Performance of pixel shaders vs. SpriteBatch: XNA

Question

Precondition: I read this question/answer about using shaders, or spritebatch, to render and mark a sprite.

I need to do something like that. I also have a 2D lighting PoC which I need to write. The way it will work will basically be something like:

• Draw all the sprites
• Draw lighting gradients to create a lighting texture
• Multiply/add the lighting texture to achieve different effects

(I use multiple passes of add/multiply the lighting texture to achieve different effects.)

My question is really about a generalization: can I say with certainty that pixel shaders are always faster than adding/multiplying textures to the SpriteBatch? Or that adding/multiplying is always faster? Or if it's not generalizable, how do I decide which approach to take, given that I can probably code either of them?

(If it matters, I'm using MonoGame 3.0 beta for Windows games)

Answer 1

if i understood correctely, you are asking about what is faster, mix colors in pixel shaders or mix colors using blend states ?

In general case, using Blender States should be faster cause Video Boards normally have specifically dedicated piece of harware to perform this operation. (In this aproach you can also use the pixel shader to perform other funcy stuff in the texture you read)

But, for example, if you need to perform 3 pass to get the result you want (each one with a different blend state), probably mixing the color in just one pass in one pixel shader would be faster.

To be honest (assuming you are targeting PC), unless you are doing MASSIVE color mixing, both approach wont be a bottleneck in your application.

just to clarify one last thing. Spritebatch is just an easy to let you handle textures and text easily (and also performatic way), behind the curtains, it is actually drawing a full screen quad, using a shader ......