The project includes following components and frameworks:
- ML Collective component
- NETPATTERNS and COMMPATTERNS common components
- BCOL framework
- SBGP framework
Note: By default the ML collective component is disabled. In order to enable
new collectives user should bump up the priority of ml component (coll_ml_priority)
=============================================
Primary Contributors (in alphabetical order):
Ishai Rabinovich (Mellanox)
Joshua S. Ladd (ORNL / Mellanox)
Manjunath Gorentla Venkata (ORNL)
Mike Dubman (Mellanox)
Noam Bloch (Mellanox)
Pavel (Pasha) Shamis (ORNL / Mellanox)
Richard Graham (ORNL / Mellanox)
Vasily Filipov (Mellanox)
This commit was SVN r27078.
technically this is a necessary thing to do, it wasn't a tragedy that
we didn't have it because err was initialize to 0 in the beginning of
the functions where this problem occurred. Also, OMPI will likely
abort if one of the MCA_PML_CALLs actually incurs an error (or, even
if it doesn't, MPI doesn't define the behavior anyway ;-) ).
But looking forward to an FT-aware world, fixing this issue is a Good
Thing. Many thanks to Hristo Iliev for pointing out the issue.
This commit was SVN r27070.
- OMPI_SUCCESS
- OMPI_ERROR
- OMPI_ERR_RESOURCE_BUSY
If an "OMPI_ERR_OUT_OF_RESOURCE" occurs, the request is added to the pending list, and will be handled later. An error message
should not be printed to the user in this case. This is not an error, but rather a notification of a possible valid condition.
Only in the case of "OMPI_ERROR" should it be printed to the user.
This commit was SVN r27065.
btl_openib_connect_udcm when notifying not to listen to an fd to ensure
that the main thread does not continue until the service thread has
processed the message
Adds ability to send message to openib async thread to tell it to
ignore the ERR state on a specific QP. Adds this call to udcm_module_finalize
so when we set the error state on the QP it doesn't cause the
openib async thread to abort the mpi program prematurely
Fixes trac:3161
This commit was SVN r27064.
The following Trac tickets were found above:
Ticket 3161 --> https://svn.open-mpi.org/trac/ompi/ticket/3161
receive queues value) so that we don't break the use of RDMA CM, and
therefore break RoCE.
This commit was SVN r27017.
The following SVN revision numbers were found above:
r26869 --> open-mpi/ompi@fe0e7f81df
aren't separated out into individual commits; they represent a few
months of work in the Mercurial branch, and it seemed error-prone to
try to break them up into multiple SVN commits.
* Remove 2nd overloaded interfaces for MPI_TESTALL, MPI_TESTSOME,
MPI_WAITALL, and MPI_WAITSOME in the "mpi" module implementations
(because we're not allowed to have them, anyway -- it causes
complications in the profiling interface). This forced an MPI-2.2
errata in the MPI Forum; we applied the errata here (the array of
statuses parameter could not have a specific dimension specified in
the dummy argument). Fixes trac:3166.
* Similarly, fix type for MPI_ARGVS_NULL in Fortran
* Add MPI_3.0 function MPI_F_SYNC_REG (Fortran interfaces only).
* Add MPI-3.0 MPI_MESSAGE_NO_PROC in the mpi_f08 module.
* Added mpi_f08 handle comparison operators, per MPI-3.0 addendum to
the F08 proposal at the last Forum meeting.
* Added missing type(MPI_File) and type(Message) in mpi_f08 module.
* Fix --disable-mpi-io configure switch with all Fortran interfaces
* Re-factor the Fortran header files to be fundamentally simpler and
easier to maintain. Fortran constant values in the header files
are now generated by a script named mpif-values.pl during
autogen.pl (they were previously generated by mpif-common.pl, but
it was quite a bit more subtle/complex). A second commit will
follow this one to update svn:ignore values (just to ensure we
don't muck up the first commit with the SVN client getting confused
by the changed ignore values and new/changed files).
* Fix some dependencies for compile ordering in
ompi/mpi/fortran/use-mpi-ignore-tkr/Makefile.am.
* Fix bad wording in several places (.m4 file name, ompi_info output,
etc.): we previoulsy said "F08 assumed shape" when we really meant
"F08 assumed rank" (for Fortran gurus, those are very different
things).
* Removed the GREEK/SVN version string from mpif.h. It really had no
purpose being there.
Still to be done:
* Handling of 2D array of strings in MPI_COMM_SPAWN_MULTIPLE still
isn't right yet. Not sure how many people really care about this
:-), but it is still broken.
This commit was SVN r26997.
The following Trac tickets were found above:
Ticket 3166 --> https://svn.open-mpi.org/trac/ompi/ticket/3166
- return MPI_ERR_OTHER instead of MPI_SUCCESS for the functions that are not
yet implemented
- add another field to the mca_io_ompio_file_t structure to point back to the
ompi_file_t structure.
This commit was SVN r26908.
- For now we'll use 8192 as a base value
- We leave the adjust_cq() as is
- For the long term we can work on an appropriate setting to expose through the INI file.
8K CQEs are 512K per process, which is 8MB for ppn=16
This commit was SVN r26877.
ibv_get_device_list_compat() and not finding it, I finally realized
that it was a function in OMPI. So let's name it with a proper ompi_
prefix, not an ibv_ prefix.
This commit was SVN r26867.
it to a negative number). Get rid of the multiplication in the critical
path, and keep the functions as simple as possible.
This commit was SVN r26864.
move). Extended common sm API with: mca_common_sm_module_create and
mca_common_sm_module_attach. Please note that the new routines aren't currently
used -- but will be...
This commit was SVN r26845.
alignment, which typically causes problems on SPARC. Further, the pointer
manipulation to access elements in a round schedule was clumsy. This change
introduces macros to facilitate addressing and make it more portable.
This commit was SVN r26802.
1000. Refs trac:3154.
IB/iWarp vendors need to get together to figure out a real fix.
This commit was SVN r26777.
The following SVN revision numbers were found above:
r26730 --> open-mpi/ompi@5315c91baf
The following Trac tickets were found above:
Ticket 3154 --> https://svn.open-mpi.org/trac/ompi/ticket/3154
Among other things, this patch deals with the following issues:
* fix ompi-checkpoint argument parsing
* ompi-restart -showme prints an extraneous "Restarted child with PID"
message. Move around the debug statement to avoid this.
* fixes for the state machine changes
This commit was SVN r26770.
- change the location where we mark the file view as contiguous and the
condition on how it is determined to be contiguous
- remove the unnecessary include statements
This commit was SVN r26763.
* If the MCA param btl_openib_cq_size is set to 0 (which is the
default), use the device CQ max size. Otherwise, use the MCA param
value (and never adjust it again).
* Remove the CQ size adjustment code. Since we default to max CQ
size, there really isn't much point in having it any more. I think
people setting an absolute CQ size is going to be rare, so let's
not do anything fancy with it.
* If the MCA param value is larger than what the device supports,
print a warning (only once per process) and default to using the
device max
* Add a BTL_VERBOSE displaying which CQ size we used
This commit was SVN r26730.
The following Trac tickets were found above:
Ticket 3152 --> https://svn.open-mpi.org/trac/ompi/ticket/3152
where the MX BTL and MTL are stepping on each other regarding the
mpool. Thanks to Yong Qin for assistance in tracking this down.
This commit was SVN r26698.
The following Trac tickets were found above:
Ticket 3130 --> https://svn.open-mpi.org/trac/ompi/ticket/3130
hook early in the setup, but ''not'' during the component register
function. And then properly unset it if was set.
This commit was SVN r26697.
The following Trac tickets were found above:
Ticket 3130 --> https://svn.open-mpi.org/trac/ompi/ticket/3130
* Use proper tag space (the negatives below the blocking communicators)
instead of the point-to-point space
* Use the PML interface instead of the MPI interface, since the MPI
interface 1) shouldn't be used by components and 2) doesn't like
negative tags
This commit was SVN r26693.
Update all the orte ess components to remove their associated APIs for retrieving proc data. Update the grpcomm API to reflect transfer of set/get modex info to the db framework.
Note that this doesn't recreate the old GPR. This is strictly a local db storage that may (at some point) obtain any missing data from the local daemon as part of an async methodology. The framework allows us to experiment with such methods without perturbing the default one.
This commit was SVN r26678.
* Properly set message to MPI_MESSAGE_NULL in the right places
* Fix double free of buffer for non-contiguous blocking sends
* Remove useless debugging output
This commit was SVN r26604.
2. Checking for 0 bytes datatypes and sending only when data available to avoid 0 byte messages being sent and received.
3. Changing timing extraction to support calculating, min, max and avg communication costs + min and avg write costs
This commit was SVN r26450.
* Need to look at slot count before flowcontrol for sending to prevent
race in restart
* Need to free pending request fragments when done with the request
* A number of branch prediction optimizations for error conditions
This commit was SVN r26430.
helper file, even if we find that the system has <infiniband/verbs.h>.
The reason is because there are some inline functions in that verbs
helper file that invoke ibv_* functions. Some linkers (e.g., Solaris
Studio Compilers) will instantiate those static inline functions --
even if we don't use them -- and therefore we need to be able to
resolve the ibv_* symbols at link time.
But since -libverbs is only specified in places where we use other
ibv_* functions (e.g., the OpenFabrics-based BTLs), that means that
linking random executables can/will fail (e.g., orterun).
So instead, introduce a new #define: OPAL_HWLOC_WANT_VERBS_HELPER. If
this macro is set to 1 before including opal/mca/hwloc/hwloc.h, then
you'll also get the hwloc OpenFabrics verbs helper header file (*if*
hwloc found <infiniband/verbs.h> -- otherwise, it'll #error).
This commit was SVN r26417.
* Remove paffinity, maffinity, and carto frameworks -- they've been
wholly replaced by hwloc.
* Move ompi_mpi_init() affinity-setting/checking code down to ORTE.
* Update sm, smcuda, wv, and openib components to no longer use carto.
Instead, use hwloc data. There are still optimizations possible in
the sm/smcuda BTLs (i.e., making multiple mpools). Also, the old
carto-based code found out how many NUMA nodes were ''available''
-- not how many were used ''in this job''. The new hwloc-using
code computes the same value -- it was not updated to calculate how
many NUMA nodes are used ''by this job.''
* Note that I cannot compile the smcuda and wv BTLs -- I ''think''
they're right, but they need to be verified by their owners.
* The openib component now does a bunch of stuff to figure out where
"near" OpenFabrics devices are. '''THIS IS A CHANGE IN DEFAULT
BEHAVIOR!!''' and still needs to be verified by OpenFabrics vendors
(I do not have a NUMA machine with an OpenFabrics device that is a
non-uniform distance from multiple different NUMA nodes).
* Completely rewrite the OMPI_Affinity_str() routine from the
"affinity" mpiext extension. This extension now understands
hyperthreads; the output format of it has changed a bit to reflect
this new information.
* Bunches of minor changes around the code base to update names/types
from maffinity/paffinity-based names to hwloc-based names.
* Add some helper functions into the hwloc base, mainly having to do
with the fact that we have the hwloc data reporting ''all''
topology information, but sometimes you really only want the
(online | available) data.
This commit was SVN r26391.
