To follow up for the list (we've had a few e-mails about this
off-list): the issue is that ROMIO's configure test for the fortran
symbols is not working properly on OS X, and therefore Bad Things
happen.
We're going to put a workaround in -- we'll post back to the list when
it's in Subversion.
On Feb 21, 2005, at 3:39 PM, Bernard Knaepen wrote:
> ahah,
> there are no fortran bindings in lammpio. The part of my configure
> script relevant to ROMIO is copied below. Apparently the script has a
> problem with the fortran compiler in the ROMIO part. The rest of the
> configure script seems to be ok with the fortran compiler though. I
> attach the complete output of configure as a separate file.
> Thanks,
> Bernard.
>
> ps: I have to leave my email now until tomorrow...
>
> =======================================================================
> =====
> == Configuring ROMIO (MPI-2 I/O) support
> =======================================================================
> =====
>
> Configuring ROMIO Version 1.2.5.1/LAM
>
> Configuring with args --prefix=/Users/bknaepen/Unix/lam/ -mpi=lam
> -from_lam -mpiincdir=/Users/bknaepen/Desktop/lam-7.1.2b12/share/include
> -I/Users/bknaepen/Desktop/lam-7.1.2b12/share/include -cc=gcc
> -fc=/Applications/Absoft/bin/f90
> -mpiolib=/Users/bknaepen/Desktop/lam-7.1.2b12/romio/lib/liblammpio.a
> -cflags=-O3 -DLAM_BUILDING=1 -fflags=-O -arch=freebsd
> checking for virtual path format... VPATH
> ROMIO home directory is /Users/bknaepen/Desktop/lam-7.1.2b12/romio
> checking for ranlib
> checking if MOUNT_NFS is defined in the include files... no
> configuring for file systems ufs and nfs
> checking for memalign
> checking if MPIR_Status_set_bytes is defined... no
> checking Fortran external names
> *# Unable to determine the form of Fortran external names
> *# Make sure that the compiler /Applications/Absoft/bin/f90 can be run
> on this system
> *# Turning off Fortran (-nof77 being assumed).
> checking for inline... yes
> checking for unistd.h... yes
> checking for size of int... 4
> checking for size of void *... 4
> checking for int large enough for pointers... yes
> defining MPI_Offset as long in C and integer in Fortran
> checking for weak symbol support... no
> checking whether file locks work with NFS... yes
> checking for strerror()... yes
> setting CC to gcc
> setting F77 to :
> setting TEST_CC to mpicc
> setting TEST_F77 to mpif77
> setting CFLAGS to -O3 -DLAM_BUILDING=1 -O3 -DHAVE_ROMIOCONF_H
> setting USER_CFLAGS to
> setting USER_FFLAGS to
> creating config.status
> creating Makefile
> creating mpi-io/Makefile
> creating mpi2-other/info/Makefile
> creating mpi2-other/array/Makefile
> creating adio/common/Makefile
> creating test/Makefile
> creating test/misc.c
> creating test/large_file.c
> creating test/runtests
> creating util/romioinstall
> creating include/mpio.h
> creating include/mpiof.h
> creating adio/ad_nfs/Makefile
> creating adio/ad_ufs/Makefile
> creating adio/ad_xfs/Makefile
> creating adio/ad_hfs/Makefile
> creating adio/ad_sfs/Makefile
> creating adio/ad_pfs/Makefile
> creating adio/ad_testfs/Makefile
> creating adio/ad_pvfs/Makefile
> creating adio/ad_piofs/Makefile
> creating adio/include/romioconf.h
> creating Makefile
> creating mpi-io/Makefile
> creating mpi2-other/info/Makefile
> creating mpi2-other/array/Makefile
> creating adio/common/Makefile
> creating test/Makefile
> creating test/misc.c
> creating test/large_file.c
> creating test/runtests
> creating util/romioinstall
> creating include/mpio.h
> creating include/mpiof.h
> creating adio/ad_nfs/Makefile
> creating adio/ad_ufs/Makefile
> creating adio/ad_xfs/Makefile
> creating adio/ad_hfs/Makefile
> creating adio/ad_sfs/Makefile
> creating adio/ad_pfs/Makefile
> creating adio/ad_testfs/Makefile
> creating adio/ad_pvfs/Makefile
> creating adio/ad_piofs/Makefile
> creating adio/include/romioconf.h
> adio/include/romioconf.h is unchanged
>
> *** SSI modules
> checking which modules should be run-time loadable... none
> checking for SSI kinds... boot coll crlam crmpi rpi
> configure: LAM configuring in share/ssi/boot/bproc
> configure: running /bin/sh './configure'
> '--prefix=/Users/bknaepen/Unix/lam/' '--with-rpi=sysv'
> '--with-fc=/Applications/Absoft/bin/f90' --cache-file=/dev/null
> --srcdir=.
>
>
>
>
> On Mon, 21 Feb 2005 15:03:57 -0500, Jeff Squyres
> <jsquyres_at_[hidden]> wrote:
>> For those of you who noticed a gap in the communication here, we had
>> an
>> exchange off list that identified that I missed the fact that *all* of
>> ROMIO's bindings are in liblammpio.* (even the fortran bindings).
>>
>> Can you send the output of your configure script?
>>
>> ROMIO's configure script should be figuring out the name mangling
>> convention for Fortran and using it. For example, on my g77-based
>> system, I see:
>>
>> ------
>> [...much snipped...]
>> checking Fortran external names
>> Fortran externals are lower case and have two trailing underscores
>> [...much snipped...]
>> ------
>>
>> Is the equivalent not happening on your system? What fortran symbols
>> are included in your liblammpio.*? E.g., can you see the file
>> set_viewf.o in the nm output, and see the symbols that it defines?
>> For
>> me, it defines:
>>
>> ------
>> 00000000 W mpi_file_set_view__
>> 00000000 T pmpi_file_set_view__
>> ------
>>
>>
> <myconfig.LOG.gz>
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
