Memory consumption increases during event generation

[sschmidt]

The memory consumption by WHIZARD as shown top by increases during the generation of events.

A mtrace=all check using nagfor showed that

LEAK: Allocation 2002495 (size 16) = Z'43B65570' at line 596 of state_matrices.f90

was the only leak shown whose occurrence increased with increasing number of events generated. The section from state_matrices.f90 is

    if (associated (state%root)) then
       if (allocated (state%me))  deallocate (state%me)
       allocate (state%me (state%n_matrix_elements))   <--This is line 596
       state%me = 0
    end if

[sschmidt]

sin file I used

[sschmidt]

forgot to mention: the source code is from the subroutine state_matrix_freeze1

[Juergen Reuter]

WK, did you have a look into this?

[kilian]

Hmm... I can't reproduce this. valgrind (with gfortran) doesn't show anything in state_matrix_freeze1, and nagfor -mtrace=all gives an output file with 1M lines, but not a single one matching LEAK. Neither do I see memory growing in top.

I used the example file, just reducing the number of events for valgrind and -mtrace to finish in finite time.

[sschmidt]

As an update: I still reproduce the increase in memory consumption:

For example the output of top at different times:

VIRT RES SHR S %CPU %MEM TIME+ COMMAND
2007m 84m 4908 R 100.3 0.2 0:40.10 whizard
2092m 168m 4908 R 100.0 0.3 1:24.61 whizard
2525m 602m 4908 R 100.2 1.2 5:03.23 whizard
2892m 968m 4908 R 100.0 2.0 8:06.25 whizard
3318m 1.4g 4912 R 99.7 2.9 11:31.84 whizard
3526m 1.6g 4912 R 100.0 3.3 13:13.79 whizard
3651m 1.7g 4912 R 100.0 3.6 14:16.97 whizard

(current trunk on theoc01.desy.de using gfortran 4.6 and ticket422.sin)

Valgrind (+gfortran4.6) gives me an increasing amount of lost memory with a slope of (2368.21 +/- 0.08)bytes/event, also with ticket422.sin.

The output when using nagfor and nagfmcheck is also unchanged...

[sschmidt]

After Updating to nagfor 5.2(775) the leak seams to disappear (nagfmcheck crashes, but some greps showed that the formerly undeallocated memory from the line mentioned above, are now deallocated)

[ohl]

Probably gfortran compiler bug (NAG works)