Android, Java faster than C++ [Solved]

luca

hi,
I have an Android application that simply calculate the first N prime numbers in both way: C++ and Java (using JNI).
The C++ code is:


...
long  m_startNum=0;
long  m_endNum=100000;
...
void Engine::primeNumbers() {
    int tot_prime=0;
    int count=0;
    for( long i=m_startNum;i<=m_endNum;i++)
    {
        for( long  j=2;j<=sqrt(i);j++)
        {
            if(i%j==0)
                count++;
        }
        if(count==0 && i!=1)
        {
            tot_prime++;
            count=0;

        }
        count=0;
    }
    emit ended();
    qDebug() << "TOT TROVATI: " << tot_prime;
    emit newPrimeNumber(tot_prime);
}

The Java version is:

void Engine::primeNumbersJava()
{
    jint tot = QAndroidJniObject::callStaticMethod<jint>(
                "com/test/calc_test_java/PrimeNumbers",
                "primeNumbers");

    qDebug() << "TOT TROVATI: " << tot;
    emit newPrimeNumber(tot);
    emit ended();
}

and my Java class:

package com.test.calc_test_java;

public class PrimeNumbers {
    static long m_startNum = 0;
    static long m_endNum = 100000;

    public static int primeNumbers() {
        System.out.println("inizio primeNumbers");
        int prime_found=0;
        int count=0;
        for(long i=m_startNum;i<=m_endNum;i++)
        {
            for(long  j=2;j<=Math.sqrt(i);j++)
            {
                if(i%j==0)
                    count++;
            }
            if(count==0 && i!=1)
            {
                count=0;
                prime_found++;
            }
            count=0;
        }

        System.out.println("fine primeNumbers");
        return prime_found;
    }
}

The strange thing is that when I execute on a tablet, the C++ version takes 4 secs to end while Java version takes only 2 secs.
I didn't expect that result...

What do you think about?

Luca

Chris Kawa

I don't see any code measuring time. Have you actually checked it? Running Java and C++ have different runtimes spinning up and lots of other overhead - loading libraries and such. It doesn't mean that particular function runs faster or slower.

luca

@Chris-Kawa
Yes, I checked time by using QElapsedTimer on the calling method of primeNumbersJava and primeNumbers.
As you write, I suppose the reason is related to library loading or something else.
But I grant you Java is very faster in my specific methods primeNumbersJava and primeNumbers...

cookie-jar

This post is deleted!

luca

Looking on the compiling output I saw that QtCreator is using "-O2" optimization flag.

I tried with -O3 by putting the following lines on .pro:

QMAKE_CXXFLAGS_RELEASE -= -O
QMAKE_CXXFLAGS_RELEASE -= -O1
QMAKE_CXXFLAGS_RELEASE -= -O2 
QMAKE_CXXFLAGS_RELEASE *= -O3

And now the C++ version on primeNumbers takes only 620 msec while Java version takes 2 sec .
That's what I expect!! ;-)