GIO is an I/O library designed for supporting accessing a shared file by multiple processes in parallel. GIO follows the I/O semantics of MPI standard, but provides only a subset of its features and a slightly different API syntax.

• Similar to MPI-IO, GIO provides collective and independent read and write APIs. It also supports I/O hints through setting the MPI info objects, and follows the MPI-IO data consistency semantics.
• Unlike MPI-IO, GIO does not support features such as, fileview, non-blocking APIs, asynchronous APIs, split APIs, or shared-file pointers.

• The latest version 1.0.0 was released on June 27, 2026.
• All previous release tarball files are available from the download page

GIO is built on top of MPI and thus requires an MPI C compiler. Shared-file access requires coordination of processes in order to achieve consistency results and a good performance. In particular, GIO adopts the two-phase I/O strategy to implement its collective read and write APIs.

The names of all GIO public APIs contain the same prefix, "GIO_". For example, the two APIs below are for opening and closing a shared file, respectively

int GIO_open(MPI_Comm    comm,      /* MPI communicator */
             const char *filename,  /* name of the file to be opened/created */
             int         amode,     /* file access mode */
             MPI_Info    info,      /* I/O hints */
             GIO_File   *fh);       /* file handler returend to the caller */

int GIO_close(GIO_File *fh);

There are 4 APIs for performing file reads and writes. Similar to MPI-IO, the ones with suffix name '_all' are collective and without independent. The collective APIs require all processes in the MPI communicator to participate the call. All APIs for reading from and writing to the shared file contain arguments describing the layouts for both file accesses and buffer in memory, in the form of a list of offset-length pairs. For example,

MPI_Offset GIO_write_all(GIO_File          fh,
                         const void       *buf,
                         MPI_Offset        file_npairs,
                         const MPI_Offset *file_offs,
                         const MPI_Offset *file_lens,
                         MPI_Offset        buf_npairs,
                         const MPI_Offset *buf_offs,
                         const MPI_Offset *buf_lens);

• File access layout is described by arguments file_npairs, file_offs, and file_lens.
  • file_npairs is the number of offset-length pairs,
  • file_offs is an array containing file_npairs number of non-contiguous file offsets, and
  • file_lens is an array containing file_npairs number of non-contiguous request lengths, each corresponding to the element in file_offs.
• Buffer layout is described by arguments buf_npairs, buf_offs, and buf_lens.
  • buf_npairs is the number of offset-length pairs,
  • buf_offs is an array containing buf_npairs number of non-contiguous offsets relative to the memory address pointed by argument buf, and
  • buf_lens is an array containing buf_npairs number of non-contiguous request lengths, each corresponding to the element in buf_offs.

• C API references
• A complete list of I/O hints supported by GIO
• GIO utility programs

• PnetCDF, a high-level parallel I/O library for accessing Unidata's NetCDF, files in classic formats.

Copyright (C) 2026, Northwestern University.

See COPYRIGHT notice in top-level directory.