[mvapich-discuss] MPI_Comm_connect/accept segfault during MPIDI_CH3I_comm_create

[Neil Spruit]

Hello,

I have a program that establishes a connection between two separate mpi
processes (client/server)
through opening an MPI Port and connecting by MPI_Comm_connect and
MPI_Comm_accept. Running this same program with mpich I have no issues, but
if I use mvapich I get the following stack trace:
#0  0x00007ffff7849cab in init_MV2_collops (comm=0x7ffff7d87040) at
src/mpid/ch3/channels/mrail/src/rdma/ch3i_comm.c:31
#1  0x00007ffff7849f40 in MPIDI_CH3I_comm_create (comm=0x7ffff7d87040) at
src/mpid/ch3/channels/mrail/src/rdma/ch3i_comm.c:107
#2  0x00007ffff7810972 in MPIDI_CH3I_Comm_create_hook (comm=0x7ffff7d87040)
at src/mpid/ch3/src/ch3u_comm.c:72
#3  0x00007ffff7816157 in MPIDI_CH3I_Initialize_tmp_comm
(comm_pptr=0x7fffffffce70, vc_ptr=0x7bb9e0, is_low_group=<value optimized
out>, context_id_offset=<value optimized out>) at
src/mpid/ch3/src/ch3u_port.c:320
#4  0x00007ffff7817e05 in MPIDI_Create_inter_root_communicator_connect
(port_name=0x7fffffffcfd0
"tag#0$description#\"#RANK:00000000(00000002:00000176:00000001)#\"$",
info=<value optimized out>, root=0, comm_ptr=0x7ffff7d86c68,
newcomm=0x7fffffffcf68) at src/mpid/ch3/src/ch3u_port.c:154
#5  MPIDI_Comm_connect (port_name=0x7fffffffcfd0
"tag#0$description#\"#RANK:00000000(00000002:00000176:00000001)#\"$",
info=<value optimized out>, root=0, comm_ptr=0x7ffff7d86c68,
newcomm=0x7fffffffcf68) at src/mpid/ch3/src/ch3u_port.c:380
#6  0x00007ffff7833ab3 in MPID_Comm_connect (port_name=0x7fffffffcfd0
"tag#0$description#\"#RANK:00000000(00000002:00000176:00000001)#\"$",
info=<value optimized out>, root=<value optimized out>, comm=<value
optimized out>, newcomm_ptr=<value optimized out>) at
src/mpid/ch3/src/mpid_port.c:188
#7  0x00007ffff7a90da6 in PMPI_Comm_connect (port_name=0x7fffffffcfd0
"tag#0$description#\"#RANK:00000000(00000002:00000176:00000001)#\"$",
info=469762048, root=0, comm=1140850689, newcomm=0x7fffffffd0ec) at
src/mpi/spawn/comm_connect.c:109
#8  0x0000000000400da8 in main ()


I tracked down the operation in init_MV2_collops  that it is failing on and
it seems to be getting a segfault due to the following call
comm->coll_fns->Ibcast_sched = MPIR_Ibcast_MV2;. Is there a configuration I
need to set or an env that must be set for this to work? I am currently
using mvapich2-2.0.

Thank you for your time!

Respectfully,
Neil Spruit
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cse.ohio-state.edu/pipermail/mvapich-discuss/attachments/20141120/eedb82d9/attachment.html>