1
1
Граф коммитов

5193 Коммитов

Автор SHA1 Сообщение Дата
Shiqing Fan
90eeba252e Make openib compile again for Windows.
Update the CMake script for checking mca subdirs.
Add windows support for __attribute__ packed structures.
Define usleep and posix_memalign with equivalent windows functions.
And a few minor fixes, type casts.

This commit was SVN r24429.
2011-02-22 15:49:27 +00:00
Doron Shoham
e5eef80364 fix type warning in openib btl
This commit was SVN r24419.
2011-02-21 15:13:30 +00:00
Mike Dubman
89ba89e812 - added support for upcomming FCA v2.1 version
This commit was SVN r24418.
2011-02-21 14:08:24 +00:00
Donald Kerr
995d46344c simplify the way IBV_ACCESS_SO is discovered
This commit was SVN r24409.
2011-02-17 04:28:56 +00:00
Jeff Squyres
3f4d4886f2 Minor update for something that has been bugging me for quite a while:
OMPI supports multiple different repository systems (SVN, hg, git).
But the VERSION file has listed "want_svn" and "svn_r" as fields, even
though the actual repo system and version may not be SVN.

So search/replace those fields (and derrivative values that come from
those fields) with "want_repo_rev" and "repo_rev", respectively.

This commit was SVN r24405.
2011-02-16 22:53:23 +00:00
Donald Kerr
2b60b165aa on Solaris, when IBV_ACCESS_SO is available, use strong ordered memory region for eager rdma connection
This commit was SVN r24395.
2011-02-16 05:37:22 +00:00
Jeff Squyres
94356e98d4 Fix from Nikolay Piskun at Rogue Wave (TotalView) -- fixes the case
where MPI jobs are launched directly via srun (i.e., there's no HNP).

This commit was SVN r24376.
2011-02-14 19:03:53 +00:00
Mike Dubman
81222e1fe7 * fix PGI compiler support which does not have __BASE_FILE__ macro
This commit was SVN r24369.
2011-02-10 06:42:37 +00:00
Ethan Mallove
c6fd141923 missing include
This commit was SVN r24368.
2011-02-09 17:59:55 +00:00
Eugene Loh
cd5c2e794f Some minor changes to help the openib BTL build and run on Solaris:
- poll() can return POLLRDNORM even if not requested (Solaris bug)
- MIN macro not defined in btl_openib.c
  and while we're at it, we clean up the MIN definition in ad_bgl_pset.h
- btl_openib_connect_rdmacm.c was calling rdma_destroy_id() twice
  leading to undefined behavior (a hang on Solaris)

This commit was SVN r24356.
2011-02-03 23:53:21 +00:00
Jeff Squyres
5bc2ad2b44 Fix some deprecated notices to refer to the correct new function names
This commit was SVN r24313.
2011-01-27 19:55:42 +00:00
Jeff Squyres
30e164e246 Fix all the problems with "make distcheck" caused by the new ROMIO import so that we can finally get a trunk nightly tarball\!
This commit was SVN r24292.
2011-01-24 21:10:14 +00:00
Jeff Squyres
700d601dfc Also need to check the "flag" value, because if flag!=true, then the
value of "local_spawn" (and "non_mpi") is not set by ompi_info_get*().

This commit was SVN r24286.
2011-01-22 16:27:58 +00:00
Jeff Squyres
89fb26eb1c Add missing line continuation character to prevent a Makefile syntax
error

This commit was SVN r24285.
2011-01-22 11:13:28 +00:00
Rolf vandeVaart
8171370287 Fix typo which broke builds when configured with hetero and
debug.

This commit was SVN r24283.
2011-01-21 17:10:09 +00:00
Abhishek Kulkarni
a1090575c2 Nitpick: Get rid of a redundant OPAL_SOS_GET_ERROR_CODE.
This commit was SVN r24282.
2011-01-20 23:48:11 +00:00
Abhishek Kulkarni
3243b16bb3 Decode SOS error code before checking it with the native error code.
This commit was SVN r24281.
2011-01-20 23:21:38 +00:00
Rolf vandeVaart
6a5ad29c36 Update configure command since it changed.
This commit was SVN r24275.
2011-01-20 14:42:12 +00:00
Sylvain Jeaugey
46b711e164 Fixes trac:1888 introduced by r24264 : make Romio autogen.sh executable.
This commit was SVN r24272.

The following SVN revision numbers were found above:
  r24264 --> open-mpi/ompi@0e921bba7f

The following Trac tickets were found above:
  Ticket 1888 --> https://svn.open-mpi.org/trac/ompi/ticket/1888
2011-01-20 09:20:34 +00:00
Rolf vandeVaart
acd38ff746 Final changes from jsquyres review. Moved configure
code from upper level into btl configure.m4.  Changed
prefix from "OMPI" to "BTL" in preprocessor macro.  Add
an mca param that shows it has been configured in.

This commit was SVN r24270.
2011-01-19 20:58:22 +00:00
Brian Barrett
4859bb82e2 * Update to support direct call
* Add missing cancel (not that it does anything useful)
* Fix bug in opal_output call

This commit was SVN r24269.
2011-01-19 20:49:28 +00:00
Brian Barrett
8f6a19b0fc export component/module interface so that direct call works again
This commit was SVN r24268.
2011-01-19 20:47:17 +00:00
Rolf vandeVaart
f22f76a6ff Add byte swapping macro for failover control message per jsquyres review.
This commit was SVN r24266.
2011-01-19 19:58:35 +00:00
Rolf vandeVaart
e75b86d3ab Fix some issues from jsquyres review.
1. Use asprintf instead of snprintf
2. Return remote_proc where possible.
3. Remove dead code.
4. Fix two comment typos.

This commit was SVN r24265.
2011-01-19 16:09:17 +00:00
Sylvain Jeaugey
0e921bba7f Romio Refresh from mpich2-1.3.1. Work by Pascal Deveze, tested through bitbucket by Jeff Squyres (https://bitbucket.org/devezep/new-romio-for-openmpi).
This commit was SVN r24264.
2011-01-19 15:55:10 +00:00
George Bosilca
5390fd6f33 Reshape the datatype engine. The basic types are built down in OPAL. MPI types are
either direct link to these basic predefined types, or a combination of them.
Anyway, the first items in the datatype list belong to OPAL, the second round
are MPI datatypes created by composing basic OPAL datatypes, and the last
batch are mapped datatype (direct correspondance between an OMPI datatype and
an OPAL one such as int -> int32_t).

Modify the op to fit this new scheme.

This commit was SVN r24247.
2011-01-13 06:08:54 +00:00
Jeff Squyres
f08433c1e1 Fixes trac:2669.
Apparently, gcc 4.4.x and 4.5.x complain about the ''possibility'' of
us calling free() on a non-heap variable.  We know that this case can
never happen because the refcount will absolutely not go to zero
here.  We think it may be gcc being a bit too aggressive on the
warnings.

However, since this happens with gcc 4.4.x and 4.5.x, and since gcc
4.5.x ship in RHEL6 and Fedora 14 (and others), someone '''will'''
complain about this in the future, so we might as well code around it
so that we don't have to keep explaining "despite the warning, it's
really ok."

The workaround is pretty simple: just OBJ_RELEASE the values from
ompi_mpi_comm_parent before it is re-assigned to the new
intercommunicator.  Then the compiler's static code analysis can't
possibly tell that it's not a heap variable, and we're ok.

So yes, we are still calling OBJ_RELEASE on a non-heap variable.  But
free() '''will never be called''' on it because of the refcount.

This commit was SVN r24214.

The following Trac tickets were found above:
  Ticket 2669 --> https://svn.open-mpi.org/trac/ompi/ticket/2669
2011-01-10 21:12:27 +00:00
Ethan Mallove
82054cb02c Include <stdlib.h> instead of <malloc.h>. This avoids a compiler error
on some systems caused by the definition of malloc in
opal_config_bottom.h getting expanded in the system malloc.h when
OPAL_ENABLE_MEM_DEBUG is set to 1.

This commit was SVN r24210.
2011-01-06 18:16:36 +00:00
Jeff Squyres
58445f3775 After being hit by "why is openib not working?" ''again'', add a
verbose statement that shows up when you --mca btl_base_verbose 100.
It clearly states that the openib BTL disqualifies itself when
MPI_THREAD_MULTIPLE is used.

This commit was SVN r24209.
2011-01-05 22:01:15 +00:00
Eugene Loh
9bbcd51c5a Properly initialize ep->btl_max_send_size, ep->btl_pipeline_send_length, and
ep->btl_send_limit in mca_bml_r2_del_proc_btl() so that the loops will correctly
compute new endpoint max/min after the BTL has been removed.  See
http://www.open-mpi.org/community/lists/devel/2011/01/8829.php

This commit was SVN r24202.
2011-01-04 20:35:33 +00:00
Jeff Squyres
7648c36023 Correct an error in MPI_FILE_SET_VIEW man page: the fh does ''not''
have to have an identical ''value'' (it must be the same file handle,
but that means nothing about its actual value).  But the datarep must
have an identical value.  Additionaly, the etype does not need to be
identical, but the extent of all the etypes supplied must be
identical.

See MPI-2.2 p401-402 for further details.

This commit was SVN r24201.
2011-01-04 18:10:52 +00:00
Josh Hursey
2bdff63e6f move the INIT to after the error handler, so it matches MPI_INIT. Thanks to Jeff for catching this
This commit was SVN r24200.
2011-01-03 18:16:53 +00:00
Mike Dubman
4a2e29eb32 updated Makefile with a new file
This commit was SVN r24199.
2011-01-01 14:11:49 +00:00
Mike Dubman
c56e3141cb fca: fix segmentation fault when no underlying collective implementation is found
This commit was SVN r24198.
2010-12-31 12:03:49 +00:00
Josh Hursey
bbfdf04a81 Fix a couple of 'unused variable' warnings, and one return value warning.
{{{
base/paffinity_base_service.c: In function ‘opal_paffinity_base_cset2mapstr’:
base/paffinity_base_service.c:623: warning: unused variable ‘range_last’
base/paffinity_base_service.c:623: warning: unused variable ‘range_first’
base/paffinity_base_service.c:622: warning: unused variable ‘count’
base/paffinity_base_service.c:622: warning: unused variable ‘m’
}}}

{{{
connect/btl_openib_connect_oob.c: In function ‘init_ud_qp’:
connect/btl_openib_connect_oob.c:1111: warning: control reaches end of non-void function
connect/btl_openib_connect_oob.c: In function ‘init_device’:
connect/btl_openib_connect_oob.c:1235: warning: unused variable ‘i’
connect/btl_openib_connect_oob.c: In function ‘get_pathrecord_sl’:
connect/btl_openib_connect_oob.c:1323: warning: unused variable ‘i’
}}}

This commit was SVN r24196.
2010-12-30 15:37:50 +00:00
Doron Shoham
834625cc51 Currently the service lever is passed as static parameter (ib_service_level), but the service level is possibly dynamic and if so the only way to get a proper value is to ask the SA.
New mca parameter is added (ib_path_rec_service_level) - positive value means that we should get the SL from the SA.

This is usable for torus topologies where different SL value is used for different endpoints.

A cache is kept of ib queue pairs used to communicate with the SA for a particular device and port and path record SL values retrieved from that SA.

The interaction with the cache assumes that there are no recursive calls to these routines. This must be solved either by code flow, by using higher level locks, or by adding a locking mechanism to these routines along with some method for avoiding deadlock.

This code use a UD queue pair to talk to the SA, and not need to chmod /dev/infiniband/umad* for use by normal users.  

The request to the SA is a SubnAdmGet(), not a SubnAdmGetTable().
In the future we might add a support of a SubnAdmGetTable(), but it will require implementing RMPP (Reliable Multi-Packet Transaction Protocol) and I'm not sure we want to do that.

This patched is based on the work of David McMillen <davem@systemfabricworks.com>.

This commit was SVN r24195.
2010-12-30 08:20:24 +00:00
Josh Hursey
0b514e234b MPI_Init_thread is used in place of MPI_Init, so for the checkpoint/restart functionality it must correctly init the C/R functionality instead of simply making a critical section. This allows the C/R thread to be started properly.
Thanks to Takayuki Seki for finding this bug.

This commit was SVN r24194.
2010-12-29 15:37:30 +00:00
Mike Dubman
3d517c0285 ABI cleanups
This commit was SVN r24193.
2010-12-28 07:11:46 +00:00
Mike Dubman
b339a7a07b Add FCA 1.2/2.0 backward compatibility, depending on OMPI_FCA_VERSION_xx macro definition.
This commit was SVN r24192.
2010-12-27 21:32:34 +00:00
Doron Shoham
bfe611d3bd This patch fixes bugs #2627 (1.5.2) and #2623 (1.4.2) - Sending large messages over RDMA fails.
The patch includes the following:
 *  Add new mca parameter - btl_openib_max_hw_msg_size - Maximum size (in bytes) of a single fragment of a long message when using the RDMA protocols (must be > 0 and <= hw capabilities).
 *  If btl_openib_max_hw_msg_size is larger than the maximum hw limitation print error message.
 *  Change the default openib flags to include only PUT and not GET.
 *  Print error message if user choose manually GET flag in openib btl.
 *  In prepare_dst: limit the message size to be the minimum of both endpoint's hw_limitation and the user limitation (if requested).

This commit was SVN r24191.
2010-12-23 11:48:43 +00:00
Brian Barrett
621344cce4 Remove duplicate DT tests
This commit was SVN r24189.
2010-12-20 23:38:36 +00:00
Brian Barrett
9876e65137 Fix race condition in unlock code, as well as a small memory leak.
Somehow they got fixed in the pt2pt implementation, but not the RDMA
implementation.  Thanks to Guillaume Thouvenin for finding this issue.

This commit was SVN r24188.
2010-12-20 22:15:29 +00:00
Rolf vandeVaart
a57e5587f6 Brief description of bfo functionality.
This commit was SVN r24186.
2010-12-17 19:12:00 +00:00
Terry Dontje
6da16ab0d7 add format parameter and layout format to OMPI_Affinity_str
This commit was SVN r24182.
2010-12-16 15:11:17 +00:00
Jeff Squyres
b113b1a382 Add the btl_tcp_if_seq MCA parameter. From the help string:
If specified, a comma-delimited list of TCP interfaces.  Interfaces
  will be assigned, one to each MPI process, in a round-robin fashion
  on each server.  For example, if the list is "eth0,eth1" and four
  MPI processes are run on a single server, then local ranks 0 and 2
  will use eth0 and local ranks 1 and 3 will use eth1.

This feature is only useful for environments with virtual ethernet
interfaces on the same network.  For example, if eth0 and eth1 are
virtual interfaces to the same NIC on the same subnet, and if the NIC
provides different hardware resources to eth0 and eth1 (not just
different kernel resources), some HOL blocking and congestion issues
can be eased in a modest fashion.

This commit was SVN r24181.
2010-12-16 00:54:32 +00:00
Jeff Squyres
741ba6518b Install ompi_config.h (and friends) into /openmpi, not /openmpi/ompi/include. This is consistent with prior releases and the OPAL and ORTE <foo>_config.h file installation locations.
This commit was SVN r24180.
2010-12-15 20:54:22 +00:00
Jeff Squyres
298a50ff3b Clarification that calling MPI_INIT_THREAD with MPI_THREAD_SINGLE is
the same as calling MPI_INIT.

This commit was SVN r24173.
2010-12-14 22:49:36 +00:00
Samuel Gutierrez
7d8d7769ee remove unneeded include in sm btl.
This commit was SVN r24165.
2010-12-13 17:30:47 +00:00
Brian Barrett
6cf74eeb03 Fix bug in looking at convertor_unpack return code. Always print debug
on error message for now.

This commit was SVN r24163.
2010-12-10 22:36:47 +00:00
Brian Barrett
a26fadb26e Bring Portals4 updates back into the trunk
This commit was SVN r24154.
2010-12-07 20:11:25 +00:00