Memory leaks?

Axel akoh... at gmail.com
Fri Jul 13 23:57:11 CEST 2007


hi just a short followup.

the memory leak problem seems related to recent changes in input
handling and are so far a problem of PGI only. i ran serial
executables
of the current cvs code compiled with intel 9.0, 9.1 and pgi 7.0.4
(latest installed/available versions/patches for each).

using a modified tests/QS/H2-md.inp (MOTION%MD%STEPS set to 10) i get
with intel:

==15836== LEAK SUMMARY:
==15836==    definitely lost: 0 bytes in 0 blocks.
==15836==      possibly lost: 0 bytes in 0 blocks.
==15836==    still reachable: 17,571,474 bytes in 14,552 blocks.
==15836==         suppressed: 0 bytes in 0 blocks.

with PGI:
==15807== LEAK SUMMARY:
==15807==    definitely lost: 0 bytes in 0 blocks.
==15807==      possibly lost: 0 bytes in 0 blocks.
==15807==    still reachable: 345,039,988 bytes in 874,172 blocks.
==15807==         suppressed: 0 bytes in 0 blocks.

==15807== ERROR SUMMARY: 324 errors from 5 contexts (suppressed: 4
from 1)
==15807==
==15807== 1 errors in context 1 of 5:
==15807== Conditional jump or move depends on uninitialised value(s)
==15807==    at 0x18DE7BB: fr_readnum (in /home/akohlmey/compile/cp2k/
tests/QS/cp2k.sopt)
==15807==    by 0x18DDE50: fr_read (in /home/akohlmey/compile/cp2k/
tests/QS/cp2k.sopt)
==15807==    by 0x18DDBA2: __f90io_fmt_read (in /home/akohlmey/compile/
cp2k/tests/QS/cp2k.sopt)
==15807==    by 0x18DDCC6: pgf90io_fmt_read (in /home/akohlmey/compile/
cp2k/tests/QS/cp2k.sopt)
==15807==    by 0x5B35C1: cp_units_cp_unit_create_ (in /home/akohlmey/
compile/cp2k/tests/QS/cp2k.sopt)
==15807==
==15807== 20 errors in context 2 of 5:
==15807== Conditional jump or move depends on uninitialised value(s)
==15807==    at 0x18FC310: pgf90_trim (in /home/akohlmey/compile/cp2k/
tests/QS/cp2k.sopt)
==15807==    by 0x893FE3: input_cp2k_restarts_section_coord_val_set_
(in /home/akohlmey/compile/cp2k/tests/QS/cp2k.sopt)
==15807==
==15807== 24 errors in context 3 of 5:
==15807== Conditional jump or move depends on uninitialised value(s)
==15807==    at 0x18FC310: pgf90_trim (in /home/akohlmey/compile/cp2k/
tests/QS/cp2k.sopt)
==15807==    by 0x58A394:
cp_output_handling_cp_print_key_generate_filename_ (in /home/akohlmey/
compile/cp2k/tests/QS/cp2k.sopt)
==15807==
==15807== 103 errors in context 4 of 5:
==15807== Conditional jump or move depends on uninitialised value(s)
==15807==    at 0x18FC310: pgf90_trim (in /home/akohlmey/compile/cp2k/
tests/QS/cp2k.sopt)
==15807==    by 0x8C83D2: input_val_types_val_get_ (in /home/akohlmey/
compile/cp2k/tests/QS/cp2k.sopt)
==15807==
==15807== 176 errors in context 5 of 5:
==15807== Conditional jump or move depends on uninitialised value(s)
==15807==    at 0x18FC310: pgf90_trim (in /home/akohlmey/compile/cp2k/
tests/QS/cp2k.sopt)
==15807==    by 0x8C9D70: input_val_types_val_write_ (in /home/
akohlmey/compile/cp2k/tests/QS/cp2k.sopt)

so there is a clear common theme with pgf90_trim(),
i.e. the TRIM intrinsic from PGI) having/being a problem.

from looking at the changelog there were some input related
changes recently, but only the people having implemented
them can tell, whether there is an alternate way to do
the same and not trigger this (compiler) bug... :-(

cheers,
   axel.


On Jul 13, 12:59 pm, lmliu <lml... at gmail.com> wrote:
> Thanks a lot for all answers.
> I check the regests on cp2k web, which also shows the memory leaks for
> recent days. This may expalin it.
> I use parallel job using PGI6.2 on linux x86_64 machine.
>
> Good Weekend
>
> Limin




More information about the CP2K-user mailing list