# Primes & Squares

Codeforces #230B

Question

Whether a number has exactly three factors. (i.e. 1, itself and its square root, thus squares of primes) [Time limit is 2 sec, Memory 256 MB]

Input

Number of testcases, n (~105) then numbers (~1012)

Output

Yes or No

Problem

I first check if it is a perfect square, then check if it's root is prime or not. Earlier I was trying with naive prime checking algorithm, but now I thought to speed up I should use sieve of Eratosthenes, which wouldn't take much time as largest number is 1012 so square root is at max 106 which is not a very large number for $O(n \log \log n)$ (778 microsec considering 1GHz ~$10^9$ ops/sec and 6000 microsec for $O(n \log n)$).

My sieve is a little bit different: (i) I have only indexed odd number to half the memory required. (0→3, 1→5, 2→7, …, x→2x+3) (ii) I use inNotPrime instead of isPrime because default value of boolean array is false, so it would be easy to avoid Arrays.fill(isPrime,true).

import java.util.Scanner;

public class Solution {
public static void main(String[] args) {
Scanner sc = new Scanner(System.in);
int n = sc.nextInt();
seivePrimes();
for (int i = 0; i < n; i++) {
long a = sc.nextLong();
long sqrt = (int) Math.sqrt(a);
if (a >= 4 && sqrt * sqrt == a) {
if (isPrime((int) sqrt)) {
System.out.println("YES");
} else {
System.out.println("NO");
}
} else {
System.out.println("NO");
}
}
sc.close();
}

static int maxlim = 1000000;
private static void seivePrimes() {
for (int i = 3; i * i <= maxlim; i += 2) {
if (!isNotPrime[(i - 3) / 2]) {
for (int j = i; i * j <= maxlim; j += 2) {
isNotPrime[(i * j - 3) / 2] = true;
}
}
}
}

static boolean[] isNotPrime = new boolean[(maxlim - 3) / 2 + 1];//isNotPrimes[x]=>2*x+3 is not prime
private static boolean isPrime(int t) {
if (t == 2)
return true;
if (t % 2 == 0)
return false;
else
return !isNotPrime[(t - 3) / 2];
}
}


What is making it slow so as that time limit is exceeded? I saw the solution to the problem afterwards and even they are using the same thing, squares and sieving.

• The linked problem doesn't state that the 3 divisors should be "1,itself and it's squareroot". It just mentions "three distinct positive divisors.". Is this a variant of your own? May 24, 2016 at 9:39
• @Tunaki It is an implication, both problems are equivalent. May 24, 2016 at 9:49
• You are calculating square roots, that is slow. Check if a number is prime, then its square matches your criteria. 7 is prime so 7 * 7 = 49 is a YES. May 24, 2016 at 21:48

What is making it slow so as that time limit is exceeded

Funnily enough, it can be the println. It's synchronized and (what's worse IIRC) it flushes the output. Accumulating it in a StringBuilder can be the solution (an answer of mine claiming this on SO was accepted).

Note that division and modulus are slow. a % 2 and a / 2 can't be in general optimized to a & 1 and a >> 2, respectively. The JIT can't do this when it can't see that a >= 0 holds. You can.

    if (a >= 4 && sqrt * sqrt == a) {
if (isPrime((int) sqrt)) {
System.out.println("YES");
} else {
System.out.println("NO");
}
} else {
System.out.println("NO");
}


That's bad! Always separate computation from IO. Write a method. When doing this, combine the conditions:

boolean isSquaredPrime(long a) {
long sqrt = (int) Math.sqrt(a);
return a >= 4 && sqrt * sqrt == a && isPrime((int) sqrt);
}


Note that it fails for a above $2^{62}$.

This part is fine (though I'd recommend against leaving out braces)

private static boolean isPrime(int t) {
if (t == 2)
return true;
if (t % 2 == 0)
return false;
else
return !isNotPrime[(t - 3) / 2];
}


but it can be a bit compact, e.g.,

private static boolean isPrime(int t) {
if ((t & 1) == 0) { // optimized evenness test
return t == 2; // the only even prime
} else {
return !isNotPrime[(t - 3) >> 1];
}
}


I wrote "compact" rather than "readable" as it's bad for people unaware of bit operations. But once you get used to them....

Method extraction is good for readability and reusability. It may sometimes make the program faster as shorter methods can be better optimized. The method call cost may be substantial, but then the method gets most probably inlined (which sort of undoes the method extraction, but readability gains remain).

• Thankyou, StrinBuilder made it to 778ms, bit operations 746ms, separate function 686ms. But I'm unsure separating in another function changes what, since the flow remains the same? May 25, 2016 at 4:11
• @ADG What was the initial timing? I don't think the method extraction really helped, it may be an measurement error. Java benchmarking is pretty hard, read about JMH or Google Caliper. May 25, 2016 at 13:06
• It ran on codeforces online judge and the verdict was time limit exceeded, >2s May 25, 2016 at 13:32
• anyways C++ code ran in about half (310 ms) compared to the fastest yet. (686 ms) May 25, 2016 at 13:32
• @ADG The JVM has quite some starting overhead and warmup time (before everything gets C2 compiled), which makes it rather unsuitable for anything taking that little. I'm sure that running 10 times as many iterations would change the speed ratio substantially. May 26, 2016 at 18:02