Files
src_old/doc/H5X_File_Format.txt
T

171 lines
4.7 KiB
Plaintext

Proposal for a more general file format used by H5Part/H5Block
Authors: Achim Gsell
Status: Draft
1. Introduction
H5X is the file format used by H5Part and H5Block. H5X is designed on
top of HDF5: A H5X file is a very special HDF5 file.
+--------------------+
| Application |
+--------------------+
| H5Part/H5Block API |
+--------------------+
| H5X API |
+--------------------+
| HDF5 API |
+--------------------+
| Filesystem of OS |
+--------------------+
2. H5X file format specification
2.1 The root group
2.1.1 File attributes
Attributes of the HDF5 root group are called "file attributes". There
are no restrictions on file attribute names. Names with "__" as prefix and
postfix are intended for internal use. Supported values for file attributes
are scalars and arrays of 64bit integer and 64bit floating point number as
well as string values. Arrays of strings are not supported.
2.1.1.1 Predefined file attributes
__stepname__ (optional)
See below.
__stepnumwidth__ (optional)
See below.
2.1.2 Members of the root group
Members of the root group are special formated HDF5 groups called
"step groups". The name of a step group is "<name>#<number>".
Whereby <name> is either defined by the optional file attribute
__stepname__ or set to "Step" by default. <number> is an
unsigned integer less than 2^63.
/
+- Step#0
+- Step#1
...
+- Step#n
+- StepProperties [NEW]
+- Mesh [NEW]
The optional file attribute __stepnumwidth__ defines the minimum width
of <number> in the name of the step group. The number is padded with
'0'. The minimum width defaults to zero, so that no paddings occurs.
Example: With __stepnumwidth__ := 5 the name of the step group with
number 42 is "Step#00042", with __stepnumwidth__ := 0 the name is
"Step#42".
Step groups may be unsorted in the HDF5 file und needn't be number
consecutively.
2.1.3 Other members of the root group
Other sub-groups of the root group are allowed, but the group names are restricted
to names not matching "Step#.*".
2.2 Format of a step group
2.2.1 Step attributes
Step attributes are HDF5 attributes asigned to a step group. There
are no restrictions on step attribute names. As with file attributes names
enclosed in "__" are intended for internal use. Supported values are the same
as for file attributes:
* 64bit integer value
* arrays of 64bit integer values
* 64bit floating point number
* array of 64bit floating point number
* strings
Note that array of strings are not supported!
2.2.2 Predefined step attributes
No predefined step attributes exists.
2.2.3 Members of a step group
Members of a step group are HDF5 datasets and optional one HDF5 group
with the name "Block". HDF5 datasets in a step group are called "step
datasets". The optional block group is called "block data".
Mesh?!
2.2.4 Step datasets
Step datasets are arrays of rank 1. The dataset size must be the same
for all datasets inside a step. The size of the dataset may vary
from step to step. Array values are 64bit integer or 64bit floating
point numbers. No limitations (other than given by HDF5) are given on
step dataset names.
2.2.5 Block Data
The block data group is a container for an arbitrary number of field
data. A field is a data structure to store arrays of rank m with
n-dimensional vector values. Fields are represented within HDF5
groups. The HDF5 group name is the unique identifier for a field in
the current step.
2.2.5.1 m-rank fields with n-dimensional-vector values
Values are stored per dimension in separate datasets. Thus we have n
datasets for a field with n-dimensional vector values. Each dataset
corresponse to one dimension. The datasets are numbered from '0' to
'n-1'. Arrays are stored in column major order (Fortran indexing
scheme).
/ # HDF5 root group
...
+- Step#<i> # HDF5 group
|
+- Block # HDF5 group
|
+- <field name> # HDF5 group
|
+- 0 # HDF5 dataset, first dim of vector
+- 1 # HDF5 dataset, second dim of vector
...
+- n-1 # HDF5 dataset, last dim of vector
...
2.2.5.2 m-rank fields with scalar values
Fields with scalar values are stored as special case of fields with
n-dimensional vector values with n := 1.
/ # HDF5 root group
...
+- Step#<i> # HDF5 group
|
+- Block # HDF5 group
|
+- <field name> # HDF5 group
|
+- 0 # HDF5 dataset, scalar value
...
2.3 Step properties [NEW]
Why? Much faster than step attributes if you need to access consecutive values
Group of root group with name "StepProperties"
Members are datasets
Supported values are the same as for step attributes
Mapping of value to step via object index of step group