From: George Bosilca (bosilca_at_[hidden])
Date: 2007-05-03 21:30:19
Usually ... I would say no. But this is a special case (of
course :)). In order to get better performances we align some fields
in our TCP header. As a result there is a small gap in the TCP
headers, which of course don't get initialized. Valgrind detect it
and complain, but it's harmless.
Thanks,
george.
On May 3, 2007, at 5:41 PM, Chudin, Eugene wrote:
> I was wondering if it is expected to have error messages from
> valgrind when checking openmpi code?
>
> For instance, I have following trivial code:
>
> #include <mpi.h>
> #include <iostream>
>
> template <typename T>
> void distribute_val(T& val, int _procid, int _np)
> {
> MPI_Bcast(&val, sizeof(T), MPI_CHAR, 0, MPI_COMM_WORLD);
> }
>
> using namespace std;
>
>
> int main(int argc, char** argv)
> {
> int procid;
> int nproc;
>
> MPI_Init(&argc, &argv);
> MPI_Comm_rank (MPI_COMM_WORLD, &procid);
> MPI_Comm_size (MPI_COMM_WORLD, &nproc);
> double val = 0;
> if(procid == 0)
> val = 3.14159;
> distribute_val(val, procid, nproc);
>
> cout << "ProcID=\t" << procid << "\tval=" << val << endl;
> MPI_Finalize();
> return 0;
> }
>
> Which produces errors in valgrind if I run it on 2 processors
> connected by network.
> If I run it on 2 pocessors located on the same node then I get no
> errors from valgrind.
> In both cases code runs as expected, but I am still worried about
> causes of valgrind errors.
>
> Below is the output from valgrind:
> > mpiCC -g -Wall test.cpp -o test
> > mpirun -np 2 --machinefile ./mpd.2 --prefix /toolbox/openmpi
> valgrind --leak-check=full ./test
> ==14823== Memcheck, a memory error detector.
> ==14823== Copyright (C) 2002-2007, and GNU GPL'd, by Julian Seward
> et al.
> ==14823== Using LibVEX rev 1732, a library for dynamic binary
> translation.
> ==14823== Copyright (C) 2004-2007, and GNU GPL'd, by OpenWorks LLP.
> ==14823== Using valgrind-3.2.3, a dynamic binary instrumentation
> framework.
> ==14823== Copyright (C) 2000-2007, and GNU GPL'd, by Julian Seward
> et al.
> ==14823== For more details, rerun with: -v
> ==14823==
> ==13545== Memcheck, a memory error detector.
> ==13545== Copyright (C) 2002-2007, and GNU GPL'd, by Julian Seward
> et al.
> ==13545== Using LibVEX rev 1732, a library for dynamic binary
> translation.
> ==13545== Copyright (C) 2004-2007, and GNU GPL'd, by OpenWorks LLP.
> ==13545== Using valgrind-3.2.3, a dynamic binary instrumentation
> framework.
> ==13545== Copyright (C) 2000-2007, and GNU GPL'd, by Julian Seward
> et al.
> ==13545== For more details, rerun with: -v
> ==13545==
> ==14823== Syscall param writev(vector[...]) points to uninitialised
> byte(s)
> ==14823== at 0x59BFA86: do_writev (in /lib64/tls/libc.so.6)
> ==14823== by 0x831771E: mca_btl_tcp_frag_send (in /toolbox64/
> openmpi/lib/openmpi/mca_btl_tcp.so)
> ==14823== by 0x83160C9: mca_btl_tcp_endpoint_send_handler (in /
> toolbox64/openmpi/lib/openmpi/mca_btl_tcp.so)
> ==14823== by 0x4F50951: opal_event_base_loop (in /toolbox64/
> openmpi/lib/libopen-pal.so.0.0.0)
> ==14823== by 0x4F509E4: opal_event_loop (in /toolbox64/openmpi/
> lib/libopen-pal.so.0.0.0)
> ==14823== by 0x4F4AE50: opal_progress (in /toolbox64/openmpi/lib/
> libopen-pal.so.0.0.0)
> ==14823== by 0x4C8014B: ompi_request_wait_all (in /toolbox64/
> openmpi/lib/libmpi.so.0.0.0)
> ==14823== by 0x873412D: ompi_coll_tuned_bcast_intra_generic (in /
> toolbox64/openmpi/lib/openmpi/mca_coll_tuned.so)
> ==14823== by 0x8734293: ompi_coll_tuned_bcast_intra_binomial
> (in /toolbox64/openmpi/lib/openmpi/mca_coll_tuned.so)
> ==14823== by 0x872EA9F: ompi_coll_tuned_bcast_intra_dec_fixed
> (in /toolbox64/openmpi/lib/openmpi/mca_coll_tuned.so)
> ==14823== by 0x4C957BA: PMPI_Bcast (in /toolbox64/openmpi/lib/
> libmpi.so.0.0.0)
> ==14823== by 0x408A9D: void distribute_val<double>(double&, int,
> int) (test.cpp:7)
> ==14823== Address 0x41EEE2C is not stack'd, malloc'd or (recently)
> free'd
> ProcID= 0 val=3.14159
> ProcID= 1 val=3.14159
> ==13545==
> ==13545== ERROR SUMMARY: 0 errors from 0 contexts (suppressed: 8
> from 5)
> ==13545== malloc/free: in use at exit: 1,920 bytes in 1 blocks.
> ==13545== malloc/free: 1 allocs, 0 frees, 1,920 bytes allocated.
> ==13545== For counts of detected errors, rerun with: -v
> ==13545== searching for pointers to 1 not-freed blocks.
> ==13545== checked 1,155,400 bytes.
> ==13545==
> ==13545== LEAK SUMMARY:
> ==13545== definitely lost: 0 bytes in 0 blocks.
> ==13545== possibly lost: 0 bytes in 0 blocks.
> ==13545== still reachable: 1,920 bytes in 1 blocks.
> ==13545== suppressed: 0 bytes in 0 blocks.
> ==13545== Reachable blocks (those to which a pointer was found) are
> not shown.
> ==13545== To see them, rerun with: --leak-check=full --show-
> reachable=yes
> ==14823==
> ==14823== ERROR SUMMARY: 2 errors from 1 contexts (suppressed: 7
> from 4)
> ==14823== malloc/free: in use at exit: 1,920 bytes in 1 blocks.
> ==14823== malloc/free: 1 allocs, 0 frees, 1,920 bytes allocated.
> ==14823== For counts of detected errors, rerun with: -v
> ==14823== searching for pointers to 1 not-freed blocks.
> ==14823== checked 1,158,440 bytes.
> ==14823==
> ==14823== LEAK SUMMARY:
> ==14823== definitely lost: 0 bytes in 0 blocks.
> ==14823== possibly lost: 0 bytes in 0 blocks.
> ==14823== still reachable: 1,920 bytes in 1 blocks.
> ==14823== suppressed: 0 bytes in 0 blocks.
> ==14823== Reachable blocks (those to which a pointer was found) are
> not shown.
> ==14823== To see them, rerun with: --leak-check=full --show-
> reachable=yes
>
> ----------------------------------------------------------------------
> --------
> Notice: This e-mail message, together with any attachments, contains
> information of Merck & Co., Inc. (One Merck Drive, Whitehouse Station,
> New Jersey, USA 08889), and/or its affiliates (which may be known
> outside the United States as Merck Frosst, Merck Sharp & Dohme or MSD
> and in Japan, as Banyu - direct contact information for affiliates is
> available at http://www.merck.com/contact/contacts.html) that may be
> confidential, proprietary copyrighted and/or legally privileged. It is
> intended solely for the use of the individual or entity named on this
> message. If you are not the intended recipient, and have received this
> message in error, please notify us immediately by reply e-mail and
> then
> delete it from your system.
>
>
> ----------------------------------------------------------------------
> --------
>
> _______________________________________________
> users mailing list
> users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/users