Class Class Reference

Detailed Description

HDF5 data source using H5 reader utilities.

on HDF5-based FLD files.

This class implements a HDF5 reader/writer based on MPI/O that is designed to operate on a single file across all processors of a simulation. The definition follows vaguely similar lines to XML output but is stored somewhat differently to accommodate parallel reading and writing. At a basic level metadata is organised as follows:

• Nektar++ data lies in the root /NEKTAR group.
• The contents of a FieldDefinitions object is hashed to construct a unique identifier for each object, which becomes the name of a group within the root group. We then use the H5TagWriter to assign the field definitions to each group.
• In a similar fashion, we create a Metadata group to contain field metadata that is written.

We then define five data sets to contain field data:

• The DATA dataset contains the double-precision modal coefficient data.
• The IDS dataset contains the element IDs of the elements that are written out.
• The POLYORDERS dataset is written if the field data contains variable polynomial order, and contains the (possibly hetergeneous) mode orders in each direction for each of the elements.
• The HOMOGENEOUSZIDS dataset contains the IDs of z-planes for homogeneous simulations, if the data are homogeneous.
• The HOMOGENEOUSYIDS dataset contains the IDs of y-planes for homogeneous simulations, if the data are homogeneous.
• The HOMOGENEOUSSIDS dataset contains the strip IDs for homogeneous simulations, if the data are homogeneous and use strips.

The ordering is defined according to the DECOMPOSITION dataset. A ‘decomposition’ in this class is essentially a single field definition with its accompanying data. Data are written into each dataset by the order of each decomposition. Each decomposition contains the following seven integers that define it per field definition per processor:

• Number of elements in this field definition (index #ELEM_DCMP_IDX).
• Number of entries in the DATA array for this field definition (index #VAL_DCMP_IDX)
• Number of entries in the POLYORDERS array for this field definition (index #ORDER_DCMP_IDX)
• Number of entries in the HOMOGENEOUSZIDS array (index #HOMZ_DCMP_IDX).
• Number of entries in the HOMOGENEOUSYIDS array (index #HOMY_DCMP_IDX).
• Number of entries in the HOMOGENEOUSSIDS array (index #HOMS_DCMP_IDX).
• Hash of the field definition, represented as a 32-bit integer, which describes the name of the attribute that contains the rest of the field definition information (e.g. field names, basis type, etc).

The number of decompositions is therefore calculated as the field size divided by #MAX_DCMPS which allows us to calculate the offsets of the data for each field definition within the arrays.

XML data source using TinyXML.

on XML FLD files.

This class is the default for Nektar++ output. It reads and writes one XML file per processor that represents the underlying field data. For serial output, the format of an XML file obeys the following structure:

<NEKTAR>
    <Metadata>
        ...
    </Metadata>
    <ELEMENT FIELDS="..." ...> data1 </ELEMENT>
    <ELEMENT FIELDS="..." ...> data2 </ELEMENT>
    ...
</NEKTAR>

• Metadata is converted as key/value pairs in the <Metadata> tag.
• There are one or more ELEMENT blocks, whose attributes correspond with the FieldDefinitions class variables.
• Element data is stored as a base64-encoded zlib-compressed binary double-precision data using the functions from CompressData.

In parallel, each process writes its contributions into an XML file of the form P0000001.fld (where 1 is replaced by the rank of the process) inside a directory with the desired output name. These files only include the ELEMENT data. Metadata are instead stored in a separate Info.xml file, which contains the Metadata and additional tags of the form

<Partition FileName="P0000000.fld"> ID list </Partition>

The ID list enumerates all element IDs on each partition's contribution. For large parallel jobs, this is used to avoid each process reading in every single partition in order to load field data.