OK, thanks for clarification and code pointers.
I'll update "rte.h" to reflect the updates.
Thomas Naughton naughtont_at_[hidden]
Research Associate (865) 576-4184
On Wed, 18 Dec 2013, Ralph Castain wrote:
> There is no relation at all between ompi_proc_t and ompi_process_info_t. The ompi_proc_t is defined in the MPI layer and is used in that layer in various places very much like orte_proc_t is used in the ORTE layer.
> If you look in ompi/mca/rte/orte/rte_orte.c, you'll see how we handle the revised function calls. Basically, we use the process name to retrieve the modex data via the opal_db, and then load a pointer to the hostname into the ompi_proc_t proc_hostname field. Thus, the definition of ompi_proc_t remains in the MPI layer.
> So there was no need to change the ompi/mca/rte/rte.h file, nor to #define anything in the component .h file - just have to modify the wrapper code inside the RTE component itself.
> On Dec 18, 2013, at 1:50 PM, Thomas Naughton <naughtont_at_[hidden]> wrote:
>> Hi Ralph,
>> Question about the MPI-RTE interface change in r29931. The change was not
>> reflected in the "ompi/mca/rte/rte.h" file.
>> I'm curious how the newly added "struct ompi_proc_t" relates to the "struct ompi_process_info_t" that is described in the "rte.h" file?
>> I understand the general motivation for the API change but it is less clear
>> to me how the information previously defined in the header changes (or does
>> not change)?
>> Thomas Naughton naughtont_at_[hidden]
>> Research Associate (865) 576-4184
>> On Mon, 16 Dec 2013, svn-commit-mailer_at_[hidden] wrote:
>>> Author: rhc (Ralph Castain)
>>> Date: 2013-12-16 22:26:00 EST (Mon, 16 Dec 2013)
>>> New Revision: 29931
>>> URL: https://svn.open-mpi.org/trac/ompi/changeset/29931
>>> Revert r29917 and replace it with a fix that resolves the thread deadlock while retaining the desired debug info. In an earlier commit, we had changed the modex accordingly:
>>> * automatically retrieve the hostname (and all RTE info) for all procs during MPI_Init if nprocs < cutoff
>>> * if nprocs > cutoff, retrieve the hostname (and all RTE info) for a proc upon the first call to modex_recv for that proc. This would provide the hostname for debugging purposes as we only report errors on messages, and so we must have called modex_recv to get the endpoint info
>>> * BTLs are not to call modex_recv until they need the endpoint info for first message - i.e., not during add_procs so we don't call it for every process in the job, but only those with whom we communicate
>>> My understanding is that only some BTLs have been modified to meet that third requirement, but those include the Cray ones where jobs are big enough that launch times were becoming an issue. Other BTLs would hopefully be modified as time went on and interest in using them at scale arose. Meantime, those BTLs would call modex_recv on every proc, and we would therefore be no worse than the prior behavior.
>>> This commit revises the MPI-RTE interface to pass the ompi_proc_t instead of the ompi_process_name_t for the proc so that the hostname can be easily inserted. I have advised the ORNL folks of the change.
>>> cmr=v1.7.4:reviewer=jsquyres:subject=Fix thread deadlock
>>> Text files modified:
>>> trunk/ompi/mca/rte/orte/rte_orte.h | 7 ++++---
>>> trunk/ompi/mca/rte/orte/rte_orte_module.c | 27 ++++++++++++++++++---------
>>> trunk/ompi/proc/proc.c | 26 ++++++++++++++++++++++----
>>> trunk/ompi/runtime/ompi_module_exchange.c | 10 +++++-----
>>> 4 files changed, 49 insertions(+), 21 deletions(-)
>> devel mailing list
> devel mailing list