MOTIOn

March 4, 2026 · View on GitHub

Modular HPC Optimized Toolkit for IO in Fortran

a pure Fortran 2003+ library providing a simple, agnostic API for parallel HDF5/XDMF I/O in MPI-based HPC applications.

💾 HDF5+XDMF output One call writes binary HDF5 data and XML metadata; MPI master/worker branching handled transparently	🧱 Three grid topologies Cartesian uniform, Cartesian, and Curvilinear — declare a block type and MOTIOn handles the geometry	🔄 Kind-generic field I/O `save_block_field`/`load_block_field` overloaded for all Fortran kinds (R8P…I1P) and shapes (3-D field, vector, tensor, matrix)	⚙️ Low-level control Access `hdf5_file_object` and `xdmf_file_object` directly for custom dataspaces, attributes, and XML topologies
🏗️ Three-layer OOP `file_base_object` → format objects → `xh5f_file_object`; extend any layer for custom formats	🖥️ MPI-parallel Designed for distributed CFD solvers; each rank writes its own HDF5 file while the XDMF master collects metadata	📦 Pre-built HDF5 GNU 14.2+ and nvfortran 25.3 HDF5 1.14.6 libraries included — no separate HDF5 build required	🔤 Pure Fortran 2003+ Standard-compliant, no external dependencies beyond MPI and HDF5

Documentation

For full documentation (guide, API reference, examples, etc...) see the MOTIOn website.

Authors

Stefano Zaghi — @szaghi

Contributions are welcome — see the Contributing page.

Copyrights

This project is distributed under a multi-licensing system:

FOSS projects: GPL v3
Closed source / commercial: BSD 2-Clause, BSD 3-Clause, or MIT

Anyone interested in using, developing, or contributing to this project is welcome — pick the license that best fits your needs.

Quick start

Write a parallel HDF5/XDMF output from a collection of Cartesian uniform blocks:

use motion
implicit none
type(xh5f_file_object) :: xh5f

call xh5f%open_file(filename_hdf5='out_rank'//trim(strz(myrank,2))//'.h5', &
                    filename_xdmf='out_procs'//trim(strz(nprocs,2))//'.xdmf')
call xh5f%open_grid(grid_name='blocks', grid_type=XDMF_PARAMETERS%XDMF_GRID_TYPE_COLLECTION_ASYNC)
do b = 1, nblocks
  call xh5f%open_block(block_type=XH5F_PARAMETERS%XH5F_BLOCK_CARTESIAN_UNIFORM, &
                       block_name='block_'//trim(strz(b,2)),                    &
                       nijk=nijk, emin=emin(:,b), dxyz=dxyz, time=time)
  call xh5f%save_block_field(xdmf_field_name='pressure',                              &
                             nijk=nijk,                                               &
                             field=pressure(1:nijk(1),1:nijk(2),1:nijk(3),b),         &
                             field_center=XDMF_PARAMETERS%XDMF_ATTR_CENTER_CELL,      &
                             field_format=XDMF_PARAMETERS%XDMF_DATAITEM_NUMBER_FORMAT_HDF)
  call xh5f%close_block
enddo
call xh5f%close_grid(grid_type=XDMF_PARAMETERS%XDMF_GRID_TYPE_COLLECTION_ASYNC)
call xh5f%close_file

See the full example in src/tests/motion_write_xdmf_file_test.F90.

Install

FoBiS

Standalone — clone and fetch dependencies, then build:

git clone https://github.com/szaghi/MOTIOn && cd MOTIOn
FoBiS.py fetch
FoBiS.py build -mode tests-gnu && ./scripts/run_tests.sh --np 2

As a project dependency — declare MOTIOn in your fobos and run fetch:

[dependencies]
deps_dir = src/third_party
MOTIOn = https://github.com/szaghi/MOTIOn

FoBiS.py fetch           # fetch and build
FoBiS.py fetch --update  # re-fetch and rebuild

Mode	Command
Shared library	`FoBiS.py build -mode shared-gnu`
Static library	`FoBiS.py build -mode static-gnu`
Tests (debug)	`FoBiS.py build -mode tests-gnu-debug`
NVIDIA Fortran	`FoBiS.py build -mode tests-nvf`