro Tosi
> the package now depends on libatlas3-base besides the dependencies on
> BLAS/LAPACK. Couldn't this be avoided?

libcblas was required to build numpy properly and, AFAICS, it's only
provided by ATLAS; is this causing any problem?

-- 
Sandro "morph" Tosi
My website: http://sandrotosi.me/
Me at Debian: http://wiki.debian.org/SandroTosi
G+: https://plus.google.com/u/0/+SandroTosi



Bug#913567: python-numpy: Why depend on libatlas3-base?

2018-11-12 Thread Jörg-Volker Peetz
Package: python-numpy
Version: 1:1.15.4-1
Severity: wishlist

Dear Sandro Tosi,

the package now depends on libatlas3-base besides the dependencies on
BLAS/LAPACK. Couldn't this be avoided?

Regards,
Jörg.