Package org.apache.lucene.util.packed

Class PackedInts

java.lang.Object
org.apache.lucene.util.packed.PackedInts
public class PackedInts extends Object
Simplistic compression for array of unsigned long values. Each value is >= 0 and <= a specified maximum value. The values are stored as packed ints, with each value consuming a fixed number of bits.

  • Field Details

    • FASTEST

      public static final float FASTEST
      At most 700% memory overhead, always select a direct implementation.
      See Also:

    • FAST

      public static final float FAST
      At most 50% memory overhead, always select a reasonably fast implementation.
      See Also:

    • DEFAULT

      public static final float DEFAULT
      At most 20% memory overhead.
      See Also:

    • COMPACT

      public static final float COMPACT
      No memory overhead at all, but the returned implementation may be slow.
      See Also:

    • DEFAULT_BUFFER_SIZE

      public static final int DEFAULT_BUFFER_SIZE
      Default amount of memory to use for bulk operations.
      See Also:

    • CODEC_NAME

      public static final String CODEC_NAME
      See Also:

    • VERSION_START

      public static final int VERSION_START
      See Also:

    • VERSION_BYTE_ALIGNED

      public static final int VERSION_BYTE_ALIGNED
      See Also:

    • VERSION_CURRENT

      public static final int VERSION_CURRENT
      See Also:

  • Constructor Details

    • PackedInts

      public PackedInts()

  • Method Details

    • checkVersion

      public static void checkVersion(int version)
      Check the validity of a version number.

    • fastestFormatAndBits

      public static PackedInts.FormatAndBits fastestFormatAndBits(int valueCount, int bitsPerValue, float acceptableOverheadRatio)
      Try to find the PackedInts.Format and number of bits per value that would restore from disk the fastest reader whose overhead is less than acceptableOverheadRatio.

      The acceptableOverheadRatio parameter makes sense for random-access PackedInts.Readers. In case you only plan to perform sequential access on this stream later on, you should probably use COMPACT.

      If you don't know how many values you are going to write, use valueCount = -1.

    • getDecoder

      public static PackedInts.Decoder getDecoder(PackedInts.Format format, int version, int bitsPerValue)
      Get a PackedInts.Decoder.
      Parameters:
      format - the format used to store packed ints
      version - the compatibility version
      bitsPerValue - the number of bits per value
      Returns:
      a decoder

    • getEncoder

      public static PackedInts.Encoder getEncoder(PackedInts.Format format, int version, int bitsPerValue)
      Get an PackedInts.Encoder.
      Parameters:
      format - the format used to store packed ints
      version - the compatibility version
      bitsPerValue - the number of bits per value
      Returns:
      an encoder

    • getReaderNoHeader

      public static PackedInts.Reader getReaderNoHeader(DataInput in, PackedInts.Format format, int version, int valueCount, int bitsPerValue) throws IOException
      Expert: Restore a PackedInts.Reader from a stream without reading metadata at the beginning of the stream. This method is useful to restore data from streams which have been created using getWriterNoHeader(DataOutput, Format, int, int, int).
      Parameters:
      in - the stream to read data from, positioned at the beginning of the packed values
      format - the format used to serialize
      version - the version used to serialize the data
      valueCount - how many values the stream holds
      bitsPerValue - the number of bits per value
      Returns:
      a Reader
      Throws:
      IOException - If there is a low-level I/O error
      See Also:

    • getReaderNoHeader

      public static PackedInts.Reader getReaderNoHeader(DataInput in, PackedInts.Header header) throws IOException
      Expert: Restore a PackedInts.Reader from a stream without reading metadata at the beginning of the stream. This method is useful to restore data when metadata has been previously read using readHeader(DataInput).
      Parameters:
      in - the stream to read data from, positioned at the beginning of the packed values
      header - metadata result from readHeader()
      Returns:
      a Reader
      Throws:
      IOException - If there is a low-level I/O error
      See Also:

    • getReader

      public static PackedInts.Reader getReader(DataInput in) throws IOException
      Restore a PackedInts.Reader from a stream.
      Parameters:
      in - the stream to read data from
      Returns:
      a Reader
      Throws:
      IOException - If there is a low-level I/O error

    • getReaderIteratorNoHeader

      public static PackedInts.ReaderIterator getReaderIteratorNoHeader(DataInput in, PackedInts.Format format, int version, int valueCount, int bitsPerValue, int mem)
      Expert: Restore a PackedInts.ReaderIterator from a stream without reading metadata at the beginning of the stream. This method is useful to restore data from streams which have been created using getWriterNoHeader(DataOutput, Format, int, int, int).
      Parameters:
      in - the stream to read data from, positioned at the beginning of the packed values
      format - the format used to serialize
      version - the version used to serialize the data
      valueCount - how many values the stream holds
      bitsPerValue - the number of bits per value
      mem - how much memory the iterator is allowed to use to read-ahead (likely to speed up iteration)
      Returns:
      a ReaderIterator
      See Also:

    • getReaderIterator

      public static PackedInts.ReaderIterator getReaderIterator(DataInput in, int mem) throws IOException
      Retrieve PackedInts as a PackedInts.ReaderIterator
      Parameters:
      in - positioned at the beginning of a stored packed int structure.
      mem - how much memory the iterator is allowed to use to read-ahead (likely to speed up iteration)
      Returns:
      an iterator to access the values
      Throws:
      IOException - if the structure could not be retrieved.

    • getDirectReaderNoHeader

      public static PackedInts.Reader getDirectReaderNoHeader(IndexInput in, PackedInts.Format format, int version, int valueCount, int bitsPerValue)
      Expert: Construct a direct PackedInts.Reader from a stream without reading metadata at the beginning of the stream. This method is useful to restore data from streams which have been created using getWriterNoHeader(DataOutput, Format, int, int, int).

      The returned reader will have very little memory overhead, but every call to NumericDocValues.get(int) is likely to perform a disk seek.

      Parameters:
      in - the stream to read data from
      format - the format used to serialize
      version - the version used to serialize the data
      valueCount - how many values the stream holds
      bitsPerValue - the number of bits per value
      Returns:
      a direct Reader

    • getDirectReaderNoHeader

      public static PackedInts.Reader getDirectReaderNoHeader(IndexInput in, PackedInts.Header header) throws IOException
      Expert: Construct a direct PackedInts.Reader from an IndexInput without reading metadata at the beginning of the stream. This method is useful to restore data when metadata has been previously read using readHeader(DataInput).
      Parameters:
      in - the stream to read data from, positioned at the beginning of the packed values
      header - metadata result from readHeader()
      Returns:
      a Reader
      Throws:
      IOException - If there is a low-level I/O error
      See Also:

    • getDirectReader

      public static PackedInts.Reader getDirectReader(IndexInput in) throws IOException
      Construct a direct PackedInts.Reader from an IndexInput. This method is useful to restore data from streams which have been created using getWriter(DataOutput, int, int, float).

      The returned reader will have very little memory overhead, but every call to NumericDocValues.get(int) is likely to perform a disk seek.

      Parameters:
      in - the stream to read data from
      Returns:
      a direct Reader
      Throws:
      IOException - If there is a low-level I/O error

    • getMutable

      public static PackedInts.Mutable getMutable(int valueCount, int bitsPerValue, float acceptableOverheadRatio)
      Create a packed integer array with the given amount of values initialized to 0. the valueCount and the bitsPerValue cannot be changed after creation. All Mutables known by this factory are kept fully in RAM.

      Positive values of acceptableOverheadRatio will trade space for speed by selecting a faster but potentially less memory-efficient implementation. An acceptableOverheadRatio of COMPACT will make sure that the most memory-efficient implementation is selected whereas FASTEST will make sure that the fastest implementation is selected.

      Parameters:
      valueCount - the number of elements
      bitsPerValue - the number of bits available for any given value
      acceptableOverheadRatio - an acceptable overhead ratio per value
      Returns:
      a mutable packed integer array

    • getMutable

      public static PackedInts.Mutable getMutable(int valueCount, int bitsPerValue, PackedInts.Format format)
      Same as getMutable(int, int, float) with a pre-computed number of bits per value and format.

    • getWriterNoHeader

      public static PackedInts.Writer getWriterNoHeader(DataOutput out, PackedInts.Format format, int valueCount, int bitsPerValue, int mem)
      Expert: Create a packed integer array writer for the given output, format, value count, and number of bits per value.

      The resulting stream will be long-aligned. This means that depending on the format which is used, up to 63 bits will be wasted. An easy way to make sure that no space is lost is to always use a valueCount that is a multiple of 64.

      This method does not write any metadata to the stream, meaning that it is your responsibility to store it somewhere else in order to be able to recover data from the stream later on:

      It is possible to start writing values without knowing how many of them you are actually going to write. To do this, just pass -1 as valueCount. On the other hand, for any positive value of valueCount, the returned writer will make sure that you don't write more values than expected and pad the end of stream with zeros in case you have written less than valueCount when calling PackedInts.Writer.finish().

      The mem parameter lets you control how much memory can be used to buffer changes in memory before flushing to disk. High values of mem are likely to improve throughput. On the other hand, if speed is not that important to you, a value of 0 will use as little memory as possible and should already offer reasonable throughput.

      Parameters:
      out - the data output
      format - the format to use to serialize the values
      valueCount - the number of values
      bitsPerValue - the number of bits per value
      mem - how much memory (in bytes) can be used to speed up serialization
      Returns:
      a Writer
      See Also:

    • getWriter

      public static PackedInts.Writer getWriter(DataOutput out, int valueCount, int bitsPerValue, float acceptableOverheadRatio) throws IOException
      Create a packed integer array writer for the given output, format, value count, and number of bits per value.

      The resulting stream will be long-aligned. This means that depending on the format which is used under the hoods, up to 63 bits will be wasted. An easy way to make sure that no space is lost is to always use a valueCount that is a multiple of 64.

      This method writes metadata to the stream, so that the resulting stream is sufficient to restore a PackedInts.Reader from it. You don't need to track valueCount or bitsPerValue by yourself. In case this is a problem, you should probably look at getWriterNoHeader(DataOutput, Format, int, int, int).

      The acceptableOverheadRatio parameter controls how readers that will be restored from this stream trade space for speed by selecting a faster but potentially less memory-efficient implementation. An acceptableOverheadRatio of COMPACT will make sure that the most memory-efficient implementation is selected whereas FASTEST will make sure that the fastest implementation is selected. In case you are only interested in reading this stream sequentially later on, you should probably use COMPACT.

      Parameters:
      out - the data output
      valueCount - the number of values
      bitsPerValue - the number of bits per value
      acceptableOverheadRatio - an acceptable overhead ratio per value
      Returns:
      a Writer
      Throws:
      IOException - If there is a low-level I/O error

    • bitsRequired

      public static int bitsRequired(long maxValue)
      Returns how many bits are required to hold values up to and including maxValue
      Parameters:
      maxValue - the maximum value that should be representable.
      Returns:
      the amount of bits needed to represent values from 0 to maxValue.

    • maxValue

      public static long maxValue(int bitsPerValue)
      Calculates the maximum unsigned long that can be expressed with the given number of bits.
      Parameters:
      bitsPerValue - the number of bits available for any given value.
      Returns:
      the maximum value for the given bits.

    • copy

      public static void copy(PackedInts.Reader src, int srcPos, PackedInts.Mutable dest, int destPos, int len, int mem)
      Copy src[srcPos:srcPos+len] into dest[destPos:destPos+len] using at most mem bytes.

    • readHeader

      public static PackedInts.Header readHeader(DataInput in) throws IOException
      Expert: reads only the metadata from a stream. This is useful to later restore a stream or open a direct reader via getReaderNoHeader(DataInput, Header) or getDirectReaderNoHeader(IndexInput, Header).
      Parameters:
      in - the stream to read data
      Returns:
      packed integer metadata.
      Throws:
      IOException - If there is a low-level I/O error
      See Also: