[mpich-discuss] Segfault with MPICH 3.2+Clang but not GCC
Kenneth Raffenetti
raffenet at mcs.anl.gov
Tue Jul 26 14:57:03 CDT 2016
Thanks for confirming. We'll make sure the fix gets applied to an
upcoming bug-fix release.
Ken
On 07/26/2016 01:56 PM, Andreas Noack wrote:
> I just built 0d6412303488428c461986655a56639cbbdbf705 and there I don't
> see the issue. It's definitely on 3.2, though.
>
> On Tue, Jul 26, 2016 at 2:17 PM, Andreas Noack
> <andreasnoackjensen at gmail.com <mailto:andreasnoackjensen at gmail.com>> wrote:
>
> Thanks for the replies and sorry for the incomplete information. I
> should have written that I get a segfault for np=2 but not np=1.
> Find the exact error message below. I've also tried the program Jeff
> sent and I get the same segfault as before for all np>1. I'll try to
> checkout and build the development version and see if I can reproduce.
>
> ===================================================================================
> = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
> = PID 7703 RUNNING AT 30-9-81.wireless.csail.mit.edu
> <http://30-9-81.wireless.csail.mit.edu>
> = EXIT CODE: 11
> = CLEANING UP REMAINING PROCESSES
> = YOU CAN IGNORE THE BELOW CLEANUP MESSAGES
> ===================================================================================
> YOUR APPLICATION TERMINATED WITH THE EXIT STRING: Segmentation
> fault: 11 (signal 11)
> This typically refers to a problem with your application.
> Please see the FAQ page for debugging suggestions
>
>
> On Tue, Jul 26, 2016 at 12:16 PM, Jeff Hammond
> <jeff.science at gmail.com <mailto:jeff.science at gmail.com>> wrote:
>
> Indeed, Rob is right. I only tested np<=2. Didn't see failure
> until nproc=16 because of how malloc works but in any case, it
> is trivial to fix this by allocating C to be nproc elements.
>
> Jeff
>
> #include <mpi.h>
>
> #include <stdio.h>
>
> #include <stdlib.h>
>
>
> int main(int argc, char *argv[])
>
> {
>
> MPI_Init(&argc, &argv);
>
>
> MPI_Comm comm = MPI_COMM_WORLD;
>
> uint64_t *A, *C;
>
> int rnk, siz;
>
>
> MPI_Comm_rank(comm, &rnk);
>
> MPI_Comm_size(comm, &siz);
>
> A = calloc(1, sizeof(uint64_t));
>
> C = calloc(siz, sizeof(uint64_t));
>
> A[0] = rnk + 1;
>
>
> MPI_Allgather(A, 1, MPI_UINT64_T, C, 1, MPI_UINT64_T, comm);
>
>
> free(C);
>
> free(A);
>
>
> MPI_Finalize();
>
> return 0;
>
> }
>
>
> On Tue, Jul 26, 2016 at 9:00 AM, Rob Latham <robl at mcs.anl.gov
> <mailto:robl at mcs.anl.gov>> wrote:
>
>
>
> On 07/26/2016 10:17 AM, Andreas Noack wrote:
>
> On my El Capitan macbook I get a segfault when running
> the program below
> with more than a single process but only when MPICH has
> been compiled
> with Clang.
>
> I don't get that good debug info but here is some of
> what I got
>
>
>
> valgrind is pretty good at sussing out these sorts of things:
>
> ==18132== Unaddressable byte(s) found during client check
> request
> ==18132== at 0x504D1D7: MPIR_Localcopy (helper_fns.c:84)
> ==18132== by 0x4EC8EA1: MPIR_Allgather_intra
> (allgather.c:169)
> ==18132== by 0x4ECA5EC: MPIR_Allgather (allgather.c:791)
> ==18132== by 0x4ECA7A4: MPIR_Allgather_impl (allgather.c:832)
> ==18132== by 0x4EC8B5C: MPID_Allgather (mpid_coll.h:61)
> ==18132== by 0x4ECB9F7: PMPI_Allgather (allgather.c:978)
> ==18132== by 0x4008F5: main (noack_segv.c:18)
> ==18132== Address 0x6f2f138 is 8 bytes after a block of
> size 16 alloc'd
> ==18132== at 0x4C2FB55: calloc (in
> /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
> ==18132== by 0x4008B0: main (noack_segv.c:15)
> ==18132==
> ==18132== Invalid write of size 8
> ==18132== at 0x4C326CB: memcpy@@GLIBC_2.14 (in
> /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
> ==18132== by 0x504D31B: MPIR_Localcopy (helper_fns.c:84)
> ==18132== by 0x4EC8EA1: MPIR_Allgather_intra
> (allgather.c:169)
> ==18132== by 0x4ECA5EC: MPIR_Allgather (allgather.c:791)
> ==18132== by 0x4ECA7A4: MPIR_Allgather_impl (allgather.c:832)
> ==18132== by 0x4EC8B5C: MPID_Allgather (mpid_coll.h:61)
> ==18132== by 0x4ECB9F7: PMPI_Allgather (allgather.c:978)
> ==18132== by 0x4008F5: main (noack_segv.c:18)
> ==18132== Address 0x6f2f138 is 8 bytes after a block of
> size 16 alloc'd
> ==18132== at 0x4C2FB55: calloc (in
> /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
> ==18132== by 0x4008B0: main (noack_segv.c:15)
>
>
>
> MPI_Comm_rank(comm, &rnk);
> A = calloc(1, sizeof(uint64_t));
> C = calloc(2, sizeof(uint64_t));
> A[0] = rnk + 1;
>
> MPI_Allgather(A, 1, MPI_UINT64_T, C, 1,
> MPI_UINT64_T, comm);
>
>
> Your 'buf count tuple' is ok for A: every process sends one
> uint64
>
> your 'buf count tuple' is too small for C if there are any
> more than 2 proceses .
>
> When you say "more than one"... do you mean 2?
>
> ==rob
>
> _______________________________________________
> discuss mailing list discuss at mpich.org
> <mailto:discuss at mpich.org>
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
>
>
>
>
> --
> Jeff Hammond
> jeff.science at gmail.com <mailto:jeff.science at gmail.com>
> http://jeffhammond.github.io/
>
> _______________________________________________
> discuss mailing list discuss at mpich.org
> <mailto:discuss at mpich.org>
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
>
>
>
>
>
> _______________________________________________
> discuss mailing list discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
>
_______________________________________________
discuss mailing list discuss at mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss
More information about the discuss
mailing list