Sorry, I forgot to mention that the "export" commands I described in my last
email are POSIX-sh syntax (i.e. sh and bash). To use csh, tcsh, etc.
syntax, use
setenv LD_INCLUDE_PATH
"/usr/mpi/gcc/mvapich-1.1.0/include\:$LD_INCLUDE_PATH"
All the best,
Jason
On Tue, May 11, 2010 at 11:14 PM, Jason Swails <jason.swails.gmail.com>wrote:
> Hello,
>
> This is strange, since mpif90 makes a soft link to mpif.h in the local
> directory. This suggests that neither the local directory nor
> /usr/mpi/gcc/mvapich-1.1.0/include are in the searched include path (I'm
> guessing LD_INCLUDE_PATH, since the libraries are searched in
> LD_LIBRARY_PATH).
>
> I see two options to fix this (though you may want to talk to your system
> admin as to why the linker can't find the mpif.h header file in the local
> directory). #1: use the following command before you begin to compile
>
> export LD_INCLUDE_PATH=/usr/mpi/gcc/mvapich-1.1.0/include\:$LD_INCLUDE_PATH
>
> Or, #2: change MPI_INCLUDE back to its default value in config.h. namely:
>
>
> MPI_INCLUDE = -I$(MPI_HOME)/include
>
> Something worth trying may also be to just add the present directory to
> LD_INCLUDE_PATH (which may be a good permanent solution considering what
> mpif90 is trying to do, though more experienced users may have advice
> against this):
>
> export LD_LIBRARY_PATH=.\:$LD_INCLUDE_PATH
>
> I think that any of the above 3 solutions should be sufficient to resolve
> that error (though of course I have no way of knowing if there will be other
> errors after this one).
>
> If this doesn't work, check the directory
> /usr/mpi/gcc/mvapich-1.1.0/include . Is there a file called mpif.h in
> there? If not, that is definitely your problem. Considering how easy this
> check is, that is probably worth doing first. If it exists in that
> directory, then any of the above should fix it. If it's not, try googling
> the issue to see what comes up (the MPI may need to be reinstalled?) I just
> checked mvapich, mvapich2, mpich2, and openmpi installations on machines I
> have access to, and every single one has mpif.h in the $MPI_HOME/include
> directories (as I would expect them to), so it should be there.
>
> Good luck!
> Jason
>
>
> On Tue, May 11, 2010 at 9:38 PM, Senthil Natesan <sen.natesan.yahoo.com>wrote:
>
>> Thanks Jason. I compiled Amber10 just few months back and applied all
>> bugfixes that time. (I tried again today and it skipped all patches).
>> ***********************************
>> # mpif90 -show
>> ln -s /usr/mpi/gcc/mvapich-1.1.0/include/mpif.h mpif.h
>> gfortran -fPIC -fPIC -L/usr/lib64 -Wall -Wl,-rpath-link
>> -Wl,/usr/mpi/gcc/mvapich-1.1.0/lib/shared
>> -L/usr/mpi/gcc/mvapich-1.1.0/lib/shared -L/usr/mpi/gcc/mvapich-1.1.0/lib
>> -lmpichf90nc -lmpichfarg -lmpich -L/usr/lib64 -libverbs -libumad -libcommon
>> -lpthread -Wall -lpthread -lrt
>> rm -f mpif.h
>> ************************************
>> so it is picking up the right compiler. Good news is that we narrowed down
>> to just one error now.
>> **************************************
>> [tripos.craycx1 pmemd]$ make install
>> cd src && make install
>> make[1]: Entering directory
>> `/home/tripos/amber10_install/amber10/src/pmemd/src'
>> /lib/cpp -traditional -P -DPUBFFT -DMPI -DDIRFRC_EFS -DDIRFRC_COMTRANS
>> -DDIRFRC_NOVEC -DFFTLOADBAL_2PROC gbl_constants.fpp gbl_constants.f90
>> mpif90 -c -O3 gbl_constants.f90
>> /lib/cpp -traditional -P -DPUBFFT -DMPI -DDIRFRC_EFS -DDIRFRC_COMTRANS
>> -DDIRFRC_NOVEC -DFFTLOADBAL_2PROC gbl_datatypes.fpp gbl_datatypes.f90
>> mpif90 -c -O3 gbl_datatypes.f90
>> /lib/cpp -traditional -P -DPUBFFT -DMPI -DDIRFRC_EFS -DDIRFRC_COMTRANS
>> -DDIRFRC_NOVEC -DFFTLOADBAL_2PROC state_info.fpp state_info.f90
>> mpif90 -c -O3 state_info.f90
>> /lib/cpp -traditional -P -DPUBFFT -DMPI -DDIRFRC_EFS -DDIRFRC_COMTRANS
>> -DDIRFRC_NOVEC -DFFTLOADBAL_2PROC file_io_dat.fpp file_io_dat.f90
>> mpif90 -c -O3 file_io_dat.f90
>> /lib/cpp -traditional -P -DPUBFFT -DMPI -DDIRFRC_EFS -DDIRFRC_COMTRANS
>> -DDIRFRC_NOVEC -DFFTLOADBAL_2PROC pmemd_lib.fpp pmemd_lib.f90
>> pmemd_lib.fpp:40: error: mpif.h: No such file or directory
>> make[1]: *** [pmemd_lib.o] Error 1
>> make[1]: Leaving directory
>> `/home/tripos/amber10_install/amber10/src/pmemd/src'
>> make: *** [install] Error 2
>> **************************************************
>>
>> $MPI_HOME/bin is very much in the path(please see the out put for mpif90
>> -show above).
>> any suggestions?
>>
>> thanks in advance..
>>
>> Senthil
>>
>>
>>
>>
>>
>>
>> Hello,
>>
>> First thing to always check:
>>
>> all bugfixes have been downloaded and applied. Then try recompiling (but
>> make clean before every fresh try!). If these don't work, see my advice
>> below:
>>
>> I would probably opt for using the compiler wrappers (as long as you know
>> they've been properly installed with your MPI). Doing this, you can get
>> rid
>> of everything in front of MPI_INCLUDE and MPI_LIBS (since mpif90 will take
>> care of that for you).
>>
>> Then, change F90 = gfortran to F90 = mpif90 and LOAD = gfortran to LOAD =
>> mpif90 . See the comments below for clarification:
>>
>> On Tue, May 11, 2010 at 6:13 PM, Senthil Natesan <sen.natesan.yahoo.com
>> >wrote:
>>
>> > Dear Amber Users,
>> >
>> > I am installing PMEMD (Amber 10) on RHEL 5.3 (x86_64) using gfortran and
>> > mvapichand got some errors. I am displaying my config.h file here and
>> > please see the attachment for error output.
>> >
>> > config.h:
>> > -----------------------------
>> > MATH_DEFINES =
>> > MATH_LIBS =
>> > FFT_DEFINES = -DPUBFFT
>> > FFT_INCLUDE =
>> > FFT_LIBS =
>> > NETCDF_HOME =
>> > NETCDF_DEFINES =
>> > NETCDF_MOD =
>> > NETCDF_LIBS =
>> > MPI_HOME = /usr/mpi/gcc/mvapich-1.1.0
>> > MPI_LIBDIR2 = /usr/mpi/gcc/mvapich-1.1.0/lib/shared
>> > MPI_DEFINES = -DMPI
>> > MPI_INCLUDE = -I$(MPI_HOME)/include
>> >
>>
>> change to
>>
>> MPI_INCLUDE =
>>
>>
>> > MPI_LIBDIR = $(MPI_HOME)/lib
>> > MPI_LIBS = -L$(MPI_LIBDIR) -lmpich -L$(MPI_LIBDIR2) -lmtl_common -lvapi
>> > -lmosal -lmpga -lpthread
>> >
>>
>> change to
>>
>> MPI_LIBS =
>>
>>
>> > DIRFRC_DEFINES = -DDIRFRC_EFS -DDIRFRC_COMTRANS -DDIRFRC_NOVEC
>> > CPP = /lib/cpp
>> > CPPFLAGS = -traditional -P
>> > F90_DEFINES = -DFFTLOADBAL_2PROC
>> >
>> > F90 = gfortran
>> >
>>
>> change to
>>
>> F90 = mpif90
>>
>>
>> > MODULE_SUFFIX = mod
>> > F90FLAGS = -c
>> > F90_OPT_DBG = -g -ftrace=full
>> > F90_OPT_LO = -O0
>> > F90_OPT_MED = -O2
>> > F90_OPT_HI = -O3
>> > F90_OPT_DFLT = $(F90_OPT_HI)
>> > CC = gcc
>> > CFLAGS =
>> >
>> > LOAD = gfortran
>> >
>>
>> change to
>>
>> LOAD = mpif90
>>
>>
>> > LOADFLAGS =
>> > LOADLIBS =
>> > -------------------------------------------------
>> >
>> > Any suggestions would be very helpful. Thanks,
>> >
>> > Senthil Natesan
>> >
>>
>> However, a simple check to make sure that mpif90 is invoking the correct
>> compiler is:
>>
>> mpif90 -show
>>
>> This should show you the include directories and library directories that
>> it's providing as well as the specific libraries it's using.
>>
>> Hope this helps,
>> Jason
>>
>> --
>> Jason M. Swails
>> Quantum Theory Project,
>> University of Florida
>> Ph.D. Graduate Student
>> 352-392-4032
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>>
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>
>
> --
> Jason M. Swails
> Quantum Theory Project,
> University of Florida
> Ph.D. Graduate Student
> 352-392-4032
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Graduate Student
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue May 11 2010 - 20:30:04 PDT
