Strategy to optimize javascript

I have written a javascript program that uses a genetic algorithm to recreate an image only using triangles. Here's the strategy:

• generate a random pool of models, each model having an array of triangles (3 points and a color)
• evaluate the fitness of each model. To do so, I compare the original image's pixel array with my model's. I use Cosine Similarity to compare arrays
• keep the best models, and mate them to create new models
• randomly mutate some of the models
• evaluate the new pool and continue

It works quite well after some iterations as you can see here:

The problem I have, is that it is very slow, most of the time is spent getting model's pixels (converting list of triangles (color + points) to a pixel array). Here's how I do so now:

My pixel-array is a 1D array, I need to be able to convert x,y coordinates to index:

static getIndex(x, y, width) {
  return 4 * (width * y + x);
}

Then I am able to draw a point:

static plot(x, y, color, img) {
  let idx = this.getIndex(x, y, img.width);

  let added = [color.r, color.g, color.b, map(color.a, 0, 255, 0, 1)];
  let base = [img.pixels[idx], img.pixels[idx + 1], img.pixels[idx + 2], map(img.pixels[idx + 3], 0, 255, 0, 1)];
  let a01 = 1 - (1 - added[3]) * (1 - base[3]);

  img.pixels[idx + 0] = Math.round((added[0] * added[3] / a01) + (base[0] * base[3] * (1 - added[3]) / a01)); // red
  img.pixels[idx + 1] = Math.round((added[1] * added[3] / a01) + (base[1] * base[3] * (1 - added[3]) / a01)); // green
  img.pixels[idx + 2] = Math.round((added[2] * added[3] / a01) + (base[2] * base[3] * (1 - added[3]) / a01)); // blue
  img.pixels[idx + 3] = Math.round(map(a01, 0, 1, 0, 255));
}

Then a line:

 static line(x0, y0, x1, y1, img, color) {
  x0 = Math.round(x0);
  y0 = Math.round(y0);
  x1 = Math.round(x1);
  y1 = Math.round(y1);
  let dx = Math.abs(x1 - x0);
  let dy = Math.abs(y1 - y0);
  let sx = x0 < x1 ? 1 : -1;
  let sy = y0 < y1 ? 1 : -1;
  let err = dx - dy;

  do {
    this.plot(x0, y0, color, img);
    let e2 = 2 * err;
    if (e2 > -dy) {
      err -= dy;
      x0 += sx;
    }
    if (e2 < dx) {
      err += dx;
      y0 += sy;
    }
  } while (x0 != x1 || y0 != y1);
}

And finally, a triangle:

static drawTriangle(triangle, img) {
  for (let i = 0; i < triangle.points.length; i++) {
    let point = triangle.points[i];
    let p1 =
      i === triangle.points.length - 1
        ? triangle.points[0]
        : triangle.points[i + 1];
    this.line(point.x, point.y, p1.x, p1.y, img, triangle.color);
  }
  this.fillTriangle(triangle, img);
}

static fillTriangle(triangle, img) {
  let vertices = Array.from(triangle.points);
  vertices.sort((a, b) => a.y > b.y);
  if (vertices[1].y == vertices[2].y) {
    this.fillBottomFlatTriangle(vertices[0], vertices[1], vertices[2], img, triangle.color);
  } else if (vertices[0].y == vertices[1].y) {
    this.fillTopFlatTriangle(vertices[0], vertices[1], vertices[2], img, triangle.color);
  } else {
    let v4 = {
      x: vertices[0].x + float(vertices[1].y - vertices[0].y) / float(vertices[2].y - vertices[0].y) * (vertices[2].x - vertices[0].x),
    y: vertices[1].y
    };
    this.fillBottomFlatTriangle(vertices[0], vertices[1], v4, img, triangle.color);
    this.fillTopFlatTriangle(vertices[1], v4, vertices[2], img, triangle.color);
  }
}

static fillBottomFlatTriangle(v1, v2, v3, img, color) {
  let invslope1 = (v2.x - v1.x) / (v2.y - v1.y);
  let invslope2 = (v3.x - v1.x) / (v3.y - v1.y);

  let curx1 = v1.x;
  let curx2 = v1.x;

  for (let scanlineY = v1.y; scanlineY <= v2.y; scanlineY++) {
    this.line(curx1, scanlineY, curx2, scanlineY, img, color);
    curx1 += invslope1;
    curx2 += invslope2;
  }
}

static fillTopFlatTriangle(v1, v2, v3, img, color) {
  let invslope1 = (v3.x - v1.x) / (v3.y - v1.y);
  let invslope2 = (v3.x - v2.x) / (v3.y - v2.y);

  let curx1 = v3.x;
  let curx2 = v3.x;

  for (let scanlineY = v3.y; scanlineY > v1.y; scanlineY--) {
    this.line(curx1, scanlineY, curx2, scanlineY, img, color);
    curx1 -= invslope1;
    curx2 -= invslope2;
  }
}

You can see full code in action here

So, I would like to know:

• is it possible to optimize this code ?
• if yes, what would be the best way to do so ? Maybe there is a library doing all of the drawing stuff way better than I did ? Or by using workers ?

Thanks !

I have tested your suggestions, here's the results:

• Use RMS instead of Cosine Similarity: I am not sur if the measure of similarity is better, but it is definitively not worse. It seems to run a little bit faster too.
• Use UInt8Array: It surely have an impact, but does not runs a lot faster. Not slower though.
• Draw to invisible canvas: Definitively faster and easier! I can remove all of my drawing functions and replace it with a few lines of code, and it runs a lot faster !

Here's the code to draw to an invisible canvas:

var canvas = document.createElement('canvas');
canvas.id = "CursorLayer";
canvas.width = this.width;
canvas.height = this.height;
canvas.display = "none";
var body = document.getElementsByTagName("body")[0];
body.appendChild(canvas);

var ctx = canvas.getContext("2d");
ctx.fillStyle = "rgba(0, 0, 0, 1)";
ctx.fillRect(0, 0, this.width, this.height);

for (let i = 0; i < this.items.length; i++) {
  let item = this.items[i];
  ctx.fillStyle = "rgba(" +item.color.r + ','+item.color.g+','+item.color.b+','+map(item.color.a, 0, 255, 0, 1)+")";
  ctx.beginPath();
  ctx.moveTo(item.points[0].x, item.points[0].y);
  ctx.lineTo(item.points[1].x, item.points[1].y);
  ctx.lineTo(item.points[2].x, item.points[2].y);
  ctx.fill();
}
let pixels = ctx.getImageData(0, 0, this.width, this.height).data;
//delete canvas
body.removeChild(canvas);
return pixels;

Before those changements, my code were running at about 1.68 iterations per second. Now it runs at about 16.45 iterations per second !

See full code here.

Thanks again !