[Mvapich-discuss] MPI_Ssend causes segfault

[Subramoni, Hari]

Hi, Kent.

Glad to know that you're able to work around this. One way to avoid the configure error would be to install the OFED libraries. We will see how we can get around the issues in the meantime.

Thx,
Hari.

From: Kent Cheung <Kent.Cheung at arm.com>
Sent: Wednesday, April 28, 2021 11:55 AM
To: Subramoni, Hari <subramoni.1 at osu.edu>
Cc: mvapich-discuss at lists.osu.edu
Subject: Re: MPI_Ssend causes segfault

Hi Hari,

Without the device flag, the configure script fails with:

configure: error: 'infiniband/verbs.h not found. Did you specify --with-ib-include=?'

However, after configuring and building with

--with-device=ch3:nemesis

the crash does not reoccur. Is it possible to fix this issue for the other device setting too? Thanks.

Kent
________________________________
From: Subramoni, Hari <subramoni.1 at osu.edu<mailto:subramoni.1 at osu.edu>>
Sent: 28 April 2021 15:48
To: Kent Cheung <Kent.Cheung at arm.com<mailto:Kent.Cheung at arm.com>>
Cc: mvapich-discuss at lists.osu.edu<mailto:mvapich-discuss at lists.osu.edu> <mvapich-discuss at lists.osu.edu<mailto:mvapich-discuss at lists.osu.edu>>; Subramoni, Hari <subramoni.1 at osu.edu<mailto:subramoni.1 at osu.edu>>
Subject: RE: MPI_Ssend causes segfault


Hi, Kent.



Can you please try to remove --with-device=ch3:sock from the configure line, reconfigure the MPI library and try again?



Thx,

Hari.



From: Mvapich-discuss <mvapich-discuss-bounces+subramoni.1=osu.edu at lists.osu.edu<mailto:mvapich-discuss-bounces+subramoni.1=osu.edu at lists.osu.edu>> On Behalf Of Kent Cheung via Mvapich-discuss
Sent: Friday, April 23, 2021 11:47 AM
To: mvapich-discuss at lists.osu.edu<mailto:mvapich-discuss at lists.osu.edu>
Subject: [Mvapich-discuss] MPI_Ssend causes segfault



Hello,



I'm currently running into an issue with this example on a Rhel 8.3 x86_64 single node machine:



#include "mpi.h"



#include <stdio.h>



int main()

{

    int rank;



    MPI_Init(NULL, NULL);

    MPI_Comm_rank(MPI_COMM_WORLD, &rank);



    int number;

    if (rank == 0) {

        number = 42;

        MPI_Ssend(&number, 1, MPI_INT, 1, 0, MPI_COMM_WORLD);

    } else if (rank == 1) {

        MPI_Recv(&number, 1, MPI_INT, 0, 0, MPI_COMM_WORLD,

                 MPI_STATUS_IGNORE);

        printf("Process 1 received number %d from process 0\n",

               number);

    }



    MPI_Finalize();



    return 0;

}



Here is the backtrace:



#0  0x00007f96df5a6e28 in __memmove_avx_unaligned_erms () from /lib64/libc.so.6

#1  0x00007f96df93c8cc in MPIDI_CH3U_Receive_data_found () from /lib/libmpi.so.12

#2  0x00007f96df95ef02 in MPIDI_CH3_PktHandler_EagerSyncSend () from /lib/libmpi.so.12

#3  0x00007f96df96eb71 in MPIDI_CH3I_Progress_handle_sock_event () from /lib/libmpi.so.12

#4  0x00007f96df96ef1e in MPIDI_CH3I_Progress () from /lib/libmpi.so.12

#5  0x00007f96df8e5713 in PMPI_Recv () from /lib/libmpi.so.12

#6  0x00000000004007d6 in main () at main.c:17





If I change MPI_Ssend to MPI_Send, the program runs to completion. If I build and test with OpenMPI 4.1.0, the program runs to completion with either call. For reference, I am using GCC 10.2.0 and the configuration options for MVAPICH2 were



--enable-debug --enable-shared --enable-debuginfo --enable-sharedlib=gcc --with-device=ch3:sock



Is there are configuration flag I require to resolve this or an issue with the example?



Thanks,

Kent



IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osu.edu/pipermail/mvapich-discuss/attachments/20210428/3a074c0a/attachment-0022.html>