H5Pset_scaleoffset
(
hid_t plist_id
,
H5Z_SO_scale_type_t scale_type
,
int scale_factor
)
H5Pset_scaleoffset
sets the scale-offset filter,
H5Z_FILTER_SCALEOFFSET
, for a dataset.
Generally speaking, scale-offset compression performs a scale and/or offset operation on each data value and truncates the resulting value to a minimum number of bits (MinBits) before storing it. The current scale-offset filter supports integer and floating-point datatypes.
For an integer datatype, the parameter scale_type
should
be set to H5Z_SO_INT (2)
.
The parameter scale_factor
denotes MinBits.
If the user sets it to H5Z_SO_INT_MINBITS_DEFAULT (0)
,
the filter will calculate MinBits.
If scale_factor
is set to a positive integer, the filter
does not do any calculation and just uses the number as MinBits.
However, if the user gives
a MinBits that is less than what would be generated by the filter, the
compression will be lossy. Also, the MinBits supplied by the user cannot
exceed the number of bits to store one value of the dataset datatype.
For a floating-point datatype, the filter adopts the GRiB data packing
mechanism, which offers two alternate methods: E-scaling and D-scaling.
Both methods are lossy compression.
If the parameter scale_type
is set to
H5Z_SO_FLOAT_DSCALE (0)
, the filter will use the D-scaling
method; if it is set to H5Z_SO_FLOAT_ESCALE (1)
, the filter
will use the E-scaling method.
Since only the D-scaling method is implemented, scale_type
should be set to H5Z_SO_FLOAT_DSCALE
or 0.
When the D-scaling method is used, the original data is "D"
scaled — multiplied by 10 to the
power of scale_factor
, and the "significant" part of
the value is moved to the left of the decimal point. Care should
be taken in setting the decimal scale_factor
so that
the integer part will have enough precision to contain the appropriate
informationof the data value. For example, if scale_factor
is set to 2, the number 104.561 will be 10456.1 after "D" scaling.
The last digit 1 is not "significant" and is thrown off in the process
of rounding.
The user should make sure that after "D" scaling and rounding,
the data values are within the range that can be represented by the
integer (same size as the floating-point type).
Valid values for scale_type
are as follows:
H5Z_SO_FLOAT_DSCALE (0) |
Floating-point type, using variable MinBits method |
H5Z_SO_FLOAT_ESCALE (1) |
Floating-point type, using fixed MinBits method |
H5Z_SO_INT (2) |
Integer type |
The meaning of scale_factor
varies according to the value
assigned to scale_type
:
scale_type value |
scale_factor description |
H5Z_SO_FLOAT_DSCALE |
Denotes the decimal scale factor for D-scaling and can be positive, negative or zero. This is the current implementation of the library. |
H5Z_SO_FLOAT_ESCALE |
Denotes MinBits for E-scaling and must be a positive integer. This is not currently implemented by the library. |
H5Z_SO_INT |
Denotes MinBits and it should be a positive integer or
H5Z_SO_INT_MINBITS_DEFAULT (0).
If it is less than 0, the library will reset it to 0
since it is not implemented. |
Like other I/O filters supported by the HDF5 library, an application using the scale-offset filter must store data with chunked storage.
hid_t plist_id |
IN: Dataset creation property list identifier. |
H5Z_SO_scale_type_t scale_type |
IN: Flag indicating compression method. |
int scale_factor |
IN: Parameter related to scale. Must be non-negative. |
Release | Change |
1.8.0 | C function introduced in this release. |
1.8.8 | Fortran90 subroutine introduced in this release. |