Package org.apache.sysds.runtime.compress.lib

Class CLALibMMChain

java.lang.Object

org.apache.sysds.runtime.compress.lib.CLALibMMChain

public final class CLALibMMChain
extends Object

Support compressed MM chain operation to fuse the following cases :

XtXv == (t(X) %*% (X %*% v))

XtwXv == (t(X) %*% (w * (X %*% v)))

XtXvy == (t(X) %*% ((X %*% v) - y))

Method Summary

Modifier and Type	Method	Description
static MatrixBlock	mmChain(CompressedMatrixBlock x, MatrixBlock v, MatrixBlock w, MatrixBlock out, MapMultChain.ChainType ctype, int k)	Support compressed MM chain operation to fuse the following cases :

Methods inherited from class java.lang.Object

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

Method Detail

mmChain

public static MatrixBlock mmChain(CompressedMatrixBlock x,
                                  MatrixBlock v,
                                  MatrixBlock w,
                                  MatrixBlock out,
                                  MapMultChain.ChainType ctype,
                                  int k)

Support compressed MM chain operation to fuse the following cases :

XtXv == (t(X) %*% (X %*% v))

XtwXv == (t(X) %*% (w * (X %*% v)))

XtXvy == (t(X) %*% ((X %*% v) - y))

Note the point of this optimization is that v and w always are vectors. This means in practice the all the compute is faster if the intermediates are exploited.

Parameters:

x - Is the X part of the chain optimized kernel

v - Is the mandatory v part of the chain

w - Is the optional w port of t the chain

out - The output to put the result into. Can also be returned and in some cases will not be used.

ctype - either XtwXv, XtXv or XtXvy

k - the parallelization degree

Returns:

The result either in the given output or a new allocation