[Dakota-users] [Dakota-developers] Re: compiling on different machines (OS X 10.4, IBM DataStar)
Adams, Brian M
briadam at sandia.gov
Sat Jul 12 23:20:39 MDT 2008
Bill and G.S.,
I have seen make get tripped up at this epetra example, but can't recall the precise reason why. It might be helpful to see the specific make error preceding "gmake[5]: Leaving directory", i.e. the origination of the error that caused make to bomb out.
That said, I believe when compiling Trilinos (and therefore DAKOTA) against the MPI2-compliant OpenMPI-1.2.6, you will need to define MPICH_IGNORE_CXX_SEEK at compile time. For example, add the following to CPPFLAGS:
-DMPICH_IGNORE_CXX_SEEK
(http://trilinos.sandia.gov/faq.html#building_13) I thought failing to do this would produce a configure-time error, not make-time, though.
So, unfortunately, I don't think this is the reason for your error, but there is some chance that it will help.
Brian
----------
Brian M. Adams, PhD (briadam at sandia.gov)
Optimization and Uncertainty Estimation
Sandia National Laboratories, Albuquerque, NM
http://www.sandia.gov/~briadam
________________________________
From: dakota-developers-bounces at development.sandia.gov [mailto:dakota-developers-bounces at development.sandia.gov] On Behalf Of Bohnhoff, William J
Sent: Saturday, July 12, 2008 10:33 PM
To: gso at physics.ucsd.edu; dakota-users at software.sandia.gov
Subject: Re: [Dakota-users] [Dakota-developers] Re: compiling on different machines (OS X 10.4, IBM DataStar)
G.S.
Thanks for your config logfiles. I agree working/solving one issue at a time is the way to go. It sounds like you're trying to tackle build with the graphics issue first, and unfortunately I can't help you with that (just yet). I too chose to tackle one issue at a time, and felt that the MPI problem would be easier to solve so I dove in there first..
Comparing your MPI-enabled config logfile with mine indicated one glaring setting that was somehow missed, i.e.:
diff config.logMINE config.logMPIyours
3321d4264
< #define HAVE_MPI 1
I think this is likely leading to the build error you're encountering in the Epetra examples:
> gmake[5]: *** [power_method_example] Error 1
> gmake[5]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/epetra/example/petra_power_method'
> gmake[4]: *** [all-recursive] Error 1
> gmake[4]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/epetra/example'
> gmake[3]: *** [all-recursive] Error 1
> gmake[3]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/epetra'
I don't know why you're not getting the "#define HAVE_MPI 1" setting in your configuration. I'll investigate further and query other team members with respect to the possibility of known issues that could lead to the "bad configuration" you've encountered. In the meantime, would it be possible for you to cd into the following directory:
/Users/gso/Documents/DAKOTA/Dakota/packages/epetra/example
And type gmake? FYI, here's what I see when I build in that directory:
>>>>>
Making all in petra_power_method
if mpic++ -DHAVE_CONFIG_H -I. -I. -I../../src -I../../src -O2 -MT
power_method_example-cxx_main.o -MD -MP -MF
".deps/power_method_example-cxx_main.Tpo" -c -o
power_method_example-cxx_main.o `test -f './cxx_main.cpp' || echo
'./'`./cxx_main.cpp; \
then mv -f ".deps/power_method_example-cxx_main.Tpo"
".deps/power_method_example-cxx_main.Po"; else rm -f
".deps/power_method_example-cxx_main.Tpo"; exit 1; fi
/bin/sh ../../libtool --tag=CXX --mode=link mpic++ -O2 -framework vecLib
-o power_method_example power_method_example-cxx_main.o
../../src/libepetra.la -L/usr/local/lib/gcc/i386-apple-darwin8.8.1/4.3.0
-L/usr/local/lib/gcc/i386-apple-darwin8.8.1/4.3.0/../../.. -lgfortranbegin
-lgfortran -lgcc_s.10.4 -lSystem mkdir .libs
mpic++ -O2 -framework vecLib -o power_method_example
power_method_example-cxx_main.o -Wl,-bind_at_load
../../src/.libs/libepetra.a
-L/usr/local/lib/gcc/i386-apple-darwin8.8.1/4.3.0
-L/usr/local/lib/gcc/i386-apple-darwin8.8.1/4.3.0/../../..
/usr/local/lib/gcc/i386-apple-darwin8.8.1/4.3.0/libgfortranbegin.a
/usr/local/lib/libgfortran.dylib -lm -lgcc_s.10.4 -lSystem
<<<<<
My configuration leads to a valid build and working MPI executables. I suspect yours will fail in the link phase for the "power_method_example" executable, so if you're able to confirm my hunch, that would probably be useful information for me as I continue to investigate problems with your MPI build configuration. Sorry this hasn't been too helpful; perhaps the next messages we exchange will help you get passed this sticking point.
--Bill
On 7/9/08 2:02 PM, "gso at physics.ucsd.edu" <gso at physics.ucsd.edu> wrote:
These are the two config for when I tried turning on MPI and when I tried
to turn on Graphics. Are there packages that I should install to get a
fresh download of DAKOTA source configured and make properly, or something
I need to configure on my mac to get this working that's not obvious?
From
G.S.
> G.S
> Sorry for the delayed reply.
>
> I'll try to duplicate your build on my MacOS X 10.4 laptop and offer some
> advice based on my findings.
> In the meantime, would you mind replying to this message with your
> complete 'config.log' file attached? (I'd like to compare it with my
> resulting config.log)
>
> Regards,
> --Bill
>
> ________________________________________
> From: dakota-developers-bounces at development.sandia.gov
> [dakota-developers-bounces at development.sandia.gov] On Behalf Of
> gso at physics.ucsd.edu [gso at physics.ucsd.edu]
> Sent: Thursday, July 03, 2008 4:36 PM
> To: dakota-users at software.sandia.gov
> Subject: [Dakota-developers] Re: [Dakota-users] compiling on different
> machines (OS X 10.4, IBM DataStar)
>
> Hi, I've been having trouble compiling Dakota 4.1 src on my mac 10.4, was
> wondering if anyone has any insights or step by step instructions on how
> they got it working on their computer. Someone already offered me a
> configure in which many things were turned off, and it works, but I was
> wondering if I can turn some of those things back on i.e. graphics and
> MPI.
>
> I have OpenMPI 1.2.6 installed, and
> openmotif3/openmotif3-bin/openmotif-shlibs installed via finks.
>
> 1. With the following configure, with no MPI:
>
> #!/bin/bash
> CPPFLAGS="-I/usr/include/malloc -DDARWIN -I/usr/X11R6/include"
> -I/sw/include \
> CXXFLAGS="-O3" \
> ../Dakota/configure \
> --disable-mpi \
> --without-dot \
> --without-npsol \
> --without-xpm
>
> I get the following error messages:
>
> Common.w: In function 'class_initialize':
> Common.w:287: error: 'cvtStringToIcon' undeclared (first use in this
> function)
> Common.w:287: error: (Each undeclared identifier is reported only once
> Common.w:287: error: for each function it appears in.)
> Common.w:289: error: 'cvtIconDestructor' undeclared (first use in this
> function)
> gmake[5]: *** [Widgets/libDGraphics_la-Common.lo] Error 1
> gmake[5]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/motif/src'
> gmake[4]: *** [all-recursive] Error 1
> gmake[4]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/motif'
> gmake[3]: *** [all] Error 2
> gmake[3]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/motif'
> gmake[2]: *** [all-recursive] Error 1
> gmake[2]: Leaving directory `/Users/gso/Documents/DAKOTA/Dakota/packages'
> gmake[1]: *** [all-recursive] Error 1
> gmake[1]: Leaving directory `/Users/gso/Documents/DAKOTA/Dakota'
> gmake: *** [all] Error 2
>
>
> I'm guessing that I am missing some sort of package so that it cannot find
> the file necessary. Where can I find that package?
>
> 2. With the following configure:
>
> #!/bin/bash
> CC=mpicc \
> CXX=mpicxx \
> F77=mpif77 \
> FC=mpif90 \
> CPPFLAGS="-I/usr/include/malloc -DDARWIN" \
> CXXFLAGS="-O3" \
> ../Dakota/configure \
> --without-dot \
> --without-npsol \
> --without-graphics \
> --without-x \
> --without-xpm
>
> I run into the following error message:
>
> gmake[5]: *** [power_method_example] Error 1
> gmake[5]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/epetra/example/petra_power_method'
> gmake[4]: *** [all-recursive] Error 1
> gmake[4]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/epetra/example'
> gmake[3]: *** [all-recursive] Error 1
> gmake[3]: Leaving directory
> `/Users/gso/Documents/DAKOTA/Dakota/packages/epetra'
> gmake[2]: *** [all-recursive] Error 1
> gmake[2]: Leaving directory `/Users/gso/Documents/DAKOTA/Dakota/packages'
> gmake[1]: *** [all-recursive] Error 1
> gmake[1]: Leaving directory `/Users/gso/Documents/DAKOTA/Dakota'
> gmake: *** [all] Error 2
>
> and I don't have the slightest idea what's going on. It shouldn't be
> MPI's problem because I have 1.2.6 OpenMPI installed and have ran parallel
> programs with it (although only with 2 processors on this laptop). I
> assumed that you needed to switch out the fortran compilers as well and
> not just the C, C++ compilers to their MPI version as well.
>
> 3. Are there tutorial examples of Dakota in the command line format that
> I can poke around with if I cannot get the graphics to work? I probably
> wouldn't be using much of the graphics anyway, so probably good to learn
> the command line operations anyway.
>
> I tried the following Dakota input files
> dakota_textbook.in
> dakota_rosenbrock.in
>
> they do not call on graphics so I thought I can to play with them, but
> they all kept giving me the same error of:
>
> Error: analysis driver "text_book" not found.
> Error: analysis driver "simulator_script" not found.
>
> and Bus error
>
> From
> G.S.
>
>
>
>
>
> _______________________________________________
> Dakota-users mailing list
> Dakota-users at software.sandia.gov
> http://software.sandia.gov/mailman/listinfo/dakota-users
> _______________________________________________
> Dakota-developers mailing list
> Dakota-developers at development.sandia.gov
> http://development.sandia.gov/mailman/listinfo/dakota-developers
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://software.sandia.gov/pipermail/dakota-users/attachments/20080712/7715c724/attachment-0001.html
More information about the Dakota-users
mailing list