Jump to content

  • Log In with Google      Sign In   
  • Create Account

We need your feedback on a survey! Each completed response supports our community and gives you a chance to win a $25 Amazon gift card!


Rendering multiple complex animated models


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
1 reply to this topic

#1 NEXUSKill   Members   -  Reputation: 466

Like
0Likes
Like

Posted 13 September 2012 - 09:27 AM

By now I think we've all seen the tank example, its a hierarchically composed model. Each part of the model has an associated effect, embedded into the asset and its own vertex and index buffer.

The provided examples always take one object to render (the tank) which has more or less the same texture and effect all over. They navigate the model as a tree rendering each part in sequence applying transformations until the tank is fully composed on screen, it works and looks nice and the code is fairly neat.

However, this looks awfully like a "laboratory conditions" kind of success that would not hold a decent FPS rate in real gameplay.
Suppose we have many objects to render, each of which has various parts, some objects are instances of the same model, some use different models, some models have parts that share the same properties as those of a part of another model, like two different models that both have rubber wheels. Most models have two or three, if not more, different effects on their different parts, like wheels, hull and cockpit glass.
The rendering cycle proposed by the tank example looks like it would be changing shaders and textures constantly in that scenario and does not apply any kind of batching.

Is the tank example good for a reason I may not be seeing or would a much better approach be grouping individual sub models by shader + texture (Materials) and rendering those in batches thus reducing render state changes?

[source lang="csharp"] public void Draw(Matrix viewMatrix, Matrix projectionMatrix) { // Apply matrices to the relevant bones, as discussed in the Simple // Animation Sample. leftBackWheelBone.Transform = wheelRollMatrix * leftBackWheelTransform; rightBackWheelBone.Transform = wheelRollMatrix * rightBackWheelTransform; leftFrontWheelBone.Transform = wheelRollMatrix * leftFrontWheelTransform; rightFrontWheelBone.Transform = wheelRollMatrix * rightFrontWheelTransform; // now that we've updated the wheels' transforms, we can create an array // of absolute transforms for all of the bones, and then use it to draw. Matrix[] boneTransforms = new Matrix[model.Bones.Count]; model.CopyAbsoluteBoneTransformsTo(boneTransforms); // calculate the tank's world matrix, which will be a combination of our // orientation and a translation matrix that will put us at at the correct // position. Matrix worldMatrix = orientation * Matrix.CreateTranslation(Position); foreach (ModelMesh mesh in model.Meshes) { foreach (BasicEffect effect in mesh.Effects) { effect.World = boneTransforms[mesh.ParentBone.Index] * worldMatrix; effect.View = viewMatrix; effect.Projection = projectionMatrix; effect.EnableDefaultLighting(); effect.PreferPerPixelLighting = true; // Set the fog to match the black background color effect.FogEnabled = true; effect.FogColor = Vector3.Zero; effect.FogStart = 1000; effect.FogEnd = 3200; } mesh.Draw(); } }[/source]
Game making is godlike

LinkedIn profile: http://ar.linkedin.com/pub/andres-ricardo-chamarra/2a/28a/272



Sponsor:

#2 Drakken255   Members   -  Reputation: 173

Like
0Likes
Like

Posted 14 September 2012 - 12:16 AM

In my experience, XNA works far better with few large batches than with a bunch of small ones. It is also best to have effect and state changes kept to as few as possible. So if you have 10 models each with 5 sub-models, then you should instance each sub-model of the same type together so you have 5 batches of 10 instead of 10 batches of 5. Group all wheels together and all doors together instead of simply grouping each car's sub-models together. If a batch of sub-models requires an effect or state change, then change between batches. Each sub-model's world should be calculated by multiplying its world relative to its parent times its parent's world.

EDIT: Also, this allows you to only need one model in memory. When you send a batch of sub-models, you just simply send that one sub-model with an array of transforms. Take a look at the instancing sample by Microsoft. This method should allow you to have upwards of 100 complex models or possibly tens of thousands of simple models rendering at once. (Minecraft uses instancing.)

Edited by Drakken255, 14 September 2012 - 12:22 AM.





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