After Writing to a RenderTarget, How to Efficiently Clone the Output?

Question

XNA noob here, learning every day. I just worked out how to composite multiple textures into one using a RenderTarget2D. However, while I can use the RenderTarget2D as a Texture2D for most purposes, there's a critical difference: these rendered textures are lost when the backbuffer is resized (and no doubt under other circumstances, like the graphics device running low on memory).

For the moment, I'm just copying the finished RenderTarget2D into a new non-volatile Texture2D object. My code to do so is pretty fugly, though. Is there a more graceful way to do this? Maybe I'm just tired but I can't find the answer on Google or SO.

Slightly simplified:

public static Texture2D  MergeTextures(int width, int height, IEnumerable<Tuple<Texture2D, Color>> textures)
    {
    RenderTarget2D  buffer = new RenderTarget2D(_device, width, height);

    _device.SetRenderTarget(buffer);
    _device.Clear(Color.Transparent);

    SpriteBatch  spriteBatch = new SpriteBatch(_device);
    spriteBatch.Begin(SpriteSortMode.Immediate, BlendState.NonPremultiplied);

    // Paint each texture over the one before, in the appropriate color
    Rectangle  rectangle = new Rectangle(0, 0, width, height);
    foreach (Tuple<Texture2D, Color> texture in textures)
        spriteBatch.Draw(texture.Item1, rectangle, texture.Item2);

    spriteBatch.End();
    _device.SetRenderTarget((RenderTarget2D)null);

    // Write the merged texture to a Texture2D, so we don't lose it when resizing the back buffer
    // This is POWERFUL ugly code, and probably terribly, terribly slow
    Texture2D  mergedTexture = new Texture2D(_device, width, height);
    Color[]    content       = new Color[width * height];
    buffer.GetData<Color>(content);
    mergedTexture.SetData<Color>(content);
    return mergedTexture;
    }

I suppose I should check for IsContentLost and re-render as needed, but this happens in the middle of my main drawing loop, and of course you can't nest SpriteBatches. I could maintain a "render TODO" list, handle those after the main SpriteBatch ends, and then they'd be available for the next frame. Is that the preferred strategy?

This code is only called a few times, so performance isn't a concern, but I'd like to learn how to do things right.

Andrew Russell Andrew Russell · Accepted Answer · 2011-04-01T05:50:03

Actually your code is not so bad if you're generating textures in a once-off process when you'd normally load content (game start, level change, room change, etc). You're transferring textures between CPU and GPU, same thing you'd be doing loading plain ol' textures. It's simple and it works!

If you're generating your textures more frequently, and it starts to become a per-frame cost, rather than a load-time cost, then you will want to worry about its performance and perhaps keeping them as render targets.

You shouldn't get ContentLost in the middle of drawing, so you can safely just respond to that event and recreate the render targets then. Or you can check for IsContentLost on each of them, ideally at the start of your frame before you render anything else. Either way everything should be checked before your SpriteBatch begins.

(Normally when using render targets you're regenerating them each frame anyway, so you don't need to check them in that case.)

After Writing to a RenderTarget, How to Efficiently Clone the Output?

3 Answers