Package org.apache.sysds.runtime.data

Class SparseBlockCSR

java.lang.Object
org.apache.sysds.runtime.data.SparseBlock
org.apache.sysds.runtime.data.SparseBlockCSR
All Implemented Interfaces:
Serializable, Block
public class SparseBlockCSR extends SparseBlock
SparseBlock implementation that realizes a traditional 'compressed sparse row' representation, where the entire sparse block is stored as three arrays: ptr of length rlen+1 to store offsets per row, and indexes/values of length nnz to store column indexes and values of non-zero entries. This format is very memory efficient for sparse (but not ultra-sparse) matrices and provides very good performance for common operations, partially due to lower memory bandwidth requirements. However, this format is slow on incremental construction (because it does not allow append/sort per row) without reshifting. Finally, the total nnz is limited to INTEGER_MAX, whereas for SparseBlockMCSR only the nnz per row are limited to INTEGER_MAX. TODO: extensions for faster incremental construction (e.g., max row) TODO more efficient fused setIndexRange impl to avoid repeated copies and updates
See Also:

  • Nested Class Summary

    Nested Classes
    Modifier and Type
    Class
    Description
    class 
    SparseBlockCSR.NonEmptyRowsIteratorCSR
     

    Nested classes/interfaces inherited from class org.apache.sysds.runtime.data.SparseBlock

    SparseBlock.Type

  • Constructor Summary

    Constructors
    Constructor
    Description
    SparseBlockCSR(int rlen)
     
    SparseBlockCSR(int[] rowPtr, int[] colInd, double[] values, int nnz)
     
    SparseBlockCSR(int rlen, int capacity)
     
    SparseBlockCSR(int rows, int[] rowInd, int[] colInd, double[] values)
    Copy constructor for COO representation
    SparseBlockCSR(int rlen, int capacity, int size)
     
    SparseBlockCSR(int rows, int nnz, int[] colInd)
    Copy constructor for given array of column indexes, which identifies rows by position and implies values of 1.
    SparseBlockCSR(SparseBlock sblock)
    Copy constructor sparse block abstraction.
    SparseBlockCSR(SparseRow[] rows, int nnz)
    Copy constructor old sparse row representation.

  • Method Summary

    Modifier and Type
    Method
    Description
    boolean
    add(int r, int c, double v)
    Add a value to a matrix cell (r,c).
    void
    allocate(int r)
    Allocate the underlying data structure holding non-zero values of row r if necessary.
    void
    allocate(int r, int nnz)
    Allocate the underlying data structure holding non-zero values of row r if necessary, w/ given size.
    void
    allocate(int r, int ennz, int maxnnz)
    Allocate the underlying data structure holding non-zero values of row r w/ the specified estimated nnz and max nnz.
    void
    append(int r, int c, double v)
    Append a value to the end of the physical representation.
    boolean
    checkValidity(int rlen, int clen, long nnz, boolean strict)
    Validate the correctness of the internal data structures of the different sparse block implementations.
    void
    compact()
    In-place compaction of non-zero-entries; removes zero entries and shifts non-zero entries to the left if necessary.
    void
    compact(int r)
    Re-allocate physical row if physical size exceeds logical size plus resize factor.
    boolean
    contains(double pattern, int rl, int ru)
    Checks if the block contains at least one value of the given pattern.
    void
    deleteIndexRange(int r, int cl, int cu)
    Deletes all non-zero values of the given column range [cl,cu) in row r.
    static long
    estimateSizeInMemory(long nrows, long nnz)
     
    static long
    estimateSizeInMemory(long nrows, long ncols, double sparsity)
    Get the estimated in-memory size of the sparse block in CSR with the given dimensions w/o accounting for overallocation.
    SparseRow
    get(int r)
    Get values of row r in the format of a sparse row.
    double
    get(int r, int c)
    Get value of matrix cell (r,c).
    long
    getExactSizeInMemory()
    Computes the exact size in memory of the materialized block
    Iterator<Integer>
    getNonEmptyRowsIterator(int rl, int ru)
     
    SparseBlock.Type
    getSparseBlockType()
    Get the type of the sparse block.
    int[]
    indexes()
    Get raw access to underlying array of column indices For use in GPU code
    int[]
    indexes(int r)
    Get the sorted array of column indexes of all non-zero entries in row r.
    void
    initSparse(int rlen, int nnz, DataInput in)
    Initializes the CSR sparse block from an ordered input stream of sparse rows (rownnz, jv-pairs*).
    void
    initUltraSparse(int nnz, DataInput in)
    Initializes the CSR sparse block from an ordered input stream of ultra-sparse ijv triples.
    boolean
    isAllocated(int r)
    Indicates if the underlying data structure for a given row is already allocated.
    boolean
    isContiguous()
    Indicates if the underlying data structures returned by values and indexes are contiguous arrays, which can be exploited for more efficient operations.
    boolean
    isEmpty(int r)
    Get information if row r is empty, i.e., does not contain non-zero values.
    boolean
    isThreadSafe()
    Indicates if the underlying implementation allows thread-safe row updates if concurrent threads update disjoint rows.
    int
    numRows()
    Get the number of rows in the sparse block.
    int
    pos(int r)
    Get the starting position of row r in the indexes/values arrays returned by indexes(r) and values(r).
    int
    posFIndexGT(int r, int c)
    Get position of first column index greater than column c in row r.
    final int
    posFIndexGTE(int r, int c)
    Get position of first column index greater than or equal column c in row r.
    int
    posFIndexLTE(int r, int c)
    Get position of first column index lower than or equal column c in row r.
    void
    reset()
    Clears the sparse block by deleting non-zero values.
    void
    reset(int ennz, int maxnnz)
    Clears the sparse block by deleting non-zero values.
    void
    reset(int r, int ennz, int maxnnz)
    Clears row r of the sparse block by deleting non-zero values.
    int[]
    rowPointers()
    Get raw access to underlying array of row pointers For use in GPU code
    boolean
    set(int r, int c, double v)
    Set the value of a matrix cell (r,c).
    void
    set(int r, SparseRow row, boolean deep)
    Set the values of row r to the given sparse row.
    void
    setIndexRange(int r, int cl, int cu, double[] v, int[] vix, int vpos, int vlen)
    Sets a sparse array of non-zeros values and indexes into the column range [cl,cu) in row r.
    void
    setIndexRange(int r, int cl, int cu, double[] v, int vix, int vlen)
    Sets a dense array of non-zeros values into the column range [cl,cu) in row r.
    void
    setIndexRange(int rl, int ru, int cl, int cu, double[] v, int vix, int vlen)
    Inserts a sorted row-major array of non-zero values into the row and column range [rl,ru) and [cl,cu).
    void
    setIndexRange(int rl, int ru, int cl, int cu, SparseBlock sb)
    Inserts a sparse block into the row and column range [rl,ru) and [cl,cu).
    long
    size()
    Get the number of non-zero values in the sparse block.
    int
    size(int r)
    Get the number of non-zero values in row r.
    long
    size(int rl, int ru)
    Get the number of non-zeros values in the row range of [rl, ru).
    long
    size(int rl, int ru, int cl, int cu)
    Get the number of non-zeros values in the row and column range of [rl/cl, ru/cu);
    void
    sort()
    Sort all non-zero value/index pairs of the sparse block by row and column index.
    void
    sort(int r)
    Sort all non-zero value/index pairs of row r column index.
    String
    toString()
     
    double[]
    values()
    Get raw access to underlying array of values For use in GPU code
    double[]
    values(int r)
    Get the array of all non-zero entries in row r, sorted by their column indexes.

    Methods inherited from class org.apache.sysds.runtime.data.SparseBlock

    contains, equals, equals, equals, equals, getIterator, getIterator, getIterator, getIterator, getNonEmptyRows, getNonEmptyRows, isAligned, isAligned

    Methods inherited from class java.lang.Object

    getClass, hashCode, notify, notifyAll, wait, wait, wait

  • Constructor Details

    • SparseBlockCSR

      public SparseBlockCSR(int rlen)

    • SparseBlockCSR

      public SparseBlockCSR(int rlen, int capacity)

    • SparseBlockCSR

      public SparseBlockCSR(int rlen, int capacity, int size)

    • SparseBlockCSR

      public SparseBlockCSR(int[] rowPtr, int[] colInd, double[] values, int nnz)

    • SparseBlockCSR

      public SparseBlockCSR(SparseBlock sblock)
      Copy constructor sparse block abstraction.
      Parameters:
      sblock - sparse block to copy

    • SparseBlockCSR

      public SparseBlockCSR(SparseRow[] rows, int nnz)
      Copy constructor old sparse row representation.
      Parameters:
      rows - array of sparse rows
      nnz - number of non-zeroes

    • SparseBlockCSR

      public SparseBlockCSR(int rows, int[] rowInd, int[] colInd, double[] values)
      Copy constructor for COO representation
      Parameters:
      rows - number of rows
      rowInd - row indices
      colInd - column indices
      values - non zero values

    • SparseBlockCSR

      public SparseBlockCSR(int rows, int nnz, int[] colInd)
      Copy constructor for given array of column indexes, which identifies rows by position and implies values of 1.
      Parameters:
      rows - number of rows
      nnz - number of non-zeros
      colInd - column indexes

  • Method Details

    • initUltraSparse

      public void initUltraSparse(int nnz, DataInput in) throws IOException
      Initializes the CSR sparse block from an ordered input stream of ultra-sparse ijv triples.
      Parameters:
      nnz - number of non-zeros to read
      in - data input stream of ijv triples, ordered by ij
      Throws:
      IOException - if deserialization error occurs

    • initSparse

      public void initSparse(int rlen, int nnz, DataInput in) throws IOException
      Initializes the CSR sparse block from an ordered input stream of sparse rows (rownnz, jv-pairs*).
      Parameters:
      rlen - number of rows
      nnz - number of non-zeros to read
      in - data input stream of sparse rows, ordered by i
      Throws:
      IOException - if deserialization error occurs

    • estimateSizeInMemory

      public static long estimateSizeInMemory(long nrows, long ncols, double sparsity)
      Get the estimated in-memory size of the sparse block in CSR with the given dimensions w/o accounting for overallocation.
      Parameters:
      nrows - number of rows
      ncols - number of columns
      sparsity - sparsity ratio
      Returns:
      memory estimate

    • estimateSizeInMemory

      public static long estimateSizeInMemory(long nrows, long nnz)

    • getExactSizeInMemory

      public long getExactSizeInMemory()
      Description copied from class: SparseBlock
      Computes the exact size in memory of the materialized block
      Specified by:
      getExactSizeInMemory in class SparseBlock
      Returns:
      the exact size in memory

    • rowPointers

      public int[] rowPointers()
      Get raw access to underlying array of row pointers For use in GPU code
      Returns:
      array of row pointers

    • indexes

      public int[] indexes()
      Get raw access to underlying array of column indices For use in GPU code
      Returns:
      array of column indexes

    • values

      public double[] values()
      Get raw access to underlying array of values For use in GPU code
      Returns:
      array of values

    • allocate

      public void allocate(int r)
      Description copied from class: SparseBlock
      Allocate the underlying data structure holding non-zero values of row r if necessary.
      Specified by:
      allocate in class SparseBlock
      Parameters:
      r - row index

    • allocate

      public void allocate(int r, int nnz)
      Description copied from class: SparseBlock
      Allocate the underlying data structure holding non-zero values of row r if necessary, w/ given size.
      Specified by:
      allocate in class SparseBlock
      Parameters:
      r - row index
      nnz - number of non-zeros

    • allocate

      public void allocate(int r, int ennz, int maxnnz)
      Description copied from class: SparseBlock
      Allocate the underlying data structure holding non-zero values of row r w/ the specified estimated nnz and max nnz.
      Specified by:
      allocate in class SparseBlock
      Parameters:
      r - row index
      ennz - estimated non-zeros
      maxnnz - max non-zeros

    • compact

      public void compact(int r)
      Description copied from class: SparseBlock
      Re-allocate physical row if physical size exceeds logical size plus resize factor.
      Specified by:
      compact in class SparseBlock
      Parameters:
      r - row index

    • compact

      public void compact()
      Description copied from class: SparseBlock
      In-place compaction of non-zero-entries; removes zero entries and shifts non-zero entries to the left if necessary.
      Specified by:
      compact in class SparseBlock

    • getSparseBlockType

      public SparseBlock.Type getSparseBlockType()
      Description copied from class: SparseBlock
      Get the type of the sparse block.
      Specified by:
      getSparseBlockType in class SparseBlock
      Returns:
      sparse block type

    • numRows

      public int numRows()
      Description copied from class: SparseBlock
      Get the number of rows in the sparse block.
      Specified by:
      numRows in class SparseBlock
      Returns:
      number of rows

    • isThreadSafe

      public boolean isThreadSafe()
      Description copied from class: SparseBlock
      Indicates if the underlying implementation allows thread-safe row updates if concurrent threads update disjoint rows.
      Specified by:
      isThreadSafe in class SparseBlock
      Returns:
      true if thread-safe row updates

    • isContiguous

      public boolean isContiguous()
      Description copied from class: SparseBlock
      Indicates if the underlying data structures returned by values and indexes are contiguous arrays, which can be exploited for more efficient operations.
      Specified by:
      isContiguous in class SparseBlock
      Returns:
      true if underlying data structures are contiguous arrays

    • isAllocated

      public boolean isAllocated(int r)
      Description copied from class: SparseBlock
      Indicates if the underlying data structure for a given row is already allocated.
      Specified by:
      isAllocated in class SparseBlock
      Parameters:
      r - row index
      Returns:
      true if already allocated

    • reset

      public void reset()
      Description copied from class: SparseBlock
      Clears the sparse block by deleting non-zero values. After this call all size() calls are guaranteed to return 0.
      Specified by:
      reset in class SparseBlock

    • reset

      public void reset(int ennz, int maxnnz)
      Description copied from class: SparseBlock
      Clears the sparse block by deleting non-zero values. After this call all size() calls are guaranteed to return 0.
      Specified by:
      reset in class SparseBlock
      Parameters:
      ennz - estimated non-zeros
      maxnnz - max non-zeros

    • reset

      public void reset(int r, int ennz, int maxnnz)
      Description copied from class: SparseBlock
      Clears row r of the sparse block by deleting non-zero values. After this call size(r) is guaranteed to return 0.
      Specified by:
      reset in class SparseBlock
      Parameters:
      r - row index
      ennz - estimated non-zeros
      maxnnz - max non-zeros

    • size

      public long size()
      Description copied from class: SparseBlock
      Get the number of non-zero values in the sparse block.
      Specified by:
      size in class SparseBlock
      Returns:
      number of non-zero values in sparse block

    • size

      public int size(int r)
      Description copied from class: SparseBlock
      Get the number of non-zero values in row r.
      Specified by:
      size in class SparseBlock
      Parameters:
      r - row index starting at 0
      Returns:
      number of non-zero values in row r

    • size

      public long size(int rl, int ru)
      Description copied from class: SparseBlock
      Get the number of non-zeros values in the row range of [rl, ru).
      Specified by:
      size in class SparseBlock
      Parameters:
      rl - row lower index
      ru - row upper index
      Returns:
      number of non-zero values in the row range

    • size

      public long size(int rl, int ru, int cl, int cu)
      Description copied from class: SparseBlock
      Get the number of non-zeros values in the row and column range of [rl/cl, ru/cu);
      Specified by:
      size in class SparseBlock
      Parameters:
      rl - row lower index
      ru - row upper index
      cl - column lower index
      cu - column upper index
      Returns:
      number of non-zero values in the row and column range

    • isEmpty

      public boolean isEmpty(int r)
      Description copied from class: SparseBlock
      Get information if row r is empty, i.e., does not contain non-zero values. Equivalent to size(r)==0. Users should do this check if it is unknown if the underlying row data structure is allocated.
      Specified by:
      isEmpty in class SparseBlock
      Parameters:
      r - row index starting at 0
      Returns:
      true if row does not contain non-zero values

    • indexes

      public int[] indexes(int r)
      Description copied from class: SparseBlock
      Get the sorted array of column indexes of all non-zero entries in row r. Note that - for flexibility of the implementing format - the returned array may be larger, where the range for row r is given by [pos(r),pos(r)+size(r)).
      Specified by:
      indexes in class SparseBlock
      Parameters:
      r - row index starting at 0
      Returns:
      sorted array of column indexes

    • values

      public double[] values(int r)
      Description copied from class: SparseBlock
      Get the array of all non-zero entries in row r, sorted by their column indexes. Note that - for flexibility of the implementing format - the returned array may be larger, where the range for row r is given by [pos(r),pos(r)+size(r)).
      Specified by:
      values in class SparseBlock
      Parameters:
      r - row index starting at 0
      Returns:
      array of all non-zero entries in row r sorted by column indexes

    • pos

      public int pos(int r)
      Description copied from class: SparseBlock
      Get the starting position of row r in the indexes/values arrays returned by indexes(r) and values(r).
      Specified by:
      pos in class SparseBlock
      Parameters:
      r - row index starting at 0
      Returns:
      starting position of row r

    • set

      public boolean set(int r, int c, double v)
      Description copied from class: SparseBlock
      Set the value of a matrix cell (r,c). This might update an existing non-zero value, insert a new non-zero value, or delete a non-zero value.
      Specified by:
      set in class SparseBlock
      Parameters:
      r - row index starting at 0
      c - column index starting at 0
      v - zero or non-zero value
      Returns:
      true, if number of non-zeros changed

    • add

      public boolean add(int r, int c, double v)
      Description copied from class: SparseBlock
      Add a value to a matrix cell (r,c). This might update an existing non-zero value, or insert a new non-zero value.
      Specified by:
      add in class SparseBlock
      Parameters:
      r - row index starting at 0
      c - column index starting at 0
      v - zero or non-zero value
      Returns:
      true, if number of non-zeros changed

    • set

      public void set(int r, SparseRow row, boolean deep)
      Description copied from class: SparseBlock
      Set the values of row r to the given sparse row. This might update existing non-zero values, insert a new row, or delete a row. NOTE: This method exists for incremental runtime integration and might be deleted in the future.
      Specified by:
      set in class SparseBlock
      Parameters:
      r - row index starting at 0
      row - sparse row
      deep - indicator to create deep copy of sparse row

    • append

      public void append(int r, int c, double v)
      Description copied from class: SparseBlock
      Append a value to the end of the physical representation. This should only be used for operations with sequential write pattern or if followed by a sort() operation. Note that this operation does not perform any matrix cell updates.
      Specified by:
      append in class SparseBlock
      Parameters:
      r - row index starting at 0
      c - column index starting at 0
      v - zero or non-zero value

    • setIndexRange

      public void setIndexRange(int r, int cl, int cu, double[] v, int vix, int vlen)
      Description copied from class: SparseBlock
      Sets a dense array of non-zeros values into the column range [cl,cu) in row r. The passed value array may be larger and the relevant range is given by [vix,vix+len).
      Specified by:
      setIndexRange in class SparseBlock
      Parameters:
      r - row index starting at 0
      cl - lower column index starting at 0
      cu - upper column index starting at 0
      v - value array
      vix - start index in value array
      vlen - number of relevant values

    • setIndexRange

      public void setIndexRange(int r, int cl, int cu, double[] v, int[] vix, int vpos, int vlen)
      Description copied from class: SparseBlock
      Sets a sparse array of non-zeros values and indexes into the column range [cl,cu) in row r. The passed value array may be larger.
      Specified by:
      setIndexRange in class SparseBlock
      Parameters:
      r - row index starting at 0
      cl - lower column index starting at 0
      cu - upper column index starting at 0
      v - value array
      vix - column index array
      vpos - start index in value and index arrays
      vlen - number of relevant values

    • setIndexRange

      public void setIndexRange(int rl, int ru, int cl, int cu, double[] v, int vix, int vlen)
      Inserts a sorted row-major array of non-zero values into the row and column range [rl,ru) and [cl,cu). Note: that this is a CSR-specific method to address performance issues due to repeated re-shifting on update-in-place.
      Parameters:
      rl - lower row index, starting at 0, inclusive
      ru - upper row index, starting at 0, exclusive
      cl - lower column index, starting at 0, inclusive
      cu - upper column index, starting at 0, exclusive
      v - right-hand-side dense block
      vix - right-hand-side dense block index
      vlen - right-hand-side dense block value length

    • setIndexRange

      public void setIndexRange(int rl, int ru, int cl, int cu, SparseBlock sb)
      Inserts a sparse block into the row and column range [rl,ru) and [cl,cu). Note: that this is a CSR-specific method to address performance issues due to repeated re-shifting on update-in-place.
      Parameters:
      rl - lower row index, starting at 0, inclusive
      ru - upper row index, starting at 0, exclusive
      cl - lower column index, starting at 0, inclusive
      cu - upper column index, starting at 0, exclusive
      sb - right-hand-side sparse block

    • deleteIndexRange

      public void deleteIndexRange(int r, int cl, int cu)
      Description copied from class: SparseBlock
      Deletes all non-zero values of the given column range [cl,cu) in row r.
      Specified by:
      deleteIndexRange in class SparseBlock
      Parameters:
      r - row index starting at 0
      cl - lower column index starting at 0
      cu - upper column index starting at 0

    • sort

      public void sort()
      Description copied from class: SparseBlock
      Sort all non-zero value/index pairs of the sparse block by row and column index.
      Specified by:
      sort in class SparseBlock

    • sort

      public void sort(int r)
      Description copied from class: SparseBlock
      Sort all non-zero value/index pairs of row r column index.
      Specified by:
      sort in class SparseBlock
      Parameters:
      r - row index starting at 0

    • get

      public double get(int r, int c)
      Description copied from class: SparseBlock
      Get value of matrix cell (r,c). In case of non existing values this call returns 0.
      Specified by:
      get in interface Block
      Specified by:
      get in class SparseBlock
      Parameters:
      r - row index starting at 0
      c - column index starting at 0
      Returns:
      value of cell at position (r,c)

    • get

      public SparseRow get(int r)
      Description copied from class: SparseBlock
      Get values of row r in the format of a sparse row.
      Specified by:
      get in class SparseBlock
      Parameters:
      r - row index starting at 0
      Returns:
      values of row r as a sparse row

    • posFIndexLTE

      public int posFIndexLTE(int r, int c)
      Description copied from class: SparseBlock
      Get position of first column index lower than or equal column c in row r. The position is relative to the indexes/values arrays returned by indexes(r) and values(r). If no such value exists, this call returns -1.
      Specified by:
      posFIndexLTE in class SparseBlock
      Parameters:
      r - row index starting at 0
      c - column index starting at 0
      Returns:
      position of the first column index lower than or equal to column c in row r

    • posFIndexGTE

      public final int posFIndexGTE(int r, int c)
      Description copied from class: SparseBlock
      Get position of first column index greater than or equal column c in row r. The position is relative to the indexes/values arrays returned by indexes(r) and values(r). If no such value exists, this call returns -1. Note if CSR the pos(r) is subtracted from the result.
      Specified by:
      posFIndexGTE in class SparseBlock
      Parameters:
      r - row index starting at 0
      c - column index starting at 0
      Returns:
      position of the first column index greater than or equal to column c in row r

    • posFIndexGT

      public int posFIndexGT(int r, int c)
      Description copied from class: SparseBlock
      Get position of first column index greater than column c in row r. The position is relative to the indexes/values arrays returned by indexes(r) and values(r). If no such value exists, this call returns -1.
      Specified by:
      posFIndexGT in class SparseBlock
      Parameters:
      r - row index starting at 0
      c - column index starting at 0
      Returns:
      position of the first column index greater than column c in row r

    • toString

      public String toString()
      Specified by:
      toString in class SparseBlock

    • checkValidity

      public boolean checkValidity(int rlen, int clen, long nnz, boolean strict)
      Description copied from class: SparseBlock
      Validate the correctness of the internal data structures of the different sparse block implementations.
      Specified by:
      checkValidity in class SparseBlock
      Parameters:
      rlen - number of rows
      clen - number of columns
      nnz - number of non zeros
      strict - enforce optional properties
      Returns:
      true if the sparse block is valid wrt the corresponding format such as COO, CSR, MCSR.

    • contains

      public boolean contains(double pattern, int rl, int ru)
      Description copied from class: SparseBlock
      Checks if the block contains at least one value of the given pattern. Implementations need to handle NaN patterns as well (note that NaN==NaN yields false).
      Overrides:
      contains in class SparseBlock
      Parameters:
      pattern - checked pattern
      rl - row lower bound (inclusive)
      ru - row upper bound (exclusive)
      Returns:
      true if pattern appears at least once, otherwise false

    • getNonEmptyRowsIterator

      public Iterator<Integer> getNonEmptyRowsIterator(int rl, int ru)
      Specified by:
      getNonEmptyRowsIterator in class SparseBlock