java.lang.Object
org.apache.sysds.runtime.compress.colgroup.offset.AOffset
All Implemented Interfaces:
Serializable
Direct Known Subclasses:
AOffsetByte, OffsetChar, OffsetEmpty, OffsetSingle, OffsetTwo

public abstract class AOffset extends Object implements Serializable
Offset list encoder abstract class.

It is assumed that the input is in sorted order, all values are positive and there are no duplicates.

The no duplicate is important since 0 values are exploited to encode an offset of max representable value + 1. This gives the ability to encode data, where the offsets are greater than the available highest value that can be represented size.

See Also:
  • Constructor Details

    • AOffset

      public AOffset()
  • Method Details

    • getIterator

      public abstract AIterator getIterator()
      Get an iterator of the offsets while also maintaining the data index pointer.
      Returns:
      AIterator that iterate through index and dictionary offset values.
    • getOffsetIterator

      public abstract AOffsetIterator getOffsetIterator()
      Get an OffsetIterator of current offsets not maintaining the data index.
      Returns:
      AIterator that iterator through the delta offsets.
    • getIterator

      public AIterator getIterator(int row)
      Get an iterator that is pointing at a specific offset.
      Parameters:
      row - The row requested.
      Returns:
      AIterator that iterate through index and dictionary offset values.
    • constructSkipList

      public void constructSkipList()
    • clearSkipList

      public void clearSkipList()
    • cacheIterator

      public void cacheIterator(AIterator it, int row)
      Cache a iterator in use, note that there is no check for if the iterator is correctly positioned at the given row
      Parameters:
      it - The Iterator to cache
      row - The row index to cache the iterator as.
    • write

      public abstract void write(DataOutput out) throws IOException
      Write the offsets to disk. If you implement another remember to write the ordinal of the new type to disk as well and add it to the OffsetFactory.
      Parameters:
      out - The output to write to
      Throws:
      IOException - Exception that happens if the IO fails to write.
    • getOffsetToFirst

      public abstract int getOffsetToFirst()
      Get the offset to the first index
      Returns:
      The first index offset
    • getOffsetToLast

      public abstract int getOffsetToLast()
      Get the offset to the last value
      Returns:
      The last values offset
    • getInMemorySize

      public abstract long getInMemorySize()
      Get the in memory size of the Offset object
      Returns:
      In memory size as a long.
    • getExactSizeOnDisk

      public abstract long getExactSizeOnDisk()
      Remember to include the ordinal of the type of offset list.
      Returns:
      the size on disk as a long.
    • getSize

      public abstract int getSize()
      Get the number of contained elements, This method iterate the entire offset list, so it is not constant lookup.
      Returns:
      The number of indexes.
    • preAggregateDenseMap

      public final void preAggregateDenseMap(DenseBlock db, double[] preAV, int rl, int ru, int cl, int cu, int nVal, AMapToData data)
      Pre aggregate the specified row and block range from a dense MatrixBlock to prepare for compressed multiplication.
      Parameters:
      db - The DenseBlock to extract values from.
      preAV - The pre aggregate row linearized double array to put the values into.
      rl - The row lower to start from (this is referring to the left matrix of the multiplication)
      ru - The row upper to end at (not inclusive) (this is referring to the left matrix of the multiplication)
      cl - The column lower to start at (this is referring to the right matrix of the multiplication)
      cu - The column upper to end at (not inclusive) (this is referring to the right matrix of the multiplication)
      nVal - The number of distinct values in the PreAV indicating number of columns in the Pre aggregate
      data - The mapping to column positions in the preAV
    • preAggSparseMap

      public final void preAggSparseMap(SparseBlock sb, double[] preAV, int rl, int ru, int nVal, AMapToData data)
    • equals

      public boolean equals(Object o)
      Overrides:
      equals in class Object
    • equals

      public boolean equals(AOffset b)
    • moveIndex

      public abstract AOffset moveIndex(int m)
      Move the index start x cells
      Parameters:
      m - The amount to move
      Returns:
      The moved index.
    • getLength

      public abstract int getLength()
      Get the length of the underlying array. This does not reflect the number of contained elements, since some of the elements can be skips.
      Returns:
      The length of the underlying arrays
    • slice

      public AOffset.OffsetSliceInfo slice(int l, int u)
      Slice the offsets based on the specified range
      Parameters:
      l - inclusive lower bound
      u - exclusive upper bound
      Returns:
      The slice info containing the new slice.
    • append

      public AOffset append(AOffset t, int s)
      Append the offsets from that other offset to the offsets in this.
      Parameters:
      t - That offset/
      s - The total length encoded in this offset.
      Returns:
      this offsets followed by thats offsets.
    • appendN

      public AOffset appendN(AOffsetsGroup[] g, int s)
      Append a list of offsets together in order.
      Parameters:
      g - The offsets to append together (note fist entry is equal to this)
      s - The standard size of each g (except last, but that does not matter)
      Returns:
      The combined offsets.
    • verify

      public void verify(int size)
      Verify that the contained AOffset is a certain size, and not bigger when iterating though it.
      Parameters:
      size - The max correct size.
    • toString

      public String toString()
      Overrides:
      toString in class Object
    • reverse

      public AOffset reverse(int numRows)
      Reverse the locations of the offsets such that the inverse offsets are returned. This means that for instance if the current offsets were 1, 3 and 5 in a 5 long list, we return 0, 2 and 4.
      Parameters:
      numRows - The total number of rows to be contained, This should be greater or equal to last.
      Returns:
      The reverse offsets.
    • removeEmptyRows

      public AOffset.RemoveEmptyOffsetsTmp removeEmptyRows(boolean[] selectV, int rOut)