MultiPass & Rendering Order

Engines, Libraries and Tools Xith3D Forums
#21

I took a quick look in the source. To configure the CanvasPeer to draw the render passes on top of each other, you have to call


// clear the buffer...
canvasPeer.setDisableClearBuffer(false);
// ...but specify, that depth-buffer clearing is not necessary
canvasPeer.setFullOverpaint(true);

The MultiPassView however only calls the first in it’s prepareRenderOnce() method:


// MultiPassView:195 ...
prepareRenderOnce(RenderPass pass, boolean forceClearBuffer, boolean forceSwapBuffers, Canvas3D canvas3D)
{
        // (...)
        if (canvas3D.get3DPeer() instanceof com.xith3d.render.jsr231.CanvasPeerImpl)
        {
            com.xith3d.render.jsr231.CanvasPeerImpl canvasPeerImpl = (com.xith3d.render.jsr231.CanvasPeerImpl)(canvas3D.get3DPeer());
            // set clear buffer state
            canvasPeerImpl.setDisableClearBuffer(!forceClearBuffer);
            // set force swap state
            canvasPeerImpl.setForceNoSwap(!forceSwapBuffers);
        }
        else if (canvas3D.get3DPeer() instanceof com.xith3d.render.lwjgl.CanvasPeerImpl)
        {
            com.xith3d.render.lwjgl.CanvasPeerImpl canvasPeerImpl = (com.xith3d.render.lwjgl.CanvasPeerImpl)(canvas3D.get3DPeer());
            // set clear buffer state
            canvasPeerImpl.setDisableClearBuffer(forceClearBuffer);
            // set force swap state
            //canvasPeerImpl.setForceNoSwap(!forceSwapBuffers);
        }
}

which results in either no clearing at all or full clearing including the color buffer.

So to solve this, a new property - named “layered” for example - should be added to MultiPassView. If this is set to true, the two canvasPeer-properties should be set like the code at the top of this post for all render passes but the first one.

I’ll try this tomorrow, maybe :wink:

#22

Ha!.. just tested it and it seems to work. The only drawback is, that you also loose the stencil buffer in “layered-mode”, but I don’t know, if someone might miss it in this context…

#23

Great! Great! ;D Thank you very much.

Isn’t multipass rendering always layered? At least it should be the default. What is the effort of multipass rendering when it’s not layered?

Marvin

#24

Dunno… maybe to better organize your scene by using multiple independent scene graphs!? Anyway, making the layered variant the default is a good idea as long as there are no compatibility concerns to change the default behaviour of the MultiPassView…

#25

cylab, great to see you contribute !

When I saw “clearBuffers” I also had suspicion that it may be not only the depth buffer.

#26

How to proceed? Should I send a patch or can I have write access to the cvs rep?

I might contibute a little (render-to-texture in MultiPassView and terrain maybe), but I have little time due to work and am not always very motivated to code after a 10hour day with J2EE :wink:

#27

Please register at xith-tk and I’ll give you Developer access as soon as possible.

#28

Great work cylab, I was waiting for multy-pass to really start working :wink: So looking forward to try it out when fix is comitted.
Thanks!

#29

The I’d suggest you to further concentrate on terrain. It is a frequently requested feature.

#30

I thought about porting this, but haven’t analysed it’s source yet.

#31

Looks really really great. Would be fine, if we had it in xith… :wink:

#32

just registered…

#33

I tried this out. But with no success. for all passes the following code is executed:


canvas3D.get3DPeer().setDisableClearBuffer([b]!isFirstPass[/b]);
canvas3D.get3DPeer().setFullOverpaint([b]!isFirstPass[/b]);
canvas3D.get3DPeer().setForceNoSwap([b]!isLastPass[/b]);

Did I misunderstand anything?

I uploaded the new version. Please refer to this for further discussion.

btw. If you’re not utilizing the layered mode, you can even not-use the MultiPassView or put everything in one RenderPass.

Marvin

EDIT: Yeehaa! 500 Posts! ;D ;D ;D

#34

:frowning:
I was just waiting for cvs-access to check my changes in…

#35

Sorry, but I made further changes of cleanup and your changes were just a matter of about three lines of code. So I just made it for quick.

But it doesn’t work anyway. So please check, what I made wrong and check it in.

Marvin

#36

Nevermind… I’ll merge it as soon as Amos gets back from whatever is keeping him away and gives me cvs access :wink:

I have also made a fix for texture-loading in the MD2Test, so…

In layered mode you have to enable the buffer clearing for every pass:


canvas3D.get3DPeer().setDisableClearBuffer(false);
canvas3D.get3DPeer().setFullOverpaint(!isFirstPass);

If you test this, don’t commit; in my changes I have already refactored the different renderOnce()-Methods to contain an additional parameter and the MultiPassView to contain the layered property along with getters and setters.

What do you mean? Why?

Mathias

#37

access just granted… :wink:

#38

It’s called Lycée.

;D

#39

[quote="<MagicSpark.org [ BlueSky ]>,post:37,topic:27917"]
access just granted… :wink:
[/quote]
Thanks!

[quote="<MagicSpark.org [ BlueSky ]>,post:38,topic:27917"]
It’s called Lycée.
[/quote]
I was just about to ask you about a photo of her, but then I noticed google only found pictures of crappy schools ;D

Anyway - school on sunday ? :stuck_out_tongue:

#40

I was just about to ask you about a photo of her, but then I noticed google only found pictures of crappy schools ;D
[/quote]
Aow ;D funny.

Oh no, no school on sunday, but err… no computer either (parents, grrr…).