MRTs, which depth is used?

[hendu]

When drawing to multiple render targets, which depth buffer is used? Both for drawing, and for checking?

From cursory reading of the irr GL driver code, it would appear that no depth is used at all with MRT's. Is this so?

[CuteAlien]

In COpenGLDriver::addRenderTargetTexture you find the line:

cpp Code: Select all

 
 success = static_cast<video::COpenGLFBODepthTexture*>(tex)->attach(rtt);
 

That attaches a depth-buffer it seems. And creation is in COpenGLTexture COpenGLFBODepthTexture::COpenGLFBODepthTexture.
Not sure which parameters are used in there in your case, maybe step through once.

[Mel]

I think he means which depth is used to perform the ZBuffering when MRTs are used.

Say for instance you have 4 projection matrices, and you use each to perform a render to each rendertarget, which Z value is used? or can be used all of them, or... ? i had the same doubt myself

[hendu]

@CuteAlien

Yep, that attaches a depth texture to that RTT. But when setting multiple RTTs, it seems only the color textures are attached:

GL_COLOR_ATTACHMENT0_EXT+i

There is no mention at all of GL_DEPTH_ATTACHMENT. I get a huge fps drop now when drawing to MRTs vs the screen, so it does look like there's no depth used at all.


@Mel

Indeed. Quick reading of the GL docs suggests that only one depth buffer can be attached at any time. It also seems that irrlicht attaches none when using MRTs.

[hendu]

Diving deeper, it seems the first FBO RTT's depth is the one that is used.

[hybrid]

I would have thought so too, as all the settings are used from the first FBO, unless they can be specified per target. And all RTTs have a depth buffer in Irrlicht (so far).