Swift SpriteKit/Quartz/Metal: How to draw low-level pixel data at 60FPS?

Question

I am working on a NES Emulator project for macOS written in Swift. However, I'm having a hard time finding a solution to render the internal VRAM into some sort of view in an efficient way.

Simplifying the problem, let's assume I have an array of RGBA pixel values representing my 256x240 pixel screen:

var viewport: Data = Data(count: width * height * 4 /* one byte each color channel */);

My code will directly change the internal pixels of this viewport changing individual pixels' color. The goal is to render these pixels 60 times per second, in the most efficient way possible.

Failed attempts

So far I tested both SpriteKit, and Quartz but I always face the same issue: the overhead is so large, that the code barely runs at 30FPS, and it's not even doing anything special: just changing pixels' color randomly.

SpriteKit code

let node = SKSpriteNode(color: NSColor.black, size: self.size);

func display() {
  self.node.texture = SKTexture(data: self.pixels, size: self.size);
}

Quartz code

// Custom NSView class

override func draw(_ dirtyRect: NSRect) {
  if let context = NSGraphicsContext.current?.cgContext {
    let bitmapInfo = CGBitmapInfo(rawValue: CGImageAlphaInfo.noneSkipLast.rawValue)
    let ctx = CGContext(data: &data,
                        width: width, height: height,
                        bitsPerComponent: 8,
                        bytesPerRow: width * 4,
                        space: CGColorSpaceCreateDeviceRGB(),
                        bitmapInfo: bitmapInfo.rawValue)
     let image = ctx?.makeImage()!
     context.draw(image, in: self.frame);
   }
}

Question

Is there a better way to implement such functionality? Or are these tools too much high-level for this kind of task?

PS: I am also looking at GLUT/OpenGL, but I know OpenGL has been deprecated and I'm wondering if there is a new "official" way to code this

Answer 1

After a day of debugging turns out the actual bottleneck was... Swift itself! I made a simple command line project to measure the time needed to update an array of 250,000 bytes and the result was a surprise (at least for me).

Swift took 58ms. So why this much? The problem is that the XCode debugging build has a terrible overhead I was not expecting: coming from the C++ world, I was expecting the same drop in performance of a Debug build in clang, but that was really not the case.

Once archived and exported the build, the problem was vastly reduced.

Final note: I then proceeded to make some tests and compare OpenGL, SpriteKit, and Quartz. The latter was the only framework I could not get the 60fps I was looking for, not even in Release mode.