Jump to content

  • Log In with Google      Sign In   
  • Create Account

Multiple Render Target Rules


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
8 replies to this topic

#1 weeska   Members   -  Reputation: 255

Like
0Likes
Like

Posted 15 August 2012 - 11:12 AM

Hi there,

in some presentations there are statements, that one can only use Render Targets of the same depth and number of channels. Is this still true for newer versions of OpenGL?

I tried looking it up in the specs, but i didn't find anything useful except framebuffer completeness rules. I wrote a sample program where i use different number of channels, different depths and so on, and it works.

I just need to know if this is officially supported or some implementation-specific "advantage".

Background is, when writing about deferred shading, some people said they're limited in what formats they can use due to resctrictions with multiple render targets. Now i need to know if that's true anymore.

thanks in advance,
weeska

Sponsor:

#2 larspensjo   Members   -  Reputation: 1561

Like
0Likes
Like

Posted 15 August 2012 - 02:16 PM

You are talking about FBO:s, aren't you? A FBO never has more than one depth buffer, but can have many render targets.

I am not sure what you mean by "channel", but I suppose you mean a render target. If you are talking about FBO:s with different sets of render targets, I can't see why and how there could be a restriction that all FBO:s have to use the same setup.

In my game, I use different FBO:s. One for the deferred lighting, one for shadow maps, one for terrain maps, etc. And they are all different.

Maybe you are thinking about the vertex attrib pointers. In that case, you should use the same numbers for these for every shader program that will be using the FBO. Do that either by using the layout directive in the shader logic (easiest), or by calling glBindAttribLocation() before the linking phase. That is, you define the index instead of query for them. If you do this, the same vertex attribute index will be used for all programs using your FBO, which means you can have a common logic for glVertexAttribPointer() and glEnableVertexAttribArray(). I do this by having a common program shader base class (in C++) that defines an enum with all render targets. The programs doesn't have to use all render targets available in the FBO. But you are still allowed to call glEnableVertexAttribArray() for all of them (making it easier to have common code).
Current project: Ephenation.
Sharing OpenGL experiences: http://ephenationopengl.blogspot.com/

#3 Ignifex   Members   -  Reputation: 507

Like
3Likes
Like

Posted 15 August 2012 - 02:46 PM

I believe the Op is talking about MRT, the use of multiple color attachments, render buffers, with a single FBO. Note that each render buffer is also considered an implementation of a Render Target, although this term is far more vague.

Yes, there are obvious restrictions to the types of the separate render buffers attached. All must have the same number of color channels. However, data types may differ. I have used 4 channel half-float and float buffers in MRT without problem.

Honestly, I am not sure about official support rules for MRT, nor do I know the current implementation details, but I thought it would be wise to clear up any confusion on the topic. That said, I would advise to keep using what layouts work for you, as it will likely work on other systems. If it ends up not working with a different driver or card, deal with it then.

Edited by Ignifex, 15 August 2012 - 02:47 PM.


#4 weeska   Members   -  Reputation: 255

Like
0Likes
Like

Posted 15 August 2012 - 04:44 PM

Sorry, i shoud have made that more clear: i'm talking about Multiple Render Targets.

In my case i'm using 3 color attachments as textures:
  • RGBA8
  • RG16F
  • RGB32F
So it's possible - at least with my gpu (GL 3.3) - to have a different number of channels, as it all works as expected. I think i assume that it's not possible in general, i think.

Thanks for your help ;)

Edited by weeska, 15 August 2012 - 04:48 PM.


#5 BitMaster   Crossbones+   -  Reputation: 4442

Like
2Likes
Like

Posted 16 August 2012 - 05:24 AM

Yes, there are obvious restrictions to the types of the separate render buffers attached. All must have the same number of color channels. However, data types may differ. I have used 4 channel half-float and float buffers in MRT without problem.


Are you absolutely certain that is still a requirements in newer versions? I remember reading (although the source currently eludes me) that it was at some time a requirement but was removed in one of the OpenGL revisions in between.
Also, I have a hobby project running on an AMD card (which generally make more of an effort to be standard compliant) which has a four channel byte render target and a one channel integer render buffer bound to the same FBO and I can render to both at the same time with the expected results.

#6 Ignifex   Members   -  Reputation: 507

Like
1Likes
Like

Posted 16 August 2012 - 06:16 AM

Hmm, well, as I said, I am not sure about official support. The last time I experimented with the possibilities on a GL 3.3 system (pre-Fermi card), I wasn't able to create such an FBO, although that has been a while.

Currently reading through the opengl Wiki pages. The framebuffer completeness rules only provide one limitation:
- All images must have the same number of multisample samples
Also, on the Image Format page, a list of required formats for render buffers is provided.
This says that 1, 2 and 4 channels with some typical formats must be supported. 3 channel textures and some other formats are optional for implementations. This does not appear to mention anything on multiple render targets.

Number of channels used to be a limitation, but clearly isn't anymore. If anyone can find a good source, it would help to clear this up.

Edited by Ignifex, 16 August 2012 - 06:27 AM.


#7 Geometrian   Crossbones+   -  Reputation: 1602

Like
1Likes
Like

Posted 17 August 2012 - 03:40 AM

  • RGBA8
  • RG16F
  • RGB32F

Technically, the number of channels is a semantic quantity--the underlying hardware will use whichever internal format it wants to. E.g., RGB is almost always some for of RGBA under the covers, because hardware operates on ones and fours. For example, I would guess that RG16F would be stored in four 8-bit channels. You probably knew that though.

I regret that I don't know for sure whether it is a portability limitation to include color attachments/rendertargets of differing internal formats. It never occurred to me, but everything I've seen seems to suggest it is possible, including its working on my card. It's an interesting concern though.

A helpful analogy might be to compare FBOs to a standard OpenGL context, wherein you can request a number of different buffers--stencil, depth, color, accumulation, etc. These are necessarily different formats, but they've all cooperated with each other for as long as they've been around. A FBO is just like a standard context, except you can choose whether you want to be able to read the buffers directly (attach a texture instead of a renderbuffer to an attachment point). I see no reason why FBOs should be different.

-G

Edited by Geometrian, 17 August 2012 - 03:41 AM.

And a Unix user said rm -rf *.* and all was null and void...|There's no place like 127.0.0.1|The Application "Programmer" has unexpectedly quit. An error of type A.M. has occurred.

#8 AgentC   Members   -  Reputation: 1420

Like
1Likes
Like

Posted 17 August 2012 - 04:44 AM

The rules should be more relaxed on OpenGL 3.0 and onwards. However if you look at http://www.opengl.org/wiki/Framebuffer_Object it still says: "There's one more rule that can trip you up: The implementation likes your combination of attached image formats. (GL_FRAMEBUFFER_UNSUPPORTED when false)."

I've only used OpenGL 2.0 with EXT_framebuffer_object extension and with that, different color formats would work on recent GPUs on Windows/Linux, but OS X liked to play it strict and gave an error when the color formats differed. So if possible, I'd suggest testing on OS X to see if you're compatible.

Every time you add a boolean member variable, God kills a kitten. Every time you create a Manager class, God kills a kitten. Every time you create a Singleton...

Urho3D (engine)  Hessian (C64 game project)


#9 weeska   Members   -  Reputation: 255

Like
0Likes
Like

Posted 17 August 2012 - 05:20 AM

The rules should be more relaxed on OpenGL 3.0 and onwards. However if you look at http://www.opengl.or...mebuffer_Object it still says: "There's one more rule that can trip you up: The implementation likes your combination of attached image formats. (GL_FRAMEBUFFER_UNSUPPORTED when false)."


It also says:
"Basically, don't concern yourself with GL_FRAMEBUFFER_UNSUPPORTED too much. Check for it, but you'll be fine as long as you stick to the required formats."

The way i understand it is, use the formats that have to be supported, and there shouldn't be a problem.


Checking for support on OS X is no option nor needed at the moment, but it's always good to know who's strict with these optional parts.

Thanks for the answers so far ;)




Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS