Emscripten - where is the performance?

Question

JavaScript

Here for a simple program there are source codes in
C ++

#include <iostream>

using namespace std;

#include <sys/time.h>
#include <unistd.h>

static double GetMilliseconds(){
  struct timeval time;
  gettimeofday(&time, NULL);
  
  return (time.tv_sec*1000.0 + time.tv_usec/1000.0);
}


int i=0;
int fib(int x) {i++;
  if (x < 2) {
    return 1;
  } else {
    return fib(x - 1) + fib(x - 2);
  }
}


int main() {

double t0 = GetMilliseconds();


int result = fib(35);



double ms = GetMilliseconds()-t0;

cout << "time ms: "<< ms << "res: "<<i<<"-"<<result<<endl;


cin.get();

  return 0;
}

and js

var i=0;
    function fib(x) {i++;
      if (x < 2) {
      return 1;
      } else {
      return fib(x - 1) + fib(x - 2);
      }
    }
  
    window.onload = function(){


  
      var time = new Date().getTime();

    
    var result = fib(35);
    


      document.write(new Date().getTime()-time+"<br>res: "+i+"-"+result);


    }

In the main block, the number of operations is identical.
After [assembly and] launch, I get the following picture:

1. pure JS 180-200ms
   
2. em++ 500-600ms
   
3. em++ -O3 180-200ms
   
4. g++ 180-200ms
   
5. g++ -O3 50-60ms
   

Why is that? Okay, there without -O3 gives out wild things. But the code compiled via em++ is far from the speed on g++ -O3 and is no faster than pure JS on the same V8. Moreover, the indicators are the same in FF, in Chrome.

Answer 1

I tried to write a recursive fibonacci in pure asm.js:

(function(std, env, heap) {
    'use asm';
    function fib(n) {
        n = n|0;
        if((n|0) < (2|0)) {
            return 1|0;
        }
        return (fib((n|0) - (1|0)) + fib((n|0) - (2|0)))|0;
    }
    return fib;
}(
    {},
    null,
    new ArrayBuffer(0)
));

Really slower than js
The reason is obvious: asm.js is not optimized in any way, these are ready-made instructions for a virtual machine (for example, v8)
Normal js is run through an optimizer before compilation, which successfully converts tail recursions into a loop
We conclude that emscripten is not yet able to optimize tail recursions
And by the way, what optimization option did you give in em++?
PS about performance, my hashing algorithm (crc64 with salt) after a complete rewrite from js to asm.js gave a performance increase of 21.7 times (node ​​6.3.1 v8 5.0.71.57)

Answer 2

With such questions, you need to look at the assembler listing. I guess gcc threw out all the calculations and planted a constant.
He definitely had to unwind the recursion.