You are not logged in.

#1 2013-02-02 14:34:35

wesker
Member
Registered: 2013-02-02
Posts: 2

Segmentation Fault during check of the atlas_lapack package

Dear All,
I am facing a problem when i try to install atlas3.10.1 with full lapack functionality (version 3.4.2). What i do:

../ATLAS/configure --with-lapack=[lapack archive path] -D c -DWall --shared
make
make check

after the last step the following error occurs:

[home]/my_ATLAS_build/bin/ATLrun.sh[home]/my_ATLAS_build/bin xdslvtst -n 167 -r 83 -O 2 c r \
                  >> [home]/my_ATLAS_build/bin/sanity.out
[home]/my_ATLAS_build/bin/ATLrun.sh: line 5:  2456 Segmentation fault      (core dumped) $atldir/$*
make[3]: *** [dsanity_test] Error 139
make[3]: Leaving directory `[home]/my_ATLAS_build/bin'
make[2]: *** [sanity_test] Error 2
make[2]: Leaving directory `[home]/my_ATLAS_build/bin'
make[1]: *** [sanity_test] Error 2
make[1]: Leaving directory `[home]/my_ATLAS_build'
make: *** [test] Error 2

The xdslvtst which segfaults is a tool to run serialised matrix test operations. Inspecting it further gives:

[bin]$ ./xdslvtst -n 90
ORD  UPLO       N    NRHS     lda     ldb       TIME     MFLOP        RESID
===  ====  ======  ======  ======  ======  =========  ========  ===========

  C     G      90       1      90      90     -1.000     -1.00  4.019164e+00

1 TESTS RUN, ALL PASSED.

[bin]$ ./xdslvtst -n 91
ORD  UPLO       N    NRHS     lda     ldb       TIME     MFLOP        RESID
===  ====  ======  ======  ======  ======  =========  ========  ===========

Segmentation fault (core dumped)

A similar problem was reported in http://sourceforge.net/p/math-atlas/bugs/184/ but was solved in atlas3.9.56. The xdslvtst tool performs matrix test operations with the atlas library. The -n parameter should be the matrix size, it works fine for all -n smaller than 91 and fails for all bigger than 90. The problem occurs when using the PKGBUILD from the AUR and when compiling from source. Expect of the mentioned thread i could not find anybody else with such a problem, therefore i believe it is my system failing on certain memory allocation calls. My limits.conf looks like this:

*               -       rtprio          0
*               -       nice            0
@audio          -       rtprio          65
@audio          -       nice           -10
@audio          -       memlock         2500000

i already increased the memlock parameter, it did not change anything. My system:
Arch Kernel 3.7.5-1 on 64bit AMD 2.3GHz dual core 4GB Memory, Laptop.
gcc-fortran is installed, as required for the full atlas test run. If anyone has a further idea where to look for the problem i would appreciate that. If i missed to give You necessary information please let me know.

Regards,

wesker

Offline

Board footer

Powered by FluxBB