[mvapich-discuss] RMA data corruption in 2.0, 2.1a
Hajime Fujita
hfujita at uchicago.edu
Fri Oct 3 17:22:42 EDT 2014
Dear MVAPICH2 team,
We found a potential bug in MVAPICH2 2.0 and 2.1a regarding RMA.
When we run the attached program on two nodes (1 process/node), it
produces the wrong result. This setting means inter-process
communications goes over InfiniBand.
# Requesting interactive job with 2 nodes
$ sinteractive -N 2
$ mpiexec ./rma_putget_test # launches 2 proc
loc_buff[1024]=-1 != 1024
If it works correctly, there would be no output.
If we run this on a single machine with multiple processes, it runs
correctly.
If I'm using MPI RMA functions in some incorrect way please let me know.
Hardware platform:
UChicago RCC Midway
http://rcc.uchicago.edu/resources/midway_specs.html
MVAPICH versions and configurations:
[hfujita at midway-login1 ~]$ mpichversion
MVAPICH2 Version: 2.1a
MVAPICH2 Release date: Sun Sep 21 12:00:00 EDT 2014
MVAPICH2 Device: ch3:mrail
MVAPICH2 configure:
--prefix=/project/aachien/local/mvapich2-2.1a-gcc-4.8 --enable-shared
MVAPICH2 CC: gcc -DNDEBUG -DNVALGRIND -O2
MVAPICH2 CXX: g++ -DNDEBUG -DNVALGRIND -O2
MVAPICH2 F77: gfortran -L/lib -L/lib -O2
MVAPICH2 FC: gfortran -O2
[hfujita at midway-login1 ~]$ mpichversion
MVAPICH2 Version: 2.0
MVAPICH2 Release date: Fri Jun 20 20:00:00 EDT 2014
MVAPICH2 Device: ch3:mrail
MVAPICH2 configure: --prefix=/software/mvapich2-2.0-el6-x86_64
--enable-shared
MVAPICH2 CC: gcc -DNDEBUG -DNVALGRIND -O2
MVAPICH2 CXX: g++ -DNDEBUG -DNVALGRIND
MVAPICH2 F77: gfortran -L/lib -L/lib -O2
MVAPICH2 FC: gfortran
Thank you,
Hajime
--
Hajime Fujita
Postdoctoral Scholar, Large-Scale Systems Group
Department of Computer Science, The University of Chicago
http://www.cs.uchicago.edu/people/hfujita
-------------- next part --------------
A non-text attachment was scrubbed...
Name: rma_putget_test.c
Type: text/x-csrc
Size: 2109 bytes
Desc: not available
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20141003/d9f8935d/attachment.bin>
More information about the mvapich-discuss
mailing list