Quality of bench

[MaxGraey]

I have some investigation into code. Actually only for CollisionDetection test and found some issues. In my opinion.

1. jsCollisionDetection is wrong functionality. It shoud be:

function jsCollisionDetection(positions, radiuses, res, n) {
  var count = 0;
  for (var i = 0; i < n; i++) {
    var p = positions[i];
    var r = radiuses[i];
    var collision = false;
    for (var j = i+1; j < n; j++) {
      var p2 = positions[j];
      var r2 = radiuses[j];
      var dx = p.x - p2.x;
      var dy = p.y - p2.y;
      var dz = p.z - p2.z;
      var d = Math.sqrt(dx*dx + dy*dy + dz*dz);
      if (r + r2 >= d) { // <---------------------------- 2 fixes ---------------------------
        collision = true;
        count++;
        break;
      }
    }
    var index = (i / 8) | 0;
    var pos = 7 - (i % 8);
    if (!collision) {
      res[index] &= ~(1 << pos);
    } else {
      res[index] |= (1 << pos);
    }
  }
  return count;
}

2. binaryen and emscripten does not yet optimize as much as modern JS-engines. So we should optimize c++ manually:

int collisionDetection(struct position *positions,
                       double *radiuses,
                       unsigned char *res, int n) {
  int count = 0;
  for (int i = 0; i < n; i++) {
    struct position p = positions[i];
    double r = radiuses[i];
    bool collision = false;
    for (int j = i+1; j < n; j++) {
      struct position  p2 = positions[j];
      double r2 = radiuses[j];
      double dx = p.x - p2.x;
      double dy = p.y - p2.y;
      double dz = p.z - p2.z;
      double d = sqrt(dx*dx + dy*dy + dz*dz);
      if (r + r2 >= d) {
        collision = true;
        count++;
        break;
      }
    }
    unsigned int index = i >> 3;
    unsigned int pos = 7 - (i & 7);
    unsigned int mask = 1 << pos;
    unsigned char current = res[index];
    res[index] = collision ? current | mask : current & ~mask;
  }
  return count;
}

3. You should be carried beyond init and fill memory routines from wsCollisionDetection and measure only functions.collisionDetection(pointer1, pointer2, pointer3, n) procedure.

4. You should use loop with 32-64 iterations for warm-up instead only one call for earlier turn on JIT hot-patching.