Hi all, Are there plans to provide official 64bit Windows installers for NumPy? I'm aware that Christoph Gohlke had been able to do this, since he offers unofficial plain builds and MKL builds for NumPy here: http://www.lfd.uci.edu/~gohlke/pythonlibs/ Regards, Peter
I believe the problem has been been 64 bit fortran for ATLAS, the mingw version has/had problems. A plain build using the MS compilers works fine without ATLAS. Chuck On Mon, Jan 17, 2011 at 7:05 AM, Peter < numpy-discussion@maubp.freeserve.co.uk> wrote:
Hi all,
Are there plans to provide official 64bit Windows installers for NumPy?
I'm aware that Christoph Gohlke had been able to do this, since he offers unofficial plain builds and MKL builds for NumPy here: http://www.lfd.uci.edu/~gohlke/pythonlibs/<http://www.lfd.uci.edu/%7Egohlke/pythonlibs/>
Regards,
Peter _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
On Tue, Jan 18, 2011 at 1:02 PM, Charles R Harris <charlesr.harris@gmail.com> wrote:
I believe the problem has been been 64 bit fortran for ATLAS, the mingw version has/had problems. A plain build using the MS compilers works fine without ATLAS.
Chuck
Do you think there would be interest/demand for official non-ATLAS binaries as a short term solution? I'm thinking also of 3rd party Python libraries that use NumPy, and if they/we can ship a win64 installer if NumPy doesn't. Peter
On Wed, Jan 19, 2011 at 1:24 AM, Peter < numpy-discussion@maubp.freeserve.co.uk> wrote:
On Tue, Jan 18, 2011 at 1:02 PM, Charles R Harris <charlesr.harris@gmail.com> wrote:
I believe the problem has been been 64 bit fortran for ATLAS, the mingw version has/had problems. A plain build using the MS compilers works fine without ATLAS.
Chuck
Do you think there would be interest/demand for official non-ATLAS binaries as a short term solution?
I doubt this would add much to what's currently available unofficially.
I'm thinking also of 3rd party Python libraries that use NumPy, and if they/we can ship a win64 installer if NumPy doesn't.
This is no problem of course. If I were you though, I would first consider if it's not better to refer your users to the Enthought version, or the builds provided by Christoph Gohlke for example. Ralf
On Sat, Jan 22, 2011 at 12:28 PM, Ralf Gommers <ralf.gommers@googlemail.com> wrote:
On Wed, Jan 19, 2011 at 1:24 AM, Peter <numpy-discussion@maubp.freeserve.co.uk> wrote:
On Tue, Jan 18, 2011 at 1:02 PM, Charles R Harris <charlesr.harris@gmail.com> wrote:
I believe the problem has been been 64 bit fortran for ATLAS, the mingw version has/had problems. A plain build using the MS compilers works fine without ATLAS.
Chuck
Do you think there would be interest/demand for official non-ATLAS binaries as a short term solution?
I doubt this would add much to what's currently available unofficially.
But it would be "official", which counts for something - especially in commercial setting.
I'm thinking also of 3rd party Python libraries that use NumPy, and if they/we can ship a win64 installer if NumPy doesn't.
This is no problem of course. If I were you though, I would first consider if it's not better to refer your users to the Enthought version, or the builds provided by Christoph Gohlke for example.
We're currently pointing people on 64 bit Windows towards Christoph Gohlke's unofficial builds. I'd be quite happy if Christoph's 64bit NumPy installer was blessed as official and distributed via the NumPy website (but there may be technical issues I'm unaware of). Peter
On Sat, Jan 22, 2011 at 10:20 PM, Peter < numpy-discussion@maubp.freeserve.co.uk> wrote:
On Sat, Jan 22, 2011 at 12:28 PM, Ralf Gommers <ralf.gommers@googlemail.com> wrote:
On Wed, Jan 19, 2011 at 1:24 AM, Peter <numpy-discussion@maubp.freeserve.co.uk> wrote:
On Tue, Jan 18, 2011 at 1:02 PM, Charles R Harris <charlesr.harris@gmail.com> wrote:
I believe the problem has been been 64 bit fortran for ATLAS, the
mingw
version has/had problems. A plain build using the MS compilers works fine without ATLAS.
Chuck
Do you think there would be interest/demand for official non-ATLAS binaries as a short term solution?
I doubt this would add much to what's currently available unofficially.
But it would be "official", which counts for something - especially in commercial setting.
I'm thinking also of 3rd party Python libraries that use NumPy, and if they/we can ship a win64 installer if NumPy doesn't.
This is no problem of course. If I were you though, I would first consider if it's not better to refer your users to the Enthought version, or the builds provided by Christoph Gohlke for example.
We're currently pointing people on 64 bit Windows towards Christoph Gohlke's unofficial builds. I'd be quite happy if Christoph's 64bit NumPy installer was blessed as official and distributed via the NumPy website (but there may be technical issues I'm unaware of). <http://mail.scipy.org/mailman/listinfo/numpy-discussion>
The plain builds don't work with scipy as I think you know, which IMHO means they should not be official. The MKL ones should not be official because they're non-free. That said, if others feel that plain official builds are useful *and* someone steps up to create and troubleshoot them, then of course that's fine with me. Cheers, Ralf
On Sat, Jan 22, 2011 at 5:43 PM, Ralf Gommers wrote:
On Sat, Jan 22, 2011 at 10:20 PM, Peter wrote:
We're currently pointing people on 64 bit Windows towards Christoph Gohlke's unofficial builds. I'd be quite happy if Christoph's 64bit NumPy installer was blessed as official and distributed via the NumPy website (but there may be technical issues I'm unaware of).
The plain builds don't work with scipy as I think you know, which IMHO means they should not be official.
I see (on closer inspection) that Christoph only has 64bit SciPy for the Intel MKL version of 64bit NumPy (not the plain version), which is presumably what you are refering to? Is this down to some problem in SciPy or NumPy (or both)? I'm not familiar with the problem.
The MKL ones should not be official because they're non-free.
Of course - Intel's licensing must be respected there.
That said, if others feel that plain official builds are useful *and* someone steps up to create and troubleshoot them, then of course that's fine with me.
Cheers, Ralf
:) Peter
On Sun, Jan 23, 2011 at 4:26 AM, Peter <numpy-discussion@maubp.freeserve.co.uk> wrote:
I see (on closer inspection) that Christoph only has 64bit SciPy for the Intel MKL version of 64bit NumPy (not the plain version), which is presumably what you are refering to? Is this down to some problem in SciPy or NumPy (or both)? I'm not familiar with the problem.
The main issue is that there is no easy way to get a working implementation of blas/lapack for windows 64 bits, nor a working, freely available fortran compiler. This is not so much a limitation of numpy/scipy, but rather one of lack of open source tools maturity on that platform. As Intel license requires developers to get a license of their tools, that's not an acceptable solution for the official version of numpy/scipy. cheers, David
But if you have an Intel license, you are allows to redistribute the MKL runtime. This is the reason why Enthought can distribute EPD which includes numpy, scipy and numexpr linked against MKL. The next EPD 7.0, will include Python 2.7.1 and numpy 1.5.1 and be released on February 8. - Ilan On Sat, Jan 22, 2011 at 6:34 PM, David Cournapeau <cournape@gmail.com> wrote:
On Sun, Jan 23, 2011 at 4:26 AM, Peter <numpy-discussion@maubp.freeserve.co.uk> wrote:
I see (on closer inspection) that Christoph only has 64bit SciPy for the Intel MKL version of 64bit NumPy (not the plain version), which is presumably what you are refering to? Is this down to some problem in SciPy or NumPy (or both)? I'm not familiar with the problem.
The main issue is that there is no easy way to get a working implementation of blas/lapack for windows 64 bits, nor a working, freely available fortran compiler.
This is not so much a limitation of numpy/scipy, but rather one of lack of open source tools maturity on that platform. As Intel license requires developers to get a license of their tools, that's not an acceptable solution for the official version of numpy/scipy.
cheers,
David _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
participants (5)
-
Charles R Harris -
David Cournapeau -
Ilan Schnell -
Peter -
Ralf Gommers