Re: [AMBER-Developers] -j x parallel builds broken if you don't build serial first

From: Ross Walker <ross.rosswalker.co.uk>
Date: Wed, 27 Mar 2013 12:36:49 -0700

It works now.


On 3/27/13 11:15 AM, "Jason Swails" <jason.swails.gmail.com> wrote:

>On Wed, Mar 27, 2013 at 2:03 PM, Ross Walker <ross.rosswalker.co.uk>
>wrote:
>
>> Something looks to be messed up with the netcdf dependencies.
>>
>> The following works:
>>
>> cd $AMBERHOME
>> git pull
>> git clean -f -d -x
>> ./configure gnu
>> make -j8 install
>> make clean
>> ./configure -cuda gnu
>> make -j8 install
>>
>> But the following does not:
>> git clean -f -d -x
>> ./configure -cuda gnu
>> make -j8 install
>>
>> gfortran -I../libsrc -I. -O3 -mtune=native -c -o typeSizes.o
>>typeSizes.f90
>> gfortran -I../libsrc -I. -O3 -mtune=native -c -o netcdf.o netcdf.f90
>> netcdf.f90:38.50:
>>
>> FourByteReal, EightByteReal
>> 1
>> Fatal Error: Can't open module file 'typesizes.mod' for reading at (1):
>>No
>> such file or directory
>> make[4]: *** [netcdf.o] Error 1
>> make[4]: *** Waiting for unfinished jobs....
>> gfortran -I../libsrc -I. -O3 -mtune=native -c netcdf.f90 -o netcdf.o
>> gfortran -I../libsrc -I. -O3 -mtune=native -c typeSizes.f90 -o
>>typeSizes.o
>> make[4]: Leaving directory
>>
>>`/server-home1/rcw/cvs_checkouts/amber.git/amber/AmberTools/src/netcdf/sr
>>c/
>> f90'
>> make[3]: *** [install-recursive] Error 1
>> make[3]: Leaving directory
>>
>>`/server-home1/rcw/cvs_checkouts/amber.git/amber/AmberTools/src/netcdf/sr
>>c'
>> make[2]: ***
>>[/home/rcw/cvs_checkouts/amber.git/amber/include/netcdf.mod]
>> Error 2
>> make[2]: Leaving directory
>> `/server-home1/rcw/cvs_checkouts/amber.git/amber/src'
>> make[1]: [cuda] Error 2 (ignored)
>>
>> Any ideas?
>>
>
>This is a race condition in the NetCDF install, I think. I don't see this
>every time (and the fewer threads you use, the less frequently this
>occurs). It also happens more often on large, parallel file systems where
>file access time is less reliable.
>
>I checked in a fix that basically hard-codes the NetCDF build (only) to a
>single thread. If someone wants to go through the NetCDF build and fix
>the
>dependencies themselves to remove this restriction, they're more than
>welcome ;).
>
>Try it now? If it fixes the issue then I'll cherry-pick that commit over
>to the amber13-with-patches branch.
>
>--Jason
>
>--
>Jason M. Swails
>Quantum Theory Project,
>University of Florida
>Ph.D. Candidate
>352-392-4032
>_______________________________________________
>AMBER-Developers mailing list
>AMBER-Developers.ambermd.org
>http://lists.ambermd.org/mailman/listinfo/amber-developers



_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Wed Mar 27 2013 - 13:00:03 PDT
Custom Search