[MITgcm-devel] Fwd: Re: linux_amd64_ifort+mpi_ice_nas update

[Jean-Michel Campin]

Hi David,

Sounds good.

I did pass few tests (run on Week-End) on acesgrid with ifort (v13)
and using standard optfile (linux_amd64_ifort11),
both with -fast and with just -ieee (testreport default).
And it seems that some experiment have problems with "-fast"
(including some failed restart).
Did not have time to try with just -O1.

Cheers,
Jean-Michel

On Fri, Oct 18, 2013 at 05:55:27PM +0100, David Ferreira wrote:
> Hi Jean-Michel,
> Pleiades is back. So about seaice_growth.F:
> I'll run a testreport on the experiments which include sea ice
> (found only 6 with a grep)
> I'll just set the NOOPTFLAGS flag to -O1 for this (the other files
> are calc_oce_mxlayer.F fizhi_lsm.F fizhi_clockstuff.F and should not
> matter)
> 
> 1D_ocean_ice_column/code/packages.conf:seaice
> global_ocean.cs32x15/code/packages.conf:seaice
> lab_sea/code/packages.conf:seaice
> offline_exf_seaice/code/packages.conf:seaice
> seaice_itd/code/packages.conf:seaice
> seaice_obcs/code/packages.conf:seaice
> 
> cheers,
> david
> 
> -------- Original Message --------
> Subject: 	Re: [MITgcm-devel] linux_amd64_ifort+mpi_ice_nas update
> Date: 	Tue, 01 Oct 2013 23:27:30 +0100
> From: 	David Ferreira <dfer at MIT.EDU>
> Reply-To: 	MITgcm-devel at mitgcm.org
> To: 	MITgcm-devel at mitgcm.org
> 
> 
> 
> How would you proceed to get seaice_growth.F at -O1, the
> NOOPTFILES at -O0, and the rest at -O2 ?
> 
> 
> On 10/1/13 11:15 PM, Jean-Michel Campin wrote:
> >Hi David & Dimitris,
> >
> >The reason I was asking is that there is a fair amount of computation
> >in seaice_growth.F, -O0 is slow (even sometimes very slow),
> >so may be if it could go through -O1 it could speed up the runs.
> >
> >>Could this be tested with testreport first?
> >Sounds like a good suggestion. We have few experiments that use
> >seaice_growth.F, not testing everything but should be
> >a good start.
> >
> >Cheers,
> >Jean-Michel
> >
> >On Tue, Oct 01, 2013 at 10:47:00PM +0100, David Ferreira wrote:
> >>I've never tried optimizing seaice_growth.F
> >>Could this be tested with testreport first?
> >> From memory, the results of the verification exp with seaice are good.
> >>Compilation problem should show up easily.
> >>david
> >>
> >>
> >>
> >>On 10/1/13 9:38 PM, Dimitris Menemenlis wrote:
> >>>Jean-Michel I have not tried.
> >>>Pleiades is inaccessible right now because of shutdown.
> >>>But when it comes back up what would be a good test to run?
> >>>I can give it a try and see what happens but the
> >>>results with O1 or O2 are not expected to be the same as with O0,
> >>>and a possible problem might not manifest itself right away.
> >>>
> >>>David any more info on this?
> >>>
> >>>Dimitris Menemenlis
> >>>
> >>>On Oct 1, 2013, at 1:19 PM, Jean-Michel Campin wrote:
> >>>
> >>>>Hi David and Dimitris,
> >>>>
> >>>>I wonder if it's still necessary to compile seaice_growth.F with -O0.
> >>>>it has been added to the NOOPTFILES list in July 2010.
> >>>>Did someone try recently with -O1 ? and if it works, is it faster
> >>>>than -O0 ?
> >>>>
> >>>>Cheers,
> >>>>Jean-Michel
> >>>>
> >>>>On Fri, Sep 27, 2013 at 05:00:58AM -0400, David Ferreira wrote:
> >>>>>Update of /u/gcmpack/MITgcm/tools/build_options
> >>>>>In directory forge:/tmp/cvs-serv18718
> >>>>>
> >>>>>Modified Files:
> >>>>>	linux_amd64_ifort+mpi_ice_nas
> >>>>>Log Message:
> >>>>>Updated with a newer compiler (comp-intel/2012.0.032) and mpi library (mpi-sgi/mpt.2.06rp16)
> >>>_______________________________________________
> >>>MITgcm-devel mailing list
> >>>MITgcm-devel at mitgcm.org
> >>>http://mitgcm.org/mailman/listinfo/mitgcm-devel
> >>
> >>_______________________________________________
> >>MITgcm-devel mailing list
> >>MITgcm-devel at mitgcm.org
> >>http://mitgcm.org/mailman/listinfo/mitgcm-devel
> >_______________________________________________
> >MITgcm-devel mailing list
> >MITgcm-devel at mitgcm.org
> >http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> 
>