- change mca_ptl_base_header_t definition to decrease the
header size for small messages. note that this requires
all ptls to be updated. tcp/self/sm/mx have been changed,
gm/ib/quadrics will be broken by this commit. george and
mitch have volunteered to make the required changes to gm/ib
- revised matching logic to reduce function call overhead
- changes to tcp/self/sm/mx ptls to support the revised headers
This commit was SVN r3602.
This may trigger a complete rebuild :(. Short overview of changes:
- reduce number of network slams at startup
- prevent gpr from hanging when doing process death code
- general gpr cleanups
This commit was SVN r3584.
the contiguous datatype correctly and fast. However in the "developer mode" (OMPI_ENABLE_DEBUG set to 1)
there are additional test that are made. If you plan to look for performances please compile OpenMPI
in normal mode.
Changes:
- correctly generate the stack when we have an initial position
- same thing when we start from ZERO.
- in contiguous mode handle the partial copy of a data (using the stack)
- in the case when not all the data is packed/unpacked leave the stack in a correct state
- take care of data with size ZERO (who really want to create such things !!!)
- general pack/unpack functions handle the new stack representation
- make sure that we never goes outside the user provided buffer
This commit was SVN r3581.
to kill the bootproxy). This should fix the leftover cruft processes
people are seeing when processes fail.
* Do some code cleanups to get process-related data types handled slightly
better. Still have to do some work here to handle transmitting the
data in a more reasonable fashion.
* Clean up some of the bootproxy code
This commit was SVN r3574.
- Add new thread function to compare current thread to an
ompi_thread_t*
- Save current thread ID in MPI_INIT*
- Add first cut of solaris thread functions
- Add stubs for missing Windows thread functions
This commit was SVN r3569.
calls to comm_get_parent shuld return comm_null. Additionally, if
it's freed, we don't want to free it again during ompi_comm_finalize.
So if we're freeing the user's handle to the parent comm (i.e., we're
not calling ompi_comm_free on ompi_mpi_comm_parent itself), then
explicitly set ompi_mpi_comm_parent to &ompi_mpi_comm_null. This does
two things:
1. subsequent calls to comm_get_parent will return comm_null
2. we won't try to ompi_comm_free the ompi_mpi_comm_parent in
ompi_comm_finalize()
This commit was SVN r3543.
