Re: [osg-users] Shadow vs reflection vs cull visitor
Hello ! I hope you don't mind I address you directly for this question, but as you implemented the view-dependent shadow techniques, you are best placed to answer it. Well... we will see if I will be able to help here. I am currently investigating a problem that seems to be caused by the ordering of RTT passes. I have a scene that uses osgOcean and osgShadow, and the shadow INSIDE the refraction (which is an RTT effect) is late compared to the shadows OUTSIDE the refraction (i.e. above the water level). What I mean by late is that the shadow moves away from the casting object if I move the eye around, but stops moving and is in the right place as soon as I stop moving the eye. So the refraction pass seems to be happening before the shadow pass. I understand. I know this problem well from my experience. The nodes are ordered like this: root ShadowedScene OceanScene the scene And in the OceanScene traverse() method, there is a check if the current camera is the shadow pass camera or analysis camera (for DrawBounds techniques) and just do a regular traversal in that case. My expectation was that the first traversal of the OceanScene would be the shadow pass, then the main pass, and so when we do the refraction RTT (during the main pass), the shadow map would already be correct for the current frame. [..] So the main pass is done (culled) before the shadow pass. I can understand that the bounds calculation needs to cull the shadow receiving scene first, or else we won't know where the shadow map should lie. However, I wonder how this even works. Culling order does not need to be the same as Rendering order. Rendering order is defined by PRE_RENDER / POST_RENDER / NESTED_RENDER camera flag. Cull visitor on the other hand simply traverses the scene tree in first encountered/first processed order. My understanding is that when the cull visitor traverses a scene graph, it accumulates the drawables and state into a list. It does this in the order it visits notes, and might reorder based on state or other criteria (distance to camera for the transparent bin, for example). Sorting is done in Draw stage as far as I know. So given the code above, how does the cullShadowCastingScene() manage to place the drawables for the shadow pass before the ones for the main pass, even though the cullShadowReceivingScene() (culling the main pass) is before? Each camera has render order flag and associated RenderStage+RenderBin set. Cull visitor simply fills these render stage bins and state graphs without checking which will be rendered first. They are waiting for Draw phase to be rendered and they are then drawn in order defined by cameras. Main question here I believe is how to ensure that PRE_RENDER cameras will render in certain desired order. And honestly I don't know definite answer to this question. My tests with analysis shadow cameras seem to suggest that first traversed by cull visitor will be rendered first. For DrawBounds method order of culling is 1: scene / 2: analysis / 3: shadow. Scene has standard order (i am not sure but suppose its NESTED) and analysis and shadow are PRE_RENDER. So if analysis comes before shadow (which I made sure it does) it means that first PRE_RENDER camera traversed by cull visitor will be drawn first. I think for osgOcean's refraction pass to contain correct shadows, I would need to do something similar, i.e. place the drawables for that pass after the shadow pass, but before the main pass, in the render list. I'm actually surprised that's not what happens already, since the sequence should be: cullShadowReceivingScene() -- eventually calls OceanScene::traverse() -- does cull for refraction RTT camera -- then does cull for main pass cullShadowCastingScene() -- does shadow pass, somehow placing results before main pass, which for all it knows should contain the refraction pass And the above order of cull visits explains the effect. I assume refraction RTT camera is PRE_RENDER cam like analisys and shadow. So its rendered first because its culled first. You would need to put refraction cull after cullShadowCastingScene() to be sure it will get rendered at the right moment, I guess. Thanks in advance, You are welcome, but I doubt I helped ;-) Cheers, Wojtek ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Shadow vs reflection vs cull visitor
Hi Wojtek, Well... we will see if I will be able to help here. Yes, I am sure you will! :-) Each camera has render order flag and associated RenderStage+RenderBin set. Ah yes, I had forgotten about the render order number (integer to order more finely between PRE_RENDER cameras and so on). Perhaps that's the answer here? And I checked, the main pass camera is POST_RENDER with a _renderOrderNum of 0. And the above order of cull visits explains the effect. I assume refraction RTT camera is PRE_RENDER cam like analisys and shadow. So its rendered first because its culled first. You would need to put refraction cull after cullShadowCastingScene() to be sure it will get rendered at the right moment, I guess. You're my hero! I just set the refraction camera as PRE_RENDER, but with a number higher than the shadow pass and analysis cameras (I chose 1 since the shadow pass camera is PRE_RENDER 0), and it now happens before the main pass (since that is POST_RENDER 0), but after the shadow and analysis passes! Thanks a million, I owe you yet another beer. Though I kind of hate hard-coding the render order in osgOcean, since the shadow pass could be any render order number imaginable and it won't necessarily work then, but at least now it works with the default shadow technique. I think we might need a render passes manager in OSG that would work out the order of all passes in all libraries whether they know about each other or not... I know, wishful thinking :-) Thanks again, J-S -- __ Jean-Sebastien Guayjean-sebastien.g...@cm-labs.com http://www.cm-labs.com/ http://whitestar02.dyndns-web.com/ ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Shadow vs reflection vs cull visitor
On 5/19/2011 2:07 PM, Wojciech Lewandowski wrote: [..] So the main pass is done (culled) before the shadow pass. I can understand that the bounds calculation needs to cull the shadow receiving scene first, or else we won't know where the shadow map should lie. However, I wonder how this even works. Culling order does not need to be the same as Rendering order. Rendering order is defined by PRE_RENDER / POST_RENDER / NESTED_RENDER camera flag. Cull visitor on the other hand simply traverses the scene tree in first encountered/first processed order. If you have multiple sibling Cameras with the same render order, their draw order is determined by their order in the scene graph (or by the CullVisitor, in other words). The CullVisitor is assembling a graph of RenderBins. RenderStage derives from RenderBin. Each Camera keeps a cache of RenderStages indexed by CullVisitor address. Every time the CullVisitor encounters a Camera node (and the order isn't NESTED_RENDER), it inserts the Camera's RenderStage for that CullVisitor into the render graph. Your Viewer object and/or SceneView has an implicit root Camera, and its corresponding RenderStage is the root node of the render graph. If that root Camera has two PRE_RENDER Camera children, child 0 and child 1, then the child 0 Camera RenderStage gets added to the root Camera RenderStage's list of pre-render stages first, and the child 1 Camera RenderStage goes into that list second. During draw, RenderStage::draw() is called on the list of pre-render stages in the order they appear in the list -- in this case, the same order they were encountered by the CullVisitor. All of the root Camera RenderStage's pre-render children are drawn before ::draw() gets called on the root Camera RenderStage. I hope that helps. -Paul ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Shadow vs reflection vs cull visitor
Hi, I am glad you could solve it. I was not aware of _renderOrderNum index. So thanks, I have learned important new stuff today ;-) I guess you had to override either Ocean or LispSM techniuque already. So you may set _renderOrderNum in overriden code. Depending on which one you overrode, I believe you may either set 1 for Refraction camera or -1 for Analysis and Shadow cams. Cheers, Wojtek Hi Wojtek, Well... we will see if I will be able to help here. Yes, I am sure you will! :-) Each camera has render order flag and associated RenderStage+RenderBin set. Ah yes, I had forgotten about the render order number (integer to order more finely between PRE_RENDER cameras and so on). Perhaps that's the answer here? And I checked, the main pass camera is POST_RENDER with a _renderOrderNum of 0. And the above order of cull visits explains the effect. I assume refraction RTT camera is PRE_RENDER cam like analisys and shadow. So its rendered first because its culled first. You would need to put refraction cull after cullShadowCastingScene() to be sure it will get rendered at the right moment, I guess. You're my hero! I just set the refraction camera as PRE_RENDER, but with a number higher than the shadow pass and analysis cameras (I chose 1 since the shadow pass camera is PRE_RENDER 0), and it now happens before the main pass (since that is POST_RENDER 0), but after the shadow and analysis passes! Thanks a million, I owe you yet another beer. Though I kind of hate hard-coding the render order in osgOcean, since the shadow pass could be any render order number imaginable and it won't necessarily work then, but at least now it works with the default shadow technique. I think we might need a render passes manager in OSG that would work out the order of all passes in all libraries whether they know about each other or not... I know, wishful thinking :-) Thanks again, J-S -- __ Jean-Sebastien Guayjean-sebastien.g...@cm-labs.com http://www.cm-labs.com/ http://whitestar02.dyndns-web.com/ ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Shadow vs reflection vs cull visitor
On 5/19/2011 2:51 PM, Paul Martz wrote: On 5/19/2011 2:07 PM, Wojciech Lewandowski wrote: [..] So the main pass is done (culled) before the shadow pass. I can understand that the bounds calculation needs to cull the shadow receiving scene first, or else we won't know where the shadow map should lie. However, I wonder how this even works. Culling order does not need to be the same as Rendering order. Rendering order is defined by PRE_RENDER / POST_RENDER / NESTED_RENDER camera flag. Cull visitor on the other hand simply traverses the scene tree in first encountered/first processed order. If you have multiple sibling Cameras with the same render order, their draw order is determined by their order in the scene graph (or by the CullVisitor, in other words). Scratch that statement, in light of the render order num discussion in this thread. :-) On the other hand, if the render order number is *also* the same... -Paul ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Shadow vs reflection vs cull visitor
Hi Wojtek, Just a quick follow-up on this: I guess you had to override either Ocean or LispSM techniuque already. I was wondering, if I ever need to override LispSM's ViewData class, how would I go about it? I guess I need to override the *ShadowMap class as well so that it instantiates my new derived ViewData class instead of the default one in the *ShadowMap class itself? I think you've said this before but it's been a while and I can't find it in the archives. Thanks, J-S -- __ Jean-Sebastien Guayjean-sebastien.g...@cm-labs.com http://www.cm-labs.com/ http://whitestar02.dyndns-web.com/ ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Shadow vs reflection vs cull visitor
Hi Wojtek, I guess you had to override either Ocean or LispSM techniuque already. So you may set _renderOrderNum in overriden code. Depending on which one you overrode, I believe you may either set 1 for Refraction camera or -1 for Analysis and Shadow cams. Actually I'm working on osgOcean a lot lately (and I'm co-developer on it so I have commit access). So this is going straight into the osgOcean code. Since the default shadow techniques in OSG all use PRE_RENDER #0 for their shadow pass cameras, it's a safe change to set the refraction one (and other PRE_RENDER passes in osgOcean) to PRE_RENDER #1. Thanks for your help, J-S -- __ Jean-Sebastien Guayjean-sebastien.g...@cm-labs.com http://www.cm-labs.com/ http://whitestar02.dyndns-web.com/ ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
Re: [osg-users] Shadow vs reflection vs cull visitor
Hi Paul, On the other hand, if the render order number is *also* the same... Thanks for going over the basics again for me, I knew all that but for some reason had completely forgotten about the render order number (_renderOrderNum)... So yeah, the RenderStages are ordered in order of _renderOrder (PRE_RENDER before NESTED_RENDER before POST_RENDER), if 2 cameras have the same _renderOrder they are ordered by _renderOrderNum, and if they have the same _renderOrder AND _renderOrderNum they are ordered in the order they were traversed by the CullVisitor (their order in the graph). That explains both why the shadow pass renders before the main pass in StandardShadowMap::ViewData::cull(), and why I was getting the results I was getting for the shadows in the refraction, as Wojtek said: cullShadowReceivingScene() (main pass camera is POST_RENDER #0) .. eventually goes into OceanScene::traverse() .. eventually culls refraction camera (PRE_RENDER #0) cullShadowCastingScene() (shadow pass camera is PRE_RENDER #0) So the order of rendering of the cameras (or RenderStages to be more correct) was: refraction (PRE_RENDER #0 traversed before shadow pass camera) shadow pass (PRE_RENDER #0 traversed after refraction camera) main pass (POST_RENDER #0) which explains why the refraction was using the shadow map from the previous frame, the shadow map for this frame wasn't rendered yet. I knew the error was ordering of the passes, I just didn't know why the order was wrong. So I'm explaining this as much for myself as for others. Now I've set the refraction pass _renderOrderNum to 1 (it could be any number 0 so it occurs after the shadow pass) and everything is fine. Thanks Wojtek and Paul for reminding me of lots of little things, and filling in gaps in my knowledge, thanks to you I was able to fix this bug and my users are happy once again :-) J-S -- __ Jean-Sebastien Guayjean-sebastien.g...@cm-labs.com http://www.cm-labs.com/ http://whitestar02.dyndns-web.com/ ___ osg-users mailing list osg-users@lists.openscenegraph.org http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
