* add MCA parameter (OMPI_MCA_mpi_yield_when_idle) to cause sched_yield()
to be called when the progress engine is called and nothing happens.
Default is to call sched_yield().
* add MCA parameter (OMPI_MCA_mpi_event_tick_rate) to adjust the rate
at which the event library is called from ompi_progress. When set
to 0, the event library will never be ticked. When set to 1, the
event library will be progressed every time. 2 every other, etc.
The MCA parameters are only in effect from end of MPI_Init to start of
MPI_Finalize.
This commit was SVN r5099.
day (to convert "void*" -> real function pointer types, and "char*
extra_state" to "MPI_Fint* extra_state").
George still gets to cleanup/finish MPI_REGISTER_DATAREP(). :-)
This commit was SVN r5097.
- Added some protection to portions that should only be used when
we're building OMPI (not, for example, when mpicc is being used to
compile a user's MPI application)
This commit was SVN r5082.
Update the unit-test-status matrix to include priority.
Add several new registry diagnostics that helped track down the above bug.
M test/mca/gpr/gpr_triggers.c
M test/Unit-Test-Status.xls
M test/Unit-Test-Status.pdf
M src/mpi/runtime/ompi_mpi_init.c
M src/mca/oob/base/oob_base_xcast.c
M src/mca/ns/base/ns_base_nds_env.c
M src/mca/gpr/replica/api_layer/gpr_replica_dump_api.c
M src/mca/gpr/replica/api_layer/gpr_replica_api.h
M src/mca/gpr/replica/communications/gpr_replica_comm.h
M src/mca/gpr/replica/communications/gpr_replica_remote_msg.c
M src/mca/gpr/replica/communications/gpr_replica_cmd_processor.c
M src/mca/gpr/replica/communications/gpr_replica_dump_cm.c
M src/mca/gpr/replica/gpr_replica_component.c
M src/mca/gpr/replica/gpr_replica.h
M src/mca/gpr/replica/functional_layer/gpr_replica_dump_fn.c
M src/mca/gpr/replica/functional_layer/gpr_replica_fn.h
M src/mca/gpr/replica/functional_layer/gpr_replica_trig_ops_fn.c
M src/mca/gpr/replica/functional_layer/gpr_replica_messaging_fn.c
M src/mca/gpr/replica/functional_layer/gpr_replica_segment_fn.c
M src/mca/gpr/proxy/gpr_proxy_dump.c
M src/mca/gpr/proxy/gpr_proxy.h
M src/mca/gpr/proxy/gpr_proxy_component.c
M src/mca/gpr/gpr_types.h
M src/mca/gpr/base/base.h
M src/mca/gpr/base/unpack_api_response/gpr_base_dump_notify.c
M src/mca/gpr/base/pack_api_cmd/gpr_base_pack_dump.c
M src/mca/gpr/gpr.h
This commit was SVN r5080.
it looks because I re-word wrapped a couple of long comments):
- remove the polling of the event library in all cases where the
condition variables already do so. The condition variables were
updated and I didn't update this code to match. This was only
causing problems because there were some cases where it was
causing deadlock-like things with the orte_wait mutex.
- unlock the orte_wait mutex once we have the status info from
the pid we were waiting on, rather than holding it until the
condition variable can be destroyed. This allows us to poll
a bit more blindly while waiting for the other thread to finish
with the condition signal
- do the condition variable-like unlock / poll / unlock cycle
when progressing the event library when the condition variable
doesn't do it for us.
This commit was SVN r5064.
ompi_mpi_init(): we no longer *query* the components to see what they
can support -- instead, we *tell* the components what they need to
support (if they can't support it, they should disqualify themselves
from selection).
This involved a bunch of things:
- pass the configure-time-decided OMPI_ENABLE_PROGRESS_THREADS and
OMPI_ENABLE_MPI_THREADS constants to each framework selection
function
- eliminate mca_base_init_select_components()
- eliminate all "allow_multi_user_threads" and "have_hidden_threads"
kinds of arguments and aggregation from various framework selection
and component/module functions
- correctly determine the max MPI thread level that we can support and
limit the final MPI thread level as appropriate during
ompi_mpi_init()
As a side effect, while editing nearly every MPI component I also did
the following:
- found 2 places (gm and ib ptls) where ompi_set_using_threads() was
erroneously being called (I think this is left over kruft from
misunderstandings by the original authors). I removed these.
This commit was SVN r5055.
built. The issue is that these tests are trying to test specific
components, and is calling the functions directly -- and therefore
needs to have the component linked in. This is fine when the
component is statically linked as part of libmpi, but presents a
problem when the component is a DSO.
GNU compilers/linkers allow us to link in the DSO as part of the test
executable (and everything "just works"), but this is not portable. A
better solution is going to involve:
- a better unit test support library that can load a DSO on demand
- using function pointers in the unit tests (rather than direct
function invocation)
This commit was SVN r5051.
