After finally getting it right (instead of the first two efforts, which were basically enhanced edge detectors) I’m unhappy with the speed hit I’m getting to keep SSAO enabled.  We’re talking 20fps on the 360, which is obviously not good.  I managed to get it up to a stable 30fps by reducing the size of the SSAO buffer (as I’m blurring it anyway it’s not much of a loss) but unless I can keep that when the game is further developed I’ll keep SSAO off.

Besides, my last change broke it anyway :)  I added directional light shadows, but I was unable to get the “reconstruction of the world position from the depth” right so I switched from keeping a depth buffer to storing positions, which mandated moving to HalfVector4 floating point surfaces.  With SSAO disabled I’m still getting 60fps on the 360 (and a not-too-bad 20fps on my 6600GT) so I’ll probably put that problem off for a while and keep with the easier-to-use FP buffers.

I’ve moved to a 32-bit depth buffer packed into R8G8B8A8, and the increased precision looks to have helped.  It’s still not perfect, but I’m reasonably happy with it for now anyway.

As you may know, I’m developing an XNA game.  While originally a 2D game, I’ve embraced the advantages of 3D* and am currently attempting to do screen-space ambient occlusion (SSAO), a technique first shown off in Crysis.

This is my first halfway-successful attempt, but there’s quite a way to go before it’s actually usable:

* I’ll make another post about why 2D just isn’t feasible for some HD games even if you’re trying to obtain a retro/classic feel.