Re: [OMPI devel] [OMPI svn] svn:open-mpi r17941
Hmmm...puzzling. It is working fine for me on TM machines and on my Mac. However, Galen reports it borked on alps as well. I'll have to dig a little to check this out and see if there is something missing on those PLMs. Will get back shortly. Sorry for problem On 3/27/08 10:28 AM, "Tim Prins"wrote: > Unfortunately now with r17988 I cannot run any mpi programs, they seem > to hang in the modex. > > Tim > > Ralph H Castain wrote: >> Thanks Tim - I found the problem and will commit a fix shortly. >> >> Appreciate your testing and reporting! >> >> >> On 3/27/08 8:24 AM, "Tim Prins" wrote: >> >>> This commit breaks things for me. Running on 3 nodes of odin: >>> >>> mpirun -mca btl tcp,sm,self examples/ring_c >>> >>> causes a hang. All of the processes are stuck in >>> orte_grpcomm_base_barrier during MPI_Finalize. Not all programs hang, >>> and the ring program does not hang all the time, but fairly often. >>> >>> Tim >>> >>> r...@osl.iu.edu wrote: Author: rhc Date: 2008-03-24 16:50:31 EDT (Mon, 24 Mar 2008) New Revision: 17941 URL: https://svn.open-mpi.org/trac/ompi/changeset/17941 Log: Fix the allgather and allgather_list functions to avoid deadlocks at large node/proc counts. Violated the RML rules here - we received the allgather buffer and then did an xcast, which causes a send to go out, and is then subsequently received by the sender. This fix breaks that pattern by forcing the recv to complete outside of the function itself - thus, the allgather and allgather_list always complete their recvs before returning or sending. Reogranize the grpcomm code a little to provide support for soon-to-come new grpcomm components. The revised organization puts what will be common code elements in the base to avoid duplication, while allowing components that don't need those functions to ignore them. Added: trunk/orte/mca/grpcomm/base/grpcomm_base_allgather.c trunk/orte/mca/grpcomm/base/grpcomm_base_barrier.c trunk/orte/mca/grpcomm/base/grpcomm_base_modex.c Text files modified: trunk/orte/mca/grpcomm/base/Makefile.am| 5 trunk/orte/mca/grpcomm/base/base.h |23 + trunk/orte/mca/grpcomm/base/grpcomm_base_close.c | 4 trunk/orte/mca/grpcomm/base/grpcomm_base_open.c| 1 trunk/orte/mca/grpcomm/base/grpcomm_base_select.c | 121 ++--- trunk/orte/mca/grpcomm/basic/grpcomm_basic.h |16 trunk/orte/mca/grpcomm/basic/grpcomm_basic_component.c |30 - trunk/orte/mca/grpcomm/basic/grpcomm_basic_module.c| 845 ++- trunk/orte/mca/grpcomm/cnos/grpcomm_cnos.h | 8 trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_component.c | 8 trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_module.c |21 trunk/orte/mca/grpcomm/grpcomm.h |45 + trunk/orte/mca/rml/rml_types.h |31 trunk/orte/orted/orted_comm.c |27 + 14 files changed, 226 insertions(+), 959 deletions(-) Diff not shown due to size (92619 bytes). To see the diff, run the following command: svn diff -r 17940:17941 --no-diff-deleted ___ svn mailing list s...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/svn >>> ___ >>> devel mailing list >>> de...@open-mpi.org >>> http://www.open-mpi.org/mailman/listinfo.cgi/devel >> >> >> ___ >> devel mailing list >> de...@open-mpi.org >> http://www.open-mpi.org/mailman/listinfo.cgi/devel > > ___ > devel mailing list > de...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/devel
Re: [OMPI devel] [OMPI svn] svn:open-mpi r17941
Unfortunately now with r17988 I cannot run any mpi programs, they seem to hang in the modex. Tim Ralph H Castain wrote: Thanks Tim - I found the problem and will commit a fix shortly. Appreciate your testing and reporting! On 3/27/08 8:24 AM, "Tim Prins"wrote: This commit breaks things for me. Running on 3 nodes of odin: mpirun -mca btl tcp,sm,self examples/ring_c causes a hang. All of the processes are stuck in orte_grpcomm_base_barrier during MPI_Finalize. Not all programs hang, and the ring program does not hang all the time, but fairly often. Tim r...@osl.iu.edu wrote: Author: rhc Date: 2008-03-24 16:50:31 EDT (Mon, 24 Mar 2008) New Revision: 17941 URL: https://svn.open-mpi.org/trac/ompi/changeset/17941 Log: Fix the allgather and allgather_list functions to avoid deadlocks at large node/proc counts. Violated the RML rules here - we received the allgather buffer and then did an xcast, which causes a send to go out, and is then subsequently received by the sender. This fix breaks that pattern by forcing the recv to complete outside of the function itself - thus, the allgather and allgather_list always complete their recvs before returning or sending. Reogranize the grpcomm code a little to provide support for soon-to-come new grpcomm components. The revised organization puts what will be common code elements in the base to avoid duplication, while allowing components that don't need those functions to ignore them. Added: trunk/orte/mca/grpcomm/base/grpcomm_base_allgather.c trunk/orte/mca/grpcomm/base/grpcomm_base_barrier.c trunk/orte/mca/grpcomm/base/grpcomm_base_modex.c Text files modified: trunk/orte/mca/grpcomm/base/Makefile.am| 5 trunk/orte/mca/grpcomm/base/base.h |23 + trunk/orte/mca/grpcomm/base/grpcomm_base_close.c | 4 trunk/orte/mca/grpcomm/base/grpcomm_base_open.c| 1 trunk/orte/mca/grpcomm/base/grpcomm_base_select.c | 121 ++--- trunk/orte/mca/grpcomm/basic/grpcomm_basic.h |16 trunk/orte/mca/grpcomm/basic/grpcomm_basic_component.c |30 - trunk/orte/mca/grpcomm/basic/grpcomm_basic_module.c| 845 ++- trunk/orte/mca/grpcomm/cnos/grpcomm_cnos.h | 8 trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_component.c | 8 trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_module.c |21 trunk/orte/mca/grpcomm/grpcomm.h |45 + trunk/orte/mca/rml/rml_types.h |31 trunk/orte/orted/orted_comm.c |27 + 14 files changed, 226 insertions(+), 959 deletions(-) Diff not shown due to size (92619 bytes). To see the diff, run the following command: svn diff -r 17940:17941 --no-diff-deleted ___ svn mailing list s...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/svn ___ devel mailing list de...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/devel ___ devel mailing list de...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/devel
Re: [OMPI devel] [OMPI svn] svn:open-mpi r17941
Thanks Tim - I found the problem and will commit a fix shortly. Appreciate your testing and reporting! On 3/27/08 8:24 AM, "Tim Prins"wrote: > This commit breaks things for me. Running on 3 nodes of odin: > > mpirun -mca btl tcp,sm,self examples/ring_c > > causes a hang. All of the processes are stuck in > orte_grpcomm_base_barrier during MPI_Finalize. Not all programs hang, > and the ring program does not hang all the time, but fairly often. > > Tim > > r...@osl.iu.edu wrote: >> Author: rhc >> Date: 2008-03-24 16:50:31 EDT (Mon, 24 Mar 2008) >> New Revision: 17941 >> URL: https://svn.open-mpi.org/trac/ompi/changeset/17941 >> >> Log: >> Fix the allgather and allgather_list functions to avoid deadlocks at large >> node/proc counts. Violated the RML rules here - we received the allgather >> buffer and then did an xcast, which causes a send to go out, and is then >> subsequently received by the sender. This fix breaks that pattern by forcing >> the recv to complete outside of the function itself - thus, the allgather and >> allgather_list always complete their recvs before returning or sending. >> >> Reogranize the grpcomm code a little to provide support for soon-to-come new >> grpcomm components. The revised organization puts what will be common code >> elements in the base to avoid duplication, while allowing components that >> don't need those functions to ignore them. >> >> Added: >>trunk/orte/mca/grpcomm/base/grpcomm_base_allgather.c >>trunk/orte/mca/grpcomm/base/grpcomm_base_barrier.c >>trunk/orte/mca/grpcomm/base/grpcomm_base_modex.c >> Text files modified: >>trunk/orte/mca/grpcomm/base/Makefile.am| 5 >>trunk/orte/mca/grpcomm/base/base.h |23 + >>trunk/orte/mca/grpcomm/base/grpcomm_base_close.c | 4 >>trunk/orte/mca/grpcomm/base/grpcomm_base_open.c| 1 >>trunk/orte/mca/grpcomm/base/grpcomm_base_select.c | 121 ++--- >>trunk/orte/mca/grpcomm/basic/grpcomm_basic.h |16 >>trunk/orte/mca/grpcomm/basic/grpcomm_basic_component.c |30 - >>trunk/orte/mca/grpcomm/basic/grpcomm_basic_module.c| 845 >> ++- >>trunk/orte/mca/grpcomm/cnos/grpcomm_cnos.h | 8 >>trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_component.c | 8 >>trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_module.c |21 >>trunk/orte/mca/grpcomm/grpcomm.h |45 + >>trunk/orte/mca/rml/rml_types.h |31 >>trunk/orte/orted/orted_comm.c |27 + >>14 files changed, 226 insertions(+), 959 deletions(-) >> >> >> Diff not shown due to size (92619 bytes). >> To see the diff, run the following command: >> >> svn diff -r 17940:17941 --no-diff-deleted >> >> ___ >> svn mailing list >> s...@open-mpi.org >> http://www.open-mpi.org/mailman/listinfo.cgi/svn > > ___ > devel mailing list > de...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/devel
Re: [OMPI devel] [OMPI svn] svn:open-mpi r17941
This commit breaks things for me. Running on 3 nodes of odin: mpirun -mca btl tcp,sm,self examples/ring_c causes a hang. All of the processes are stuck in orte_grpcomm_base_barrier during MPI_Finalize. Not all programs hang, and the ring program does not hang all the time, but fairly often. Tim r...@osl.iu.edu wrote: Author: rhc Date: 2008-03-24 16:50:31 EDT (Mon, 24 Mar 2008) New Revision: 17941 URL: https://svn.open-mpi.org/trac/ompi/changeset/17941 Log: Fix the allgather and allgather_list functions to avoid deadlocks at large node/proc counts. Violated the RML rules here - we received the allgather buffer and then did an xcast, which causes a send to go out, and is then subsequently received by the sender. This fix breaks that pattern by forcing the recv to complete outside of the function itself - thus, the allgather and allgather_list always complete their recvs before returning or sending. Reogranize the grpcomm code a little to provide support for soon-to-come new grpcomm components. The revised organization puts what will be common code elements in the base to avoid duplication, while allowing components that don't need those functions to ignore them. Added: trunk/orte/mca/grpcomm/base/grpcomm_base_allgather.c trunk/orte/mca/grpcomm/base/grpcomm_base_barrier.c trunk/orte/mca/grpcomm/base/grpcomm_base_modex.c Text files modified: trunk/orte/mca/grpcomm/base/Makefile.am| 5 trunk/orte/mca/grpcomm/base/base.h |23 + trunk/orte/mca/grpcomm/base/grpcomm_base_close.c | 4 trunk/orte/mca/grpcomm/base/grpcomm_base_open.c| 1 trunk/orte/mca/grpcomm/base/grpcomm_base_select.c | 121 ++--- trunk/orte/mca/grpcomm/basic/grpcomm_basic.h |16 trunk/orte/mca/grpcomm/basic/grpcomm_basic_component.c |30 - trunk/orte/mca/grpcomm/basic/grpcomm_basic_module.c| 845 ++- trunk/orte/mca/grpcomm/cnos/grpcomm_cnos.h | 8 trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_component.c | 8 trunk/orte/mca/grpcomm/cnos/grpcomm_cnos_module.c |21 trunk/orte/mca/grpcomm/grpcomm.h |45 + trunk/orte/mca/rml/rml_types.h |31 trunk/orte/orted/orted_comm.c |27 + 14 files changed, 226 insertions(+), 959 deletions(-) Diff not shown due to size (92619 bytes). To see the diff, run the following command: svn diff -r 17940:17941 --no-diff-deleted ___ svn mailing list s...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/svn