Hacker News Reader

Comment by tlrobinson | Hacker News Reader

It had to be done:

    setInterval(function() {
      var n = parseInt(document.querySelector("#n").textContent);
      document.querySelector(is_prime(n) === "prime" ? "#yes" : "#no").click();
    }, 0);
    
    document.querySelector("#start").click();

EDIT: replaced my isPrime with the one already in the game.

EDIT 2: There appears to be some sort of increasing delay (pretty sure it's not just `is_prime` being slow) starting around 500 points. Maybe the game itself trying to compute primes?

View on HN

Ah, you beat me to the minute (game comes with a function is_prime):

    var yes = document.getElementById('yes'),
        no = document.getElementById('no'),
        n = document.getElementById('n'),
        start = document.getElementById('start');


    function tick() {
        console.log('tick');
        if (is_prime(n.innerHTML) === 'prime') {
            yes.click();
        } else {
            no.click();
        }
        window.setTimeout(tick, 500);
    }

    start.click();
    tick();

Or, do away with the whole "checking" nonsense:

  game.check = function (answer) {
  	this.streak += 1;
  	this.next_n();
  }

  setInterval(function() {
    document.querySelector("#yes").click();
  }, 0);
  
  document.querySelector("#start").click();

(High score of 15,000!)

this. IF you cheat, you might as well take the easiest path of all :D

"Inspect Element"

24607 (with the cheat), 621 (without) :D Fun evening challenge :P

  function isPrime(n){
    if ((n<=1) || (!n % 2)) {
      return false;
    }

  for (var k = 3; k < Math.round(Math.sqrt(n) + 0.5); k = k+2) {
      if (n % k == 0) {
        return false;
      }
    } 
    return true;
  }
  var nSpan = $("span#n")
  var yesButton = $("button#yes");
  var noButton = $("button#no");
  var body = $("body");
  function solve(){
    if (isPrime(parseInt(nSpan.innerHTML)) {
      yesButton.click();
    } else {
      noButton.click();
    }
    if (body.className != "end") {
      setTimeout(solve,0);
    }
  }
  solve();

Why only add 1? Why add anything at all?

    game.check = function() {
        this.streak = Number.MAX_VALUE;
        this.next_n();
    };
  
    document.querySelector("#start").click();
    document.querySelector("#yes").click();

Decreasing the loop's upper bound to Math.sqrt(value) and only testing odd values after 2 are really simple ways to improve performance.

Beyond that it'd be nice to build a list of previously determined primes to reduce your number of checks, but I guess at that point it might just be better to implement classic primality tests...

It's faster to check that the value isn't less than the loop counter squared than to perform a square root.

No it isn't. You only calculate the square root once, vs calculating the square at each iteration.

Calculating a square is ~1. A square root it ~N. Much harder to do. Its a common optimization in graphics code to compare squares for this reason.

She's right about squaring on each iteration, though. And granted, a square root is much more expensive, though done only once. Which method is faster would depend on the number of iterations.

The latency for sqrtss on broadwell is 11 cycles with a throughput of 4, where mul is a latency of 3 and throughput of 1. So, using some concrete numbers, sqrt is more expensive, but not polynomially or even an order of magnitude.

Right with modern floating point implementations its not the old guess-and-iterate method any more. SQRT is probably now on the order of an inverse?

I honestly haven't looked into it in that much detail, but if I had to guess... probably. I'm not sure what the implementation used in hardware currently is, but at very least it's constant bounded,

Conversions are still somewhat expensive, but I do know compared to polynomial time or a large enough constant, it can be a better choice for an optimization. For example, computing log10 of an integer.

Since you're claiming that square root is "~N", the N here must be the number of digits. The number of multiplications required for the "optimization" you're proposing is exponential in this N.

If you go that way, it may be even faster (depending on you CPU, of course) to use a recursion for computing the squares:

  (n+1)^2 = n^2 + 2n + 1

That gets you:

  def squares:
    n = 1
    nSquared = 1
    delta = 1
    while nSquared <= n:
      n += 1
      yield n
      delta += 2
      nSquared += delta

but as others said, once your limit is large enough, the relatively large cost of computing a square root once may be lower than the O(sqrt(n)) additions in this loop.

(Quick check: if you use a binary search to find sqrt(n), you need 2log(n) iterations, each of which requires a division by two (for integers, that is a bit shift) and one addition. A linear search takes sqrt(n) iterations, each with three additions. For whatever ratio of speed of operations you pick, the former will be faster for sufficiently large n)

You would normally use something like newton-raphson to compute a square root, which still isn't O(1) but converges fairly quickly. I would be very skeptical that n muls/adds would be faster than an sqrt, at very least for x86-64 since it is implemented as an instruction.

My mental model of a CPU is a 6502 or, at best, a 68000 without FPU :-)

When asked about modern day performance, I would just say “I do not know”. Thing is that those muls, on modern CPUs, might be pipelined in parallel with the adds, so that, time wise, you would get them for free.

But of course, it might be possible to get that sqrt for free, too. Even if it takes 50 cycles, you might start one, do a few iterations without checking for ‘reached sqrt(n) yet’, and then start a loop testing for the limit.

If you really want to know for x_64, http://www.agner.org/optimize/ probably has the answers, but then, you would have to know the x86 instruction set, which is horrendous (compared to that 6502 or 68000). Even realising that there will be single, double and vector variants, the number of different instructions with ‘SQRT’ in their name I find in http://www.agner.org/optimize/instruction_tables.pdf is insane.

And of course, probably, something completely out of left field could well be the fastest way to do this (by a few ns, probably). For example, modern CPUs can count leading zeros in an integer. If that instruction is fast (for x86, that is not a given; ‘bit scan reverse’ was slow on some CPUs) subtract from 64/32/16, and halve, and you have a decent approximation to 2log(sqrt(n)) (using ‘if n has b bits, sqrt(n) has about b/2’)

Incidentally, even if you are using Newton's method to do square roots, it helps significantly to start with a guess of the right size. If your initial guess is way too large, each step will cut it down by about a factor of 1/2; if it's way too small, it's equivalent (for finding sqrt(N)) to choosing N/guess. Thus, if you're taking the square root of 2^64, and you start with 1, your subsequent guesses will be roughly 2^63, 2^62, 2^61, 2^60, ..., taking a total of 36 steps to reach the right integer; whereas if you started with 2^33, it takes (by my computation) only 5 steps to get to the right integer.

The effect is more significant with larger numbers:

  arc> (newton-steps (round (* 3/2 (expt 2 1000))) 1)
  508
  arc> (newton-steps (round (* 3/2 (expt 2 1000))) (expt 2 500))
  8

My mental model of a CPU unfortunately tends toward x86.

> But of course, it might be possible to get that sqrt for free, too. Even if it takes 50 cycles, you might start one, do a few iterations without checking for ‘reached sqrt(n) yet’, and then start a loop testing for the limit.

On the CPU side, the branch predictor might make the sqrt free by pipelining it, but that starts to make the analysis a bit harder. Strange things can happen when you introduce branch prediction. The performance would also depend on ALU/FPU contention, hyperthreads, etc...

> If you really want to know for x_64, http://www.agner.org/optimize/ probably has the answers, but then, you would have to know the x86 instruction set, which is horrendous (compared to that 6502 or 68000). Even realising that there will be single, double and vector variants, the number of different instructions with ‘SQRT’ in their name I find in http://www.agner.org/optimize/instruction_tables.pdf is insane.

I've only just started reading Agner Fog's documentation (which is awesome), but it definitely seems like the place the find low level information of the ilk when it's needed. If you read through the instruction table though, you probably noticed there are fsqrt and (v)sqrt(p)(s/d). For most purposes, x87 is actually deprecated and SSE2, the latter, is preferred for scalar floating point. I would guess the Mill devs might have some insightful comments.

> And of course, probably, something completely out of left field could well be the fastest way to do this (by a few ns, probably). For example, modern CPUs can count leading zeros in an integer. If that instruction is fast (for x86, that is not a given; ‘bit scan reverse’ was slow on some CPUs) subtract from 64/32/16, and halve, and you have a decent approximation to 2log(sqrt(n)) (using ‘if n has b bits, sqrt(n) has about b/2’)

I'm not sure that's quite right. Counting leading zeros is normally very fast if you have the instruction, but those two functions diverge pretty quick. For a constant cost, my guess is that the accuracy lost wouldn't be performance gained. The other point is that accuracy is fairly important if you're testing for primality.

It's an FPU instruction with some setup and teardown instructions associated.

I usually bow to Agner Fog on this:

"On Core2 65nm, FSQRT takes 9 to 69 cc's (with almost equal reciprocal throughput), depending on the value and precision bits. For comparison, FDIV takes 9 to 38 cc's (with almost equal reciprocal throughput), FMUL takes 5 (recipthroughput = 2) and FADD takes 3 (recipthroughput = 1). SSE performance is about equal, but looks faster because it can't do 80bit math. SSE has a super fast approximate reciprocal and approximate reciprocal sqrt though.

On Core2 45nm, division and square root got faster; FSQRT takes 6 to 20 cc's, FDIV takes 6 to 21 cc's, FADD and FMUL haven't changed. Once again SSE performance is about the same."

x87 is largely deprecated. Try inspecting the code emitted for floating point by all the major compilers.

You are correct. It looks like SSE sqrt matches or beats fsqrt too.

SSE performance is better than x87, for a number of reasons. Latency and throughput are generally better, but SSE also supports SIMD which allows more than one op per instruction. SSE also has more registers (x87 only has 8 80bit compared to at least 16 128bit). If I'm not mistaken, the number of available execution units with SSE is also larger, which means more opportunities for instruction level parallelism.

These are why modern compilers don't emit x87 for floating point.

Dynamic programming (testing divisibility by the primes already found) would greatly improve the algorithm. And of course sieving is absurdly fast compared to checking primality individually.

If you don't want the browser to freeze ...

    var target = document.getElementById('n');
    function detect() {
      n = parseInt(target.firstChild.nodeValue);
      if (typeof n === 'number' && is_prime(n) === 'prime') {
        console.log('>> YES,' + n + ' is prime, clicking "yes" ...');
        document.getElementById('yes').click();
      } else {
        console.log('>> NO,' + n + ' is not prime, clicking "no" ...');
        document.getElementById('no').click();
      }
    }
    var observer = new MutationObserver(function (mutations) {
      mutations.forEach(function (mutation) {
        window.setTimeout(detect, 500);
      });
    });
    observer.observe(target, {
      attributes: true,
      childList: true,
      characterData: true
    });

Or just:

    setInterval(function() {game.check(is_prime(game.current_n) === 'prime')}, 0);

Now the game is, Who can write most efficient primality check algorithm!

    window.is_prime = function () { return "prime"; }

At least write your own prime checking algorithm!

He initially did. Then decided to reuse the one already provided. The goal of cheating in this manner, after all, would be to match exactly the implementation the original programmer used. Even if it was buggy.

Reply on news.ycombinator.com