- Fully support REAL*N, INTEGER*N, and COMPLEX*N in the MPI_Op
reduction operations.
- Update ddt to fully support these types as well, to include using
the results of sizes and alignments determined by configure
- Discover the goodness of m4 and consolidate a LOT of configure code
(i.e., remove a lot of essentially duplicated code and
m4-subroutine-ize it). The big kicker was figuring out how to
parameterize AC_DEFINE_UNQUOTED, which you can do if you use m4
properly.
- If we don't support a given INTEGER*N, REAL*N, or COMPLEX*N, don't
error. Just set the right flags so that we don't support them in
the MPI layer.
This commit was SVN r5788.
1. Added a new function to launch head node processes on remote nodes.
2. Added new tool "orteprobe" that checks to see if a daemon is running on a node. If so, it reports the contact info back to the requestor. If not, it will (eventually - but not now) fork/exec a daemon on the node, report the contact info back to requestor, and then die.
3. Modified orted to handle universe name parameters, and added separate command line flags for debugging the daemon and saving daemon debugging output in a file. The "debug" flag now turns on the runtime debug info instead of the daemon debug - thus, you can now just get daemon debug info if you like.
4. Fix the dps to handle zero length strings correctly.
5. Modify the fork and rsh launchers to pass required environmental variables to the daemons and processes
6. Pulled the redirection of stdin/stdout/stderr for the daemon out of orted and put it into the daemon_init function to simplify orted logic.
7. Modified sys_info to correctly deal with passed mca param
8. Modified univ_info to parse incoming universe location information.
This commit was SVN r5705.
- Check for icc-8.1 segfaulting with varargs.
- Add check for size of REAL*4,8,16 types and purge,
if smaller
- Fix configure-output with regard to checking for c-types
for the REAL*-types.
This commit was SVN r5661.
for the size. If it does not fit, purge from the list, i.e. reset all
values to zero.
Fixes pgf90 problem, which silently accepts INTEGER*16, but internally
using INTEGER...
This commit was SVN r5594.
Merged in from:
svn merge -r5506:5553 https://svn.open-mpi.org/svn/ompi/tmp/hetero .
This commit was SVN r5552.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r5506
r5553
Merged in from:
svn merge -r5448:5496 https://svn.open-mpi.org/svn/ompi/tmp/hetero .
This commit was SVN r5550.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r5448
r5496
from:
svn merge -r5440:5448 https://svn.open-mpi.org/svn/ompi/tmp/hetero .
This commit was SVN r5549.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r5440
r5448
MPI_COMPLEX*x, and some optional C datatypes in MPI reduction
operations. These types are not technically supported by the letter
of the MPI standard, but are implied by the spirit of it (and there
are definitely users that use them in real applications)
- Add checks in configure for back-end C types for MPI_INTEGER*x and
MPI_REAL*x
- Create C data structs for MPI_COMPLEX*x
- Fixed typo for MPI_INTEGER8 in mpi.h
- Updated configure macros to create MPI_FORTRAN_INTEGER* defines, as
opposed to MPI_FORTRAN_INT, which was causing [me] lots of confusion
(between C "*_INT" names and Fortran "*_INT" names). This caused
some trivial updates in ddt, ompi_info, and the MPI layer to match.
- Update ompi_info to show whether we have each MPI_INTEGER*x,
MPI_REAL*x, and MPI_COMPLEX*x
- Extended reduction operations for optional datatypes:
- "C integer" now includes long long int, long long, and unsigned
long long
- "Fortran integer" now includes MPI_INTEGER*x
- "Floating point" now includes MPI_REAL*x
- "Complex" now includes MPI_COMPLEX*x
This commit was SVN r5511.
we are part of the source tree and not defined otherwise, we are going
with an always defined if ompi_config.h is included policy. If
ompi_config.h is included before mpi.h or before OMPI_BUILDING is set,
it will set OMPI_BUILDING to 1 and enable all the internal code that
is in ompi_config_bottom.h. Otherwise, it will only include the
system configuration data (enough for defining the C and C++ interfaces
to MPI, but not perturbing the user environment).
This should fix the problems with bool and the like that the Eclipse
folks were seeing. It also cleans up some build system hacks that
we had along the way.
Also, don't use int64_t as the default size of MPI_Offset, because it
requires us including stdint.h in mpi.h, which is something we really
shouldn't be doing.
And finally, fix a ROMIO Makefile that didn't set -DOMPI_BUILDING=1,
as ROMIO includes mpi.h, but not ompi_config.h
This commit was SVN r5430.
having it in config/ompi_setup_cxx.m4
2. Adding --enable-coverage option. This will add teh flags -ftest-coverage
and -fprofile-arcs to the flags. Also, one needs to compile with
debug and static only to enable code coverage
3. Adding the coverage flag options to WRAPPER_*_FLAGS so that mpicc and co.,
will add these to teh executables when they are compiled
This commit was SVN r5416.
Monday with Craig:
- split two f90 modules into two separate .f90 files so that
dependencies can be satisfied properly in parallel builds (i.e.,
mpi_kinds.f90 can be compiled and its f90 module can be generated
before any others are compiled, because it is needed by all of the
other f90 source files).
- rename mpi.i.h* to be mpi-f90-interfaces.h, just to be a little more
clear and less name-confusing with mpi.h and mpif.h
- update the build process in src/mpi/f90, including printing out a
warning that compiling mpi.f90 may take quite a while :-\
- update the wrapper compilers to add in the Right Things for
compiling F90 MPI applications
- add a check in configure to find what flag the F90 compiler uses to
identify where f90 module files live
This commit was SVN r5297.
- remove all tabs from source code (replace with spaces)
- use size of size_t and int to determine some equivalent ORTE types
in a header file so that it's always done consistently (at compile
time)
- be consistent about using size_t for number of elements that are
packed
- don't assume that size_t is uint32_t (it is uint64_t on some
platforms)
- simplify some logic
- implement packing/unpacking for INT64 and UINT64 (so that we can do
size_t on platforms where it's 64 bits)
This commit was SVN r5172.
- the dirent.dt_type field is non-portable. Use stat() if there isn't
a dt_type field in the dirent struct
- Make sure -laio is added to LIBS in the romio component configure
script so that static builds don't result in missing symbols
- Add missing header file for the signal constants in pls_fork_module
This commit was SVN r5152.
more obvious (and documented)
- VERSION now has lots of comments in it
- separate out whether we want the SVN r number in the version and
what the SVN r number is
- clearly describe the process for building the full version number
string
- make ompi info have a separate line item specifically for the SVN r
number, regardless of what the version number is
- update "make dist" process to substitute in the SVN r version
This commit was SVN r5022.
build / run. Only things that actually build / run right now are the
asm and class tests. The mca tests probably will with a static build
but that hasn't been verified
This commit was SVN r4979.
* SPARC Assembly fixes:
- Use SPARC assembly type only when not able to use the SPARC v9
instruction set (instead of previous whenever sizeof(void*) == 4
- Use SPARCV9_32 (aka sparc v8plus) and SPARCV9_64 (aka sparc v9) when
possible. Already were doing so for the 64bit, but on 32bit we
were falling back to SPARC.
- Implemented SPARCV9_32 atomic operations
- Emit warning if user is going to get SPARC atomic, as that probably
isn't what the user wants for best performance.
* Started shell of MIPS assembly. Does not work, but wanted something
so that everyone would only have to re-autogen once today....
This commit was SVN r4465.
- Use fcntl.h, not sys/fcntl.h (man pages on every platform I could find
say fcntl.h and solaris/AIX don't provide sys/fcntl.h)
- Make timeradd macro available on platforms where sys/time.h exists but
timeradd macro doesn't (Solaris)
- Include util/printf.h from ompi_config_bottom.h so that ompi_asprintf
and friends are always available.
This commit was SVN r4441.
* Make sure the update.sh scripts all use OMPI_WANT_SMP_LOCKS
* Add sparc (32bit) assembly
- Memory barriers
- spinlocks
- emulate add/sub using the spinlock table with hashed lookups, as
suggested by the linux kernel folk (better than the other option,
requiring the counters only use 24 bits)
This commit was SVN r4429.
script in components and distributes them as appropriate throughout
the tree (previously, we only did WRAPPERS_* flags). These flags are
*only* used when libmpi is being build statically (--disable-shared
--enable-static).
I also added another classification of flags -- LIBMPI_ALWAYS_* and
WRAPPERS_ALWAYS_* for flags that should always be added to the libmpi
/ wrapper linker lines, not just when compiling libmpi statically.
This commit was SVN r4322.
- Fix up the checks for selected_[int|real]_kind so that we only check
for their sizes/alignment if the compiler actually supports them
This commit was SVN r4279.
during testing. In particular, we can't have a cmpset macro that can
deal with pointers without some really evil voodoo. So have a
different macro for pointers
* Add more detailed testing of the atomics using AM's test framework.
More to come...
This commit was SVN r4191.
of assembler format
* Fix minor bugs in AMD64, PPC, and IA32 assembly for atomic operations
* Clean up the #defines to look for when examining level of atomic operation
support
This commit was SVN r4183.
after MPI-startup.
For this a new mpirun-parameter "mpi_signal" is added, one may specify a
comma-separated list of signals to grab, e.g. mpirun --mca mpi_signal 8,11
will check for SIGFPE and SIGSEGV.
It only finds the first fault (SA_ONESHOT), as after the return the same
fault will occur again.
As printout, the data provided by siginfo_t is printed to STDOUT (yes,
it calls printf ,-]).
Additionally, with glibc, it uses backtrace and backtrace_symbols to
print the calling stack up to the function in which the signal was raised:
(Rank:0) Going to write to RD_ONLY mmaped shared mem
Signal:11 info.si_errno:0(Success) si_code:2(SEGV_ACCERR)
Failing at addr:0x4020c000
[0] func:/home/rusraink/ompi-gcc/lib/libmpi.so.0 [0x40121afe]
[1] func:./t0 [0x42029180]
[2] func:./t0(__libc_start_main+0x95) [0x42017589]
[3] func:./t0(__libc_start_main+0x49) [0x8048691]
This commit was SVN r4170.
Note: in bringing this across, I have obeyed the mca-prefix-rule in defining functions and variables. However, I left the name of the component call as "orte_errmgr.xxx" so that any use of the code will directly "merge" with the new RTE branch without changes.
This commit was SVN r4099.
