de.jungblut.math.cuda

Class JCUDAMatrixUtils

java.lang.Object

de.jungblut.math.cuda.JCUDAMatrixUtils

public final class JCUDAMatrixUtils
extends Object

Matrix utilities for CUDA graphics card greater version 400, e.g. Nvidia 480gtx.
-Djava.library.path="/lib/;${env_var:PATH}" must be added to the running VM. If you have the cuda libs in the /lib folder or under windows in your path variables.

Author:

thomas.jungblut

Field Summary

Fields

Modifier and Type	Field and Description
static boolean	CUBLAS2_AVAILABLE
static boolean	EXCEPTIONS_ENABLED

Constructor Summary

Constructors

Constructor and Description
JCUDAMatrixUtils()

Method Summary

Modifier and Type	Method and Description
static void	freePointer(jcuda.Pointer p) Frees the given pointer.
static de.jungblut.math.dense.DenseDoubleMatrix	getMatrix(jcuda.Pointer src, int rows, int columns) Read a matrix from device memory.
static void	main(String[] args)
static jcuda.Pointer	memcpyMatrix(de.jungblut.math.dense.DenseDoubleMatrix a) Copies the given matrix to the device memory in column major format.
static de.jungblut.math.dense.DenseDoubleMatrix	multiply(de.jungblut.math.dense.DenseDoubleMatrix a, de.jungblut.math.dense.DenseDoubleMatrix b) Multiplies matrix A with matrix B and returns a new matrix.
static de.jungblut.math.dense.DenseDoubleMatrix	multiply(de.jungblut.math.dense.DenseDoubleMatrix a, de.jungblut.math.dense.DenseDoubleMatrix b, boolean transposeA, boolean transposeB) Multiplies matrix a with matrix b and returns a new matrix.
static de.jungblut.math.dense.DenseDoubleMatrix	multiply(jcuda.Pointer a, jcuda.Pointer b, MatrixDimension dim) Multiplies matrix A with matrix B (these are pointers, thus the dimension must be passed and returns a new matrix.

Methods inherited from class java.lang.Object

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

Field Detail

EXCEPTIONS_ENABLED

public static boolean EXCEPTIONS_ENABLED

CUBLAS2_AVAILABLE

public static boolean CUBLAS2_AVAILABLE

Constructor Detail

JCUDAMatrixUtils

public JCUDAMatrixUtils()

Method Detail

multiply

public static de.jungblut.math.dense.DenseDoubleMatrix multiply(de.jungblut.math.dense.DenseDoubleMatrix a,
                                                                de.jungblut.math.dense.DenseDoubleMatrix b)

Multiplies matrix A with matrix B and returns a new matrix.

multiply

public static de.jungblut.math.dense.DenseDoubleMatrix multiply(jcuda.Pointer a,
                                                                jcuda.Pointer b,
                                                                MatrixDimension dim)

Multiplies matrix A with matrix B (these are pointers, thus the dimension must be passed and returns a new matrix.

multiply

public static de.jungblut.math.dense.DenseDoubleMatrix multiply(de.jungblut.math.dense.DenseDoubleMatrix a,
                                                                de.jungblut.math.dense.DenseDoubleMatrix b,
                                                                boolean transposeA,
                                                                boolean transposeB)

Multiplies matrix a with matrix b and returns a new matrix. You can add transpose flags for both matrices.

memcpyMatrix

public static jcuda.Pointer memcpyMatrix(de.jungblut.math.dense.DenseDoubleMatrix a)

Copies the given matrix to the device memory in column major format.

Returns:

a pointer to this matrix.

getMatrix

public static de.jungblut.math.dense.DenseDoubleMatrix getMatrix(jcuda.Pointer src,
                                                                 int rows,
                                                                 int columns)

Read a matrix from device memory.

Parameters:

src - the head pointer to the matrix.

rows - the number of rows.

columns - the number of columns

Returns:

a new matrix with the results from device.

freePointer

public static void freePointer(jcuda.Pointer p)

Frees the given pointer.

Parameters:

p - the pointer to free

main

public static void main(String[] args)