Java Vector API

Java Vector API provides the ability to explicitly write vector (SIMD) computations that compiles to vector instructions on supported CPUs. Although one of the JVM JIT compiler optimizations (called auto-vectorization) uses the vector instructions too, it cannot always benefit from them because it cannot know as much about the algorithm as the programmer.

The feature is still under development; however, it is mature enough to be a part of the JDK library as an incubator module from Java 16. You need to explicitly include the module by the following JVM argument:

--add-modules jdk.incubator.vector

Exercises

Sum of an Array

Try to implement the following code using vector API (solution will be presented during the seminar and later provided in slides):

public static int sum(int[] array) {
	int sum = 0;
	for (int i : array) {
		sum += i;
	}
	return sum;
}

Dot Product

Implement dot product using vector API. Scalar implementation can be seen here:

public static int dot(int[] v1, int[] v2) {
	if (v1.length != v2.length) throw new IllegalArgumentException();
	int sum = 0;
	for (int i = 0; i < v1.length; i++) {
		sum += v1[i] * v2[i];
	}
	return sum;
}

Matrix Multiplication

You can also try to improve the multiplication with the second matrix transposed.

Materials

The documentation can be seen in the appropriate JEPs: 338, 414, 417

Or you can look at this video

Slides from the seminar (will be provided later)

Table of Contents