The good news is that the issue reported in R25290 is fixed in the latest Intel compilers release (2011.7.256).  The bad news is that both the 2011.6.233 and 2011.7.256 releases identify themselves as V12.1.0 from the command line.  (I reported this bug to Intel already.)  They can only be reliably distinguished using the predefined __INTEL_COMPILER_BUILD_DATE macro.  I verified that the build dates for all three compilers we have -- Linux, Mac OS X, and Windows -- are the same.

I developed a more targeted patch (attached) for OpenMPI 1.4.3 opal/mca/memory/ptmalloc2/malloc.c which disables vectorization for _int_malloc() only if an Intel compiler with the 2011.6.233 release build date is found (__INTEL_COMPILER_BUILD_DATE == 20110811).  This patch could presumably make its way into all the copies of opal/mca/memory/ptmalloc2/malloc.c in the various versions of OpenMPI that are still being maintained.

Larry Baker
US Geological Survey
650-329-5608

On 17 Oct 2011, at 8:18 PM, George Bosilca wrote:

Larry,

Sorry for not updating this thread. The issue was identified and fixed by Rainer in r25290 (https://svn.open-mpi.org/trac/ompi/changeset/25290). Please read the comments and the linked thread on the Intel forum for more info about.

I couldn't find a trace of this being fixed in the 1.4 series, so I would wait upgrading until this issue gets resolved.

  Thanks,
    george.

On Oct 17, 2011, at 23:00 , Larry Baker wrote:

George,

I have not had time to look over the 1.4.3 make check failure for Intel 2011.6.233 compilers.  Have you?

I had planned to get 1.4.3 compiled on all six of our compilers using the latest compiler releases.  I was putting off upgrading to 1.4.4 or 1.5.x until after that to minimize the number of things that could go wrong.  Do you recommend otherwise?

Larry Baker
US Geological Survey
650-329-5608

On 7 Oct 2011, at 6:46 PM, George Bosilca wrote:

The may_alias attribute was part of a forward-looking attribute checking, at a time where few compiler supported them. This explains why they are not widely used in the library itself. Moreover, as they do not affect the compilation itself (as your test highlights this is not the issue with the icc 2011.6.233 compiler), there is no urge to remove the may_alias support.

I just got that particular version of the compiler installed on one of our machines. I'll give it a try over the weekend.

  george.

On Oct 7, 2011, at 20:21 , Larry Baker wrote:

The test for the __may_alias_ attribute uses the following short code snippet:

int * p_value __attribute__ ((__may_alias__));
int
main ()
{

  ;
  return 0;
}

Indeed, for Intel 2011 compilers prior to 2011.6.233, this results in a warning:

root@hydra openmpi-1.4.3]# module load compilers/intel/2011.5.220
[root@hydra openmpi-1.4.3]# icc -c may_alias_test.c 
may_alias_test.c(123): warning #1292: attribute "__may_alias__" ignored
  int * p_value __attribute__ ((__may_alias__));
                                ^

[root@hydra openmpi-1.4.3]# module unload compilers/intel/2011.5.220

[root@hydra openmpi-1.4.3]# module load compilers/intel/2011.6.233
[root@hydra openmpi-1.4.3]# icc -c may_alias_test.c 

I modified ./configure to force

ompi_cv___attribute__may_alias=0

Then I compiled and tested the library.  Unfortunately, the results were exactly the same:

make  check-TESTS
make[3]: Entering directory `/state/partition1/root/src/openmpi-1.4.3/test/datatype'
/bin/sh: line 4: 26326 Segmentation fault      ${dir}$tst
FAIL: checksum
/bin/sh: line 4: 26359 Segmentation fault      ${dir}$tst
FAIL: position
========================================================
2 of 2 tests failed
========================================================

I could not find any use of the may_alias attribute, other than in a #define in opal/include/opal_config_bottom.h.  Is OMPI_HAVE_ATTRIBUTE_MAY_ALIAS just cruft that can be removed?

Larry Baker
US Geological Survey
650-329-5608

On 7 Oct 2011, at 11:08 AM, Larry Baker wrote:

I ran into a problem this past week trying to upgrade our OpenMPI 1.4.3 for the latest Intel 2011 compiler, 2011.6.233.

make check fails with Segmentation Fault errors:

[root@hydra openmpi-1.4.3]# tail -20 ../openmpi-1.4.3-check-intel.6.233.log
/bin/sh ../../libtool --tag=CC   --mode=link icc  -DNDEBUG -g -O3 -finline-functions -fno-strict-aliasing -restrict -pthread -fvisibility=hidden -shared-intel -export-dynamic -shared-intel  -o ddt_pack ddt_pack.o ../../ompi/libmpi.la -lnsl -lutil  
libtool: link: icc -DNDEBUG -g -O3 -finline-functions -fno-strict-aliasing -restrict -pthread -fvisibility=hidden -shared-intel -shared-intel -o .libs/ddt_pack ddt_pack.o -Wl,--export-dynamic  ../../ompi/.libs/libmpi.so /usr/local/src/openmpi-1.4.3/orte/.libs/libopen-rte.so /usr/local/src/openmpi-1.4.3/opal/.libs/libopen-pal.so -ldl -lnsl -lutil -pthread -Wl,-rpath -Wl,/usr/local/lib
make[3]: Leaving directory `/state/partition1/root/src/openmpi-1.4.3/test/datatype'
make  check-TESTS
make[3]: Entering directory `/state/partition1/root/src/openmpi-1.4.3/test/datatype'
/bin/sh: line 4:  6322 Segmentation fault      ${dir}$tst
FAIL: checksum
/bin/sh: line 4:  6355 Segmentation fault      ${dir}$tst
FAIL: position
========================================================
2 of 2 tests failed
========================================================
make[3]: *** [check-TESTS] Error 1
make[3]: Leaving directory `/state/partition1/root/src/openmpi-1.4.3/test/datatype'
make[2]: *** [check-am] Error 2
make[2]: Leaving directory `/state/partition1/root/src/openmpi-1.4.3/test/datatype'
make[1]: *** [check-recursive] Error 1
make[1]: Leaving directory `/state/partition1/root/src/openmpi-1.4.3/test'
make: *** [check-recursive] Error 1

Before trying to track down the problem, I thought I'd describe what I see here in case someone recognizes what might be happening.

We have been using OpenMPI 1.4.3 compiled with the Intel 2011.3.174 compiler.  I've updated the Intel 2011 compilers as they have come out with new versions: 2011.4.191, 2011.5.220, and now 2011.6.233.  However, I've not recompiled OpenMPI 1.4.3 until now.

Since the original compilation of OpenMPI 1.4.3 with the Intel 2011.3.174 compilers, I have installed libnuma and libnuma-devel RPMs on our cluster front end.  I noticed that changed the OpenMPI 1.4.3 ./configure output.  To test that this was not the cause of the problem, I recompiled OpenMPI 1.4.3 using both the CentOS/Rocks GNU compilers and the Intel 2011.3.174 compilers.  They both passed all the make check tests.

To find out when this problem first occurs, I systematically configured, compiled, and checked OpenMPI 1.4.3 with all four versions of the Intel 2011 compilers we have.  We use the modules package to load the compiler environment:

[root@hydra openmpi-1.4.3]# env | grep /opt/intel
LD_LIBRARY_PATH=/opt/intel/composer_xe_2011_sp1.6.233/compiler/lib/intel64:/opt/intel/composer_xe_2011_sp1.6.233/mkl/lib/intel64
PATH=/opt/intel/composer_xe_2011_sp1.6.233/bin/intel64:/usr/kerberos/sbin:/usr/kerberos/bin:/usr/java/latest/bin:/usr/local/sbin:/usr/local/bin:/sbin:/bin:/usr/sbin:/usr/bin:/opt/eclipse:/opt/ganglia/bin:/opt/ganglia/sbin:/opt/maui/bin:/opt/torque/bin:/opt/torque/sbin:/opt/rocks/bin:/opt/rocks/sbin:/root/bin

Here's the steps I use to make and test OpenMPI 1.4.3 (I use a patched version to accommodate the six compilers we have; I've submitted those patches here in the past):

# cd /usr/local/src
# tar -xjf openmpi-1.4.3-patched.tar.bz2
# cd openmpi-1.4.3
# module load compilers/intel/2011.6.233
# ./configure >../openmpi-1.4.3-configure-intel.6.233.log 2>&1 --with-tm --with-openib --without-valgrind --without-udapl --enable-contrib-no-build=vt --with-wrapper-ldflags="-shared-intel" CC="icc" CFLAGS="-g -O3" CXX="icpc" CXXFLAGS="-g -O3" FC="ifort" FCFLAGS="-g -O3" F77="ifort" FFLAGS="-g -O3" LDFLAGS="-shared-intel"
# make >../openmpi-1.4.3-make-intel.6.233.log 2>&1
# make check >../openmpi-1.4.3-check-intel.6.233.log 2>&1

(When I generate the OpenMPI 1.4.3 library we actually use, I also add a --prefix.  But, that complicates diff's of the stdout files for these steps, so it is not used here.  Thus, I do NOT proceed to make install any of these libraries.)

The three earlier versions of the Intel 2011 compilers all pass the make check tests.  When I compare the ./configure stdout files, they are all identical.  However, the ./configure stdout file for the Intel 2011.6.233 compilers has one difference:

[root@hydra openmpi-1.4.3]# diff ../openmpi-1.4.3-configure-intel.{5.220,6.233}.log
178c178
< checking for __attribute__(may_alias)... no
---
> checking for __attribute__(may_alias)... yes

That is obviously where I will start looking for the source of the problem.

Maybe someone reading this list knows what the purpose of that test is, whether the Intel 2011 compilers are expected to have this feature enabled, and whether the code this enables can cause this problem if the Intel 2011.6.233 compilers do not fully support whatever this test is intended to discern.

Larry Baker
US Geological Survey
650-329-5608

_______________________________________________
devel mailing list
devel@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

_______________________________________________
devel mailing list
devel@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

_______________________________________________
devel mailing list
devel@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

_______________________________________________
devel mailing list
devel@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel

_______________________________________________
devel mailing list
devel@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel