On 03 Oct 2011, at 14:06, Dominique Dhumieres wrote:
> Now what worries me more is that benchmark_dtmm(1000,1000,10) gives
>
> Time for A'*B (A,B m-by-n matrices): 1.72362 on a 2.53Ghz Core2Duo
>
> and
>
> Time for A'*B (A,B m-by-n matrices): 0.508563 on a 1.8Ghz G5
>
> If I understand the meaning of these numbers, this means ~4GFlops
> on the G5 (increasing to 5GFlops for n=m=4000, IIRC Atlas gives
> ~5.6GFlops
> out of the peak 7.2GFlops), but only 1.2GFlops for the Core2Duo
> (Atlas or
> the accelerated framework give ~11GFlops out of the peak 20GFlops).
>
> 1.1GFlops on my intel box corresponds to what I get when compiling
> lapack
> with gfortran which make me suspicious that octave (plain variant)
> is not
> linked against -Wl,-framework,Accelerate. Could someone check what
> is wrong
> with the above (my conclusions or the lapack library used)?
Since 'dead_strip_dylibs' is used in the build of octave, the following
command list all binaries of the octave-atlas family that link against
atlas.
# dpkg -L octave-atlas octave324-atlas-shlibs octave324-atlas-dev
octave324-atlas|xargs file|fgrep 'Mach-O'|cut -f1 -d:|xargs otool -L|
fgrep -B4 -e blas -e lapack
/sw64/lib/octave-3.2.4/libcruft.3.2.4.dylib:
/sw64/lib/octave-3.2.4/libcruft.dylib (compatibility version 0.0.0,
current version 0.0.0)
/sw64/lib/liblapack.dylib (compatibility version 0.0.0, current
version 3.9.32)
/sw64/lib/libf77blas.dylib (compatibility version 0.0.0, current
version 3.9.32)
--
/sw64/lib/octave-3.2.4/liboctave.3.2.4.dylib:
/sw64/lib/octave-3.2.4/liboctave.dylib (compatibility version 0.0.0,
current version 0.0.0)
/sw64/lib/octave-3.2.4/libcruft.dylib (compatibility version 0.0.0,
current version 0.0.0)
/sw64/lib/libqrupdate.1.dylib (compatibility version 0.0.0, current
version 0.0.0)
/sw64/lib/liblapack.dylib (compatibility version 0.0.0, current
version 3.9.32)
/sw64/lib/libf77blas.dylib (compatibility version 0.0.0, current
version 3.9.32)
--
/sw64/lib/octave/3.2.4/oct/x86_64-apple-darwin/eigs.oct:
/sw64/lib/octave-3.2.4/libcruft.dylib (compatibility version 0.0.0,
current version 0.0.0)
/sw64/lib/octave-3.2.4/liboctave.dylib (compatibility version 0.0.0,
current version 0.0.0)
/sw64/lib/octave-3.2.4/liboctinterp.dylib (compatibility version
0.0.0, current version 0.0.0)
/sw64/lib/libf77blas.dylib (compatibility version 0.0.0, current
version 3.9.32)
--
/sw64/lib/octave/3.2.4/oct/x86_64-apple-darwin/qz.oct:
/sw64/lib/octave-3.2.4/libcruft.dylib (compatibility version 0.0.0,
current version 0.0.0)
/sw64/lib/octave-3.2.4/liboctave.dylib (compatibility version 0.0.0,
current version 0.0.0)
/sw64/lib/octave-3.2.4/liboctinterp.dylib (compatibility version
0.0.0, current version 0.0.0)
/sw64/lib/liblapack.dylib (compatibility version 0.0.0, current
version 3.9.32)
Dominique, do you get the same binaries with the analogous command for
the non-atlas version ?
JF
------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure contains a
definitive record of customers, application performance, security
threats, fraudulent activity and more. Splunk takes this data and makes
sense of it. Business sense. IT sense. Common sense.
http://p.sf.net/sfu/splunk-d2dcopy1
_______________________________________________
Fink-users mailing list
Fink-users@lists.sourceforge.net
List archive:
http://news.gmane.org/gmane.os.macosx.fink.user
Subscription management:
https://lists.sourceforge.net/lists/listinfo/fink-users
