Skip to main content
GameDev.net gamedev.net
🔒 Locked

depth stencil framebuffer problem

Started by Ponkx Apr 16, 2010 at 2:57 AM 4 replies 3.7k views
Original Post
Ponkx
Ponkx
Hello, I have setup FBO depth stencil texture, and everything works fine, but problem is, when I try to bind that depth texture, and later bind that FBO again, then contents of stencil buffer is damaged (it looks, that it is rewriten by some depth values). My code looks somehow like this: Generation of FBO: glGenFramebuffersEXT(1, &m_gl_framebuffer); glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, m_gl_framebuffer); glGenTextures(1,&depthTextureID); glBindTexture(GL_TEXTURE_2D,depthTextureID); glTexImage2D (GL_TEXTURE_2D,0,GL_DEPTH24_STENCIL8_EXT,800,600,0,GL_DEPTH_STENCIL_EXT,GL_UNSIGNED_INT_24_8_EXT, NULL); glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_DEPTH_ATTACHMENT_EXT, GL_TEXTURE_2D, depthTextureID, 0); glFramebufferTexture2DEXT(GL_FRAMEBUFFER_EXT, GL_STENCIL_ATTACHMENT_EXT, GL_TEXTURE_2D, depthTextureID, 0); test for rewrite: glBindFramebufferEXT(GL_FRAMEBUFFER_EXT, m_gl_framebuffer); glClearDepth(0.500000); glClearStencil(1); glClear(GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT); glBindFramebufferEXT(GL_FRAMEBUFFER_EXT,0); //Binding of texture - if commented, then stencil is ok! glActiveTextureARB(GL_TEXTURE0); glBindTexture(GL_TEXTURE_2D,depthTextureID); glActiveTextureARB(GL_TEXTURE0); glBindTexture(GL_TEXTURE_2D,0); glBindFramebufferEXT(GL_FRAMEBUFFER_EXT,m_gl_framebuffer); //<--- here stencil buffer is damaged! glBindFramebufferEXT(GL_FRAMEBUFFER_EXT,0); If I dont bind that texture, then everything is OK :( Did anyone encounter this problem too? Thanks for help.
Spikx
Spikx
I think I have exactly the same problem. But it happens only on ATi cards (or at least on a Radeon HD 4870). In one of my projects I use a deferred shading engine. I have an FBO with a packed depth stencil buffer as texture which is configured as a Multiple Rendering Target. In the geometry pass this MRT is filled with color, depth and stencil values (I stencil the geometry, so that later on only the actual geometry fragments get shaded).

Then in the lighting stage, a screen aligned rectangle is drawn, in order to do the shading computation for each fragment. The fragment shaders gets the diffuse texture, normal texture, specular texture and of course the depth texture from the MRT. However, when I activate the stencil test, I get something like this:

http://dl.dropbox.com/u/2309215/tree_stencil.gif

(I used a fragment shader here, that simply puts out white. Thus, all pixels where the stencil test passes are white, the other ones are black.)

It looks like the stencil test is using the depth buffer values instead of the actual stencil buffer values that were presumably written before. Or, as you said, the stencil buffer gets somehow "destroyed" when the depth texture is bound (which is the case for me in the lighting stage, since I need the depth values from the depth texture there).

[Edited by - Spikx on April 21, 2010 4:16:41 AM]
Spikx
Spikx
I just checked in my application and I can confirm what you said:

When I bind the depth stencil buffer texture, the stencil buffer gets either "destroyed" or the stencil test actually accesses the depth buffer part of the packed depth stencil buffer (as seen in the picture above for example). Though only on ATi cards.


I don't know how to solve this. The only workaround I could think of (for me at least), is to do the stenciling in an additional rendering pass, in a separate stencil buffer. This way I can later on bind a depth buffer only texture from the previous rendering pass and attach this separate buffer to the FBO (instead of attaching and binding it). This would be a bit tedious. Damn you ATi! ;)

Hm... maybe it works, if I only make the stencil attachment in my lighting pass and no depth attachment, have to check that now. // edit: nah, didn't work.
Ponkx
Ponkx
Well, I tried this on nvidia, and it worked, so problem is ATI only. So, I installed new catalyst drivers, and now it looks OK.

BTW: The reason why I didn't have the newest drivers is, that I work on sony notebook with ATI 5650 and when I was checking for driver updates using some sony vaio updates, it reported that no new update is available. Then I looked right into the catalyst control center and found, that I have 9.11 drivers. So, I tried to install newest drivers directly from the ATI, but it said, that I dont have ATI hardware, or something like that. Finally I read this post and installed drivers from that page and now it works. ...Goood work sony, really...

But still there is a problem, when I want to bind depth buffer texture in shader and at the same time use stencil buffer to do some stenciling. I hoped, that switch to OGL30 (which I wanted to do anyway) could solve this, but is seems, that it doesn't.
Spikx
Spikx
Hmmm... I already tried Catalyst 10.3, but it didn't change anything. Maybe I'll try a clean driver reinstall with Catalyst 10.4 which just came out.

// edit: a just read that the problem didn't really get fixed. I also have to bind the texture and use stenciling at the same time.
Ponkx
Ponkx
Hm.. I was unable to solve that (do stenciling and bind same depth texture in shader) - maybe it is the same limitation as in DX10, so now I'm using some ugly copy with glBlitFramebuffer.

It even drives my crazy, that I cannot have my own FBO depth buffer when I want to render into the GL back buffer (which is possible in DX10).

If someone have some clever trick how to do it I would like to hear...

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.

Sign in to reply to this topic.