Package org.apache.sysds.runtime.matrix.data

Class LibMatrixMult

java.lang.Object

org.apache.sysds.runtime.matrix.data.LibMatrixMult

public class LibMatrixMult
extends Object

MB: Library for matrix multiplications including MM, MV, VV for all combinations of dense, sparse, ultrasparse representations and special operations such as transpose-self matrix multiplication.

In general all implementations use internally dense outputs for direct access, but change the final result to sparse if necessary. The only exceptions are ultra-sparse matrix mult, wsloss and wsigmoid.

Field Summary

Fields

Modifier and Type	Field	Description
static int	L2_CACHESIZE
static int	L3_CACHESIZE

Method Summary

Modifier and Type	Method	Description
static long	copyUpperToLowerTriangle(MatrixBlock ret)
static long	copyUpperToLowerTriangleDense(MatrixBlock ret)	Used for all version of TSMM where the result is known to be symmetric.
static long	copyUpperToLowerTriangleSparse(MatrixBlock ret)
static double	dotProduct(double[] a, double[] b, int[] aix, int ai, int bi, int len)
static double	dotProduct(double[] a, double[] b, int ai, int bi, int len)
static MatrixBlock	emptyMatrixMult(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret)
static boolean	isOuterProductTSMM(int rlen, int clen, boolean left)
static boolean	isSkinnyRightHandSide(long m1rlen, long m1clen, long m2rlen, long m2clen, boolean inclCacheSize)
static boolean	isSparseOutputMatrixMult(MatrixBlock m1, MatrixBlock m2)
static boolean	isSparseOutputTSMM(MatrixBlock m1)
static boolean	isSparseOutputTSMM(MatrixBlock m1, boolean ultraSparse)
static boolean	isUltraSparseMatrixMult(MatrixBlock m1, MatrixBlock m2, boolean m1Perm)
static MatrixBlock	matrixMult(MatrixBlock m1, MatrixBlock m2)	Performs a matrix multiplication All variants use a IKJ access pattern, and internally use dense output.
static MatrixBlock	matrixMult(MatrixBlock m1, MatrixBlock m2, int k)	Performs a matrix multiplication All variants use a IKJ access pattern, and internally use dense output.
static MatrixBlock	matrixMult(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret)	Performs a matrix multiplication and stores the result in the output matrix.
static MatrixBlock	matrixMult(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret, boolean fixedRet)	This method allows one to disabling exam sparsity.
static MatrixBlock	matrixMult(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret, boolean fixedRet, int k)	Performs a matrix multiplication and stores the result in the output matrix.
static MatrixBlock	matrixMult(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret, int k)	Performs a multi-threaded matrix multiplication and stores the result in the output matrix.
static void	matrixMultChain(MatrixBlock mX, MatrixBlock mV, MatrixBlock mW, MatrixBlock ret, MapMultChain.ChainType ct)	Performs a matrix multiplication chain operation of type t(X)%*%(X%*%v) or t(X)%*%(w*(X%*%v)).
static void	matrixMultChain(MatrixBlock mX, MatrixBlock mV, MatrixBlock mW, MatrixBlock ret, MapMultChain.ChainType ct, int k)	Performs a parallel matrix multiplication chain operation of type t(X)%*%(X%*%v) or t(X)%*%(w*(X%*%v)).
static void	matrixMultDenseDenseMM(DenseBlock a, DenseBlock b, DenseBlock c, int n, int cd, int rl, int ru, int cl, int cu)
static void	matrixMultDenseDenseMMDedup(DenseBlockFP64DEDUP a, DenseBlock b, DenseBlockFP64DEDUP c, int n, int cd, int rl, int ru, ConcurrentHashMap<double[],double[]> cache)
static void	matrixMultPermute(MatrixBlock pm1, MatrixBlock m2, MatrixBlock ret1, MatrixBlock ret2)
static void	matrixMultPermute(MatrixBlock pm1, MatrixBlock m2, MatrixBlock ret1, MatrixBlock ret2, int k)
static void	matrixMultTransposeSelf(MatrixBlock m1, MatrixBlock ret, boolean leftTranspose)
static void	matrixMultTransposeSelf(MatrixBlock m1, MatrixBlock ret, boolean leftTranspose, boolean copyToLowerTriangle)
static void	matrixMultTransposeSelf(MatrixBlock m1, MatrixBlock ret, boolean leftTranspose, int k)	TSMM with optional transposed left side or not (Transposed self matrix multiplication)
static void	matrixMultWCeMM(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, double eps, MatrixBlock ret, WeightedCrossEntropy.WCeMMType wt)
static void	matrixMultWCeMM(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, double eps, MatrixBlock ret, WeightedCrossEntropy.WCeMMType wt, int k)
static void	matrixMultWDivMM(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock mX, MatrixBlock ret, WeightedDivMM.WDivMMType wt)	NOTE: This operation has limited NaN support, which is acceptable because all our sparse-safe operations have only limited NaN support.
static void	matrixMultWDivMM(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock mX, MatrixBlock ret, WeightedDivMM.WDivMMType wt, int k)	NOTE: This operation has limited NaN support, which is acceptable because all our sparse-safe operations have only limited NaN support.
static void	matrixMultWSigmoid(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock ret, WeightedSigmoid.WSigmoidType wt)
static void	matrixMultWSigmoid(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock ret, WeightedSigmoid.WSigmoidType wt, int k)
static void	matrixMultWSLoss(MatrixBlock mX, MatrixBlock mU, MatrixBlock mV, MatrixBlock mW, MatrixBlock ret, WeightedSquaredLoss.WeightsType wt)
static void	matrixMultWSLoss(MatrixBlock mX, MatrixBlock mU, MatrixBlock mV, MatrixBlock mW, MatrixBlock ret, WeightedSquaredLoss.WeightsType wt, int k)
static void	matrixMultWuMM(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock ret, WeightedUnaryMM.WUMMType wt, ValueFunction fn)
static void	matrixMultWuMM(MatrixBlock mW, MatrixBlock mU, MatrixBlock mV, MatrixBlock ret, WeightedUnaryMM.WUMMType wt, ValueFunction fn, int k)
static MatrixBlock	prepMatrixMultTransposeSelfInput(MatrixBlock m1, boolean leftTranspose, boolean par)
static boolean	satisfiesMultiThreadingConstraints(MatrixBlock m1, boolean checkMem, boolean checkFLOPs, long FPfactor, int k)
static boolean	satisfiesMultiThreadingConstraints(MatrixBlock m1, int k)
static boolean	satisfiesMultiThreadingConstraints(MatrixBlock m1, MatrixBlock m2, boolean checkMem, boolean checkFLOPs, long FPfactor, int k)
static void	vectAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen)
static void	vectAdd(double[] a, double[] c, int ai, int ci, int len)
static void	vectAdd(double[] a, double bval, double[] c, int ai, int ci, int len)
static void	vectAddInPlace(double aval, double[] c, int ci, int len)
static void	vectMultiplyAdd(double aval, double[] b, double[] c, int[] bix, int bi, int ci, int len)
static void	vectMultiplyAdd(double aval, double[] b, double[] c, int bi, int ci, int len)
static void	vectMultiplyInPlace(double aval, double[] c, int ci, int len)
static void	vectMultiplyWrite(double[] a, double[] b, double[] c, int[] bix, int ai, int bi, int ci, int len)
static void	vectMultiplyWrite(double[] a, double[] b, double[] c, int ai, int bi, int ci, int len)
static void	vectMultiplyWrite(double aval, double[] b, double[] c, int bi, int ci, int len)

Methods inherited from class java.lang.Object

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

Field Detail

L2_CACHESIZE

public static final int L2_CACHESIZE

See Also:

Constant Field Values

L3_CACHESIZE

public static final int L3_CACHESIZE

See Also:

Constant Field Values

Method Detail

matrixMult

public static MatrixBlock matrixMult(MatrixBlock m1,
                                     MatrixBlock m2)

Performs a matrix multiplication All variants use a IKJ access pattern, and internally use dense output. After the actual computation, we recompute nnz and check for sparse/dense representation.

Parameters:

m1 - first matrix

m2 - second matrix

Returns:

ret Matrix Block

matrixMult

public static MatrixBlock matrixMult(MatrixBlock m1,
                                     MatrixBlock m2,
                                     int k)

Performs a matrix multiplication All variants use a IKJ access pattern, and internally use dense output. After the actual computation, we recompute nnz and check for sparse/dense representation.

Parameters:

m1 - first matrix

m2 - second matrix

k - maximum parallelism

Returns:

ret Matrix Block

matrixMult

public static MatrixBlock matrixMult(MatrixBlock m1,
                                     MatrixBlock m2,
                                     MatrixBlock ret)

Performs a matrix multiplication and stores the result in the output matrix. All variants use a IKJ access pattern, and internally use dense output. After the actual computation, we recompute nnz and check for sparse/dense representation.

Parameters:

m1 - first matrix

m2 - second matrix

ret - result matrix

Returns:

ret Matrix Block

matrixMult

public static MatrixBlock matrixMult(MatrixBlock m1,
                                     MatrixBlock m2,
                                     MatrixBlock ret,
                                     boolean fixedRet)

This method allows one to disabling exam sparsity. This feature is useful if matrixMult is used as an intermediate operation (for example: LibMatrixDNN). It makes sense for LibMatrixDNN because the output is internally consumed by another dense instruction, which makes repeated conversion to sparse wasteful. This should be used in rare cases and if you are unsure, use the method 'matrixMult(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret)' instead.

Parameters:

m1 - first matrix

m2 - second matrix

ret - result matrix

fixedRet - if true, output representation is fixed and nnzs not recomputed

Returns:

ret Matrix Block

matrixMult

public static MatrixBlock matrixMult(MatrixBlock m1,
                                     MatrixBlock m2,
                                     MatrixBlock ret,
                                     int k)

Performs a multi-threaded matrix multiplication and stores the result in the output matrix. The parameter k (k>=1) determines the max parallelism k' with k'=min(k, vcores, m1.rlen).

Parameters:

m1 - first matrix

m2 - second matrix

ret - result matrix

k - maximum parallelism

Returns:

ret Matrix Block

matrixMult

public static MatrixBlock matrixMult(MatrixBlock m1,
                                     MatrixBlock m2,
                                     MatrixBlock ret,
                                     boolean fixedRet,
                                     int k)

Performs a matrix multiplication and stores the result in the output matrix. All variants use a IKJ access pattern, and internally use dense output. After the actual computation, we recompute nnz and check for sparse/dense representation. This method allows one to disabling exam sparsity. This feature is useful if matrixMult is used as an intermediate operation (for example: LibMatrixDNN). It makes sense for LibMatrixDNN because the output is internally consumed by another dense instruction, which makes repeated conversion to sparse wasteful. This should be used in rare cases and if you are unsure, use the method 'matrixMult(MatrixBlock m1, MatrixBlock m2, MatrixBlock ret)' instead. The parameter k (k>=1) determines the max parallelism k' with k'=min(k, vcores, m1.rlen).

Parameters:

m1 - first matrix

m2 - second matrix

ret - result matrix

fixedRet - if true, output representation is fixed and nnzs not recomputed

k - maximum parallelism

Returns:

ret Matrix Block

emptyMatrixMult

public static MatrixBlock emptyMatrixMult(MatrixBlock m1,
                                          MatrixBlock m2,
                                          MatrixBlock ret)

matrixMultChain

public static void matrixMultChain(MatrixBlock mX,
                                   MatrixBlock mV,
                                   MatrixBlock mW,
                                   MatrixBlock ret,
                                   MapMultChain.ChainType ct)

Performs a matrix multiplication chain operation of type t(X)%*%(X%*%v) or t(X)%*%(w*(X%*%v)). All variants use a IKJ access pattern, and internally use dense output. After the actual computation, we recompute nnz and check for sparse/dense representation.

Parameters:

mX - X matrix

mV - v matrix

mW - w matrix

ret - result matrix

ct - chain type

matrixMultChain

public static void matrixMultChain(MatrixBlock mX,
                                   MatrixBlock mV,
                                   MatrixBlock mW,
                                   MatrixBlock ret,
                                   MapMultChain.ChainType ct,
                                   int k)

Performs a parallel matrix multiplication chain operation of type t(X)%*%(X%*%v) or t(X)%*%(w*(X%*%v)). The parameter k (k>=1) determines the max parallelism k' with k'=min(k, vcores, m1.rlen). NOTE: This multi-threaded mmchain operation has additional memory requirements of k*ncol(X)*8bytes for partial aggregation. Current max memory: 256KB; otherwise redirect to sequential execution.

Parameters:

mX - X matrix

mV - v matrix

mW - w matrix

ret - result matrix

ct - chain type

k - maximum parallelism

matrixMultTransposeSelf

public static void matrixMultTransposeSelf(MatrixBlock m1,
                                           MatrixBlock ret,
                                           boolean leftTranspose)

matrixMultTransposeSelf

public static void matrixMultTransposeSelf(MatrixBlock m1,
                                           MatrixBlock ret,
                                           boolean leftTranspose,
                                           boolean copyToLowerTriangle)

matrixMultTransposeSelf

public static void matrixMultTransposeSelf(MatrixBlock m1,
                                           MatrixBlock ret,
                                           boolean leftTranspose,
                                           int k)

TSMM with optional transposed left side or not (Transposed self matrix multiplication)

Parameters:

m1 - The matrix to do tsmm

ret - The output matrix to allocate the result to

leftTranspose - If the left side should be considered transposed

k - the number of threads to use

matrixMultPermute

public static void matrixMultPermute(MatrixBlock pm1,
                                     MatrixBlock m2,
                                     MatrixBlock ret1,
                                     MatrixBlock ret2)

matrixMultPermute

public static void matrixMultPermute(MatrixBlock pm1,
                                     MatrixBlock m2,
                                     MatrixBlock ret1,
                                     MatrixBlock ret2,
                                     int k)

matrixMultWSLoss

public static void matrixMultWSLoss(MatrixBlock mX,
                                    MatrixBlock mU,
                                    MatrixBlock mV,
                                    MatrixBlock mW,
                                    MatrixBlock ret,
                                    WeightedSquaredLoss.WeightsType wt)

matrixMultWSLoss

public static void matrixMultWSLoss(MatrixBlock mX,
                                    MatrixBlock mU,
                                    MatrixBlock mV,
                                    MatrixBlock mW,
                                    MatrixBlock ret,
                                    WeightedSquaredLoss.WeightsType wt,
                                    int k)

matrixMultWSigmoid

public static void matrixMultWSigmoid(MatrixBlock mW,
                                      MatrixBlock mU,
                                      MatrixBlock mV,
                                      MatrixBlock ret,
                                      WeightedSigmoid.WSigmoidType wt)

matrixMultWSigmoid

public static void matrixMultWSigmoid(MatrixBlock mW,
                                      MatrixBlock mU,
                                      MatrixBlock mV,
                                      MatrixBlock ret,
                                      WeightedSigmoid.WSigmoidType wt,
                                      int k)

matrixMultWDivMM

public static void matrixMultWDivMM(MatrixBlock mW,
                                    MatrixBlock mU,
                                    MatrixBlock mV,
                                    MatrixBlock mX,
                                    MatrixBlock ret,
                                    WeightedDivMM.WDivMMType wt)

NOTE: This operation has limited NaN support, which is acceptable because all our sparse-safe operations have only limited NaN support. If this is not intended behavior, please disable the rewrite. In detail, this operator will produce for W/(U%*%t(V)) a zero intermediate for each zero in W (even if UVij is zero which would give 0/0=NaN) but INF/-INF for non-zero entries in V where the corresponding cell in (Y%*%X) is zero.

Parameters:

mW - matrix W

mU - matrix U

mV - matrix V

mX - matrix X

ret - result type

wt - weighted divide matrix multiplication type

matrixMultWDivMM

public static void matrixMultWDivMM(MatrixBlock mW,
                                    MatrixBlock mU,
                                    MatrixBlock mV,
                                    MatrixBlock mX,
                                    MatrixBlock ret,
                                    WeightedDivMM.WDivMMType wt,
                                    int k)

NOTE: This operation has limited NaN support, which is acceptable because all our sparse-safe operations have only limited NaN support. If this is not intended behavior, please disable the rewrite. In detail, this operator will produce for W/(U%*%t(V)) a zero intermediate for each zero in W (even if UVij is zero which would give 0/0=NaN) but INF/-INF for non-zero entries in V where the corresponding cell in (Y%*%X) is zero.

Parameters:

mW - matrix W

mU - matrix U

mV - matrix V

mX - matrix X

ret - result matrix

wt - weighted divide matrix multiplication type

k - maximum parallelism

matrixMultWCeMM

public static void matrixMultWCeMM(MatrixBlock mW,
                                   MatrixBlock mU,
                                   MatrixBlock mV,
                                   double eps,
                                   MatrixBlock ret,
                                   WeightedCrossEntropy.WCeMMType wt)

matrixMultWCeMM

public static void matrixMultWCeMM(MatrixBlock mW,
                                   MatrixBlock mU,
                                   MatrixBlock mV,
                                   double eps,
                                   MatrixBlock ret,
                                   WeightedCrossEntropy.WCeMMType wt,
                                   int k)

matrixMultWuMM

public static void matrixMultWuMM(MatrixBlock mW,
                                  MatrixBlock mU,
                                  MatrixBlock mV,
                                  MatrixBlock ret,
                                  WeightedUnaryMM.WUMMType wt,
                                  ValueFunction fn)

matrixMultWuMM

public static void matrixMultWuMM(MatrixBlock mW,
                                  MatrixBlock mU,
                                  MatrixBlock mV,
                                  MatrixBlock ret,
                                  WeightedUnaryMM.WUMMType wt,
                                  ValueFunction fn,
                                  int k)

matrixMultDenseDenseMMDedup

public static void matrixMultDenseDenseMMDedup(DenseBlockFP64DEDUP a,
                                               DenseBlock b,
                                               DenseBlockFP64DEDUP c,
                                               int n,
                                               int cd,
                                               int rl,
                                               int ru,
                                               ConcurrentHashMap<double[],double[]> cache)

matrixMultDenseDenseMM

public static void matrixMultDenseDenseMM(DenseBlock a,
                                          DenseBlock b,
                                          DenseBlock c,
                                          int n,
                                          int cd,
                                          int rl,
                                          int ru,
                                          int cl,
                                          int cu)

dotProduct

public static double dotProduct(double[] a,
                                double[] b,
                                int ai,
                                int bi,
                                int len)

dotProduct

public static double dotProduct(double[] a,
                                double[] b,
                                int[] aix,
                                int ai,
                                int bi,
                                int len)

vectMultiplyAdd

public static void vectMultiplyAdd(double aval,
                                   double[] b,
                                   double[] c,
                                   int bi,
                                   int ci,
                                   int len)

vectMultiplyAdd

public static void vectMultiplyAdd(double aval,
                                   double[] b,
                                   double[] c,
                                   int[] bix,
                                   int bi,
                                   int ci,
                                   int len)

vectMultiplyWrite

public static void vectMultiplyWrite(double aval,
                                     double[] b,
                                     double[] c,
                                     int bi,
                                     int ci,
                                     int len)

vectMultiplyInPlace

public static void vectMultiplyInPlace(double aval,
                                       double[] c,
                                       int ci,
                                       int len)

vectMultiplyWrite

public static void vectMultiplyWrite(double[] a,
                                     double[] b,
                                     double[] c,
                                     int ai,
                                     int bi,
                                     int ci,
                                     int len)

vectMultiplyWrite

public static void vectMultiplyWrite(double[] a,
                                     double[] b,
                                     double[] c,
                                     int[] bix,
                                     int ai,
                                     int bi,
                                     int ci,
                                     int len)

vectAdd

public static void vectAdd(double[] a,
                           double bval,
                           double[] c,
                           int ai,
                           int ci,
                           int len)

vectAdd

public static void vectAdd(double[] a,
                           double[] c,
                           int ai,
                           int ci,
                           int len)

vectAdd

public static void vectAdd(double[] a,
                           double[] c,
                           int[] aix,
                           int ai,
                           int ci,
                           int alen)

vectAddInPlace

public static void vectAddInPlace(double aval,
                                  double[] c,
                                  int ci,
                                  int len)

copyUpperToLowerTriangle

public static long copyUpperToLowerTriangle(MatrixBlock ret)

copyUpperToLowerTriangleDense

public static long copyUpperToLowerTriangleDense(MatrixBlock ret)

Used for all version of TSMM where the result is known to be symmetric. Hence, we compute only the upper triangular matrix and copy this partial result down to lower triangular matrix once.

Parameters:

ret - matrix

Returns:

number of non zeros

copyUpperToLowerTriangleSparse

public static long copyUpperToLowerTriangleSparse(MatrixBlock ret)

prepMatrixMultTransposeSelfInput

public static MatrixBlock prepMatrixMultTransposeSelfInput(MatrixBlock m1,
                                                           boolean leftTranspose,
                                                           boolean par)

isSkinnyRightHandSide

public static boolean isSkinnyRightHandSide(long m1rlen,
                                            long m1clen,
                                            long m2rlen,
                                            long m2clen,
                                            boolean inclCacheSize)

satisfiesMultiThreadingConstraints

public static boolean satisfiesMultiThreadingConstraints(MatrixBlock m1,
                                                         int k)

satisfiesMultiThreadingConstraints

public static boolean satisfiesMultiThreadingConstraints(MatrixBlock m1,
                                                         boolean checkMem,
                                                         boolean checkFLOPs,
                                                         long FPfactor,
                                                         int k)

satisfiesMultiThreadingConstraints

public static boolean satisfiesMultiThreadingConstraints(MatrixBlock m1,
                                                         MatrixBlock m2,
                                                         boolean checkMem,
                                                         boolean checkFLOPs,
                                                         long FPfactor,
                                                         int k)

isUltraSparseMatrixMult

public static boolean isUltraSparseMatrixMult(MatrixBlock m1,
                                              MatrixBlock m2,
                                              boolean m1Perm)

isSparseOutputMatrixMult

public static boolean isSparseOutputMatrixMult(MatrixBlock m1,
                                               MatrixBlock m2)

isSparseOutputTSMM

public static boolean isSparseOutputTSMM(MatrixBlock m1)

isSparseOutputTSMM

public static boolean isSparseOutputTSMM(MatrixBlock m1,
                                         boolean ultraSparse)

isOuterProductTSMM

public static boolean isOuterProductTSMM(int rlen,
                                         int clen,
                                         boolean left)