org.apache.flink.streaming.examples.gpu

Class MatrixVectorMul

java.lang.Object

org.apache.flink.streaming.examples.gpu.MatrixVectorMul

public class MatrixVectorMul
extends Object

Implements the matrix-vector multiplication program that shows how to use GPU resources in Flink.

The input is a vector stream from a RandomVectorSource, which will generate random vectors with specified dimension. The data size of the vector stream could be specified by user. Each vector will be multiplied with a random dimension * dimension matrix in Multiplier and the result would be emitted to output.

Usage: MatrixVectorMul [--output <path>] [--dimension <dimension> --data-size <data_size>]

If no parameters are provided, the program is run with default vector dimension 10 and data size 100.

This example shows how to:

• leverage external resource in operators,
• accelerate complex calculation with GPU resources.

Notice that you need to add JCuda natives libraries in your Flink distribution by the following steps:

• download the JCuda native libraries bundle for your CUDA version from http://www.jcuda.org/downloads/
• copy the native libraries jcuda-natives and jcublas-natives for your CUDA version, operating system and architecture to the "lib/" folder of your Flink distribution

Constructor Summary

Constructors

Constructor and Description
MatrixVectorMul()

Method Summary

Modifier and Type	Method and Description
static void	main(String[] args)

Methods inherited from class java.lang.Object

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

Constructor Detail

MatrixVectorMul

public MatrixVectorMul()

Method Detail

main

public static void main(String[] args)
                 throws Exception

Throws:

Exception