the ompi_message_t structure to properly initialize convertor (the peer
is available in the request in OB1, and wasn't needed when I did the
original implementation).
* Implement matched probe for the Portals4 MTL and add NULL function pointers
for the other MTLs.
* Add add_comm and del_comm functions to portals4 MTL so that direct call
almost works again.
* Add NEWS item that we've implemented matched probe
This commit was SVN r26180.
Note that the previous patch allowed the following test to -pass-:
ompi-tests/mpich_tester/mpich_pt2pt/truncmult.c
This patch makes that test -fail- due to the assumption that MPI_Wait will update the status.MPI_ERROR field. In Open MPI we do not do this, so the MPI_ERROR field being inspected will remain set to MPI_ERR_PENDING. See comments in req_wait.c for why we do this.
If we change the test to not inspect the MPI_ERROR field after calling MPI_Wait successfully, then the test would pass correctly with this patch.
This change was made per discussion on the below email thread:
http://www.open-mpi.org/community/lists/devel/2012/03/10753.php
This commit was SVN r26177.
The following SVN revision numbers were found above:
r26172 --> open-mpi/ompi@03a33417d5
Worked with Oracle to verify that hwloc PCI detection is correctly
disabled on the Suse 10/64 bit platform and is enabled by default on
all other platforms. The --[en|dis]able-hwloc-pci switch is also
available for manual override of the configure decision about hwloc
PCI support.
This commit was SVN r26175.
* split eq into send and receive eqs so that we can control the number
of outstanding events in send eq and ensure we never lose an ack
* Shouldn't ever truncate on short unexpected receive bocks, so don't set
the truncate bit
* Track active vs. waiting for free short unexpected receive blocks so
to ensure an active short unexpected receive block is posted coming out
of flow control. Also allow creation of "temporary" blocks which should
be released once FREE event is received.
* Slight reorganization of some code in preparation for more flow control
work.
This commit was SVN r26174.
This commit was SVN r26165.
The following SVN revisions from the original message are invalid or
inconsistent and therefore were not cross-referenced:
r4402
- configure:
- changed default CUPTI library path to $CUPTI-DIR/lib64
- VT Libs:
- corrected prototype of MPI_Get_address in Fortran MPI wrappers (the second parameter should be an MPI_Aint* instead of MPI_Fint*)
- temporary removed MPI_<Comm|Type|Win>_<get|set>_attr and MPI_Attr_<get|put> from the Fortran MPI wrappers due to missing conversion of the attribute value parameter
- Docu:
- latex doc \usepackage[T1]{fontenc} so that _ can be searched and copied
- smaller font in Environment Variables section
- some improvements in CUDA section
- removed GPU idle time as official feature for CUPTI tracing method
This commit was SVN r26161.
- general:
- added missing entry in ChangeLog
- vtunify[-mpi]:
- fixed possibly uninitialized global token for the predefined Node and "All" process groups
This commit was SVN r26147.
- general:
- corrected OTF version number
- otfprofile:
- removed leading '=' from CSV lines to make it loadable into spreadsheets (e.g. Open Office)
- fixed process naming in CSV output of collective operation statistics
Changes to VT:
- configure:
- added *_FOR_BUILD variables to CrayXE's default configure options; required for cross-building
- VT libs:
- fixed GPU communication, due to new process ID splitting
- fixed parsing of PAPI native events in VT_METRICS; use strtok_r instead of strtok which is successively called inPAPI_event_name_to_code
- added VT_METRICS_SEP to definition comments (-> Vampir's trace info)
- Docu:
- fixed link to TAU Reference Guide
This commit was SVN r26137.
This feature can be enabled at compile time with --with-cma passed
to configure.
At runtime it is also necessary to add "--mca btl btl_sm_use_cma 1"
to the mpirun command.
If both CMA and KNEM are compiled in and enabled at runtime then
KNEM will take precedence and CMA will disable itself
This commit was SVN r26134.
Note: I am working on a system that buffered all output until the application fishished due to a default of 'fully buffered.' This makes debugging painful. This switch fixed the problem by allowing me to adjust the buffering.
This commit was SVN r26119.
Changes to OTF:
- general:
- updated copyright information (2011->2012)
- otfmerge-mpi:
- use the MPI-2 versions of MPI_Address and MPI_Type_struct
- otfdump:
- don't abort when reading events fails - the input tracefile might only have statistics
Changes to VT:
- general:
- updated version number to 5.12.2openmpi
- updated copyright information (2011->2012)
- configure:
- added configure switches to enable/disable CUPTI and CUDA wrapping
- fixed detection of C++ runtime libraries for Cray and PGI v11.x compilers
- fixed detection of Cray compiler's OpenMP flag
- fixed detection of MPI_IN_PLACE
- disable support for RTLD_DEFAULT on CrayX? platforms; it's provided by dlfcn.h but not working
- added '-force_flat_namespace' to linker flags of compiler wrappers on MacOS (causes that OpenMPI's libmpi_f77 calls the VT MPI wrapper functions - not the original ones)
- default configure options on Cray platforms: use compiler option '--target=$XTPE_COMPILE_TARGET' only if the environment variable is set
- VT libs:
- added support for CUDA tracing via CUPTI callbacks and activities (runtime and driver API, kernels, memory copies, GPU idle time and GPU memory usage)
- added support for cudaMemcpyDefault and synchronous peer-to-peer memory copies in CUDA library wrapper
- fixed a bug in CUDA runtime wrapper initialization and thread creation
- fixed a build bug occurred if CUDA and CUPTI found, but support for library tracing is disabled
- use stack-allocated char-array when composing vtunify command; on some platforms system() results in exit code 127 when using a dynamically allocated char-array
- fixed bug in async. counter plugin
- fixed handling of empty MPI groups (MPI_GROUP_EMPTY)
- fixed handling of MPI groups implicitly generated by MPI_Win_create
- fixed conversion from MPI_Fint-arrays to MPI_Aint-arrays in Fortran MPI wrappers
- fixed order of OpenMP threads based on its id (omp_get_thread_num)
- fixed parsing of filter file to consider non-rank-specific filter rules appearing after a rank selection for disabling
- fixed handling of 'errno' in LIBC[-I/O] wrappers for statically linked applications (set application's errno to the errno defined in the external LIBC which is used for calling the real functions)
- suppress warnings about usage of deprecated MPI functions (OMPI_WANT_MPI_INTERFACE_WARNING=0)
- vtunify[-mpi]:
- fixed potential memory corruption during enqueuing recv. messages for p2p message matching
- vtunify-mpi:
- use the MPI-2 versions of MPI_Address and MPI_Type_struct
- removed unused MPI wrappers
- fixed assertion in p2p message matching which occurred when processing local traces with disabled ranks
- vtdyn:
- load user-specified shared libraries (-s SHLIB) into the mutatee before starting the instrumentation; adds support for instrumenting shared libraries which are loaded during runtime
- compiler wrappers:
- fixed detection of MPI library linked in the path form (e.g. libmpi.a instead of -lmpi)
- fixed corrupt library order when using vtnvcc for linking MPI/CUDA mixed program
- OPARI:
- fixed Fortran parsing for detecting end of block DO loops
This commit was SVN r26114.
