Hello Curtis,
yes, done with ompi-trunk:
Apart from --enable-mpi-threads --enable-progress-threads, You need to compile
Open MPI with --enable-mca-no-build=memory-ptmalloc2 ; and of course the
usual options for debugging (--enable-debug) and the options for
icc/ifort/icpc:
CFLAGS='-debug all -inline-debug-info -tcheck'
CXXFLAGS='-debug all -inline-debug-info -tcheck'
FFLAGS='-debug all -tcheck'
LDFLAGS='-tcheck'
Then, as You already noted, run the application with --mca btl tcp,sm,self:
mpirun --mca tcp,sm,self -np 2 \
tcheck_cl \
--reinstrument \
-u all \
-c \
-d '/tmp/hpcraink_$$__tc_cl_cache' \
-f html \
-o 'tc_mpi_test_suite_$$.html' \
-p 'file=tc_mpi_test_suite_%H_%I, \
pad=128, \
delay=2, \
stall=2' \
-- \
./mpi_test_suite -j 2 -r FULL -t 'Ring Ibsend' -d MPI_INT
-- the reinstrument is not really necessary, also setting the padding and
delay for startup of threads; shortenign the delay for stalls to 2 seconds
alos does not trigger any deadlocks.
This was with icc-9.1 and itt-3.0 23205.
Hope this helps,
Rainer
On Friday 23 March 2007 05:22, Curtis Janssen wrote:
> I'm interested in getting OpenMPI working with a multi-threaded
> application (MPI_THREAD_MULTIPLE is required). I'm trying the trunk
> from a couple weeks ago (1.3a1r14001) compiled for multi-threading and
> threaded progress, and have had success with some small cases. Larger
> cases with the same algorithms fail (they work with MPICH2 1.0.5/TCP and
> other thread-safe MPIs, so I don't think it is an application bug). I
> don't mind doing a little work to track down the problem, so I'm trying
> to use the Intel Thread Checker. I have the thread checker working with
> my application when using Intel's MPI, but with OpenMPI it hangs.
> OpenMPI is compiled for OFED 1.1, but I'm overriding communications with
> "-gmca btl self,tcp" in the hope that OpenMPI won't do anything funky
> that would cause the thread checker problems (like RMDA or writes from
> other processes into shared memory segments). Has anybody used the
> Intel Thread Checker with OpenMPI successfully?
>
> Thanks,
> Curt
--
----------------------------------------------------------------
Dipl.-Inf. Rainer Keller http://www.hlrs.de/people/keller
High Performance Computing Tel: ++49 (0)711-685 6 5858
Center Stuttgart (HLRS) Fax: ++49 (0)711-685 6 5832
POSTAL:Nobelstrasse 19 email: kel...@hlrs.de
ACTUAL:Allmandring 30, R.O.030 AIM:rusraink
70550 Stuttgart
