org.apache.sysml.runtime.compress

Class ColGroupOffset

java.lang.Object

org.apache.sysml.runtime.compress.ColGroup

org.apache.sysml.runtime.compress.ColGroupValue

org.apache.sysml.runtime.compress.ColGroupOffset

All Implemented Interfaces:

Serializable

Direct Known Subclasses:

ColGroupOLE, ColGroupRLE

public abstract class ColGroupOffset
extends ColGroupValue

Base class for column groups encoded with various types of bitmap encoding. NOTES: * OLE: separate storage segment length and bitmaps led to a 30% improvement but not applied because more difficult to support both data layouts at the same time (distributed/local as well as w/ and w/o low-level opt)

See Also:

Serialized Form

Nested Class Summary

Nested classes/interfaces inherited from class org.apache.sysml.runtime.compress.ColGroup

ColGroup.CompressionType

Field Summary

Fields

Modifier and Type	Field and Description
protected char[]	_data
protected int[]	_ptr Bitmaps, one per uncompressed value in ColGroupValue._values.
protected int[]	_skiplist
protected boolean	_zeros
static boolean	ALLOW_CACHE_CONSCIOUS_ROWSUMS
protected static boolean	CREATE_SKIPLIST
protected static int	READ_CACHE_BLKSZ
static int	WRITE_CACHE_BLKSZ

Fields inherited from class org.apache.sysml.runtime.compress.ColGroupValue

_values, LOW_LEVEL_OPT, SORT_VALUES_BY_LENGTH

Fields inherited from class org.apache.sysml.runtime.compress.ColGroup

_colIndexes, _numRows

Constructor Summary

Constructors

Modifier	Constructor and Description
	ColGroupOffset()
protected	ColGroupOffset(int[] colIndices, int numRows, boolean zeros, double[] values) Constructor for subclass methods that need to create shallow copies
	ColGroupOffset(int[] colIndices, int numRows, UncompressedBitmap ubm) Main constructor.

Method Summary

Modifier and Type	Method and Description
protected abstract void	computeColSums(MatrixBlock result, org.apache.sysml.runtime.functionobjects.KahanFunction kplus)
protected int[]	computeOffsets(boolean[] ind) Utility function of sparse-unsafe operations.
protected abstract void	computeRowMxx(MatrixBlock result, org.apache.sysml.runtime.functionobjects.Builtin builtin, int rl, int ru)
protected abstract void	computeRowSums(MatrixBlock result, org.apache.sysml.runtime.functionobjects.KahanFunction kplus, int rl, int ru)
protected abstract void	computeSum(MatrixBlock result, org.apache.sysml.runtime.functionobjects.KahanFunction kplus)
protected void	createCompressedBitmaps(int numVals, int totalLen, char[][] lbitmaps)
void	decompressToBlock(MatrixBlock target, int colpos) Decompress to block.
void	decompressToBlock(MatrixBlock target, int[] colIndexTargets) Decompress the contents of this column group into uncompressed packed columns
void	decompressToBlock(MatrixBlock target, int rl, int ru) Decompress the contents of this column group into the specified full matrix block.
long	estimateInMemorySize() Note: Must be overridden by child classes to account for additional data and metadata
double	get(int r, int c) Get the value at a global row/column position.
int[]	getBitmapOffsets()
char[]	getBitmaps()
abstract Iterator<Integer>	getDecodeIterator(int k)
long	getExactSizeOnDisk() Returns the exact serialized size of column group.
boolean	hasZeros()
protected int	len(int k)
protected double	mxxValues(int bitmapIx, org.apache.sysml.runtime.functionobjects.Builtin builtin)
void	readFields(DataInput in) Deserializes column group from data input.
protected void	sumAllValues(double[] b, double[] c)
void	unaryAggregateOperations(org.apache.sysml.runtime.matrix.operators.AggregateUnaryOperator op, MatrixBlock result, int rl, int ru)
void	write(DataOutput out) Serializes column group to data output.

Methods inherited from class org.apache.sysml.runtime.compress.ColGroupValue

allocDVector, allocIVector, applyScalarOp, applyScalarOp, cleanupThreadLocalMemory, computeColMxx, computeMxx, containsAllZeroValue, getNumValues, getValues, preaggValues, preaggValues, setupThreadLocalMemory, sumAllValues, sumAllValues, sumValues, sumValues, sumValues, unaryAggregateOperations

Methods inherited from class org.apache.sysml.runtime.compress.ColGroup

countNonZerosPerRow, getColIndex, getColIndices, getCompType, getNumCols, getNumRows, leftMultByRowVector, rightMultByVector, scalarOperation, shiftColIndices

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

CREATE_SKIPLIST

protected static final boolean CREATE_SKIPLIST

See Also:

Constant Field Values

READ_CACHE_BLKSZ

protected static final int READ_CACHE_BLKSZ

See Also:

Constant Field Values

WRITE_CACHE_BLKSZ

public static final int WRITE_CACHE_BLKSZ

See Also:

Constant Field Values

ALLOW_CACHE_CONSCIOUS_ROWSUMS

public static boolean ALLOW_CACHE_CONSCIOUS_ROWSUMS

_ptr

protected int[] _ptr

Bitmaps, one per uncompressed value in ColGroupValue._values.

_data

protected char[] _data

_zeros

protected boolean _zeros

_skiplist

protected int[] _skiplist

Constructor Detail

ColGroupOffset

public ColGroupOffset()

ColGroupOffset

public ColGroupOffset(int[] colIndices,
                      int numRows,
                      UncompressedBitmap ubm)

Main constructor. Stores the headers for the individual bitmaps.

Parameters:

colIndices - indices (within the block) of the columns included in this column

numRows - total number of rows in the parent block

ubm - Uncompressed bitmap representation of the block

ColGroupOffset

protected ColGroupOffset(int[] colIndices,
                         int numRows,
                         boolean zeros,
                         double[] values)

Constructor for subclass methods that need to create shallow copies

Parameters:

colIndices - raw column index information

numRows - number of rows in the block

zeros - indicator if column group contains zero values

values - set of distinct values for the block (associated bitmaps are kept in the subclass)

Method Detail

len

protected final int len(int k)

createCompressedBitmaps

protected void createCompressedBitmaps(int numVals,
                                       int totalLen,
                                       char[][] lbitmaps)

estimateInMemorySize

public long estimateInMemorySize()

Description copied from class: ColGroup

Note: Must be overridden by child classes to account for additional data and metadata

Overrides:

estimateInMemorySize in class ColGroupValue

Returns:

an upper bound on the number of bytes used to store this ColGroup in memory.

decompressToBlock

public void decompressToBlock(MatrixBlock target,
                              int rl,
                              int ru)

Description copied from class: ColGroup

Decompress the contents of this column group into the specified full matrix block.

Specified by:

decompressToBlock in class ColGroup

Parameters:

target - a matrix block where the columns covered by this column group have not yet been filled in.

rl - row lower

ru - row upper

decompressToBlock

public void decompressToBlock(MatrixBlock target,
                              int[] colIndexTargets)

Description copied from class: ColGroup

Decompress the contents of this column group into uncompressed packed columns

Specified by:

decompressToBlock in class ColGroup

Parameters:

target - a dense matrix block. The block must have enough space to hold the contents of this column group.

colIndexTargets - array that maps column indices in the original matrix block to columns of target.

decompressToBlock

public void decompressToBlock(MatrixBlock target,
                              int colpos)

Description copied from class: ColGroup

Decompress to block.

Specified by:

decompressToBlock in class ColGroup

Parameters:

target - dense output vector

colpos - column to decompress, error if larger or equal numCols

get

public double get(int r,
                  int c)

Description copied from class: ColGroup

Get the value at a global row/column position.

Specified by:

get in class ColGroup

Parameters:

r - row

c - column

Returns:

value at the row/column position

sumAllValues

protected final void sumAllValues(double[] b,
                                  double[] c)

mxxValues

protected final double mxxValues(int bitmapIx,
                                 org.apache.sysml.runtime.functionobjects.Builtin builtin)

getBitmaps

public char[] getBitmaps()

getBitmapOffsets

public int[] getBitmapOffsets()

hasZeros

public boolean hasZeros()

getDecodeIterator

public abstract Iterator<Integer> getDecodeIterator(int k)

Parameters:

k - index of a specific compressed bitmap (stored in subclass, index same as ColGroupValue.getValues())

Returns:

an object for iterating over the row offsets in this bitmap. Only valid until the next call to this method. May be reused across calls.

computeOffsets

protected int[] computeOffsets(boolean[] ind)
                        throws DMLRuntimeException

Utility function of sparse-unsafe operations.

Parameters:

ind - row indicator vector of non zeros

Returns:

offsets

Throws:

DMLRuntimeException - if DMLRuntimeException occurs

readFields

public void readFields(DataInput in)
                throws IOException

Description copied from class: ColGroup

Deserializes column group from data input.

Specified by:

readFields in class ColGroup

Parameters:

in - data input

Throws:

IOException - if IOException occurs

write

public void write(DataOutput out)
           throws IOException

Description copied from class: ColGroup

Serializes column group to data output.

Specified by:

write in class ColGroup

Parameters:

out - data output

Throws:

IOException - if IOException occurs

getExactSizeOnDisk

public long getExactSizeOnDisk()

Description copied from class: ColGroup

Returns the exact serialized size of column group. This can be used for example for buffer preallocation.

Specified by:

getExactSizeOnDisk in class ColGroup

Returns:

exact serialized size for column group

unaryAggregateOperations

public void unaryAggregateOperations(org.apache.sysml.runtime.matrix.operators.AggregateUnaryOperator op,
                                     MatrixBlock result,
                                     int rl,
                                     int ru)
                              throws DMLRuntimeException

Specified by:

unaryAggregateOperations in class ColGroupValue

Parameters:

op - aggregation operator

result - output matrix block

rl - row lower index, inclusive

ru - row upper index, exclusive

Throws:

DMLRuntimeException - on invalid inputs

computeSum

protected abstract void computeSum(MatrixBlock result,
                                   org.apache.sysml.runtime.functionobjects.KahanFunction kplus)

computeRowSums

protected abstract void computeRowSums(MatrixBlock result,
                                       org.apache.sysml.runtime.functionobjects.KahanFunction kplus,
                                       int rl,
                                       int ru)

computeColSums

protected abstract void computeColSums(MatrixBlock result,
                                       org.apache.sysml.runtime.functionobjects.KahanFunction kplus)

computeRowMxx

protected abstract void computeRowMxx(MatrixBlock result,
                                      org.apache.sysml.runtime.functionobjects.Builtin builtin,
                                      int rl,
                                      int ru)