Remark that allocreportlevel does not enable you to check for memory leaks!

If you only find the memory leak with intel MPI and not with OpenMPI I
would presume siesta isn't the cause of it. siesta does not
differ on implementation between MPI vendors.

Also, you may want to try with allocreportlevel 3 which should still do
what you want, but only when a change in peak memory (note that this will
only be on the main node).



2016-01-20 14:19 GMT+01:00 Matthias Neuer <[email protected]>:

> Hi.
>
> On 01/20/2016 02:08 PM, Nick Papior wrote:
>
>> Simply do not use allocreportlevel 4.
>> Its main use is to check for memory consumption in siesta (mainly
>> developers will use it), I would not advice to use it in production runs
>> as it also has some overhead.
>>
>
> This was exactly the reason why I set the option to 4. I have a very high
> memory consumption when I run siesta on multiple nodes. I only get this
> behavior with Intel MPI not with OpenMPI and I am not sure if my MPI
> implementation or siesta has a memory leak.
> Actually Intel Inspector shows me some leaks in siesta but I don't know if
> they are false positives.
>
> Regards,
> Matthias
>
>
>> 2016-01-20 13:51 GMT+01:00 Matthias Neuer <[email protected]
>> <mailto:[email protected]>>:
>>
>>     Hi.
>>
>>     I'm getting at lot of warning messages when setting the option
>>     AllocReportLevel to 4. They look all the same:
>>
>>     WARNING: alloc-realloc-dealloc name mismatch
>>               Name: b
>>               Size:            -35968. Bytes
>>     Subsequent mismatches will not be reported
>>
>>     What does this mean and what can I do to get rid of these messages?
>>
>>     I compiled the code with the Intel compiler 14 and 15.
>>
>>     Thank you very much
>>     Matthias
>>
>>     --
>>     Matthias Neuer
>>     Universität Ulm
>>     kiz / Abteilung Infrastruktur
>>
>>
>>
>>
>> --
>> Kind regards Nick
>>
>
>
> --
> Matthias Neuer
> Universität Ulm
> kiz / Abteilung Infrastruktur
>



-- 
Kind regards Nick

Responder a