this doc describes the latest version of the data file format, which is numbered "5"

All data files consist of a single snapshot, and they can be used for restart and/or further conversion to other data formats directly usable for visualization. Note that restart is possible on a differing number of CPUs, and may suddenly allow more refinement levels. Also, note that the individual snapshots will typically have different lengths, as the number of grid blocks will vary dynamically. The data is saved in binary format (double precision). You can find the exact implementation in src/mod_input_output.t, more specifically in snapshot_write_header() and snapshot_read_header().

A snapshot (.dat file) contains a header, mesh tree information, and block data, in the following order:

Header

integer           :: Version number
integer           :: Byte offset where tree information starts
integer           :: Byte offset where block data starts
integer           :: nw
integer           :: ndir
integer           :: ndim
integer           :: levmax
integer           :: nleafs
integer           :: nparents
integer           :: it
double precision  :: global_time
double precision  :: xprobmin(ndim)
double precision  :: xprobmax(ndim)
integer           :: domain_nx(ndim)
integer           :: block_nx(ndim)
logical           :: periodic(ndim)
character(len=16) :: geometry
logical           :: staggered
character(len=16) :: w_names(nw)
character(len=16) :: physics_type
! The physics parameters, such as gamma
integer           :: n_params
double precision  :: parameters(n_params)
character(len=16) :: parameter_names(n_params)
! Indexes for file output (for restarting)
integer           :: snapshotnext
integer           :: slicenext
integer           :: collapsenext

Tree information

logical :: leaf(nleafs+nparents)
integer :: refinement_level(nleafs)
integer :: spatial_index(ndim, nleafs)
integer(kind=MPI_OFFSET_KIND) :: offset_block(nleafs)

Block 1 to nleafs

integer :: n_ghost_lo(ndim) ! number of ghost cells on lower boundaries
integer :: n_ghost_hi(ndim) ! number of ghost cells on upper boundaries
! block_shape = 1-n_ghost_lo:block_nx+n_ghost_hi
double precision :: w(block_shape, nw)

Version history

Version 1

Version 1 contained the following information

1. block data (nw variables)
2. leaf/parent logical array
3. header:
    nx^D
    domain_nx^D
    xprobmin^D
    xprobmax^D
    nleafs
    levmax
    ndim
    ndir
    nw
    it
    global_time

The idea is that you can reconstruct the full grid when you know the Morton order used for the leaf/parent logical array.

Version 2

Version 2 had the same information as version 1, but changes were made to the Morton order on the coarse grid, causing incompatibility.

Version 5 (current)

Version 5 introduced the geometry parameter (e.g. "polar_2D", "cartesian_1.75D"... ), as well as a periodic and a staggered flags. The periodic flag a ndir-long 1D boolean array. For each direction, it defaults to .false. but is set to .true. if at least one quantity is using a periodic boundary condition in the corresponding direction. This change was motivated by improving the (upcoming) compatibility with yt. The staggered flag is added to support staggered grid for constrained transport MHD.