SparseBlockCSR (SystemML 0.14.0 API)

java.lang.Object
- org.apache.sysml.runtime.matrix.data.SparseBlock
- - org.apache.sysml.runtime.matrix.data.SparseBlockCSR

All Implemented Interfaces:

Serializable
```
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:

Serialized Form

Nested Class Summary
- Nested classes/interfaces inherited from class org.apache.sysml.runtime.matrix.data.SparseBlock
  SparseBlock.Type

Field Summary
- Fields inherited from class org.apache.sysml.runtime.matrix.data.SparseBlock
  INIT_CAPACITY, RESIZE_FACTOR1, RESIZE_FACTOR2

Constructor Summary

Constructors
Constructor and 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(SparseBlock sblock)` Copy constructor sparse block abstraction.
`SparseBlockCSR(SparseRow[] rows, int nnz)` Copy constructor old sparse row representation.

Method Summary

All Methods Static Methods Instance Methods Concrete Methods
Modifier and Type	Method and Description
`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.
`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`	`estimateMemory(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).
`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.
`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.
`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 vlen)` Sets a sorted 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.sysml.runtime.matrix.data.SparseBlock
getIterator, getIterator, getIterator, isAligned, isAligned

Methods inherited from class java.lang.Object
clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

- Constructor Detail
  - SparseBlockCSR
```
public SparseBlockCSR(int rlen)
```
  - SparseBlockCSR
```
public SparseBlockCSR(int rlen,
                      int capacity)
```
  - 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
- Method Detail
  - estimateMemory
```
public static long estimateMemory(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
  - 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
  - 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
  - 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:
    
    ?
  - 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 sorted 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 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 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. NOTE: This method exists for incremental runtime integration and might be deleted in the future.
    
    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 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.
    
    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
  - 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

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

Fields inherited from class org.apache.sysml.runtime.matrix.data.SparseBlock

Methods inherited from class org.apache.sysml.runtime.matrix.data.SparseBlock

Methods inherited from class java.lang.Object

SparseBlockCSR

SparseBlockCSR

SparseBlockCSR

SparseBlockCSR

SparseBlockCSR

SparseBlockCSR

estimateMemory

allocate

allocate

allocate

numRows

isThreadSafe

isContiguous

reset

reset

reset

size

size

size

size

isEmpty

indexes

values

pos

set

set

append

setIndexRange

setIndexRange

setIndexRange

deleteIndexRange

sort

sort

get

get

posFIndexLTE

posFIndexGTE

posFIndexGT

toString

rowPointers

indexes

values