Hi,

On Fri, May 9, 2014 at 4:06 AM, David Cournapeau <cournape@gmail.com> wrote:
>
>
>
> On Fri, May 9, 2014 at 11:49 AM, Julian Taylor
> <jtaylor.debian@googlemail.com> wrote:
>>
>> On 09.05.2014 12:42, David Cournapeau wrote:
>> >
>> >
>> >
>> > On Fri, May 9, 2014 at 1:51 AM, Matthew Brett <matthew.brett@gmail.com
>> > <mailto:matthew.brett@gmail.com>> wrote:
>> >
>> >     Hi,
>> >
>> >     On Mon, Apr 28, 2014 at 3:29 PM, David Cournapeau
>> >     <cournape@gmail.com <mailto:cournape@gmail.com>> wrote:
>> >     >
>> >     >
>> >     >
>> >     > On Sun, Apr 27, 2014 at 11:50 PM, Matthew Brett
>> >     <matthew.brett@gmail.com <mailto:matthew.brett@gmail.com>>
>> >     > wrote:
>> >     >>
>> >     >> Aha,
>> >     >>
>> >     >> On Sun, Apr 27, 2014 at 3:19 PM, Matthew Brett
>> >     <matthew.brett@gmail.com <mailto:matthew.brett@gmail.com>>
>> >     >> wrote:
>> >     >> > Hi,
>> >     >> >
>> >     >> > On Sun, Apr 27, 2014 at 3:06 PM, Carl Kleffner
>> >     <cmkleffner@gmail.com <mailto:cmkleffner@gmail.com>>
>> >     >> > wrote:
>> >     >> >> A possible option is to install the toolchain inside
>> >     site-packages and
>> >     >> >> to
>> >     >> >> deploy it as PYPI wheel or wininst packages. The PATH to the
>> >     toolchain
>> >     >> >> could
>> >     >> >> be extended during import of the package. But I have no idea,
>> >     whats the
>> >     >> >> best
>> >     >> >> strategy to additionaly install ATLAS or other third party
>> >     libraries.
>> >     >> >
>> >     >> > Maybe we could provide ATLAS binaries for 32 / 64 bit as part
>> >     of the
>> >     >> > devkit package.  It sounds like OpenBLAS will be much easier to
>> >     build,
>> >     >> > so we could start with ATLAS binaries as a default, expecting
>> >     OpenBLAS
>> >     >> > to be built more often with the toolchain.  I think that's how
>> >     numpy
>> >     >> > binary installers are built at the moment - using old binary
>> >     builds of
>> >     >> > ATLAS.
>> >     >> >
>> >     >> > I'm happy to provide the builds of ATLAS - e.g. here:
>> >     >> >
>> >     >> > https://nipy.bic.berkeley.edu/scipy_installers/atlas_builds
>> >     >>
>> >     >> I just found the official numpy binary builds of ATLAS:
>> >     >>
>> >     >> https://github.com/numpy/vendor/tree/master/binaries
>> >     >>
>> >     >> But - they are from an old version of ATLAS / Lapack, and only
>> >     for 32-bit.
>> >     >>
>> >     >> David - what say we update these to latest ATLAS stable?
>> >     >
>> >     >
>> >     > Fine by me (not that you need my approval !).
>> >     >
>> >     > How easy is it to build ATLAS targetting a specific CPU these days
>> >     ? I think
>> >     > we need to at least support nosse and sse2 and above.
>> >
>> >     I'm getting crashes trying to build SSE2-only ATLAS on 32-bits, I
>> >     think Clint will have some time to help out next week.
>> >
>> >     I did some analysis of SSE2 prevalence here:
>> >
>> >     https://github.com/numpy/numpy/wiki/Window-versions
>> >
>> >     Firefox crash reports now have about 1 percent of machines without
>> >     SSE2.  I suspect that people running new installs of numpy will have
>> >     slightly better machines on average than Firefox users, but it's
>> > only
>> >     a guess.
>> >
>> >     I wonder if we could add a CPU check on numpy import to give a
>> > polite
>> >     'install from the exe' message for people without SSE2.
>> >
>> >
>> > We could, although you unfortunately can't do it easily from ctypes only
>> > (as you need some ASM).
>> >
>> > I can take a quick look at a simple cython extension that could be
>> > imported before anything else, and would raise an ImportError if the
>> > wrong arch is detected.
>> >
>>
>> assuming mingw is new enough
>>
>> #ifdef __SSE2___
>> raise_if(!__builtin_cpu_supports("sse"))
>> #endof
>
>
> We need to support it for VS as well, but it looks like win32 API has a
> function to do it:
> http://msdn.microsoft.com/en-us/library/ms724482%28VS.85%29.aspx
>
> Makes it even easier.

Nice.  So all we would need is something like:

try:
    from ctypes import windll, wintypes
except (ImportError, ValueError):
    pass
else:
   has_feature = windll.kernel32.IsProcessorFeaturePresent
   has_feature.argtypes = [wintypes.DWORD]
   if not has_feature(10):
       msg = ("This version of numpy needs a CPU capable of SSE2, "
              "but Windows says - not so.\n",
              "Please reinstall numpy using a superpack installer")
       raise RuntimeError(msg)

At the top of numpy/__init__.py

What would be the best way of including that code in the 32-bit wheel?
 (The 64-bit wheel can depend on SSE2).

Cheers,

Matthew

_______________________________________________
NumPy-Discussion mailing list
NumPy-Discussion@scipy.org
http://mail.scipy.org/mailman/listinfo/numpy-discussion