com.nativelibs4java.opencl.util

Class ParallelRandom

java.lang.Object

com.nativelibs4java.opencl.util.ParallelRandom

public class ParallelRandom
extends Object

Parallel Random numbers generator (Xorshift adapted to work in parallel).
This class was designed as a drop-in replacement for java.util.Random (albeit with a more limited API) :

 ParallelRandom r = new ParallelRandom();
 r.setPreload(true); // faster
 while (true) {
     System.out.println(r.nextDouble());
 }
 

It is also possible to get entire batches of random integers with next() or next(Pointer).
The preload feature precomputes a new batch in background as soon as one starts to consume numbers from the current batch.

Author:

ochafik

Constructor Summary

Constructors

Constructor and Description
ParallelRandom()
ParallelRandom(CLQueue queue, int parallelSize, long seed)

Method Summary

Modifier and Type	Method and Description
CLEvent	doNext()
int	getParallelSize() Number of random numbers generated at each call of next() or next(Pointer) The numbers might not all be generated exactly in parallel, the level of parallelism is implementation-dependent.
CLQueue	getQueue()
CLBuffer<Integer>	getSeeds()
boolean	isPreload() If true, a new batch of parallel random numbers is automatically precomputed in background as soon as one starts to consume numbers from the current batch (this gives faster random numbers, at the risk of computing more values than needed)
Pointer<Integer>	next() Returns a direct NIO buffer containing the next getParallelSize() random integers.
void	next(Pointer<Integer> output) Copies the next getParallelSize() random integers in the provided output buffer
double	nextDouble()
float	nextFloat()
int	nextInt()
int	nextInt(int n)
long	nextLong()
void	setPreload(boolean preload) If true, a new batch of parallel random numbers is automatically precomputed in background as soon as one starts to consume numbers from the current batch (this gives faster random numbers, at the risk of computing more values than needed)

Methods inherited from class java.lang.Object

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

Constructor Detail

ParallelRandom

public ParallelRandom()
               throws IOException

Throws:

IOException

ParallelRandom

public ParallelRandom(CLQueue queue,
                      int parallelSize,
                      long seed)
               throws IOException

Throws:

IOException

Method Detail

nextInt

public int nextInt()

isPreload

public boolean isPreload()

If true, a new batch of parallel random numbers is automatically precomputed in background as soon as one starts to consume numbers from the current batch (this gives faster random numbers, at the risk of computing more values than needed)

setPreload

public void setPreload(boolean preload)
                throws CLBuildException

If true, a new batch of parallel random numbers is automatically precomputed in background as soon as one starts to consume numbers from the current batch (this gives faster random numbers, at the risk of computing more values than needed)

Throws:

CLBuildException

nextLong

public long nextLong()

nextInt

public int nextInt(int n)

nextFloat

public float nextFloat()

nextDouble

public double nextDouble()

getSeeds

public CLBuffer<Integer> getSeeds()

getQueue

public CLQueue getQueue()

getParallelSize

public int getParallelSize()

Number of random numbers generated at each call of next() or next(Pointer)
The numbers might not all be generated exactly in parallel, the level of parallelism is implementation-dependent.

Returns:

size of each buffer returned by next()

doNext

public CLEvent doNext()

next

public void next(Pointer<Integer> output)

Copies the next getParallelSize() random integers in the provided output buffer

Parameters:

output -

next

public Pointer<Integer> next()

Returns a direct NIO buffer containing the next getParallelSize() random integers.
This buffer is read only and will only be valid until any of the "next" method is called again.

Returns:

output buffer of capacity ; see getParallelSize()