Something is not right.

[Jerry DeLisle]

Steve Kargl wrote:
> On Sat, Mar 03, 2007 at 12:51:03AM +0100, Tobias Burnus wrote:
>>> I updated to today's trunks source and fired off a build
>>> in my private testsuite, whcih now includes CP2K.  I managed
>>> to exhaust 8 GB of memory during the computational of 
>>> all_cp2k_gfortran.f90.  Unfortunately, this is a 432K LOC
>>> program, so reducing this to a manageable PR isn't going
>>> to happen.  During a rerun of the build I found
>>>
>>>   PID USERNAME    THR PRI NICE   SIZE    RES STATE  C   TIME   WCPU COMMAND
>>> 48973 sgk           1 113    0  1715M  1708M CPU1   0   4:36 99.07% f951
>>>   
>> Did you manage to compile it before? Here, with merely 2 GB physical
>> memory, I had lots of problems with all_cp2k_gfortran.f90 because of its
>> size. (ifort -O0 crashed with out of memory, I forgot whether it worked
>> with gfortran or not, I think at that time it stopped before due to a
>> now fixed bug.)
> 
> Yes, it built without a problem a week or two ago.  From watching
> top, I have a sinking feeling that someone other than a gfortran
> developer has introduced a memory leak.
> 
>> I therefore decided to use the CVS version of CP2K. (One should not
>> update CP2k and gfortran at the same time, otherwise one does not see
>> which of the two programs causes the regression.)
>>
>> CP2k is available from: http://cp2k.berlios.de/download.html
>>
> 
> No thanks.  I wasted a few hours trying to get it build from
> source before Joost provided the all_cp2k_gfortran.f90 file.
> Its build system is a nightmare.
> 
Joost has added gfortran into the makefiles so its fairly straight now.

Steve, could you send me the all_cp2k file.  I managed to delete it.  I would 
like to give it a valgrind run for its money.

( OT:  I see now that the latest release of lapack now provides a gfortran 
make.inc file and it builds flawlessly.  Nice. )

Jerry