[MITgcm-support] optimization: lapack? flags?

Andrea Cimatoribus andrea.cimatoribus at epfl.ch
Wed Apr 5 11:22:51 EDT 2017


Thanks for the responses, I know optimization is machine dependent, but 
I was just seeking some general advice. In any case, if you have more 
specific comments, they're of course more than welcome! The machine is a 
rather old and very small workstation, with 8 Intel Core i7-3770K CPU @ 
3.50GHz and 16Gb of memory. Input and output is done on an SSD. I am 
running linux 4.4.0-72 (ubuntu 16.04) and using gfortran 5.4. As you 
see, it's a very small machine, that is why I want to try to be as 
efficient as possible. My intention is to load it with something heavy 
and just wait while I do other things on other machines.

In any case, at least now I know I can ignore LAPACK (btw, I did switch 
on the LAPACK test, but that didn't have any effect).
Thanks,

Andrea


Andrea Cimatoribus
postdoctoral researcher
EPFL ENAC IIE ECOL
https://people.epfl.ch/andrea.cimatoribus

On 05/04/17 16:18, Martin Losch wrote:
> Hi Andrea,
>
> Micheal is right about the optimization. The flags useSingleCpuIO and
> globalFiles only change the I/O, which can also depend on the machine
> that you are working on. True code optimization has to be done with
> compiler flags (in the build options file), or if you can spot
> something seriously inefficient, directly in the code.
>
> I think you always get an “No” on LAPACK because this test is never
> done in genmake2. I am not aware of any forward code that uses
> Lapack. I think some of the very specialized adjoint applications use
> Lapack for singular value decompositions.
>
> Martin
>
>> On 5 Apr 2017, at 16:13, Michael Schaferkotter
>> <schaferk at bellsouth.net> wrote:
>>
>> A.
>>
>> Optimization choices are machine dependent.
>>
>> Your intended audience would need to know the machine architecture
>> that you are targeting.
>>
>> Please provide.
>>
>> michael
>>
>> Sent from my iPhone
>>
>>> On Apr 5, 2017, at 07:42, Andrea Cimatoribus
>>> <andrea.cimatoribus at epfl.ch> wrote:
>>>
>>> Dear all, I am starting to do some more ambitious simulations, in
>>> terms of computational effort, and it is thus time for me to
>>> optimize the compilation a bit. I know it is mostly a matter of
>>> trial and error with compiler options, but I would like to ask
>>> you if you have some suggestion on two specific points:
>>>
>>> - Is it possible to compile using gfortran and use intel MKL? My
>>> genmake2 now reports that it is compiling without LAPACK. I tried
>>> to link to the Intel MKL I have on my machine (while still using
>>> gfortran, we don't have intel compilers for the moment), but
>>> without much success. Is anyone actually using LAPACK? Does it
>>> make a difference? I could find hardly anything in the archives.
>>> - Apart from the "classical" useSingleCpuIO and globalFiles, is
>>> there any other runtime flag which has a significant impact on
>>> execution time (but does not affect the results)?
>>>
>>> Thank you very much, Andrea
>>>
>>> -- Andrea Cimatoribus postdoctoral researcher EPFL ENAC IIE ECOL
>>> https://people.epfl.ch/andrea.cimatoribus
>>>
>>> _______________________________________________ MITgcm-support
>>> mailing list MITgcm-support at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>>
>>
>> _______________________________________________ MITgcm-support
>> mailing list MITgcm-support at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-support
>
>
> _______________________________________________ MITgcm-support
> mailing list MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
>



More information about the MITgcm-support mailing list