The V8 Engine and JavaScript Optimization Tips

Paul Ryan

V8 is Google’s engine for compiling our JavaScript. Firefox has it’s own engine called SpiderMonkey, it’s quite similar to V8 but there are differences. We will be discussing the V8 engine in this article.

A couple of facts about the V8 engine:

• Written in C++ and used in Chrome and Node.js (and the newest release of Microsoft Edge)
• Implements ECMAScript as specified in ECMA-262

The JavaScript Journey

So what exactly happens when we send our JavaScript to be parsed by the V8 engine (this is after it is minified, uglified and whatever other crazy stuff you do to your JavaScript code)?

I’ve created the following diagram that shows all the steps, we will then discuss each step in detail:

In this article we’ll discuss how the JavaScript code gets parsed and how to get as much of your JavaScript to the Optimizing Compiler as possible. The Optimizing Compiler (aka Turbofan) takes our JavaScript code and converts it to high performance Machine Code, so the more code we can give it the faster our application will be. As a side note, the interpreter in Chrome is called Ignition.

Parsing JavaScript

So the first treatment of our JavaScript code is to parse it. Let’s discuss exactly what parsing is.

There are two phases to parsing which are:

• Eager (full-parse) - this parses each line right away
• Lazy (pre-parse)- do the bare minimum, parse what we need and leave the rest until later

Which is better? It all depends.

Let’s look at some code.

// eager parse declarations right away
const a = 1;
const b = 2;

// lazily parse this as we don't need it right away
function add(a, b) {
  return a + b;
}

// oh looks like we do need add so lets go back and parse it
add(a, b);

So here our variable declarations will be eager parsed but then our function is lazily parsed. This is great until we get to add(a, b) as we need our add function right away so it would have been quicker to eager parse add right away.

To eager parse the add function right away, we can do:

// eager parse declarations right away
const a = 1;
const b = 2;

// eager parse this too
var add = (function(a, b) {
  return a + b;
})();

// we can use this right away as we have eager parsed
// already
add(a, b);

This is how most modules you use are created.

Function Inlining

Chrome will sometimes essentially rewrite your JavaScript, one example of this is inlining a function that is being used.

Let’s take the following code as an example:

const square = (x) => { return x * x }

const callFunction100Times = (func) => {
  for(let i = 0; i < 100; i++) {
    // the func param will be called 100 times
    func(2)
  }
}

callFunction100Times(square)

The above code will be optimized by the V8 engine as follows:

const square = (x) => { return x * x }

const callFunction100Times = (func) => {
  for(let i = 100; i < 100; i++) {
    // the function is inlined so we don't have 
    // to keep calling func
    return x * x
  }
}

callFunction100Times(square)

As you can see from the above, V8 is essentially removing the step where we call func and instead inlining the body of square. This is very useful as it will improve the performance of our code.

Function inlining gotcha

There is a little gotcha with this approach, let’s take the following code example:

const square = (x) => { return x * x }
const cube = (x) => { return x * x * x }

const callFunction100Times = (func) => {
  for(let i = 100; i < 100; i++) {
    // the function is inlined so we don't have 
    // to keep calling func
    func(2)
  }
}

callFunction100Times(square)
callFunction100Times(cube)

So this time after we have called the square function 100 times, we will then call the cube function 100 times. Before cube can be called, we must first de-optimize the callFunction100Times as we have inlined the square function body. In cases like this, the square function will seem like it’s faster than the cube function but what is happening is the de-optimization step makes the execution longer.

Objects

When it comes to objects, V8 under the hood has a type system to differentiate your objects:

Monomorphism

The objects have the same keys with no differences.

// mono example
const person = { name: 'John' }
const person2 = { name: 'Paul' }

Polymorphism

The objects share a similar structure with some small differences.

// poly example
const person = { name: 'John' }
const person2 = { name: 'Paul', age: 27 }

Megamorphism

The objects are entirely different and cannot be compared.

// mega example
const person = { name: 'John' }
const building = { rooms: ['cafe', 'meeting room A', 'meeting room B'], doors: 27 }

So now we know the different objects in V8, let’s see how V8 optimizes our objects.

Hidden classes

Hidden classes are how V8 identifies our objects.

Let’s break this down into steps.

We declare an object:

const obj = { name: 'John'}

V8 will then declare a classId for this object.

const objClassId = ['name', 1]

Then our object is created as follows:

const obj = {...objClassId, 'John'}

Then when we access the name property on our object like so:

obj.name

V8 does the following lookup:

obj[getProp(obj[0], name)]

This is process V8 goes through when creating our objects, now let’s see how we can optimize our objects and reuse classIds.

Tips for Creating Objects

If you can, you should declare your properties in the constructor. This will ensure the object structure stays the same so V8 can then optimize your objects.

class Point {
  constructor(x,y) {
    this.x = x
    this.y = y
  }
}

const p1 = new Point(11, 22) // hidden classId created
const p2 = new Point(33, 44)

You should keep the property order constant, take the following example:

const obj = { a: 1 } // hidden class created
obj.b = 3

const obj2 = { b: 3 } // another hidden class created
obj2.a = 1

// this would be better
const obj = { a: 1 } // hidden class created
obj.b = 3

const obj2 = { a: 1 } // hidden class is reused
obj2.b = 3

General Optimization Tips

So now let’s get into some general tips that will help your JavaScript code be better optimized.

Fix function argument types

When arguments are being passed to a function it’s important they are the same type. Turbofan will give up trying to optimize your JavaScript after 4 tries if the argument types are different.

Take the following example:

function add(x,y) {
  return x + y
}

add(1,2) // monomorphic
add('a', 'b') // polymorphic
add(true, false)
add({},{})
add([],[]) // megamorphic - at this stage, 4+ tries, no optimization will happen

Another tip is to make sure to declare classes in the global scope:

// don't do this
function createPoint(x, y) {
  class Point {
    constructor(x,y) {
      this.x = x
      this.y = y
    }
  }

  // new point object created every time
  return new Point(x,y)
}

function length(point) {
  //...
}

Conclusion

So I hope you learned a few things about how V8 works under the hood and how to write better optimized JavaScript code.