My apologies if I have asked this before, or if this info is available in the documentation; I have searched through the online documentation and cannot find the answer. The documentation for H5Pset_fapl_mpio says:
' H5Pset_fapl_mpio stores the user-supplied MPI IO parameters comm, for communicator, and info, for information, in the file access property list fapl_id. That property list can then be used to create and/or open a file. H5Pset_fapl_mpio is available only in the parallel HDF5 library and is not a collective function. comm is the MPI communicator to be used for file open, as defined in MPI_FILE_OPEN of MPI-2. This function makes a duplicate of the communicator, so modifications to comm after this function call returns have no effect on the file access property list. ... If the file access property list already contains previously-set communicator and Info values, those values will be replaced and the old communicator and Info object will be freed.' Does this mean that: 1) HDF5 now has a copy of the communicator passed to H5Pset_fapl_mpio so that I am free to use the communicator I passed in for computations without worrying about communication collisions? 2) If 'H5Pset_fapl_mpio...is not a collective function,' does that mean it should only be called by one MPI rank? 3) If I want to open and work with multiple files simultaneously, using the same or different communicators, then the only way to do this is to create one file access property list per file using H5Pset_fapl_mpio, right? One does this instead of reusing the same file access property list (if the communicator is the same) or placing another call to H5Pset_fapl_mpio and passing in the the old property list, but the new communicator (if the communicator is different), correct? (If I am reading the documentation correctly then opening multiple files with the same access property list means that they will all use the same communicator and messages may collide, alternatively if one uses a new file-access-property list then each open file has a unique communicator/communication context.) Dimension scales: 4) Also, are there Fortran bindings for the high-level dimension scales API? If not, one should be able to follow the specification document to implement dimension scales using lower level HDF5 API calls, right? Sorry to have so many questions, i just want to make sure that I am understanding the semantics correctly. Izaak Beekman =================================== (301)244-9367 Princeton University Doctoral Candidate Mechanical and Aerospace Engineering [email protected] UMD-CP Visiting Graduate Student Aerospace Engineering [email protected] [email protected]
_______________________________________________ Hdf-forum is for HDF software users discussion. [email protected] http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
