Contents:

 


Remove Inaccessible Objects and Unused Space in a File

HDF5 files may accumulate unused space when they are read and rewritten to or if objects are deleted within them. With many edits and deletions this unused space can add up to a sizable amount.

The h5repack tool can be used to remove unused space in an HDF5 file. If no options other than the input and output HDF5 files are specified on the h5repack command line, it will write the file to the new file, getting rid of the unused space:

h5repack <input file> <output file>


Change a Dataset's Storage Layout

The h5repack utility can be used to change a dataset's storage layout. By default, the storage layout of a dataset is defined at creation time and it cannot be changed. However, with h5repack you can write an HDF5 file to a new file and change the layout for objects in the new file.

The -l option in h5repack is used to change the layout for an object. The string following the -l option defines the layout type and parameters for specified objects (or all objects):

h5repack -l [list of objects:]<layout type>=<layout parameters> <input file> <output file>

 

If no object is specified, then everything in the input file will be written to the output file with the specified layout type and parameters. If objects are specified then everything in the input file will be written to the output file as is, except for those specified objects. They will be written to the output file with the given layout type and parameters.

Following is a description of the dataset layouts and the h5repack options to use to change a dataset:

Storage Layouth5repack OptionDescription
Contiguous
CONTI
Data is stored physically together
Chunked
CHUNK=DIM[xDIM...xDIM]
Data is stored in DIM[xDIM...xDIM] chunks
Compact
COMPA
Data is stored in the header of the object (less I/O)

If you type h5repack -h on the command line, you will see a detailed usage statement with examples of modifying the layout.

In the following example, the dataset  /dset  in the file dset.h5 is contiguous, as shown by the h5dump -pH command. The h5repack utility writes dset.h5 to a new file, dsetrpk.h5, where the dataset  dset  is chunked. This can be seen by examining the resulting dsetrpk.h5 file with h5dump, as shown:

$ h5dump -pH dset.h5
HDF5 "dset.h5" {
GROUP "/" {
   DATASET "dset" {
      DATATYPE  H5T_STD_I32BE
      DATASPACE  SIMPLE { ( 4, 6 ) / ( 4, 6 ) }
      STORAGE_LAYOUT {
         CONTIGUOUS
         SIZE 96
         OFFSET 1400
      }
      FILTERS {
         NONE
      }
      FILLVALUE {
         FILL_TIME H5D_FILL_TIME_IFSET
         VALUE  0
      }
      ALLOCATION_TIME {
         H5D_ALLOC_TIME_LATE
      }
   }
}
}

$ h5repack -l dset:CHUNK=4x6 dset.h5 dsetrpk.h5

$ h5dump -pH dsetrpk.h5
HDF5 "dsetrpk.h5" {
GROUP "/" {
   DATASET "dset" {
      DATATYPE  H5T_STD_I32BE
      DATASPACE  SIMPLE { ( 4, 6 ) / ( 4, 6 ) }
      STORAGE_LAYOUT {
         CHUNKED ( 4, 6 )
         SIZE 96
      }
      FILTERS {
         NONE
      }
      FILLVALUE {
         FILL_TIME H5D_FILL_TIME_IFSET
         VALUE  0
      }
      ALLOCATION_TIME {
         H5D_ALLOC_TIME_INCR
      }
   }
}
}

 

There can be many reasons that the storage layout needs to be changed for a dataset. For example, there may be a performance issue with a dataset due to a small chunk size.


Apply Compression Filter to a Dataset

The h5repack utility can be used to compress or remove compression from a dataset in a file. By default, compression cannot be added to or removed from a dataset once it has been created. However, with h5repack you can write a file to a new file and specify a compression filter to apply to a dataset or datasets in the new file.

To apply a filter to an object in an HDF5 file, specify the -f option, where the string following the -f option defines the filter and its parameters (if there are any) to apply to a given object or objects:

h5repack -f [list of objects:]<name of filter>=<filter parameters> <input file> <output file>

 

If no objects are specified then everything in the input file will be written to the output file with the filter and parameters specified. If objects are specified, then everything in the input file will be written to the output file as is, except for the specified objects. They will be written to the output file with the filter and parameters specified.

If you type h5repack --help on the command line, you will see a detailed usage statement with examples of modifying a filter. There are actually numerous filters that you can apply to a dataset:

FilterOptions
GZIP compression (levels 1-9)
GZIP=<deflation level>
SZIP compression
SZIP=<pixels per block,coding>
Shuffle filter
SHUF
Checksum filter
FLET
NBIT compression
NBIT
HDF5 Scale/Offset filter
SOFF=<scale_factor,scale_type>
User defined filter
UD=<filter_number,cd_value_count,value_1[,value_2,...,value_N]>
Remove ALL filters
NONE

 

Be aware that a dataset must be chunked to apply compression to it. If the dataset is not already chunked, then h5repack will apply chunking to it. Both chunking and compression cannot be applied to a dataset at the same time with h5repack.

In the following example,

 

Example:

$ h5dump -pH dset.h5
HDF5 "dset.h5" {
GROUP "/" {
   DATASET "dset" {
      DATATYPE  H5T_STD_I32BE
      DATASPACE  SIMPLE { ( 12, 18 ) / ( 12, 18 ) }
      STORAGE_LAYOUT {
         CONTIGUOUS
         SIZE 864
         OFFSET 1400
      }
      FILTERS {
         NONE
      }
      FILLVALUE {
         FILL_TIME H5D_FILL_TIME_IFSET
         VALUE  0
      }
      ALLOCATION_TIME {
         H5D_ALLOC_TIME_LATE
      }
   }
}
}

$ h5repack -f dset:GZIP=5 dset.h5 dsetrpk.h5

$ h5dump -pH dsetrpk.h5
HDF5 "dsetrpk.h5" {
GROUP "/" {
   DATASET "dset" {
      DATATYPE  H5T_STD_I32BE
      DATASPACE  SIMPLE { ( 12, 18 ) / ( 12, 18 ) }
      STORAGE_LAYOUT {
         CHUNKED ( 12, 18 )
         SIZE 160 (5.400:1 COMPRESSION)
      }
      FILTERS {
         COMPRESSION DEFLATE { LEVEL 5 }
      }
      FILLVALUE {
         FILL_TIME H5D_FILL_TIME_IFSET
         VALUE  0
      }
      ALLOCATION_TIME {
         H5D_ALLOC_TIME_INCR
      }
   }
}
}

 


Copy Objects to Another File

The h5copy utility can be used to copy an object or objects from one HDF5 file to another or to a different location in the same file. It uses the  H5O_COPY  and  H5L_COPY  APIs in HDF5.

Following are some of the options that can be used with h5copy.

h5copy OptionsDescription
 -i, --input 
Input file name
 -o, --output 
Output file name
 -s, --source 
Source object name
 -d, --destination 
Destination object name
 -p, --parents 
Make parent groups as needed
 -v, --verbose 
Verbose mode
 -f, --flag 
Flag type

 

For a complete list of options and information on using h5copy, type:

h5copy -help

 

In the example below, the dataset  /MyGroup/Group_A/dset2  in groups.h5 gets copied to the root ("/") group of a new file, newgroup.h5, with the name  dset3 :

$h5dump -H groups.h5
HDF5 "groups.h5" {
GROUP "/" {
   GROUP "MyGroup" {
      GROUP "Group_A" {
         DATASET "dset2" {
            DATATYPE  H5T_STD_I32BE
            DATASPACE  SIMPLE { ( 2, 10 ) / ( 2, 10 ) }
         }
      }
      GROUP "Group_B" {
      }
      DATASET "dset1" {
         DATATYPE  H5T_STD_I32BE
         DATASPACE  SIMPLE { ( 3, 3 ) / ( 3, 3 ) }
      }
   }
}
}

$ h5copy -i groups.h5 -o newgroup.h5 -s /MyGroup/Group_A/dset2 -d /dset3

$ h5dump -H newgroup.h5
HDF5 "newgroup.h5" {
GROUP "/" {
   DATASET "dset3" {
      DATATYPE  H5T_STD_I32BE
      DATASPACE  SIMPLE { ( 2, 10 ) / ( 2, 10 ) }
   }
}
}

 

There are also h5copy flags that can be specified with the -f option. In the example below, the -f shallow option specifies to copy only the immediate members of the group  /MyGroup  from the groups.h5 file mentioned above to a new file mygrouponly.h5:

h5copy -v -i groups.h5 -o mygrouponly.h5 -s /MyGroup -d /MyGroup -f shallow

 

The output of the above command is shown below. The verbose option -v describes the action that was taken, as shown in the highlighted text.

Copying file <groups.h5> and object </MyGroup> to file <mygrouponly.h5> and object </MyGroup>
Using shallow flag

$ h5dump -H mygrouponly.h5
HDF5 "mygrouponly.h5" {
GROUP "/" {
   GROUP "MyGroup" {
      GROUP "Group_A" {
      }
      GROUP "Group_B" {
      }
      DATASET "dset1" {
         DATATYPE  H5T_STD_I32BE
         DATASPACE  SIMPLE { ( 3, 3 ) / ( 3, 3 ) }
      }
   }
}
}

 


Add or Remove User Block from File

The user block is a space in an HDF5 file that is not interpreted by the HDF5 library. It is a property list that can be added when creating a file. See the H5P_SET_USERBLOCK API in the HDF5 Reference Manual for more information regarding this property.

Once created in a file, the user block cannot be removed. However, you can use the h5jam and h5unjam utilities to add or remove a user block from a file into a new file.

These two utilities work similarly, except that h5jam adds a user block to a file and h5unjam removes the user block. You can also overwrite or delete a user block in a file.

Specify the -h option to see a complete list of options that can be used with h5jam and h5unjam. For example:

   h5jam -h
   h5unjam -h

 

Below are the basic options for adding or removing a user block with h5jam and h5unjam:

OptionsDescription
-i
Input File
-o
Output File
-u
File to add or remove from user block

 

Let's say you wanted to add the program that creates an HDF5 file to its user block. As an example, you can take the h5_crtgrpar.c program from the HDF5 Introductory Tutorial Examples and add it to the file it creates, groups.h5. This can be done with h5jam, as follows:

h5jam -i groups.h5 -u h5_crtgrpar.c -o groupsub.h5

 

You can actually view the file with more groupsub.h5 to see that the h5_crtgrpar.c file is indeed included.

To remove the user block that was just added, type:

h5unjam -i groupsub.h5 -u h5_crtgrparNEW.c -o groups-noub.h5

 

This writes the user block in the file groupsub.h5 into h5_crtgrparNEW.c. The new HDF5 file, groups-noub.h5, will not contain a user block.