The ABINIT Users Mailing List ( CLOSED )

["Anglade Pierre-Matthieu" <anglade@gmail.com>]

The weird point here is that abinit is not a threaded program. It uses
mpi parrallelization. Then for as much as I know there is no "fork" or
threading instructions into it. So a priori no compiler should ask for
them when compiling abinit.
Yet ...
1) Such instructions might be part of  an mpi layer.  But you are not
compiling your mpi lib when getting this message, aren't you?
2) Some compilers in order to increase the sequential speed of
programs on smp or virtual smp machine (hyperthreaded for instance)
try to parallelize some instructions. This may result in such call to
functions like pthread_atfork.  For instance when we compile abinit on
power5 machine with xlf90 and maximum optimization we run into this
exact same problem... I see potentialy two solutions:
a) you can try including the pthread (posix thread) library when
compiling.  For this the flag is usally -pthread.  Yet I would guess
you'll get some others problem. At least you'd have to test your newly
compile abinit for accuracy.
b) you can try to reduce the optimization level you have chosen. This
might be both the simplest and most efficient solution.

PMA

On 7/25/06, ilukacevic@fizika.unios.hr <ilukacevic@fizika.unios.hr> wrote:
> I was very ignorant in this problem. I linked the libraries from a directory 
> /lib/32/ instead of /lib/em64t/ (there is also a directory /lib/64/, but in 
> this one there are no libmkl_def.so and libmkl_p4.so).
> Despite this, I again get a stopped compile, with a message
>
> /opt/intel_fce_80/lib/libguide.so: undefined reference to `pthread_atfork'
> chmod: failed to get attributes of `abinip': No such file or directory
> make[1]: *** [abinip] Error 1
> make[1]: Leaving directory `/home/lukacevic/abinit-4.6.5'
> make: *** [abinip] Error 2
>
> As I see from man pthread_atfork registers handler functions to be called 
> just before and just after a new process is created with fork(2). Also, as I 
> have insufficient knowledge about this, I would again ask for help in 
> explanation about this error.
>
> Thank you in advance
>
> I. Lukacevic


--
Pierre-Matthieu Anglade