Will be replaced by PRRTE. Ensure that OMPI and OPAL layers build
without reference to ORTE. Setup opal/pmix framework to be static.
Remove support for all PMI-1 and PMI-2 libraries. Add support for
"external" pmix component as well as internal v4 one.
remove orte: misc fixes
- UCX fixes
- VPATH issue
- oshmem fixes
- remove useless definition
- Add PRRTE submodule
- Get autogen.pl to traverse PRRTE submodule
- Remove stale orcm reference
- Configure embedded PRRTE
- Correctly pass the prefix to PRRTE
- Correctly set the OMPI_WANT_PRRTE am_conditional
- Move prrte configuration to the end of OMPI's configure.ac
- Make mpirun a symlink to prun, when available
- Fix makedist with --no-orte/--no-prrte option
- Add a `--no-prrte` option which is the same as the legacy
`--no-orte` option.
- Remove embedded PMIx tarball. Replace it with new submodule
pointing to OpenPMIx master repo's master branch
- Some cleanup in PRRTE integration and add config summary entry
- Correctly set the hostname
- Fix locality
- Fix singleton operations
- Fix support for "tune" and "am" options
Signed-off-by: Ralph Castain <rhc@pmix.org>
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
Automake's Fortran compilation rules inexplicably use CPPFLAGS and
AM_CPPFLAGS. Unfortunately, this can cause problems in some cases
(e.g., picking up already-installed mpi.mod in a system-default
include search path).
So in relevant module-using Fortran compilation Makefile.am's, zero
out CPPFLAGS and AM_CPPFLAGS.
This has a side-effect of requiring that we compile the one .c file in
the F08 library in a new, separate subdirectory (with its own
Makefile.am that does _not_ have CPPFLAGS/AM_CPPFLAGS zeroed out).
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
Fix the C types for the following:
* MPI_UNWEIGHTED
* MPI_WEIGHTS_EMPTY
* MPI_ARGV_NULL
* MPI_ARGVS_NULL
* MPI_ERRCODES_IGNORE
There is lengthy discussion on
https://github.com/open-mpi/ompi/pull/7210 describing the issue; the
gist of it is that the C and Fortran types for several MPI global
sentenial values should agree (specifically: their sizes must(**)
agree). We erroneously had several of these array-like sentinel
values be "array-like" values in C. E.g., MPI_ERRCODES_IGNORE was an
(int *) in C while its corresponding Fortran type was "integer,
dimension(1)". On a 64 bit platform, this resulted in C expecting the
symbol size to be sizeof(int*)==8 while Fortran expected the symbol
size to be sizeof(INTEGER, DIMENSION(1))==4.
That is incorrect -- the corresponding C type needed to be (int).
Then both C and Fortran expect the size of the symbol to be the same.
(**) NOTE: This code has been wrong for years. This mismatch of types
typically worked because, due to Fortran's call-by-reference
semantics, Open MPI was comparing the *addresses* of these instances,
not their *types* (or sizes) -- so even if C expected the size of the
symbol to be X and Fortran expected the size of the symbol to be Y
(where X!=Y), all we really checked at run time was that the addresses
of the symbols were the same. But it caused linker warning messages,
and even caused errors in some cases.
Specifically: due to a GNU ld bug
(https://sourceware.org/bugzilla/show_bug.cgi?id=25236), the 5 common
symbols are incorrectly versioned VER_NDX_LOCAL because their
definitions in Fortran sources have smaller st_size than those in
libmpi.so.
This makes the Fortran library not linkable with lld in distributions
that ship openmpi built with -Wl,--version-script
(https://bugs.llvm.org/show_bug.cgi?id=43748):
% mpifort -fuse-ld=lld /dev/null
ld.lld: error: corrupt input file: version definition index 0 for symbol
mpi_fortran_argv_null_ is out of bounds
>>> defined in /usr/lib/x86_64-linux-gnu/openmpi/lib/libmpi_usempif08.so
...
If we fix the C and Fortran symbols to actually be the same size, the
problem goes away and the GNU ld bug does not come into play.
This commit also fixes a minor issue that MPI_UNWEIGHTED and
MPI_WEIGHTS_EMPTY were not declared as Fortran arrays (not fully fixed
by commit 107c0073dd).
Fixesopen-mpi/ompi#7209
Signed-off-by: Fangrui Song <i@maskray.me>
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
Forgot to include a fix for the fortran test used to check if
new dtags is supported.
Related to #7268
This patch is already included on v4.0.x branch.
Signed-off-by: Howard Pritchard <howardp@lanl.gov>
* Additionally, fixes the `NULL` option to `OMPI_MCA_plm_rsh_agent`
would would also lead to a segv. Now it operates as intended by
disqualifying the `rsh` component and falling back onto the `isolated`
component.
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
OpenMPI doesn't compile anymore with IME because the header
file "ompi/mca/fs/base/base.h" needs to be include in every
file where mca_fs_base_get_mpi_err() is used.
Signed-off-by: Sylvain Didelot <sdidelot@ddn.com>
Improves the performance when excess non-blocking operations are posted
by periodically calling progress on ucx workers.
Co-authored with:
Artem Y. Polyakov <artemp@mellanox.com>,
Manjunath Gorentla Venkata <manjunath@mellanox.com>
Signed-off-by: Tomislav Janjusic <tomislavj@mellanox.com>
Previously we used a fairly simple algorithm in
mca_btl_tcp_proc_insert() to pair local and remote modules. This was a
point in time solution rather than a global optimization problem (where
global means all modules between two peers). The selection logic would
often fail due to pairing interfaces that are not routable for traffic.
The complexity of the selection logic was Θ(n^n), which was expensive.
Due to poor scalability, this logic was only used when the number of
interfaces was less than MAX_PERMUTATION_INTERFACES (default 8). More
details can be found in this ticket:
https://svn.open-mpi.org/trac/ompi/ticket/2031 (The complexity estimates
in the ticket do not match what I calculated from the function)
As a fallback, when interfaces surpassed this threshold, a brute force
O(n^2) double for loop was used to match interfaces.
This commit solves two problems. First, the point-in-time solution is
turned into a global optimization solution. Second, the reachability
framework was used to create a more realistic reachability map. We
switched from using IP/netmask to using the reachability framework,
which supports route lookup. This will help many corner cases as well as
utilize any future development of the reachability framework.
The solution implemented in this commit has a complexity mainly derived
from the bipartite assignment solver. If the local and remote peer both
have the same number of interfaces (n), the complexity of matching will
be O(n^5).
With the decrease in complexity to O(n^5), I calculated and tested
that initialization costs would be 5000 microseconds with 30 interfaces
per node (Likely close to the maximum realistic number of interfaces we
will encounter). For additional datapoints, data up to 300 (a very
unrealistic number) of interfaces was simulated. Up until 150
interfaces, the matching costs will be less than 1 second, climbing to
10 seconds with 300 interfaces. Reflecting on these results, I removed
the suboptimal O(n^2) fallback logic, as it no longer seems necessary.
Data was gathered comparing the scaling of initialization costs with
ranks. For low number of interfaces, the impact of initialization is
negligible. At an interface count of 7-8, the new code has slightly
faster initialization costs. At an interface count of 15, the new code
has slower initialization costs. However, all initialization costs
scale linearly with the number of ranks.
In order to use the reachable function, we populate local and remote
lists of interfaces. We then convert the interface matching problem
into a graph problem. We create a bipartite graph with the local and
remote interfaces as vertices and use negative reachability weights as
costs. Using the bipartite assignment solver, we generate the matches
for the graph. To ensure that both the local and remote process have
the same output, we ensure we mirror their respective inputs for the
graphs. Finally, we store the endpoint matches that we created earlier
in a hash table. This is stored with the btl_index as the key and a
struct mca_btl_tcp_addr_t* as the value. This is then retrieved during
insertion time to set the endpoint address.
Signed-off-by: William Zhang <wilzhang@amazon.com>
We initially thought it was a safe bet that opal_gethostname() would
never be called before opal_init(). However, it turns out that there
are some cases -- e.g., developer debugging -- where it is useful to
call opal_output() (which calls opal_gethostname()) before
opal_init().
Hence, we need to guarantee that opal_gethostname() always returns a
valid value. If opal_gethostname() finds NULL in
opal_process_info.nodename, simply call the internal function to
initialize opal_process_info.nodename.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
Advance to hwloc-2.1.0rc2-33-g38433c0f, which includes a .gitignore
update that we want here in Open MPI.
Be warned; this is actually 33 commits beyond the hwloc v2.1.0 tag.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
