Sort.java

import org.openjdk.jmh.annotations.*;

import java.util.Arrays;
import java.util.Random;
import java.util.concurrent.TimeUnit;

@BenchmarkMode(Mode.AverageTime)
@OutputTimeUnit(TimeUnit.MICROSECONDS)
@State(Scope.Thread)
public class Sort {

    @Param({"10", "1000", "100000", "10000000" })
    int size;

    @Param({"1", "2", "3"})
    int seed;

    int[] array;

    int[] arrayToSort;

    @Setup
    public void setUp() {
        this.array = new int[this.size];
        final Random random = new Random(this.seed);
        for (int i = 0; i < this.array.length; i++) {
            this.array[i] = random.nextInt();
        }
    }

    @Benchmark
    public int dualPivot() {
        this.arrayToSort = this.array.clone();
        Arrays.sort(this.arrayToSort);
        return this.arrayToSort[0];
    }

    @Benchmark
    public int radix() {
        this.arrayToSort = this.array.clone();
        RadixSort.sort(this.arrayToSort);
        return this.arrayToSort[0];
    }
}

Maybe you should return array[0] as array.length is an invariant that hotspot could optimize (dead code elimination)

@bourgesl Thanks. Fixed.

Your test seems correct, I suppose arrayToSort (cloning) should stay out of the benchmark. How costly is it ?