Had group discussion about this on the weekly call. The decision was
that we should pass the real error code to the back-end exception
handler because it's pretty useless to pass MPI_ERR_IN_STATUS to the
back-end exception handler (because exception handlers don't have
access to the request or the status - this has potential issues for
fault tolerance kinds of scenarios). So in TESTALL, TESTSOME,
WAITALL, and WAITSOME, we examine the error code and if it's not
MPI_SUCCESS, return MPI_ERR_IN_STATUS.
This commit was SVN r12389.
The following Trac tickets were found above:
Ticket 549 --> https://svn.open-mpi.org/trac/ompi/ticket/549
mca_btl_openib_endpoint_connect_eager_rdma() is called recursively. He also
noticed that orte_pointer_array_add() can't fail because we allocate max number
of elements at init time. So just remove error handling and locking. No locking
- no deadlocks.
This commit was SVN r12388.
something is going wrong down in the code it is removed from the array. So add
mutex to prevent concurrent access to the array from different threads.
This commit was SVN r12385.
Setup subscriptions to correctly return the MPI_APPNUM attribute.
Fix an unreported bug that was found. The universe size was incorrectly defined in the attributes code. As coded, it looked for size_t values and based its size computation on those numbers. Unfortunately, the node_slots value had been changed to an orte_std_cntr_t awhile back! So the universe size was never updated.
Update the hello_nodename test to check for MPI_APPNUM.
Add a definition to ns_types for ORTE_PROC_MY_NAME - just a shortcut for orte_process_info.my_name. Brought over from ORTE 2.0 as it will be used extensively there.
This commit was SVN r12377.
What's happening is that we're holding openib_btl->eager_rdma_lock when
we call mca_btl_openib_endpoint_send_eager_rdma() on
btl_openib_endpoint.c:1227. This in turn calls
mca_btl_openib_endpoint_send() on line 1179. Then, if the endpoint
state isn't MCA_BTL_IB_CONNECTED or MCA_BTL_IB_FAILED, we call
opal_progress(), where we eventually try to lock
openib_btl->eager_rdma_lock at btl_openib_component.c:997.
The fix removes this lock altogether. Instead we atomically set local RDMA
pointer to prevent other threads to create rdma buffer for the same endpoint.
And we increment eager_rdma_buffers_count atomically thus polling thread doesn't
need lock around it.
This commit was SVN r12369.
Have no idea why this function always returns a failure. It should
always return SUCCESS (provided the status is value).
This commit was SVN r12364.
The following Trac tickets were found above:
Ticket 496 --> https://svn.open-mpi.org/trac/ompi/ticket/496
* For MPI_TEST, MPI_TESTANY, MPI_WAIT, and MPI_WAITANY (i.e., the
TEST/WAIT functions that return up to exactly one completed
request), return the actual error code.
* For MPI_TESTALL, MPI_TESTSOME, MPI_WAITALL, MPI_WAITSOME, (i.e.,
the TEST/WAIT functions that can return more than one completed
request), return MPI_ERR_IN_STATUS.
This commit was SVN r12355.
The following Trac tickets were found above:
Ticket 549 --> https://svn.open-mpi.org/trac/ompi/ticket/549
This commit essentially caches the invoking comm/win/file on the
ompi_request_t. This, paired with the req_type field, allows us to
retrieve the invoking MPI object and invoke the proper errhandler.
The patch is missing most updates for the MPI-2 one-sided stuff (i.e.,
the patch mainly fixes comms and files); I didn't really understand
that code and didn't want to hazard trying to figure it out when Brian
can probably do it much more quickly.
So #250 will still stay open, pending MPI-2 one-sided updates for this
stuff.
This commit was SVN r12339.
The following Trac tickets were found above:
Ticket 250 --> https://svn.open-mpi.org/trac/ompi/ticket/250
* Create a new request type: NOOP (described below)
* For all MPI_*_INIT functions, OBJ_NEW an ompi_request_t and set its
type to NOOP
* Ensure that the NOOP requests are OBJ_RELEASE'd when they are done
* MPI_START looks at the request type; if NOOP, just return success. If
not, call the PML start() function
* MPI_STARTALL always pass the entire array of requests back to the PML
(see next point)
* Make the PMLs only process PML requests (i.e., ignore/skip anything
that isn't of type PML -- such as the NOOP requests)
* Add a little more param error checking in STARTALL
This commit was SVN r12338.
The following Trac tickets were found above:
Ticket 529 --> https://svn.open-mpi.org/trac/ompi/ticket/529
* Remove an extra OMPI_REQUEST_INIT() from the grequest constructor
(it was already invoked by the parent MPI_Request constructor)
* Set the state of the generalized request to ACTIVE (because this is
invoked from MPI_GREQUEST_START -- analogous to MPI_START)
* Before invoking the query function in MPI_REQUEST_COMPLETE, set the
status on the base request to ompi_status_empty. This gives a set
of default values for the request, including one for
status.MPI_ERROR = MPI_SUCCESS (because we check the value of
MPI_ERROR in MPI_TEST* and MPI_WAIT* processing, and use it to
determine whether the upper-level API call should raise an MPI
exception or not).
This commit was SVN r12337.
The following Trac tickets were found above:
Ticket 532 --> https://svn.open-mpi.org/trac/ompi/ticket/532
allocation logic is completely done outside the data-type engine (in the PML) there is
no need for any special case inside the data-type engine. There is less arguments for
the ompi_convertor_pack and ompi_convertor_unpack as well (the last field free_after is
not required anymore as there is no memory allocated in the engine itself). This change
affect all components using datatypes. I test most of them, but it might happens that I
miss some ... If it's the case please let me know (don't shoot the pianist!!).
This commit was SVN r12331.
the default decision functions (for broadcast, reduce and barrier) are based on a
high performance network (not TCP). It should give good performance (really good) for
any network having the following caracteristics: small latency (5 microseconds) and good
bandwidth (more than 1Gb/s).
+ Cleanup of the reduce algorithms, plus 2 new algorithms (binary and binomial). Now most
of the reduce algorithms use a generic tree based function for completing the reduce.
+ Added macros for computing the trees (they are used for bcast and reduce right now).
+ Allow the usage of all 5 topologies.
+ Jelena's implementation of a binary tree that can be used for non commutative operations.
Right now only the tree building function is there, it will get activated soon.
+ Some others minor cleanups.
This commit was SVN r12326.
A segfault would occur in mca_pml_ob1_recv_request_progress() when trying to prepare the convertor for unpacking, because the request's req_proc field was NULL.
Turns out that we weren't setting the req_proc field in the MCA_PML_OB1_CHECK_SPECIFIC_AND_WILD_RECEIVES_FOR_MATCH macro. Instead of just setting it there I removed the other place req_proc was being set correctly, and instead took care of all the cases at once in mca_pml_ob1_recv_frag_match().
This commit was SVN r12323.
parameter. For optimisation purpose only this BTL is used to send packet
through instead of trying to send packets through all BTLs. But actually the
code was wrong. It simply used provided bml_btl and it may represent different
endpoint from packet's destination. The fixed code checks if packet's
destination is reachable through the BTL, finds appropriate bml_btl and only
then tries to send it through correct bml_btl.
This commit was SVN r12319.
is done to assure alignment so strictly aligned CPUs (like SPARC) do not
sigbus. This also may benefit other platforms too.
This commit fixes trac:494.
This commit was SVN r12312.
The following Trac tickets were found above:
Ticket 494 --> https://svn.open-mpi.org/trac/ompi/ticket/494
If you want to look at our launch and MPI process startup times, you can do so with two MCA params:
OMPI_MCA_orte_timing: set it to anything non-zero and you will get the launch time for different steps in the job launch procedure. The degree of detail depends on the launch environment. rsh will provide you with the average, min, and max launch time for the daemons. SLURM block launches the daemon, so you only get the time to launch the daemons and the total time to launch the job. Ditto for bproc. TM looks more like rsh. Only those four environments are currently supported - anyone interested in extending this capability to other environs is welcome to do so. In all cases, you also get the time to setup the job for launch.
OMPI_MCA_ompi_timing: set it to anything non-zero and you will get the time for mpi_init to reach the compound registry command, the time to execute that command, the time to go from our stage1 barrier to the stage2 barrier, and the time to go from the stage2 barrier to the end of mpi_init. This will be output for each process, so you'll have to compile any statistics on your own. Note: if someone develops a nice parser to do so, it would be really appreciated if you could/would share!
This commit was SVN r12302.
"this is bogus" kind of answer. Passing in bad error codes should
only happen in erroneous sections of the OMPI code base, but still,
it's far more social to print a message saying, "hey, you messed up!"
rather than seg faulting.
Reviewed by Edgar.
This commit was SVN r12295.
mentioned in the comment the completion/callback of the triggered
send operation can happen before the call returns. If this happens and
if the pipeline depth is 0 before we triggered the send operation and
this is the last send operation of the request then the completion detection
code will decrement the pipeline depth and check it for equality to 0.
Because (0-1) != 0 the pml completion function for this request will
*not* be called.
This part 2 of the fix for ticket #246.
This commit was SVN r12292.
