On Feb 17, 3:50 pm, Gerhard Ertaler <[EMAIL PROTECTED]> wrote:
> I would like to report a minor problem compiling sage
> -----------
> * my environment:
Hi Gerhard,
> laptop running Fedora 2.6.23.15-137.fc8
>
> with the intle mkl libraries (blas, lapack) installed
>
> environmental variables CPPFLAGS and LDFLAGS
>
> are set to point to the intel include and lib directories respectively.
>
> ---------------------------
> problems building sage 2.10.1
>
> * two packages configured themselves to use intel's libraries,
>
> but tried to link (both!) -lmkl_lapack64 and -lmkl_lapack32
> (I lost the log, but I believe they were numpy and scipy)
>
> work-around: I created links in /usr/local/lib to point libmkl_lapack32.so
> and libmkl_lapack64.so to $INTEL_MKL/lib/32/libmkl_lapack.so
>
> * a question about the R package:
> I would prefer to have R installed only once, but I require some config
> switches.
> My current method is to
> 1) use "sage -sh" to set the environmental variables,
> 2) untar the R package,
> 3) modify the configure line to read
>
> ./configure --prefix="$SAGE_LOCAL" \
> --with-recommended-packages=yes \
> --enable-R-shlib \
> --with-x=$XSUPPORT \
> --with-readline="$SAGE_LOCAL" \
> --with-blas="-lmkl -lguide -lpthread" \
> --with-lapack="-lmkl_lapack"
> 4) make and install R
> 5)
> restore a saved version of the R library directory
> (where R installs optional packages)
>
> Questions:
> - is there a way to streamline this? (I could always
> write a shell script...)
> Having the script not erase the R/library subdirectory
> in the first place would be a big plus
We do erase that directory to nuke anything that might hurt on an
upgrade. I understand that this causes work for people who install
custom packages.
> - since sage developers readily modify packages,
> is there an easy way to find out whether R might
> have been
> customized
> (possibly requiring reinstallation
> of all my additional R packages?)
We currently ship a "bare bone" version of R, not even the recommended
standard packages are installed. This will change over time.
> ----------------------------------
> PS: thank you all for all the work you are doing, sage is proving
> very useful indeed!
Ok, now the final question: Why do you prefer the Intel MKL? With
ATLAS tuned for your CPU (which Sage does) ATLAS usually beats the
Intel MKL by a couple percent [2-3] for most code I have benchmarked
(LinBox, and especially the rational solver there). The more modern
the Intel CPU the smaller the gap becomes, but the Intel MKL doesn't
even beat ATLAS on Core Duos or QuadCores.
In my experience ATLAS is also superior to the Intel MKL regarding
memory usage (In case of the RationalSolver tests the Intel MKL used
consistently four times as much memory as ATLAS and GotoBLAS) and I
have been able to crash the Intel MKL 9.1 with LinBox code where ATLAS
(and also GotoBLAS) was fine. So if you do this for performance
reasons I would like to hear about it.
What we currently do not do is compile ATLAS for threaded workloads,
but judging from the FLAGS you supply to R you aren't using that
anyway. But the threaded option for ATLAS is something I am working on
and it might be done by SD8 by the end of the month.
For all the above reasons it seems unlikely that somebody will come up
with the needed bits in the build to support the Intel MKL (since it
isn't free and also "not faster"). I would be the canonical person to
do this (I have licenses for Intel MKL on Linux, Windows and MacOSX)
and I have worked a lot with different BLAS/Lapack implementations in
the past. Getting build support for the Intel MKL isn't particularly
hard to do and I would be willing to merge it under the assumption
that you would need to explicitly enable it. Note that the Intel MKL's
libraries you need to link vary wildly between even minor releases,
the 9.0->9.1 transition was especially painful.
Cheers,
Michael
> Jetzt Mails schnell in einem Vorschaufenster überfliegen. Dies und viel
> mehr bietet das neue Yahoo! Mail -www.yahoo.de/mail
--~--~---------~--~----~------------~-------~--~----~
To post to this group, send email to sage-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at http://groups.google.com/group/sage-devel
URLs: http://www.sagemath.org
-~----------~----~----~----~------~----~------~--~---
