1
1
Граф коммитов

9092 Коммитов

Автор SHA1 Сообщение Дата
Ralph Castain
a0ef517550 Fix some errors in the bproc components that prevented compiling. Thought I had already done this, but either those changes were lost when I did the merge, or my old man's memory is fading....
Whaz-at??? :-)

This commit was SVN r12874.
2006-12-15 19:40:04 +00:00
Brian Barrett
01e8fc5f91 Redo of r12871, without the preconnect code change:
Move the req_mtl structure back to the end of each of the structures in 
the CM PML. The req_mtl structure is cast into a mtl_*_request_structure 
for each MTL, which is larger than the req_mtl itself. The cast will cause
the *_request to overwrite parts of the heavy requests if the req_mtl
isn't the *LAST* thing on each structure (hence the comment). This was 
moved as an optimization at some point, which caused buffer sends to fail...

Refs trac:669

This commit was SVN r12873.

The following SVN revision numbers were found above:
  r12871 --> open-mpi/ompi@597598b712

The following Trac tickets were found above:
  Ticket 669 --> https://svn.open-mpi.org/trac/ompi/ticket/669
2006-12-15 17:54:14 +00:00
Brian Barrett
bdf0b231b2 Undo r12871, as it contained some code in ompi/runtime that shouldn't have been
committed

Refs trac:669

This commit was SVN r12872.

The following SVN revision numbers were found above:
  r12871 --> open-mpi/ompi@597598b712

The following Trac tickets were found above:
  Ticket 669 --> https://svn.open-mpi.org/trac/ompi/ticket/669
2006-12-15 17:52:13 +00:00
Brian Barrett
597598b712 Move the req_mtl structure back to the end of each of the structures in the
CM PML.  The req_mtl structure is cast into a mtl_*_request_structure for
each MTL, which is larger than the req_mtl itself.  The cast will cause
the *_request to overwrite parts of the heavy requests if the req_mtl
isn't the *LAST* thing on each structure (hence the comment).  This was
moved as an optimization at some point, which caused buffer sends to
fail...

Refs trac:669

This commit was SVN r12871.

The following Trac tickets were found above:
  Ticket 669 --> https://svn.open-mpi.org/trac/ompi/ticket/669
2006-12-15 17:46:53 +00:00
Ralph Castain
1e1d0e8a89 Set the app_num attribute into the process environment so we pick it up on the other end
This commit was SVN r12868.
2006-12-15 16:43:52 +00:00
Ralph Castain
677d1260aa cleanup nicely if we don't launch
This commit was SVN r12867.
2006-12-15 14:03:53 +00:00
Ralph Castain
cbb660504c Retain the ability to run valgrind on the bproc launcher - do not call bproc_version if "nolaunch" is specified.
This commit was SVN r12866.
2006-12-15 14:01:21 +00:00
Ralph Castain
64ec238b7b Repair support for Bproc 4 on 64-bit systems. Update the SMR framework to actually support the begin_monitoring API. Implement the get/set_node_state APIs.
This commit was SVN r12864.
2006-12-15 02:34:14 +00:00
Rainer Keller
b99e5a71d1 - Help message in case of MPI-application with two init or
calling init functions after finalize.

This commit was SVN r12858.
2006-12-14 19:58:04 +00:00
Brad Benton
18da4c40d3 Set the QP's static rate from the associated MCA parameter, rather
than just defaulting to 0.

Fixes trac:675

This commit was SVN r12855.

The following Trac tickets were found above:
  Ticket 675 --> https://svn.open-mpi.org/trac/ompi/ticket/675
2006-12-14 19:42:24 +00:00
Rolf vandeVaart
c51c36c4a2 These changes fix two issues.
1. For OS's without the dirent.d_type field, we were potentially
not initializing a filename string.  This could result in a 
directory not being cleaned up.
2. Potential memory leaks in filename strings that were allocated.

Refs trac:678

This commit was SVN r12853.

The following Trac tickets were found above:
  Ticket 678 --> https://svn.open-mpi.org/trac/ompi/ticket/678
2006-12-14 18:27:27 +00:00
Brian Barrett
38c2e43ac2 Print out error string rather than errno for TCP-related errors, making it easier for both the user and us to debug issues with BTL and OOB issues...
This commit was SVN r12852.
2006-12-14 18:20:43 +00:00
Jeff Squyres
0ca8cb35b7 Fixes trac:366
Add ability for ini files to recognize "use_eager_rdma" flag.  Set the
default to "no" (because we should assume that HCAs cannot support the
property necessary for using RDMA for eager messages -- that the last
byte of the message is guaranteed to be written to memory last --
unless proven otherwise.  For example, iWARP cards apparently do not
provide this guarantee), and then set all Mellanox and IBM HCAs to
override the default to enable this behavior on these cards.

This commit was SVN r12851.

The following Trac tickets were found above:
  Ticket 366 --> https://svn.open-mpi.org/trac/ompi/ticket/366
2006-12-14 15:52:13 +00:00
Dan Lacher
e3f749acc4 Ticket: #673
Submitted by: Dan Lacher

This commit was SVN r12844.
2006-12-13 20:01:16 +00:00
Ralph Castain
7b8f445e13 Modify the "--display-map-at-launch" option to just "--display-map". Now that we have a "--do-not-launch" option, the "-at-launch" part of the display-map option was confusing. "--display-map" displays the resulting process map before we launch anyway, so this is clearer.
This commit was SVN r12840.
2006-12-13 13:49:15 +00:00
Ralph Castain
82946cb220 Add a new option to orterun: "--do-not-launch" directs the system to do the allocation, map, job setup, etc., but don't actually launch the job. This lets us test all the setup portions of the code.
Also, take the first step in updating how we handle mca params in ORTE - bring it closer to how it is done in the other two layers. Much more work to be done here.

This commit was SVN r12838.
2006-12-13 04:51:38 +00:00
George Bosilca
80bc0c8868 Allow the MX to survive if we are unable to connect to a peer. The PML will
try to find another route.

This commit was SVN r12837.
2006-12-13 01:12:07 +00:00
Mohamad Chaarawi
cae083dec6 replaced the old CID allocation algorithm with the blocked algorithm. The
impace in the communicator directory is still not great since the interface
for allocating a Cid has not changed..

This commit was SVN r12836.
2006-12-12 22:01:39 +00:00
Brian Barrett
10af8ab454 Corrections for when threading is enabled.
Refs trac:564

This commit was SVN r12830.

The following Trac tickets were found above:
  Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
2006-12-12 18:48:42 +00:00
Ralph Castain
3b064a624e For convenience, revise the orte_job_map_t object so it includes the vpid start/range values, the number of nodes, and the number of processes on each node. These values are all used in various places in the code base - we currently re-compute them multiple times. Since these values do not change and are already being computed by the RMAPS framework, we might as well just save them for re-use.
This commit was SVN r12829.
2006-12-12 16:07:23 +00:00
Brad Benton
337116d5fd Added IBM eHCA vendor and part id info
This commit was SVN r12827.
2006-12-12 14:12:39 +00:00
Ralph Castain
28ce8e5e5e Extend the mpirun options to support "--npernode N". This option tells the system to spawn N procs/node across all nodes in the allocation. If N is greater than the number of allocated slots, then the usual oversubscription logic will apply (i.e., the system will error out if oversubscription is not allowed, otherwise it will run with the sched_yield set to non-aggressive behavior).
In "--npernode" operation, the "-np" command line parameter is ignored.

This commit was SVN r12826.
2006-12-12 00:54:05 +00:00
Brian Barrett
cf196ce420 Instead of an unknown proc list that requires ownership transfer of data (which, in turn, requires a complex series of locks to be held during the transfer), use a modex backing store with backpointers from the proc to the backing store. The proc structures no longer own the modex data, which greatly simplifies locking when an unknown proc suddenly becomes known.
Refs trac:564

This commit was SVN r12822.

The following Trac tickets were found above:
  Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
2006-12-11 21:27:30 +00:00
Ralph Castain
8314e8dbb9 Modify the pernode option so it can accept a request for the number of processes to be launched. We now check three use-cases for pernode:
1. no -np provided - put one proc/node across all allocated nodes

2. -np N provided, N > #nodes - we print a pretty error message and exit

3. -np N provided, N <= #nodes - put one proc/node across N nodes

I also added a new orte constant (ORTE_ERR_SILENT) that allows us to pass up the chain that an error was encountered, but NOT print ORTE_ERROR_LOG messages. This is intended to be used for cases where the error we encounter is NOT an orte error, but rather is one associated with incorrect user input (e.g., the preceding case 2). In such cases, there is no point in printing an ORTE_ERROR_LOG chain of messages as it isn't an orte error.

This commit was SVN r12821.
2006-12-11 18:07:07 +00:00
Ralph Castain
0a5d41857a Complete next round of message size reduction: "strip" the descriptive info from the returned values. I have now added a flag to the gpr address mode (ORTE_GPR_STRIPPED) that instructs the gpr to not include segment names or tokens in the returned gpr_value_t objects.
I found only two places that were looking at the tokens:

1. the odls - we used the tokens to separately process the globals container data from everything else. In this case, I left the subscription that returned the globals data alone, but "stripped" the subscription that returned the launch data for the procs. These subscriptions have nothing to do with the xcast message.

2. the pml_base_modex - the callback function was getting process names from the returned tokens. Actually, this function was doing a very bad thing - it was assuming that the first token returned was *always* the process name. This is currently true, but is one of those assumptions that someone could have easily changed - and suddenly found the system inexplicably failing. I modified the function to (a) get the name sent back to us, (b) "stripped" the value structures of tokens and segment strings, and (c) correctly obtained process names from the returned values. I also reindented the heck out of the code so it was legible (at least, to my old eyes).

This commit was SVN r12813.
2006-12-09 23:10:25 +00:00
Jeff Squyres
e70ef98ea6 Update the help message to be a bit more specific and refer to the web
FAQ.

This commit was SVN r12812.
2006-12-09 15:13:03 +00:00
Jeff Squyres
c7282855e7 Fixes trac:659
This commit fixes several aspects regarding MPI conformance of requests.

 * Eliminate the last argument of ompi_errhandler_request_invoke(); we
   ''always'' want to invoke the back-end exception handler with the
   real error code.
 * Make it clear in comments that we only invoke the ''first''
   exception in a given array of requests, even if there's more than
   one request with a non-MPI_SUCCESS value for MPI_ERROR.
 * Defer the freeing of requests upon exception in the back-end
   functions to MPI_WAIT* and MPI_TEST* until later; the requests are
   kept so that we know what handler to invoke when we actually invoke
   the exception.  After figuring that out, ''then'' we free requests
   with pending exceptions on them.
 * Clean up return codes from the back-end MPI_TEST* and MPI_WAIT*
   functions.
 * Slightly modify ompi_errcode_get_mpi_code() to return unity if it
   receives an MPI error code (vs. an OMPI error code).

This commit was SVN r12810.

The following Trac tickets were found above:
  Ticket 659 --> https://svn.open-mpi.org/trac/ompi/ticket/659
2006-12-09 14:20:08 +00:00
Ralph Castain
58569546ed Fix the fix to remove compiler warning - an incorrect "\" was placed in the command string.
This commit was SVN r12805.
2006-12-08 04:17:38 +00:00
Jeff Squyres
568d20de37 Add Myricom to the aggregated list.
This commit was SVN r12804.
2006-12-08 00:11:51 +00:00
Patrick Geoffray
58c6f8c8e1 Copyright update. Thanks to Jeff to remind me.
This commit was SVN r12803.
2006-12-07 23:55:00 +00:00
Patrick Geoffray
6e09b0c23f lval is not defined when pval is assigned on 32 bit systems. this
usually is ok on little-endian systems, as the upper 32 bits will likely
be ignored, but on 32-bit big-endian systems, lval is complete junk.
Use ival if 32 bit mode, lval if 64.

Mixing of 32 and 64 bit architectures won't work without more changes.

This commit was SVN r12802.
2006-12-07 23:34:04 +00:00
Brian Barrett
98884e45e4 Clean up the way procs are added to the global process list after MPI_INIT:
* Do not add new procs to the global list during modex callback or
    when sharing orte names during accept/connect.  For modex, we
    cache the modex info for later, in case that proc ever does get
    added to the global proc list.  For accept/connect orte name
    exchange between the roots, we only need the orte name, so no
    need to add a proc structure anyway.  The procs will be added
    to the global process list during the proc exchange later in 
    the wireup process
  * Rename proc_get_namebuf and proc_get_proclist to proc_pack
    and proc_unpack and extend them to include all information
    needed to build that proc struct on a remote node (which
    includes ORTE name, architecture, and hostname).  Change
    unpack to call pml_add_procs for the entire list of new
    procs at once, rather than one at a time.
  * Remove ompi_proc_find_and_add from the public proc
    interface and make it a private function.  This function
    would add a half-created proc to the global proc list, so
    making it harder to call is a good thing.

This means that there's only two ways to add new procs into the global proc list at this time: During MPI_INIT via the call to ompi_proc_init, where my job is added to the list and via ompi_proc_unpack using a buffer from a packed proc list sent to us by someone else.  Currently, this is enough to implement MPI semantics.  We can extend the interface more if we like, but that may require HNP communication to get the remote proc information and I wanted to avoid that if at all possible.

Refs trac:564

This commit was SVN r12798.

The following Trac tickets were found above:
  Ticket 564 --> https://svn.open-mpi.org/trac/ompi/ticket/564
2006-12-07 19:56:54 +00:00
Brian Barrett
b07dfa7841 * remove unused variable in ompi_comm_get_rprocs
* don't load data into a buffer until we have the data, as
    the data contains some header information needed to
    properly load the data

This commit was SVN r12792.
2006-12-07 16:19:44 +00:00
Sven Stork
78173a697a Replace the test opertion "-e" with "-r" to improve the protability.
Refs: #392

This commit was SVN r12790.
2006-12-07 12:14:40 +00:00
Ralph Castain
62d7826e01 Helps if we total up the correct field to get the total number of slots in the universe
This commit was SVN r12789.
2006-12-07 03:17:12 +00:00
Ralph Castain
a1153fdc8f Eliminate virtually all of the attribute_predefined data from the STG1 message. We now compute the total number of slots allocated to us and save that in the registry - the attributed_predefined then retrieves it via the STG1 message. The app_num is passed via the process_info structure, which gets the value from the ODLS in the environment.
Obviously, people like bproc will have to get the app_num via another avenue...but that's a problem for another day. Several options are easily available.

This commit was SVN r12788.
2006-12-07 03:11:20 +00:00
Brian Barrett
41a70a8f01 indent, this time with the right coding standards...
This commit was SVN r12787.
2006-12-07 00:24:01 +00:00
Brian Barrett
f9ec8d6f2a reindent file to make it easier to deal with...
This commit was SVN r12786.
2006-12-07 00:21:25 +00:00
Brian Barrett
8f68764e5e A number of heterogeneous fixes for the dss with the new buffer options:
* When using the load/unload interface, stash away the current buffer
    type so that it can be properly unpacked on the receiving side if
    the buffer type is other than the receiver default
  * Include type information for unsized types (bool, int, size_t,
    pid_t) so that they can be properly unpacked by the receiver
    in the heterogeneous case.
  * Restore the NON_DESC type as the default for optimized builds,
    since it looks like this fixes the known issues with the
    non-described buffers

Refs trac:587

This commit was SVN r12784.

The following Trac tickets were found above:
  Ticket 587 --> https://svn.open-mpi.org/trac/ompi/ticket/587
2006-12-06 23:19:06 +00:00
Brian Barrett
cfeac5581a temporarily always use described buffers as the non-described causes all
kinds of problems for heterogeneous environments

This commit was SVN r12783.
2006-12-06 20:22:31 +00:00
Ralph Castain
d4bd60c9fe Restore the paffinity capability, along with all the required logic to ensure we "do the right thing" when the user gives us inaccurate information about the number of slots on a remote node.
This commit was SVN r12780.
2006-12-06 15:59:34 +00:00
Ralph Castain
b1e16fffac Add the C++ doo-hicky stuff around the odls framework definitions just in case somebody, somewhere, on some remote planet where only goats can feed needs it.
This commit was SVN r12777.
2006-12-06 13:58:04 +00:00
Ralph Castain
8ca415a0c5 Remove duplicate orte_odls declaration
This commit was SVN r12776.
2006-12-06 13:44:41 +00:00
Jeff Squyres
122b8553ef Sync against 1.1 tree
This commit was SVN r12767.
2006-12-05 19:15:56 +00:00
Edgar Gabriel
1359ba9b13 Rewriting much of the errorcode and errorclass code, since
- we have to be able to attach a string to an error class, not just to an
 error code
 - according to MPI-2 the attribute MPI_LASTUSEDCODE has to be updated
  everytime you add a new code or a new class. Thus, you have to have single
  list for both. 

Thus, we got rid of the error_class structure. In the error-code structure, we
can distinguish whether we are dealing with an error code or an error class by
looking at the err->code element of the structure. In case its value is
MPI_UNDEFINED, the according entry is a class, else it is an error code. All
predefined error codes have the code and the class field set to the same
value.

The test MPI_Add_error_class1 passes now.

Fixes trac:418

This commit was SVN r12764.

The following Trac tickets were found above:
  Ticket 418 --> https://svn.open-mpi.org/trac/ompi/ticket/418
2006-12-05 19:07:02 +00:00
Jeff Squyres
73696be5ac Add bullets about library name changes (because some users are
undoubtedly not using the wrapper compilers) and one-sided support
fixes.

This commit was SVN r12763.
2006-12-05 18:41:40 +00:00
Brian Barrett
6f8b366acb Rename liborte to libopen-rte and libopal to libopen-pal per telecon today
and bug #632.

Refs trac:632

This commit was SVN r12762.

The following Trac tickets were found above:
  Ticket 632 --> https://svn.open-mpi.org/trac/ompi/ticket/632
2006-12-05 18:27:24 +00:00
Dan Lacher
ba16157f7e Removed local zone only requirment from solaris packages
Submitted by: Dan Lacher
Reviewed by: Rolfv Vandavaart

This commit was SVN r12759.
2006-12-05 14:37:44 +00:00
Tim Prins
08d5ca821f Don't get the node architecture when useing the LoadLevleer RAS. It is slow (about a second for ~300 nodes) and we don't even use the value.
This commit was SVN r12758.
2006-12-05 13:47:53 +00:00
Ralph Castain
eb941d8ae2 Fix a bug that declared a node as "oversubscribed" a little early during the mapper procedure. This only affected the mapping procedure, and only if you had set the "--no-oversubscribe" flag.
Kudos to Tim Prins for finding it.

This commit was SVN r12757.
2006-12-05 13:04:27 +00:00