areas:
- The MPI attribute functions in f77 (the MPI-2 set, get,
keyval_create functions for each of the 3 types [comm, datatype,
win], and the MPI-1 put/get functions)
- the src/attribute/* back-end implementation
- MPI_TYPE_DUP implementation
- back-end ddt implementation
Here are the foxes:
- F77 attributes were previously being saved by reference (!)
- Until a few days ago, I didn't realize that the Fortran callback
functions between MPI-1 and MPI-2 had different signatures. These
changes update to handle all 3 kinds of callback functions: C, MPI-1
Fortran, and MPI-2 fortran.
- Also clean up a few related bugs with respect to MPI_TYPE_DUP (i.e.,
ensure that all attribute callbacks are properly invoked) and
ompi_ddt_duplicate (ensure that flags, f2c index, and keyhash are
all set properly).
This commit was SVN r6099.
mpirun -np 1 uptime : -np 1 hostname
would fail, claiming that it was out of resources. Turns out that the
code was incorrectly checking for "no more nodes". This patch fixes
that, both for the schedule-by-node and schedule-by-slot algorithms.
This commit was SVN r6088.
frontend work and bproc_orted which sets up io for the processes on the backend.
The launcher is fairly complete and has been tested on fairly large jobs.
However, the finalization is not done so right now it just blocks in finalize.
This commit was SVN r6085.
the RDS selection logic, which is, unfortunately, not yet well
supported by the testing infrastructure (it causes false failures in
the nightly build).
This commit was SVN r6073.
VAPI_SEND was not being set in mca_bmi_ib_endpoint_post_send so when the frag
had been used for an RMDA the opcode was incorrect.
Various other misc fixes related to small frag rdma.
This commit was SVN r6072.
We should refrain from using the variable 'new' since it is a
keyword in C++, and some compilers are really picky about it,
namely cl :)
Did a bit of funny wrapping in convertor.h so that the
ompi_convertor_prepare function is only exported when
doing a windows build. Talked with George about this
change and we both agreed that the windows build is
messy, but shouldn't corrupt the main line precausions.
Also wrapped convertor.h in the __cplusplus defines
to help with exporting the inline functions under
windows.
This commit was SVN r6054.
All that was required was to change the subscription request in module_exchange to specify "don't start notifying me until after the trigger". The trigger is set to fire when all processes have completed their initial stage_gate, so that means the process will receive notifications only of any changes from the base starting point.
Specifically:
1. I changed the "don't start" flag to ORTE_GPR_NOTIFY_STARTS_AFTER_TRIG to make it more obvious as to its function
2. Added that flag to the module_exchange subscription, and changed the name elsewhere it was used (I had already fixed this problem for the OOB)
3. Found an unrelated compiler warning of an unused variable in ns_proxy, so I fixed it too.
This commit was SVN r6041.
