[MPICH2-dev] MPICH2/packef fails on testing

[Derbunovich, Andrey]

Dear All,

 

We have found a bug in the test mpi/f77/datatype/packef.f. The bug can
appear on 64-bit platforms like IA64 or EM64T. This bug can cause
collective abort of all ranks at execution time. The error message may
look like this

 

rank 0 in job 236  knsem010_35400   caused collective abort of all ranks
 exit status of rank 0: killed by signal 11

 

The problem appeared when test has been compiled by gcc with non default
optimization flags options tuned on. For instance, you can get this
error message when packef.f test has been compiled with -O2 switch. 

 

The problem is that variable asizev(2) initialized implicitly before
first call of MPI_Pack_external() function. The value of this variable
may be not initialized to zero at this case. This variable passed to
MPI_Parck_external() function as value of current position in the
buffer. So, the unexpected value in this parameter causes invalid memory
reference at first call of MPI_Pack_external() in this test. We suggest
do initialization for this variable obviously before first call of
MPI_Pack_external(). See the attached patch for more details.

 

Best regards,

Andrey

 

-- 

Andrey Derbunovich           E-mail:   Andrey.Derbunovich at intel.com
<blocked::mailto:Andrey.Derbunovich at intel.com> 

Intel                        Phone:    +7 (831 30)70 700

Sarov, Russia                iNET:     313-2480

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.mcs.anl.gov/mailman/private/mpich2-dev/attachments/20051212/73befd39/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: packef.patch
Type: application/octet-stream
Size: 384 bytes
Desc: packef.patch
URL: <https://lists.mcs.anl.gov/mailman/private/mpich2-dev/attachments/20051212/73befd39/attachment.obj>