[mvapich-discuss] Segmentation fault in MPI_Reduce using MPI_IN_PLACE and root != 0

Markus Geimer mg1102 at web.de
Tue May 19 10:55:29 EDT 2015


Dear MVAPICH developers,

While testing my MPI application with MVAPICH2 2.1, I ran
into a segmentation fault.  After an in-depth investigation
I could boil down the issue to the attached minimal example.

Calling MPI_Reduce using MPI_IN_PLACE and root != 0 multiple
times causes the segfault.  The example works for me with
ITER <= 16, but fails with ITER >= 17.  If root == 0, it
works regardless of the number of iterations. Note that the
example also works fine with MPICH 3.1 and Open MPI 1.8.3.

The test system is running Debian 7 with kernel 3.2.65; the
compiler used is GCC 4.8.3 (built from vanilla source).
Please find some more detailed information below.  If you
need more -- or have a patch for me to try out -- please let
me know.

Thanks,
Markus

----- 8< ----- 8< ----- 8< ----- 8< ----- 8< ----- 8< -----

$ mpiexec -n 2 ./a.out
[host:mpi_rank_1][error_sighandler] Caught error: Segmentation fault
(signal 11)
[host:mpi_rank_1][print_backtrace]   0:
/opt/mvapich2-2.1/lib/libmpi.so.12(print_backtrace+0x20) [0x7f553b6b404c]
[host:mpi_rank_1][print_backtrace]   1:
/opt/mvapich2-2.1/lib/libmpi.so.12(error_sighandler+0x77) [0x7f553b6b419e]
[host:mpi_rank_1][print_backtrace]   2:
/lib/x86_64-linux-gnu/libc.so.6(+0x321e0) [0x7f553ae331e0]
[host:mpi_rank_1][print_backtrace]   3:
/lib/x86_64-linux-gnu/libc.so.6(+0x12abd0) [0x7f553af2bbd0]
[host:mpi_rank_1][print_backtrace]   4:
/opt/mvapich2-2.1/lib/libmpi.so.12(+0x3f6857) [0x7f553b582857]
[host:mpi_rank_1][print_backtrace]   5:
/opt/mvapich2-2.1/lib/libmpi.so.12(MPIR_Localcopy+0x357) [0x7f553b5833b4]
[host:mpi_rank_1][print_backtrace]   6:
/opt/mvapich2-2.1/lib/libmpi.so.12(MPIR_Reduce_shmem_MV2+0x385)
[0x7f553b2a3d50]
[host:mpi_rank_1][print_backtrace]   7:
/opt/mvapich2-2.1/lib/libmpi.so.12(MPIR_Reduce_two_level_helper_MV2+0x594)
[0x7f553b2a57ec]
[host:mpi_rank_1][print_backtrace]   8:
/opt/mvapich2-2.1/lib/libmpi.so.12(MPIR_Reduce_index_tuned_intra_MV2+0x97b)
[0x7f553b2a6af3]
[host:mpi_rank_1][print_backtrace]   9:
/opt/mvapich2-2.1/lib/libmpi.so.12(MPIR_Reduce_MV2+0x9f) [0x7f553b2a6ea0]
[host:mpi_rank_1][print_backtrace]  10:
/opt/mvapich2-2.1/lib/libmpi.so.12(MPIR_Reduce_impl+0x8c) [0x7f553b21534b]
[host:mpi_rank_1][print_backtrace]  11:
/opt/mvapich2-2.1/lib/libmpi.so.12(PMPI_Reduce+0x1709) [0x7f553b216b9b]
[host:mpi_rank_1][print_backtrace]  12: ./a.out() [0x400838]
[host:mpi_rank_1][print_backtrace]  13:
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xfd) [0x7f553ae1fead]
[host:mpi_rank_1][print_backtrace]  14: ./a.out() [0x4006c9]

$ mpiname -a
MVAPICH2 2.1 Fri Apr 03 20:00:00 EDT 2015 ch3:mrail

Compilation
CC: gcc    -g -O0
CXX: g++   -g -O0
F77: gfortran -L/lib -L/lib   -g -O0
FC: gfortran   -g -O0

Configuration
--prefix=/opt/mvapich2-2.1 --enable-shared --enable-g=dbg --disable-fast
--enable-debuginfo

$ ibv_devinfo
hca_id: qib0
        transport:                      InfiniBand (0)
        fw_ver:                         0.0.0
        node_guid:                      0011:7500:00ff:d76b
        sys_image_guid:                 0011:7500:00ff:d76b
        vendor_id:                      0x1175
        vendor_part_id:                 16
        hw_ver:                         0x2
        board_id:                       InfiniPath_QLE7140
        phys_port_cnt:                  1
                port:   1
                        state:                  PORT_ACTIVE (4)
                        max_mtu:                4096 (5)
                        active_mtu:             2048 (4)
                        sm_lid:                 1
                        port_lid:               1
                        port_lmc:               0x00
                        link_layer:             InfiniBand
-------------- next part --------------
A non-text attachment was scrubbed...
Name: inplace_bug.c
Type: text/x-csrc
Size: 490 bytes
Desc: not available
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20150519/e2cdf650/attachment.bin>


More information about the mvapich-discuss mailing list