mpi_io(3)

NAME
     mpi_io - Introduction to the ROMIO implementation of MPI I/O routines

DESCRIPTION
     MPI I/O routines are based on MPI-2 MPI I/O routines, in which derived
     data types are used to express data partitioning.  These MPI-2 I/O
     routines are derived from the ROMIO 1.0.1 source code. The current IRIX
     implementation contains all interfaces defined in the MPI-2 I/O chapter
     of the MPI-2 standard except shared file pointer functions (Sec. 9.4.4),
     split collective data access functions (Sec. 9.4.5), support for file
     interoperability (Sec. 9.5), I/O error handling (Sec. 9.7), and I/O error
     classes (Sec. 9.8). Because shared file pointer functions are not
     supported, the MPI_MODE_SEQUENTIAL file access mode to MPI_File_open is
     also not supported.

   Limitations
     The MPI I/O routines have the following limitations:

     *   Beginning with MPT 1.6, the status argument is set following all
         read, write, MPIO_Test, and MPIO_Wait functions. Consequently,
         MPI_Get_count and MPI_Get_elements will now work when passed the
         status object from these operations.  Previously, they did not.

     *   All nonblocking I/O functions use a MPIO_Request object instead of
         the usual MPI_Request object. Accordingly, two functions, MPIO_Test
         and MPIO_Wait are provided to wait and test on these MPIO_Request
         objects. They have the same semantics as MPI_Test and MPI_Wait, as
         shown in the following synopsis:

         int MPIO_Test(MPIO_Request *request, int *flag,
         MPI_Status *status);

         int MPIO_Wait(MPIO_Request *request, MPI_Status *status);


         The usual functions, MPI_Test, MPI_Wait, MPI_Testany, and so on, do
         not work for nonblocking I/O.

     *   All functions return only two possible error codes:  MPI_SUCCESS on
         success and MPI_ERR_UNKNOWN on failure.

     *   End-of-file is not detected.  The individual file pointer is
         increased by the requested amount of data and not by the actual
         amount of data read.  Therefore, after end-of-file is reached,
         MPI_File_get_position(3) returns a wrong offset.

   Direct I/O
     MPI I/O supports direct access to files stored in XFS filesystems. Direct
     access bypasses the system's buffer cache, leading to better performance
     in some specialized cases.  You can enable direct I/O for read and write
     operations by setting the appropriate environment variable(s),
     MPIO_DIRECT_READ or MPIO_DIRECT_WRITE, to the string "TRUE," as in the
     following example:

          setenv MPIO_DIRECT_READ TRUE
          setenv MPIO_DIRECT_WRITE TRUE


   List of Routines
     The MPI I/O routines are as follows:

          MPI_File_c2f(3)
          MPI_File_close(3)
          MPI_File_delete(3)
          MPI_File_f2c(3)
          MPI_File_get_amode(3)
          MPI_File_get_atomicity(3)
          MPI_File_get_byte_offset(3)
          MPI_File_get_group(3)
          MPI_File_get_info(3)
          MPI_File_get_position(3)
          MPI_File_get_size(3)
          MPI_File_get_type_extent(3)
          MPI_File_get_view(3)
          MPI_File_iread(3)
          MPI_File_iread_at(3)
          MPI_File_iwrite(3)
          MPI_File_iwrite_at(3)
          MPI_File_open(3)
          MPI_File_preallocate(3)
          MPI_File_read(3)
          MPI_File_read_all(3)
          MPI_File_read_at(3)
          MPI_File_read_at_all(3)
          MPI_File_seek(3)
          MPI_File_set_atomicity(3)
          MPI_File_set_info(3)
          MPI_File_set_size(3)
          MPI_File_set_view(3)
          MPI_File_sync(3)
          MPI_File_write(3)
          MPI_File_write_all(3)
          MPI_File_write_at(3)
          MPI_File_write_at_all(3)