Change summary from the previous version


Up: Contents Next: Deprecated Features Previous: Mpirun Usage

This section lists the changes to mpich since the previous version. These changes are taken from the CVS log entries that accompany each change and may not be complete or entirely accurate.

=1pt
* '#elif defined(FOO)' is better than '#elif FOO' if FOO might be undefined
* (romio) clean up the output a bit more
* (various configure.in files) for autoconf 2.13 and on, AC_SEARCH_LIBS adds any necessary libraries to the LIBS variable, so I modified our configure scripts to no longer rely on internal autoconf variables (ac_vc_search_*) and instead use LIBS.
* a start at adding pvfs2-specific hints
* Add DESTDIR to installation paths
* Add NETBSD uniformly to timer directory
* add PVFS2 to the list of filesystems that don't support shared file pointers
* Add a test for no f90 compiler
* Add code to save/set status on split collectives
* Add locations for Java
* Add mpi include path when testing for MPI_Offset size and only perform that test for Fortran
* Add mpichversion to MPICH and ensure that change directory commands in Makefile are successful before executing the next step
* Add new methods for data type SLOG_strlist_t
* Add support for io tests and upgrade runbase
* added 'struct gridftp_params' for MPICH-G2's use of Globus GridFTP
* Added .deps to .cvsignore
* Added AC_OBJEXT and AC_EXEEXT changes to Makefiles to support cygwin.
* Added AC_OBJEXT and AC_EXEEXT to configure.in and related changes to aclocal_mpi.m4, this is done to support cygwin.
* added API for IBM
* Added BUILDING_MPI symbol to suppress link tests in the midst of building
* Added CPPFLAGS to PAC_TEST_PROGRAM
* Added EXEEXT to move targets for cygwin
* Added FREEBSD_PPC machine time (like FREEBSD except for data representation
* Added FREEBSD_PPC to known arches for p4
* Added FREEBSD_PPC to known systems
* Added Fortran binding for MPE_Log_send() and MPE_Log_receive()
* Added LINUX_PPC to known arches
* Added MPE_LOGFILE_PREFIX to allow specification of MPE output filename prefix.
* added MPI_Pcontrol for illustration
* Added RANLIB
* Added Rabenseifner's algorithm for long-message Reduce and Allreduce. Also added a lgp recursive halving algorithm for Reduce_scatter in the case of commutative ops
* Added TAGS targets
* added __LINE__ to MPIR_Err_create_code
* Added a check for time.h and that sys/time.h and time.h may be used together
* Added a comment about determining the largest fd
* Added a default for F77 if no architecture is recognized
* Added a dummy definition of GREQUEST to make the romioconf.h.in files the same for configure and configure2
* Added a few comments about fields in File structure
* Added a fix for zap_p4_processes
* Added a freebsd_ppc type (likely MacOSX)
* Added a macro defn so that other parts of MPICH can tell if MPIO_Request has been defiend
* Added a more forgiving check for a valid arch in the MACHINES file
* Added a new test with an increment of two for group creation
* Added a note about problems with the AC_EXEEXT macro in autoconf 2.13
* Added a short comment about MPI_CC
* Added a test for READ_ONLY file
* Added a test for a failure in ROMIO configure
* Added a test that struct statfs can be compiled, because of problems with the Intel C compiler and glibc
* Added a way to skip the does-it-run-cleanly step, needed for coverage tests (which rarely run cleanly)
* added ad_iopen.c
* Added additional positional args for gcc
* Added additional search dirs
* added an MPI_Type_free for an MPI_Allgather subroutine
* Added an additional step to drain incoming messages and a comment about ordering head-to-head rendezvous messages
* Added an empty coverage target to MPE
* Added an option for an environment variable to set the stopfile
* Added another Linux synonym
* Added another test for cancel of Issend based on an MPICH bug report (currently MPICH1 fails this test with a SEGV in ch_p4 device)
* Added autom4te.cache to .cvsignore
* added barrier to MPI_File_close to avoid deletion of shared file pointer file while others were still accessing it.
* added bug fix sent in by Wei-keng on atomicity in collective writes (Req #516)
* added casts to int from functions that return size_t and added MPIU_dbg_init() to initthread
* Added check for getopt.h
* Added code to determine the argc/argv support needed for the ROMIO Fortran tests
* added code to propagate environment variables for Nick. Nothing happens unless P4_SETS_ALL_ENVVARS is set
* Added code to provide accurate MPI error classes for failure to open file (e.g., path too long, missing directory, no such file)
* Added comment about incompatible demon options in mpd1 and mpd2
* Added comments about setting the working directory
* Added copyright statements to ROMIO test programs
* Added coverage options and standard enable/disable for f77
* Added coverage output files to cvsignore
* Added coverage targets
* Added definitions to allow ROMIO programs that use MPIO_Wait to work with ROMIO under MPICH2, where generalized requests allow MPI_Wait et al to work
* added documentation comments, reported memory errors, put the pmi database on the root process
* added error checking blocks
* added error code for local success but remote file_open failure
* added error message catalog dll to ch_nt device
* Added exec_prefix and sbindir to mpirun variables so that we can invoke tools like cleanipcs in sbindir
* Added explanation when error seen
* Added export of OPTFLAGS and OPTFLAGSF so that all parts of MPICH get -opt=flags
* Added failure return if manager cannot be found
* Added fixes for autoconf later than 2.13; also make a failure to build slogsdk non-fatal
* Added function prototype and stdlib.h header to eliminate compiler warning on exit()
* added globus_core to globus-makefile-header command as a workaroundto newly introduced bug in either globus-makefile-header (notfinding the implicit dependency for it) or gpt-build (not seeingthe stuff from globus-makefile-header)
* Added hooks for ROMIO to propagate shared library information
* Added hooks for generating shared libraries for ROMIO adio/common directory
* Added if test to ROMIO configure to bypass check for Fortran if --disable-f77 is selected
* Added ifc to f90 compilers
* Added large file support for 32-bit OS: Added test to configure.in to check if O_LARGEFILE is available (Ref: http://ftp.sas.com/standards/large.file/xopen/x_open.05Mar96.html) and required macros to mpeconf.h.in.
* Added markers for error-checking code to aid in coverage analysis
* Added master_top_srcdir and man page documentation targets
* Added missing test for MPI_STATUS_IGNORE
* Added more support for shared libraries
* Added most of the multiple completion waits and tests in MPIO_Waitany etc. form
* Added mpicxx to installed files, and handle case-insensitive file systems (for Mac OS X)
* Added mpicxx to known compiler names
* added new algorithms for alltoall, allgather, and allgatherv
* Added optional definition of S_ISLNK if strict ANSI selected for C
* Added patch sent in by multiple people for uninitialized data bug
* Added pointer in ADIO_File structure for file system specific use.
* Added quotes to protect against an empty variable
* Added some tests for struct flock problems
* Added special support for TFLOPS architecture and devices that do not have MPID_Status_set_bytes
* Added stdlib.h
* Added stdlib.h header to eliminate compiler warning on exit()
* Added support for gcc dependency creation
* Added support for including patch information in mpichversion
* Added support for mpichversion
* Added support for setting group id
* Added support to make SLOG1 API's library and related tools depending on the successful return of the SLOG1 API's configure. This is added to support Cray. Also, modify some global subroutines that have CLOG_ prefix to C2S1_ instead( abbi's code ).
* Added temporary ac_subst items to enable the rest of the ROMIO tests
* Added test for CPP
* Added test for errno == 0 because of Solaris bug (return -1 but errno not set)
* Added test for pvfs header and int64_t definition fix
* added text on needing lockd daemon to NFS file lock error message
* Added the package slog2sdk to the configure.in and Makefile.in. Also added support for drop & configure support for slog_api, viewers and slog2sdk.
* adding ch_p4mpd device files
* adding files for lib in p4mpd
* adding files in include dir for p4mpd
* adding mpd source configure- and make-related files
* adding mpd source files
* adding top-level p4mpd files
* adjusted error handling block
* Allow -n as a synonym for -np in mpirun
* Allow a path list for mpd and make sure that it is propagated from the top-level configure
* Allow compile commands with options in command tests
* Allow empty arguments in mpeinstall
* Allow make to be set from the environment
* Assert that we support funneled thread safety
* autoconf already substitutes @prefix@, so don't manually AC_SUBST(PREFIX)
* Be sure to copy all data out of a short received packet before returning it to the available pool (where another process may use it)
* Better handle misssing comparison file
* cache local ip string
* Capability now there to use dataloop code for flattening, but it is currently disabled (at the top of flatten.c).
* ch_nt: added ad_set_view.c
* ch_nt: added gencheck.c to the project files
* ch_nt: added host name to error output
* ch_nt: added some retry code to avoid thread creation failures
* ch_nt: added synchronization to shmem queue initialization
* ch_nt: added the ability to share memory accross local windows workstation boundaries
* ch_nt: fixed mpptest in the examples \\nt directory
* ch_nt: updated error messages
* Change := to = in LIBS assignment
* Change mpptest output to use gnuplot by default
* Change p4_dprintf to p4_dprintfl in p4_sock_cr.c so that messages about environment are not printed by default
* Changed AUX_DIR to point to local version
* changed MPIR_ALLREDUCE_TAG to MPIR_REDUCE_SCATTER_TAG in a few places in reduce_scatter
* changed MPI_File_open to fail on all processes if any process returns error
* changed NULL to MPI_DATATYPE_NULL in ad_init.c
* Changed error message for intracomm expected
* changed faulty var name hostname -> host_name in DEBUG
* Changed max msg queues from 8 to 64 in p4_MD.h
* changed maximum number of processes from 32 to 256
* changed minimum spanning tree to binary tree in algorithm description for collectives
* changes to utilize GridFTP
* Check for PRINT_ERR_MSG defined
* Check for STATUS_IGNORE
* Check for and use errno.h
* Check for and use stdlib.h
* Check for broken autoconf (without support for AM macros)
* Check for f90 include files before trying to install them
* Check for mpichversion.exe for cygwin installations
* Check for needs int64_t definition
* Check for read-only access to the file before attempting to write
* Check the status of most MPI_File operations.. added some commented-out hints we should use whenever we teach coll_test how to handle more command-line arguments than just -fname
* Clean up ROMIO makefile
* Clean up ROMIOs aclocal.m4
* Cleanup use of getopt and use header file when available
* close a small memory leak
* closes romioreq #515: with pete wycoff's suggested fix: don't override user-specified include path.
* configure test for fstypename means we don't have to test for specific bsd versions anymore
* correct the markup for the deferred open hints
* corrected bug in mpirun.globus2 script: -np X with X>255 should work on any platform -- bug uncovered by [email protected], sent to [email protected] on Tue, 25 Nov 2003 13:19:57 +0800 (CST)
* Corrected message about number of allowed processes in test
* Create the links for the sharedlib names for use in-place
* DESTDIR for mpiinstall.in
* declared integer toterrs in isendf.f
* deferred open: now open the file if independent io attempted
* deleted trailing spaces in fortran/configure.in
* Deleted useless symbols: double precision MPE_WTIME, MPE_PTIME
* derrive the 'deferred_open' hint if the "no independent io" and "use collecitve buffering" hints are turned on.
* Detect a wtick of 0
* Do not define MPI_Info_xxx as PMPI_Info_xxx when building the MPI_Info routines in mpi2-other/info
* do not do link tests if building withing an MPI implementation
* Do not fail if we cannot remove a link to mpif.h that we believe we have created
* do not override includedir
* document no_indep_rw a bit better
* documentation to use SLOG-API to add arrow
* don't assume MPI_INFO_NULL is zero. explicitly compare against MPI_INFO_NULL
* don't mung MPI_LIB if you can help it. if ROMIO needs extra (non-mpi) libraries to link, use ROMIO_LIBLIST
* don't set C_OPT_FLAGS if user provides CFLAGS (closes: mpich2-req 238)
* Eliminating common uninitialized symbols in ROMIO common/ad_init.c.
* Ensure that CFLAGS is exported as well as CC in case the user has placed flags that change the compilers behavior in significant ways into CFLAGS instead of CC
* Ensure that MPI->PMPI mapping is used for internal routines
* Ensure that PAC macros are all expanded or not present in generated configure (like AC macro names, they should not even be in comments to simplify error checks for unexpanded macros)
* Ensure that all tests use FFLAGS or F90FLAGS
* Ensure that definitions of HAVE_MPI_INFO have the same value
* Ensure that long long is 8 bytes before using it for int64_t
* Ensure that sockets are only closed if they have been opened (some logic paths might not set end_1 and/or end_2)
* Ensure that status is set correctly for count == 0
* Ensure that temporary test files are removed
* Ensure that the ENABLE_SHLIBS variable is none if not building shared libraries in ROMIO
* Ensure that the Fortran interface libraries have the ranlib step applied
* Filter additional processes from aborttest checks
* fine-tuning hint precedence for deferred open
* first cut at using generalized requests, if avaliable
* Fix Makefile.in so it won't build mpereconfig from mpereconfig.in
* Fix ROMIOs scheme for detecting Fortran naming convention to use - in strings (-a isnt enough under Mac OSX)
* Fix [] bug in C++ autoconf macros (missing changequote statements)
* Fix array length
* Fix begin->end in error handling comment
* Fix for --enable-boot for ch_p4mpd testing
* Fix for MPICH variable in mpeinstall
* Fix for ROMIO under MPICH1
* Fix for exclude and machine files with ncpus specifiers
* Fix for gcc 3.2 weak symbols
* Fix for mpich-1 and romio
* fix for new mpiinstall
* Fix for read of uninitialized storage and error in the macro to check if a send has been cancelled
* Fix for shared libraris in ROMIO
* Fix for some broken automakes
* Fix for the second use of aborttest
* Fix ifdef for verbose output
* Fix incorrect use of config.log
* Fix invokation of romio configure
* Fix long-standing problem with IO tests in test suite
* Fix missing ierr on Allreduce call
* Fix missing semicolon
* Fix problems with dependency generation code
* Fix test for send datatype to apply only at the root process
* fix the unparameterized MPIR_Err_create_code calls
* Fix to allow totalview to work when -nolocal is selected
* Fix to ensure that include path is included in tests
* Fix typo in one of the wtime declarations
* fixed -hosts option bug
* fixed MPI_FILE_NULL warning and configuration bugs in ROMIO Fortran tests
* fixed ROMIO Fortran interface for gcc 3.2
* Fixed a bug in init_g.c caused by globus_duroc_intra_subjob_sendthat put a limit on the number of procs that could be startedANDcommented out "sanity checking" code that prevented MPICH-G2 fromlinking against the Globus Toolkit 2.4.3 libraries that were distributedin GT v3.2.
* fixed a logic problem comparing fd->info to MPI_INFO_NULL
* fixed bug in MPID_Comm_init so that it will now work with new MPI-2 functions MPI_Comm_connect,accept on machines with vMPI.
* fixed bug in MPI_Reduce_scatter
* fixed bug in alltoall
* fixed bug in bnr.c: reset global fence flag
* fixed bug in file_set_errhandler
* Fixed bug in noncontig memory/contig file listio code (not calling status set bytes).Added new combining write functionality into non-listio strided write, removing writev calls.
* fixed bug in reduce scatter
* Fixed bug in semop call when interrupts received
* fixed bug in type_create_darray
* fixed bug in weak symbols test in ROMIO's configure.in
* fixed bug related to MPI_DISPLACEMENT_CURRENT
* fixed bug with 0-size messages in collectives
* Fixed bugs in the new ROMIO configure
* Fixed build for configure due to new files in confdb
* Fixed documentation that refered to MPI_STATUS_NULL or MPI_STATUSES_NULL (should be IGNORE, not NULL)
* Fixed emacs header and added comments about f77 failures
* Fixed error in pmpi version of object comparisons that caused the Solaris C++ compiler to give fatal errors
* fixed list freeing code
* Fixed longstanding typo in a define.
* fixed some error printint statements
* fixed string for executable staging ... added '#' which is RSL concat operator
* Fixed the GCC ever changing treatment of va_arg by changing the test from AC_TRY_COMPILE to AC_TRY_RUN to avoid the unexpected coredump due to non int va_arg.
* Fixed the GCC3XX test
* Fixed the test for the master_top_srcdir
* fixes a segfault in noncontig_coll2 test
* Fixes for ROMIO tests from MPICH test suite
* Fixes for caching and enable strict plus fix for bogus GNU version of head and tail
* Fixes for mpichversion
* Fixes for the multiple completion routines in ROMIO
* fixup handling of profiling target
* Flatten now supports subarray and darray combiners.
* Further improve error reporting from perftest/makeconfigure
* Further update the ROMIO aclocal.m4 for autoconf 2.57
* get value directly from the hint structure instead of parsing the Info object
* getting all Globus header files above all others to fix annoyingOS-based compiler bugs/warnings are various archs ... most notably AIX
* hack around the fact that := isn't avaliable everywhere, and added a .o suffix rule to clean up some of the targets
* Handle more bugs in aclocal
* honor setting romio_ds_write to "disable"
* if indepenent io attempted during deferred open, abort
* If we use generalized requests, these conversion functions don't have to do anything
* Ignore config.cache in ROMIO since autoconf does not correctly detect critical changes to the environment
* implement scalable MPI_File_set_size for pvfs1
* implement scalable sync for pvfs1
* implement set_view as a function, not as fs-specific fcntl.
* Improve filtering for irrelevant processes
* Improve handling of IO tests
* Improve test for alternate libraries to support getarg
* Improve the handling of the IO tests in the mpich test suite when romio is not built
* improved configure to handle missing Fortran compiler
* Improved handling of shared libraries
* Improved spec file
* Increase secure server timeout to 2 seconds and add more debugging output
* Increase wait time in secure server code to 5 seconds
* Increased max keyvals from 64 to 256.
* initial deferred open work. skips the open for now. will actually defer later
* Initial deposit, to get autoheader to work correctly
* Initialize autodx and autorel (not actually needed) to keep some compilers happy
* introduce a 'worker_rank' member to ADIO_FileD.
* Issue a configure-time error message if comm=bproc but bproc is not available
* lam-submitted patch to use 'unsinged int' to hold the umask(2) output
* lam-supplied portability fixes
* Let C++ configure know about versions 1.2.5 and 1.2.6
* look for large file summit headers and defines; add proper magic to CFLAGS if we find them
* MPI_File_seek won't actually seek: we keep track of where we should be and seek before reading or writing. now only has to open the file in the MPI_SEEK_END case
* Made sure that all tests guard variables that might be empty with both a -n test and an && concatenation, along with quotes everywhere
* Made the MPIO_Request routines optional (not built for MPICH2 which should be using generalized requests instead) and added LIBTOOL definitions
* make clean_execer routine have a conforming internal name
* Make error test use standard output format
* Make test for error message less sensitive to the exact wording
* Make the C++ configure understand mpicxx as well as mpiCC
* Make the include of mpiof.h optional
* Make the release level current so that patches can be more easily generated
* Minor fix to HPUX stuff.
* Minor fix to man page for Type_indexed
* minor fixes to MPIR_Err_create_code() calls
* missing prototype for MPI_Status_set_elements
* modified CheckRecvErr() to check the error class in status->MPI_ERROR rather than the whole error code
* Modified ROMIO tests to generate standard output on no errors
* Move BasicTypes to a better place
* Move use of USE_STDARG after the point where it is defined by mpidefs.h
* moved MPID_Type_free from the "default" case to just after thewhole switch statement so that it is now called for _all_ the casesin the switch statement ... not just the "default" case
* Moved compiler selection ahead of other arch-specific tests to workaround a limitation in AC_PROG_CC and anything that requires it, including AC_CHECK_HEADERS
* Moved conf file to top
* must edit mpio.h to define HAVE_MPI_GREQUEST if avaliable
* new error handling
* new versions of pvfs have fewer members in the pvfs_filestat structure. icctreats 'too many initializer values' as an error.
* no need for worker_rank after all. ranklist[0] has what we need
* OS-based compiler bugs/warnings are various archs ... most notably AIX
* one more deferred open file
* Pass CFLAGS to mpd configure to ensure shared library options are propagated
* Pass includedir to mpe configure
* Pass libdir and bindir to mpe configure
* passed the correct size of the buffer into MPI_Buffer_attach()
* Preliminary update of MPE README for inclusion of slog2sdk
* Protect results from pwd step in case it fails
* prototopye for ADIOI_GEN_WriteStrided_naive
* Provide a default if MPIRUN_DEVICE not set
* Provide an ifdef for using Generalized requests in MPI-IO code (e.g., use MPI_Wait instead of MPIO_Wait)
* Provide more detailed error messages and error classes on open failures
* Put #if 0 #endifpair around doing chdir since that is handled by mpd in the ch_p4mpd device.
* Re-establish a connection with the listener before sending the DIE message
* Rearranged AC_PROG_CC to the beginning of configure.in so AC macros that needs CC can be invoked without complaints when running autoconf.
* Remember status return on IO operation so that status field on split collective can be set correctly on the xxx_end routine
* Remove .lo files as well as .o files in clean step
* Remove MPI_Type_count
* Remove bogus setting of default file system
* Remove duplicate ioreq_c2f and ioreq_f2c
* Removed all mpereconfig references. Explicitly creating examplesbuild_dir/Makefile using build_dir/contrib/test/Makefile as a template.
* removed all the symbols that were commented out with #if 0-#endif block
* Removed mpich2_fileutil.c from list of files to compile with the name-shifting interface
* removed mpiimpl.h
* Removed rmprog target for mpereconfig.*
* Removed spurious \\ from line
* Removed test for file locks under NFS
* Renamed internal function with p4 prefix
* replaced --enable-slog2source by --enable-slog2=build.
* fixes for the HP bug (closes romio req #570)
* romio: a few more error code fixes
* romio: bandaid fix for mpiimpl.h include problem
* romio: nfs error code updates
* romio: ufs error code updates
* romio: updated error handling for ad_ntfs - needs to be done for all the other ad_xxx's and the common files
* Run cleanipcs if possible after running comm=shared p4 jobs
* SL: corrected bug for allgather, allgatherv, alltoall when called with buffer=NULL and counts=0
* Save output from aborttest for the case where there is a problem
* Set error class to MPI_ERR_ARG for improper filetype
* Set group
* setbuf with a null arg does not always turn off buffering; also add tests that write to socket succeeded
* slog2sdk is installed in -copymode=core
* Some aggregator-aware operations need to know the rank of the 'io_worker', sointroduce a 'worker_rank' member to ADIO_FileD.
* Some enhancements for shared libraries
* Start adding casts to handle compiler warnings about varags and uint_16 items
* statuses[?].MPI_ERROR was not being reset when waiting after a MPI_ERR_PENDING was returned from the MPI_Waitall()
* Support .S as well as .s
* Switch from MPI_ERR_UNSUPPORTED to ERR_ACCESS for access failures
* Switch from MPI_INFO_SRC to HAVE_MPI_INFO_SRC definition name for codes that need the ROMIO implementation of MPI Info
* Switch to using autoconf2 configure for ROMIO
* teach romio about new pvfs2 device
* Test for getwd and only define getwd in terms of getcwd if necessary (to avoid problems with different versions of getcwd)
* that put a limit on the number of procs that could be startedANDcommented out "sanity checking" code that prevented MPICH-G2 fromlinking against the Globus Toolkit 2.4.3 libraries that were distributedin GT v3.2.
* the chek for fstypename cannot be AC_TRY_COMPILE. autoconf1.7 uses AC_COMPILE_CHECK
* the fd is inside a request structure
* the fd is part of the request structure, not standing by itself
* Took out line setting ind_wr_buffer_size to -1 in disabled case. Using thatvalue in a more generic sense now.
* Turn off the test for the MPI library when building as part of MPICH
* typedefed the ADIOI_FileD structure
* umode(2) returns a mode_t, not an int. changed type of old_mask accordingly
* Undefine MPI_xxx_c2f and f2c before defining them if necessary
* Update .cvsignore for 2.57 cache
* Update ancient nupshot configure
* Update configuration of MPE for LAM
* Update makeconfigure to be more robust
* update man page comment
* updated #define file to rename global symbols
* updated ROMIO code to take into account deletion of SGI files
* updated ROMIO's configure for weak symbols
* Updated User's Guide to include additional information on ROMIO and NFS.
* updated algo cutoff point in bcast
* updated algorithms for allgather, allgatherv, and alltoall
* updated bcast algorithm
* updated collective algorithms
* Updated configure.in in MPE for better autoheader and for mpich2
* updated description of collective algorithms
* Updated java search paths
* updated romio for mpich-2
* Updated some error handling in ROMIO
* updated test to use prototype form
* Updated the config.guess script
* Updates to get the ROMIO_HOME srcdir
* Upgrade the PAC_TEST_PROGRAM macro
* Use F90FLAGS instead of FFLAGS for f90 compilation/linking
* Use Fortran to link Fortran programs
* Use IPC_PRIVATE instead of getpid to get a unique id for new sysv mem and semaphores
* use LDFLAGS and LIBS if provided by the environment variables
* Use MPI_Wait for requests when generalized requests available for ROMIO
* Use XSL Transform xmlns for stylesheet
* Use a more careful check for problems in MPI_Wtime
* use default cflags on alpha that work with either gcc or native compiler
* Use distinct send and receive buffers
* use mpicxx in preference to mpiCC for Mac OS X
* Use mpicxx instead of mpiCC in examples/basic makefile
* Use relative links
* Use rm instead of /bin/rm for consistency with the other packages
* Use rm instead of /bin/rm in MPE makefiles
* use the PVFS readstrided function so we can use listio
* use the proper type for oldtype_extent
* use the real function name, since myname isn't always avaliable
* Used env perl instead of a hard-coded perl path
* Used environmental variables in enabling features of slog2sdk.
* we now have a PVFS2_SUPER_MAGIC
* winmpd: added -localroot option to mpirun to spawn the root process w/o mpd. This allows the root process to create visible windows and it allows the debugger to be attached to the root process
* winmpd: added -nocache option to mpiregister
* winmpd: added -nodetails flag to mpiregister
* winmpd: added TCP_NODELAY to easy_sockets
* winmpd: added check for WSAEINTT and WSAEINPROGRESS as non-fatal errors
* winmpd: added check for error return value from ContextWriteString
* winmpd: added checking of wall clock time to verify that functions don't timeout too early
* winmpd: added error checking
* winmpd: added host name to error output
* winmpd: added new microsoft project files
* winmpd: added smart caching to mpdupdate
* winmpd: added smpd
* winmpd: added the ability to specify the root host and the output redirection host names as ip addresses instead of host names
* winmpd: added timeout to stat command to prevent hangs
* winmpd: added try/catch to prevent crashing while traversing a bogus list
* winmpd: better tmp file cleanup for -localonly option
* winmpd: changed function call so it can still run on WindowsNT
* winmpd: fixed -priority option
* winmpd: fixed buffer overrun
* winmpd: fixed errors caught by Insure
* winmpd: fixed errors detected by insure++
* winmpd: fixed launch counter by atomically incrementing it
* winmpd: fixed multiply defined symbols error
* winmpd: fixed redirectio port and ip string lookup
* winmpd: made IsHostLocal() more robust and misc updates
* winmpd: made the thread creation code more reliable
* winmpd: more robust io thread creation
* winmpd: mpirun now can launch both mpich1 and mpich2 applications with the addition of two new flags -mpich1 and -mpich2. The default is mpich1
* winmpd: mpirun now uses the mpd database to determine if mpich2 applications have made it to mpi_finalize
* winmpd: prints error code along with string
* winmpd: remove cached password when 'mpiregister -remove' issued
* winmpd: remove current user registry entry when uninstalling mpd
* winmpd: reordered launch result message to avoid a potential timeout
* winmpd: replaced stricmp with getopt
* winmpd: reset variables after freeing to prevent use after release
* winmpd: robust io thread creation
* winmpd: the update tool now puts the new mpich dlls in the system32 directory if it cannot find the old dlls in the path
* winmpd: update now always gets the files from the ftp site ignoring locally cached copies



Up: Contents Next: Deprecated Features Previous: Mpirun Usage