Reference Page

From CSE231 Wiki
Jump to: navigation, search

This is a collection of all the the different tools we use in this course and their syntax.

While this page is here to assist you whenever you forget how to create a lambda or register a phaser, try not to become too reliant on it. If you learn and understand the syntax for the tools we build in this class, it will make the assignments easier and faster to get through. Also, this page will not be available during exams, and while there is always leniency with written code, there must be some demonstration of understanding how to use our parallel constructs!

Java Basics

Sometimes the problem isn't with doing things in parallel, but just getting Java to cooperate with what you want it to do. This section should hopefully help you figure out how to use the Java concepts needed to complete the assignments in the class.

This page mainly just covers syntax though. For a better understanding of how some of these tools actually function and why we use them, check out Rundown on Everything Java.

Object Constructors

There will be a lot of classes for students to make/complete throughout the semester. Many times, you will be responsible for filling out the constructor. If given something like this:

private int value;

public Foo(int parameter){
	//TO DO
}

While all sorts of code and commands may be used in the constructor, before it is over, all of the class's private variables should have been assigned.

public Foo(int parameter){
	this.value = parameter;
}

Then, outside of this class, you can create an object of type Foo by doing the following:

Foo userMadeObject = new Foo(1);


Ternary Operators

A handy and quick way to do things based on a boolean operation. The two blocks of code are equivalent.

if(testCondition){
	int finalResult = valueIfTrue;
} else{
	int finalResult = valueIfFalse;
}
int finalResult = (testCondition) ? valueIfTrue : valueIfFalse

For-each Loop

For loops don't always need to go from one integer to the next. In fact, if we have any iterable object (array, list, set, etc.), we can just loop through all the entries available in that object. Use this to make your life much simpler!

List<T> filledList;
for(T singleEntry: filledList){
	doSomething(singleEntry);
}

Runtime

In order to get the number of processors your computer has:

int numProcessors = Runtime.getRuntime().availableProcessors();

Generics

Check out the dedicated Generics page for a deep dive on how Java's generic types work!

Interfaces

Check out the dedicated Interfaces page for an in-depth discussion on interfaces in Java!

Lambda

Lambdas are very complex things, and they are also incredibly important when it comes to understanding how our code works in this course. If you are looking for more information on what lambdas actually are, go check out the Lambdas wiki page, or the lambdas section in the Rundown on Everything Java page!

The syntax for a lambda can be seen as:

(T parameter) -> {
	doSomething();
}

Within the curly braces, you will have access to that parameter variable. Most times you use lambdas though, there are no parameters to pass through, so you can just do

() -> {
	doSomething();
}

It's strongly encouraged to check out the Eclipse tip on how to use content assist lambdas.

Anonymous Classes

We use anonymous classes throughout the course as a convenient way to instantiate objects of classes that have methods we need to define. There are two ways to do this, one manually labeling the method we're defining, and one with lambda. These are two similar creations of a Function object (which has the method "apply") that adds one to whatever number is passed in:

Function<Integer, Integer> increment = new Function<Integer, Integer>{
	public Integer apply(Integer oldValue){
		return oldValue + 1;
	}
};
Function<Integer, Integer> increment = (oldValue) -> {
	return oldValue + 1;
};

Since the class we are defining only has one method, if we use a lambda, it knows that the lambda refers to that one specific method. Check out the section on [Lambdas#Anonymous_Classes|Anonymous Classes] for more information!

Runnable

A Java Runnable is an object with just a .run() method. As discussed above, that means we can instantiate a Runnable using an anonymous class created by a lambda.

Runnable body = () -> {
	doSomething();
}
body.run();

Callable

A Java Callable is very similar to a Runnable. It is an object with only a .call() method. A key difference between this and a Runnable is that a Callable has a return value when called.

Callable<V> body = () -> {
	doSomething();
	return foo;
}
V result = body.call();

Java Basic Util Data Structures

Java's Data Structures are a core part of this class, as we will be constantly working with data and making our own data. In no time at all will you become very familiar with some of the more useful data structures!

Remember that mainly syntax is shown here, with Rundown on Everything Java being the best place to go for more detailed information on certain topics.

Lists

Many times you will need to make some kind of list to keep information in. The two we recommend using are Linked Lists and Array Lists.

List<T> listA = new LinkedList<>();
List<T> listB = new ArrayList<>();

List Iterator

All lists (and many other objects) have iterators we can use to easily traverse through the list.

List<T> list;
ListIterator<T> iterator = list.listIterator();
while(iterator.hasNext){
	T itemA = iterator.next();
	doSomething(itemA);
}
while(iterator.hasPrevious){
	T itemB = iterator.previous();
	doSomething(itemB);
}

Maps

Many times you will need to make some kind of map to store information in. Depending on the race conditions of the assignment, you usually will use Hash Maps or Concurrent Hash Maps.

Map<K,V> map = new HashMap<>();

Compute

A very important application of a map is its compute method. This allows us to change a value in a map rather than having to remove it and add it back. Here is the code for the default implementation of the compute method, taken from the Map JavaDocs.

V oldValue = map.get(key);
V newValue = remappingFunction.apply(key, oldValue);
if (oldValue != null ) {
	if (newValue != null)
		map.put(key, newValue);
	else
		map.remove(key);
} else {
	if (newValue != null)
		map.put(key, newValue);
	else
		return null;
}

Most students tend to get tripped up on the use of a BiFunction. This BiFunction is essential giving the map a formula to use in order to get the new value it needs to set in the map. If we wanted to call compute on a key and increase its value by one (assuming the value is an integer), it would look like this:

map.compute(keyName, (key, value) -> {
	return value + 1;
}

Note that the BiFunction is simply given as a lambda with two parameters passed in. Also important is that this lambda has a return value. Whatever is returned is what the new value in the map will become.

Other methods in Map that you may find helpful are computeIfAbsent, merge, and replace. Check out the Map JavaDocs for more detailed information on how those methods work.

Entries

In a lot of assignments, you will need to make entries of different types. Almost always, the KeyValuePair<K,V> or KeyMutableValuePair<K,V> will be your best options.

KeyValuePair<String, int> newEntry = new KeyValuePair<>("A", 1);

Looping through entries is also very easy to do!

Map<K,V> map = mapFilledWithEntries();
for(Entry<K,V> entry: map.entrySet()){
	K entryKey = entry.getKey();
	V entryValue = entry.getValue();
	doSomething(entryKey, entryValue);
}

V5/Habanero

Habanero is Rice's collection of parallel tools. When you watch the RiceX videos, it is the Habanero version of things you will be seeing. For this course, we've designed V5. It is similar to Habanero in most ways, but has small changes to hopefully make your lives easier when it comes to using the tools in this class. Listed below is the syntax for V5's version, but the differences between the two versions are noted to help clarify.

Async

Asyncs are a fundamental building block of task level parallelism. The execution of an async will spawn a new child task, and any code written within the braces of an async will be done in that new task parallel to the parent task. The actual async method is called like this: async(Runnable body)

But as discussed above in the Runnable section, a Runnable can be defined just using a lambda. For the runnables in asyncs, there are no parameters passed in, so an async with the lambda is as follows:

async( () -> {
	doSomething();
});
doSomethingElse();

In the above example, both doSomething() and doSomethingElse() are being done at the same time as one another.

It is also worth noting that with V5, we can use asyncs without having a new task spawn. This can be really helpful when building and debugging your programs, as it will give you quick and simple means to switch back and forth between parallel and sequential. To do this, use the following version of the method: async(Boolean isParallelDesired, Runnable body). In the following example, switching the first line can switch whether or not the two things are run in parallel or not!

boolean isParallelDesired = true;
async(isParallelDesired, () -> {
	doSomething();
});
doSomethingElse();

Finish

Finish blocks will wait until everything within the brackets (no matter the number of tasks) executes and completes before moving on to any of the code after it. This is essential in preventing data races and will be used frequently in the course.

finish(Runnable body)

finish( () -> {
	async( () -> {
		doSomething();
	});
	asynch( () -> {
		doSomethingElse();
	});
});
doMoreSomethingElses();

In this example, doSomething(); and doSomethingElse(); will run in parallel with each other, but doMoreSomethingElses(); will not begin until the first two things have fully completed.

Parallel Loops

Attention niels epting.svg Warning: CSE 231s is all exclusive max, all the time. forall, forasync, and forseq are all exclusive max.

Forall

A forall loop is a simplified way to do the following:

finish(()-> {
	for(int i = minInclusive; i < maxExclusive; i++){
		final int i_ = i;
		async( () -> {
			doSomething(i_);
		});
	}
});

To do that same code with a forall loop, we simply do

forall(minInclusive, maxExclusive, (i) -> {
	doSomething(i);
});

Note how this time in the lambda, we are passing a parameter through in the parentheses. This 'i' passed through is the variable that is being incremented. You can also do a forall loop iterating through an iterable object, such as an array or list. That would look like this:

T[] array = new T[n];
forall(array, (item) -> {
	doSomething(item);
});

In this case, item is something of type T from the array. This loop will iterate through all the indicies in the array and spawn a new task for each one.

Similar to general asyncs, by doing forall(boolean isParallelDesired, int minInclusive, int maxExclusive, Consumer body), we can control whether the code runs in parallel or sequentially. Use this to make debugging a lot simpler!

Forall Chunked

Sometimes, you don't want to make a task for each any everything in the iteratable object, as the overhead of spawning tasks outweighs the benefits from parallelism. Chunking allows the creation of the tasks to be manually decided (either by the user or the computer). To use chunking, just pass a ChunkedOption before the object/values to iterate over.

forall(chunked(), array, (item) -> {
	doSomething(item);
});

You can also do chunked(100) to specify the size each chunk should be.

Passing through boolean isParallelDesired as the first parameter can cause the loop to run in parallel or sequential.

Forall 2d

Instead of doing two forall loops, we can achieve the same result using a single forall2d loop.

forall(1stMinInclusive, 1stMaxExclusive, (i) -> {
	forall(2ndMinInclusive, 2ndMaxExclusive, (j) -> {
		array[i][j] = doSomething();
	});
});

Is the exact same thing as

forall2d(1stMinInclusive, 1stMaxExclusive, 2ndMinInclusive, 2ndMaxExclusive, (i, j) -> {
	array[i][j] = doSomething();
});

Chunking can also be done on forall2d loops.

forall2d(chunked(), minA, maxExclusiveA, minB, maxExclusiveB, (i, j) -> {
	array[i][j] = doSomething();
});

Passing through boolean isParallelDesired as the first parameter can cause the loop to run in parallel or sequential.

Forasync

Functions similarly to a forall loop, but does not have the finish block surrounding the for loop. So in order to do this:

for(int i = minInclusive; i < maxExclusive; i++){
	final int _i = i;
	async( () -> {
		doSomething(_i);
	});
}

We use the forasync:

forasync(minInclusive, maxExclusive, (i) -> {
	doSomething(i);
});
T[] array = new T[n];
forasync(array, (item) -> {
	doSomething(item);
});

This style of loop does have a 2d option (forasync2d), passing through boolean isParallelDesired as the first parameter can cause the loop to run in parallel or sequential, and it is possible to use chunking options with it.

Forseq

Forseq functions just like a regular for loop (nothing is done in parallel), but it looks similar in structure to a forall loop. Use this when you want to easily switch back and forth between the parallel and sequential versions for testing purposes.

With the recent introduction of the boolean isParallelDesired option, Forseq has become a less useful tool since its purpose can be achieved elsewhere.

forseq(minInclusive, maxExclusive, (i) -> {
	doSomething(i);
});
T[] array = new T[n];
forseq(array, (item) -> {
	doSomething(item);
});

There also is a forseq2d loop that works just like its parallel counterpart. It is also possible to pass in a chunking option in a forseq, but it does not affect the performance of the code.

Future

A future is a parallel task that has a return value. This means that once the work is completed (in another thread), we can call the future and get its returned value. The basic structure of a future is future(Callable<R> body). Note that this is similar to an async, but uses a Callable rather than a Runnable. In practice, making a future and storing it so the value can be gotten later looks like this:

Future<R> futureWithTypeR = future(() -> {
	R item = doSomething();
	return item;
});

As shown, future(Callable<R> body) returns something of type Future<R>, which is a Java X10 construct.

Then, get the value later on by calling .get() on the future.

R result = futureWithTypeR.get();

Finish Accumulators

A finish accumulator allows you to do operations (such as counting) within a finish block and make them thread-safe. Then the results can be accessed outside of the finish block. First you have to make the finish accumulator, and then you have to register it with your finish.

FinishAccumulator<Integer> acc = newIntegerFinishAccumulator(NumberReductionOperator.SUM);
finish(register(acc), () -> {
	int value = doSomething();
	acc.put(value);
});
Integer result = acc.get();

Some important notes:

  • register() is a V5 method that was created for this course to make setting up accumulators a bit easier.
  • Some other NumberReductionOperators you can use are PRODUCT, MIN, and MAX.
  • Some other types of accumulators are newDoubleFinishAccumulator() and newReducerFinishAccumulator

Java Util Concurrent

X10 is Java's built in parallel tools. These are some things you will always be able to use outside of this course. If you have questions on how certain things work, searching for the javadocs for any object or class will normally provide the answers.

Once again, you can find more in-depth explanations of how some of these tools work by checking out these pages:

Threads

The most simple way to do something in parallel in Java is to create and run a new thread. The constructor needs to have a run() method, so we can just use a lambda.

Thread threadToRun = () -> {
	doSomething();
}
threadToRun.start();
doSomethingElse();
threadToRun.join();

Executors

The ExecutorService class has a submit() function that returns a Future. From there, it isn't to different from our version of Future. In the following example, ExecutorService executor is passed to us.

Future<Integer> future = executor.submit(() -> {
	int value = doSomething();
	return value;
};
doSomethingElse();
future.get()

Phasers

There are a lot of things you can do with Phasers. Check out the official Javadocs for a breakdown of all the options. Here is an example that uses some of the methods we'll call when utilizing phasers.

Phaser phaserA = new Phaser();
Phaser phaserB = new Phaser();
phaserA.register()
phaserB.register()
finish(() -> {
	async(() -> {
		for(int i = 0; i < 10; i++){
			doSomething();
			phaserA.arrive();
			PhaserUtils.awaitAdvancedForPhase(phaserB, i);
		}
	});
	
	for(int j = 0; j < 10; j++){
		doSomethingElse();
		phaserB.arrive();
		PhaserUtils.awaitAdvancedForPhase(phaserA, j);
	}
});

PhaserUtils.awaitAdvancedForPhase(Phaser phaser, int phaseNumber) is a tool we have created to make it safer to wait for other phasers. You have have multiple things wait on one phaser by using a bulk register.

Phaser phaser = new Phaser();
phaser.bulkRegister(3);
forall(0, 3, (i) -> {
	doSomething(i);
	phaser.arriveAndAwaitAdvance();
}

Synchronized

Synchronizing on something prevents on any other thread using (and thus modifying) the synchronized object until it has finished executing the code in the block. You can synchronize on as big or small an object you want. The example below, both uses of synchronized will make the class thread safe, but they have different performance implications.

public class Foo{
	public List<T> list;

	public Foo(){
		this.list = new LinkedList<>();
	}

	public void addToList(){
		synchronized(this){
			this.list.add(new T());
		}
	}

	public void removeFromList(){
		synchronized(this.list){
			this.list.poll();
		}
	}
}

Locks

Locks are an alternative to synchronized. With proper use of locks, we can make our methods thread safe. Here's a couple examples that showcase the syntax for locks.

Reentrant lock = ...
lock.lock();
doSomething();
lock.unlock();
Lock lock = ...
if (lock.tryLock()) {
	try {
		doSomething();
	} finally {
		lock.unlock();
	}
}