The blas implementation you are using may be slow. Here's my ldd on
_dotblas.so, that shows it is using libblas (this is on Ubuntu 11.04):<br><br> linux-vdso.so.1 => (0x00007fffad5ff000)<br> libblas.so.3gf => /usr/lib/libblas.so.3gf (0x00007fc608ea4000)<br>
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.<div id=":2ao">6 (0x00007fc608b10000)<br> libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fc60882b000)<br> libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fc6085a6000)<br>
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fc608390000)<br> /lib64/ld-linux-x86-64.so.2 (0x00007fc609352000)<br><br>My guess is BLAS is implemented in libsunmath.so but is a slow implementation. It's really just a guess though...<br>
<br>-=- Olivier<br></div><br><div class="gmail_quote">2011/9/20 David Cottrell <span dir="ltr"><<a href="mailto:david.cottrell@gmail.com">david.cottrell@gmail.com</a>></span><br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
Thanks, just getting back to this. I just checked again, and after<br>
setting my LD_LIBRARY_PATH properly, ldd shows _dotblas.so pointing<br>
and the sunmath and sunperf libraries. However the test_03.py still<br>
runs at about 8-9 seconds ... far too slow.<br>
<br>
~/local/lib/python3.1/site-packages/numpy/core $ ldd _dotblas.so | sed<br>
-e 's/$me/$USERNAME/g'<br>
libsunperf.so.8 =><br>
/home/$USERNAME/local/archive/SolarisStudio12.2-solaris-sparc-tar-ML/solstudio12.2/lib//libsunperf.so.8<br>
libsunmath.so.1 =><br>
/home/$USERNAME/local/archive/SolarisStudio12.2-solaris-sparc-tar-ML/solstudio12.2/lib//libsunmath.so.1<br>
libgcc_s.so.1 => /usr/sfw/lib/libgcc_s.so.1<br>
libfsu.so.1 =><br>
/home/$USERNAME/local/archive/SolarisStudio12.2-solaris-sparc-tar-ML/solstudio12.2/lib//libfsu.so.1<br>
libfui.so.2 =><br>
/home/$USERNAME/local/archive/SolarisStudio12.2-solaris-sparc-tar-ML/solstudio12.2/lib//libfui.so.2<br>
libpicl.so.1 => /usr/lib/libpicl.so.1<br>
libmtsk.so.1 => /lib/libmtsk.so.1<br>
libm.so.2 => /lib/libm.so.2<br>
libc.so.1 => /lib/libc.so.1<br>
libm.so.1 => /lib/libm.so.1<br>
libdl.so.1 => /lib/libdl.so.1<br>
libdoor.so.1 => /lib/libdoor.so.1<br>
libthread.so.1 => /lib/libthread.so.1<br>
libkstat.so.1 => /lib/libkstat.so.1<br>
libpthread.so.1 => /lib/libpthread.so.1<br>
librt.so.1 => /lib/librt.so.1<br>
libaio.so.1 => /lib/libaio.so.1<br>
libmd.so.1 => /lib/libmd.so.1<br>
/platform/SUNW,Sun-Fire-V490/lib/libc_psr.so.1<br>
/platform/SUNW,Sun-Fire-V490/lib/libmd_psr.so.1<br>
<br>
~/local/lib/python3.1/site-packages/numpy/core $ ~/python/numpy/B/test_03.py<br>
No ATLAS:<br>
8.49377894402<br>
(1000, 1000) (1000,) (1000, 1000)<br>
<br>
<br>
On Wed, Sep 7, 2011 at 9:08 AM, Samuel John <<a href="mailto:scipy@samueljohn.de">scipy@samueljohn.de</a>> wrote:<br>
><br>
> On 06.09.2011, at 22:13, David Cottrell wrote:<br>
><br>
>> Thanks, I didn't realize dot was not just calling dgemm or some<br>
>> variant which I assume would be reasonably fast. I see dgemm appears<br>
>> in the numpy code in various places such as the lapack_lite module.<br>
>><br>
>> I ran the svd test on the solaris setup and will check the OSX run<br>
>> when back at my laptop. 8.4 seconds is slightly slower than matlab but<br>
>> still seems reasonable.<br>
>><br>
>> $ ./test_03.py<br>
>> No ATLAS:<br>
>> (1000, 1000) (1000,) (1000, 1000)<br>
>> 8.17235898972<br>
><br>
> I just ran your benchmark code on OSX 10.7.1 on an 2011 MacBook Pro (core-i7) with numpy.version.version '2.0.0.dev-900d82e':<br>
> Using ATLAS:<br>
> ((1000, 1000), (1000,), (1000, 1000))<br>
> 0.908223152161<br>
><br>
> cheers,<br>
> Samuel<br>
> _______________________________________________<br>
> NumPy-Discussion mailing list<br>
> <a href="mailto:NumPy-Discussion@scipy.org">NumPy-Discussion@scipy.org</a><br>
> <a href="http://mail.scipy.org/mailman/listinfo/numpy-discussion" target="_blank">http://mail.scipy.org/mailman/listinfo/numpy-discussion</a><br>
><br>
<font color="#888888"><br>
<br>
<br>
--<br>
David Cottrell<br>
<a href="tel:%2B1%20416%20995%209860" value="+14169959860">+1 416 995 9860</a><br>
<a href="http://ca.linkedin.com/in/dcottrell" target="_blank">http://ca.linkedin.com/in/dcottrell</a><br>
_______________________________________________<br>
NumPy-Discussion mailing list<br>
<a href="mailto:NumPy-Discussion@scipy.org">NumPy-Discussion@scipy.org</a><br>
<a href="http://mail.scipy.org/mailman/listinfo/numpy-discussion" target="_blank">http://mail.scipy.org/mailman/listinfo/numpy-discussion</a><br>
</font></blockquote></div><br>