Mark,
Actually if the build finish without a segfaut, you're fine.
I would build again the Reference BLAS first and then ATLAS.
Here is a little script that may help you.
You should have only Numerical Errors (between 200 and 500) Those are torture
tests and we know some are failing.
cat *.out > testing_results.txt
echo " ****************************************************** "
echo " RESULTS"
echo " - numerical failure: "
grep -i "out of" testing_results.txt | grep drivers
grep -i "out of" testing_results.txt | grep -v drivers
grep -i "largest test error" testing_results.txt | grep -v "D\-" | grep -v
"E\-" | grep -v "+00"
grep -i "Error in" testing_results.txt
nb_illegal=`grep -c -i illegal testing_results.txt`
echo " - illegal error: $nb_illegal "
info_error=`grep -i -c -h "error code" testing_results.txt`
echo " - info error: $info_error "
nb_signal=`grep -i -h signal testing_results.txt | grep -c -v toto`
nb_exception=`grep -i -h exception testing_results.txt | grep -c -v "flags
raised"`
severe_error=`expr $nb_signal + $nb_exception`
echo " - severe error $severe_error"
Regards,
Julie
On Mar 22, 2011, at 9:16 AM, Mark Dixon wrote:
On Mon, 21 Mar 2011, julie langou wrote:
Thank you Mark for reporting the dead link. This has been corrected.
Thanks Julie.
It seems that you already went through :
http://www.netlib.org/lapack/faq.html#_how_do_i_interpret_lapack_testing_failures
and it seems it did not help. What are your output errors? maybe I can help
you understand them
Sincerely,
Julie
Thanks for the offer - I must admit that I feel that I'm thrashing about a
bit on this one.
I've been trying to install a BLAS/LAPACK stack, based on ATLAS 3.8.3 and
LAPACK 3.3.0, on our main compute cluster running RHEL5 64-bit. Perhaps being
a little too eager, I trying to build it against every variant of compiler on
the system and in both 32-bit and 64-bit modes.
After reading the FAQ and some of the forum entries, I've hit upon the
following method for analysing the test output files to see if a build is
"good" or not:
1) grep for 'failed to pass'
2) grep for ' result ' to obtain the results of test failures
3) Take (2) and remove minor errors - any result that is a small number
(where small is defined here as a number not in scientific notation)
4) Ignore anything from files [sdcz]gd.out
Is this a reasonable definition?
Based on this, I'm still working-through the advice in the FAQ (varying
compiler flags and BLAS implementation).
Thanks,
Mark
--
-----------------------------------------------------------------
Mark Dixon Email : m.c.dixon@Domain.Removed
HPC/Grid Systems Support Tel (int): 35429
Information Systems Services Tel (ext): +44(0)113 343 5429
University of Leeds, LS2 9JT, UK
-----------------------------------------------------------------
**********************************************
Julie Langou; Research Associate in Computer Science
Innovative Computing Laboratory;
University of Tennessee from Denver, Colorado ;-)
julie@Domain.Removed; http://www.cs.utk.edu/~julie/
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
http://lists.eecs.utk.edu/mailman/private/lapack/attachments/20110322/f8e97d80/attachment.html
|