[MITgcm-devel] Beaufort experiment on mac os x

[Menemenlis, Dimitris (3248)]

Torge, could you try downloading and running a fresh copy of Beaufort experiment on your laptop and workstation.
If you run into problem, edit bin/Makefile and replace "-O3" with "-O2", as Martin suggests.
Then, "make clean" and "make -j 16"
Let us know how it works.

Martin, thank you for testing beaufort experiment and for suggestions below.
I tried running using ifort and with array bound checks, etc., and did not find anything funny.
Art, may take a further look at the code.

Using "ifort" I get a much more meaningful "size mitgcmuv"
   text    data     bss     dec     hex filename
7154280   53292 170757160       177964732       a9b86bc mitgcmuv

which is more consistent with memory requirement reported by "top" command.

Cheers

Dimitris Menemenlis

On Mar 14, 2012, at 6:09 AM, Martin Losch wrote:

> Hi Dimitris, Torge,
> 
> the problem is obviously connected to the degree of optimization. 
> This works:
> FFLAGS = -O0/-O1/-O3
> this gives a segmentation fault:
> FFLAGS= -O3
> 
> (Caveat: I was to lazy to download all of the pickup and forcing data and had to turn off obcs etc.)
> 
> Here are the sizes of the executables (although they don't tell me anything):
> ize mitgcmuv_*
> __TEXT	__DATA	__OBJC	others	dec	hex
> 1937408	160423936	0	4295045120	4457406464	109aea000	mitgcmuv_O1
> 2039808	160423936	0	4295045120	4457508864	109b03000	mitgcmuv_O2
> 2510848	160423936	0	4295045120	4457979904	109b76000	mitgcmuv_O3
> 3387392	160423936	0	4295045120	4458856448	109c4c000	mitgcmuv_ieee
> 
> Maybe updating the gcc/gfortran compiler beyond version 4.6.0 might solve the problem?
> 
> Martin
> 
> On Mar 14, 2012, at 7:49 AM, Menemenlis, Dimitris (3248) wrote:
> 
>> Torge, sorry that sea ice code is changing so much.
>> Hopefully things will stabilize now.
>> 
>> I have updated Beaufort experiment to work with latest
>> MITgcm code, including the fixes that you mention below.
>> 
>> You can run the experiment using the instructions from
>> README_beaufort.txt for convenience.
>> I will keep digging for the segmentation-fault
>> large-memory footprint problem.
>> 
>>>>>>>>>>>>> 
>> 
>> cvs co -d beaufort MITgcm_contrib/MPMice/beaufort
>> cd beaufort
>> cvs co MITgcm_code
>> 
>>>> Obtain copies of following directories and put them in beaufort
>>>> ftp://ecco2.jpl.nasa.gov/data1/beaufort/run_template
>>>> ftp://ecco2.jpl.nasa.gov/data1/data/atmos/blend_forcing/cube78_forcing
>> 
>> cd MITgcm
>> mkdir bin exe
>> cd bin
>> \cp ../../code/* .
>> \mv CPP_EEOPTIONS.h_UNCOUPLED CPP_EEOPTIONS.h
>> ../tools/genmake2
>> make depend
>> make -j 16
>> 
>> cd ../exe
>> cp ../../run_template/* .
>> cp ../../input/* .
>> cp ../bin/mitgcmuv .
>> ./mitgcmuv >& output.txt &