draw chosen nodes ? multipass ?

[poulpi33]

Hello,

I'm trying to implement shadowmaps, planar reflection and postfx in my app, but i don't know how to make a group of nodes to draw for each pass, as there is only "drawAll()" fonction in irrlicht.

For exemple, in my scene, there is objects (but no all) that need to be drawn in the shadowmap pass with specific shaders, same for reflection and for main render.
In the render to texture tutorial we are using node->setVisible(false), but that would be unnecessary heavy to do a loop on arrays of hundreds of nodes to set visibility and materials before each pass at each frame, isn't it ?

Any advise of a good workflow to do that ? Multiples scenes managers, one for each pass, maybe ?

[poulpi33]

Ive found the following in the api doc:
scene::E_SCENE_NODE_RENDER_PASS
scene::ISceneManager::registerNodeForRendering()
scene::ISceneManager::getSceneNodeRenderPass()

I think it is for multipass rendering of objects to the same target, but maybe i can use that to do what i want, by rendering each pass to a different target ?
I can't find any tutorial or exemple about this in the forum.

To resume what i want to do:

Code: Select all

driver->beginScene(true, true, video::SColor(255,0,0,0));

// shadow depth pass
driver->setRenderTarget(shadowRT, true, true, video::SColor(0,0,0,255));
smgr->setActiveCamera(shadowCam);
draw nodes registered for this pass with their materials registered for this pass

// reflection pass
driver->setRenderTarget(reflectRT, true, true, video::SColor(0,0,0,255));
smgr->setActiveCamera(reflectionCam);
draw nodes registered for this pass with their materials registered for this pass

// main pass
driver->setRenderTarget(mainRT, true, true, video::SColor(0,0,0,255));
smgr->setActiveCamera(mainCam);
draw nodes registered for this pass with their materials registered for this pass

// post fx
driver->setRenderTarget(0, true, true, video::SColor(0,0,0,255));
draw my fullscreen quad with mainRT as texture...

driver->endScene();

[hybrid]

Just use multiple scene managers. The SVN trunk version now also has some additional render pass callback methods which might come in handy if you need some additional control. However, all these things are not really intended for multipass rendering.

[BlindSide]

Just use multiple scene managers.

That would duplicate the meshes right? Or do scene managers share a global mesh cache of some kind?

It's really not a big deal poulpi, the simplest way to do it now is to just use a single scene manager and change the material types of the scene nodes before each drawAll().

but that would be unnecessary heavy to do a loop on arrays of hundreds of nodes to set visibility and materials before each pass at each frame, isn't it ?

No, not really. You have to loop over them to render them in the first place, don't forget that.

[sudi]

Hey there is a much simpler way.
With the LightManager stuff in the svn trunk u have a callback before each renderpass and each scenenode. maybe try that.

[poulpi33]

Thanks for your answers !
I will try to create multiples nodes for each meshs for each renderpass, with one scene manager, so i won't need to modify materials and flags in a loop for each nodes, only set visible the nodes i need.

[poulpi33]

I get it to work, and i use a global parent node for each pass with nodes that need to be drawn in actual pass as childs, so it become very easy to hide/show nodes at each pass (see code below).

But i've got a problem:

The animation of the camera fps is jerky (nothing to do with fps wich is around 3000). I think this is beacause i call drawAll() multiple times (one for each pass), and since this function also do all the animation, the cameraFPS get animated 3 times per frames.
(if i delete the first 2 drawAll() then animation come back normal)
I've tried to set invisible the cameraFPS in passes in wich it is not used, but that doesn't not work, since animation is also done to nodes wich are invisible (wich is normal).
Did not tried but all animations should have this problem.

Any way to work around this ? so that every animations are done once before, or only at, the first drawAll() function ?

[EDIT]
Just found this post, talking about a modifiction of the drawAll() function with a bool parameter to tell if it should call OnAnimate or not. Has it be done since ? Can't find this on the api doc.

Code: Select all

driver->beginScene(true, true, video::SColor(255,0,0,0));

// shadow
driver->setRenderTarget(shadowRT, true, true, video::SColor(0,0,0,255));
smgr->setActiveCamera(shadowCam);
shadowNode->setVisible(true);
reflectNode->setVisible(false);
mainNode->setVisible(false);
smgr->drawAll();

// reflection
driver->setRenderTarget(reflectRT, true, true, video::SColor(0,0,0,255));
smgr->setActiveCamera(reflectCam);
shadowNode->setVisible(false);
reflectNode->setVisible(true);
mainNode->setVisible(false);
smgr->drawAll();

// main
driver->setRenderTarget(0, true, true, video::SColor(0,0,0,255));
smgr->setActiveCamera(fpCam);
shadowNode->setVisible(true);
reflectNode->setVisible(false);
mainNode->setVisible(false);
smgr->drawAll();

driver->endScene();

[hybrid]

No, this has not yet been done. But it's also a problem because we need to update the skinned meshes in order to draw them in the correct pose in each render loop. Hence, Luke's working on this issue.

[poulpi33]

Got it to work.
I first call drawAll() once with all my root nodes set to invisible, so it draw nothing but do animations. Then i manually render my nodes with render(), so i also need to manually cull scene nodes.
Camera animation is now smooth, here is my code:
(note: all my nodes are directs childs of the "root" nodes, that why my simple loops for firsts childs is enough here)

Code: Select all

shadowNode->setVisible(false);
reflectNode->setVisible(false);
skyNode->setVisible(false);
mainNode->setVisible(false);
postNode->setVisible(false);
gui3dNode->setVisible(false);

core::list<scene::ISceneNode*> nodes;
core::list<scene::ISceneNode*>::Iterator iNode;
scene::ISceneNode* node = NULL;

int lastFPS = -1;
int fps = 0;
int primitives = 0;

const scene::SViewFrustum* viewFrustrum;
core::aabbox3d<f32> viewBbox;
core::aabbox3d<f32> nodeBbox;

while (device->run())
    if (device->isWindowActive())
    {
        driver->beginScene(true, true);
        
        smgr->drawAll(); //animes

        // shadow
        driver->setRenderTarget(shadowRT, true, true, video::SColor(255, 255, 255, 255));
        smgr->setActiveCamera(shadowCam);
        smgr->getActiveCamera()->render();
        viewFrustrum = shadowCam->getViewFrustum();
        viewBbox = viewFrustrum->getBoundingBox();
        nodes = shadowNode->getChildren();
        for (iNode=nodes.begin(); iNode!=nodes.end(); ++iNode)
        {
            node = *iNode;
            nodeBbox = node->getBoundingBox();
            if (nodeBbox.isFullInside(viewBbox) || nodeBbox.intersectsWithBox(viewBbox))
            {
                node->render();
            }
        }

        // reflection
        driver->setRenderTarget(reflectRT, true, true);
        smgr->setActiveCamera(reflectCam);
        smgr->getActiveCamera()->render();
        viewFrustrum = shadowCam->getViewFrustum();
        viewBbox = viewFrustrum->getBoundingBox();
        nodes = reflectNode->getChildren();
        for (iNode=nodes.begin(); iNode!=nodes.end(); ++iNode)
        {
            node = *iNode;
            nodeBbox = node->getBoundingBox();
            if (nodeBbox.isFullInside(viewBbox) || nodeBbox.intersectsWithBox(viewBbox))
            {
                node->render();
            }
        }

        // main
        driver->setRenderTarget(mainRT, false, true);
        smgr->setActiveCamera(fpCam);
        smgr->getActiveCamera()->render();
        viewFrustrum = shadowCam->getViewFrustum();
        viewBbox = viewFrustrum->getBoundingBox();
        skyNode->render();
        nodes = mainNode->getChildren();
        for (iNode=nodes.begin(); iNode!=nodes.end(); ++iNode)
        {
            node = *iNode;
            nodeBbox = node->getBoundingBox();
            if (nodeBbox.isFullInside(viewBbox) || nodeBbox.intersectsWithBox(viewBbox))
            {
                node->render();
            }
        }

        // post
        driver->setRenderTarget(0, false, false);
        postNode->render();

        // 3D gui
        driver->clearZBuffer ();
        nodes = gui3dNode->getChildren();
        for (iNode=nodes.begin(); iNode!=nodes.end(); ++iNode)
        {
            node = *iNode;
            nodeBbox = node->getBoundingBox();
            if (nodeBbox.isFullInside(viewBbox) || nodeBbox.intersectsWithBox(viewBbox))
            {
                node->render();
            }
        }

        // 2D gui
        gui->drawAll();

        driver->endScene();/

getPrimitiveCountDrawn() now always return the same value. I first thought my culling was not working, but i disabled nodeBbox.intersectsWithBox(viewBbox) for test, and it work. So my question is, do getPrimitiveCountDrawn() work only with drawAll() ?
If so then this function should have been in scene::ISceneManager and not in video::IVideoDriver, should it ?

Would it be faster to first loop all nodes to check culling, adding them to arrays, and then loop render() on those arrays ?