[MPICH] mpich 1.2.7 and application profiling freeze

[William Gropp]

I don't know.  Does the gnu profiling use SIGUSR1?  If so, that will  
cause the problem. There is a way to rebuild mpich1 to use a  
different signal (the ch_p4/configure should tell you with --help) if  
that is the problem.  MPICH1 uses SIGUSR1 to tell a process that  
another process is trying to establish a connection with it.

Bill

On Jan 31, 2007, at 8:37 AM, Jan Wagner wrote:

>
> Haven't yet tried. It's too large a piece of code that is using  
> MPICH-1 to
> be migrated so fast...  :-|
>
> So is there some special way to get gnu gcc profiling to work with  
> MPICH-1
> dependent code?
>
> thanks,
>  - Jan
>
> On Tue, 30 Jan 2007, Rajeev Thakur wrote:
>>
>> Is there any reason you need to use MPICH-1? Can you try MPICH2  
>> instead?
>>
>> Rajeev
>>
>>
>>> -----Original Message-----
>>> From: owner-mpich-discuss at mcs.anl.gov
>>> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Jan Wagner
>>> Sent: Tuesday, January 30, 2007 3:01 AM
>>> To: mpich-discuss at mcs.anl.gov
>>> Subject: [MPICH] mpich 1.2.7 and application profiling freeze
>>>
>>> Hi,
>>>
>>> I'm using mpich 1.2.7b compiled from sources, and I try to run
>>> applications over 'mpirun' that have been compiled with the
>>> '-pg' flag to
>>> enable profiling.
>>>
>>> The compiler used for mpich and for compiling the application is:
>>>
>>> [jwagner at ps3-001 ~]$ g++ -v
>>> Using built-in specs.
>>> Target: ppc64-yellowdog-linux
>>> Configured with: ../configure --prefix=/usr --mandir=/usr/share/man
>>> --infodir=/usr/share/info --enable-shared --enable-threads=posix
>>> --enable-checking=release --with-system-zlib --enable-__cxa_atexit
>>> --disable-libunwind-exceptions --enable-libgcj-multifile
>>> --enable-languages=c,c++,objc,obj-c++,java,fortran
>>> --enable-java-awt=gtk
>>> --disable-dssi
>>> --with-java-home=/usr/lib/jvm/java-1.4.2-gcj-1.4.2.0/jre
>>> --enable-secureplt --with-long-double-128 --host=ppc64-yellowdog- 
>>> linux
>>> --build=ppc64-yellowdog-linux --target=ppc64-yellowdog-linux
>>> --with-cpu=default32
>>> Thread model: posix
>>> gcc version 4.1.1 20060525 (Yellow Dog 4.1.1-1)
>>>
>>> When the application is compiled without -pg it works just fine with
>>> mpirun.
>>>
>>> But when the -pg compile and link flag are set, mpirun will  
>>> simply not
>>> progress:
>>>
>>> running /home/jr/correlator/mpifxcorr-GNU-gprof on 8 LINUX
>>> ch_p4 processors
>>> Created /home/jr/correlator/PI3664
>>>
>>> There are no RSH or SSH connection attempts in auth.log.
>>>
>>>
>>> Checking with strace -f, at "stuck" time there are just a lot of
>>>
>>> [pid  6226] sigreturn()                 = ? (mask now [])
>>> [pid  6226] --- SIGPROF (Profiling timer expired) @ 0 (0) ---
>>> [pid  6226] sigreturn()                 = ? (mask now [ABRT
>>> BUS FPE KILL
>>> USR1 SEGV USR2 PIPE TERM STKFLT CONT STOP TSTP TTIN TTOU URG XCPU  
>>> XFSZ
>>> VTALRM PROF WINCH])
>>>
>>>
>>> Any ideas how to get gprof profiling to work?
>>>
>>> (btw I don't want to profile MPI, or MPI message passing -
>>> just profile my
>>> application internally when it runs a distributed test case)
>>>
>>> thanks,
>>>  - Jan
>>>
>>>
>>
>