🔒 Locked

SkinnedMesh sample

Started by Brisco Dec 27, 2004 at 5:06 AM 0 replies 550+ views

Original Post

Dec 27, 2004 at 5:06 AM

Hello! I try to implement skinned meshes for character animation in my engine and I understand the theory behind it so far. I looked into the SDK SkinnedMesh sample and found the following code in the .fx file: static const int MAX_MATRICES = 26; float4x3 mWorldMatrixArray[MAX_MATRICES] : WORLDMATRIXARRAY; I'm not very experienced using HLSL, so my first question is: what exactly does WORLDMATRIXARRAY mean? I didn't find anything about it in the SDK documentation... but I think it is nothing else than the matrices set with the D3DTS_WORLDMATRIX(n) macro... well, if this is the case, I have some problem. I am currently working on a GeForce4 Ti4600 (until my GeForce6 arrives) and MaxVertexBlendMatrixIndex=0 on this card... so, it should not be possible to set 26 different world matrices and access it within a shader, or am I misunderstanding it? Next problem is, that I need some fallback version for pre-shader hardware. I thought of switching to software vertex processing for rendering of skinned meshes but don't know if this is really useful. I also asked myself how it is possible to use a different number of blend matrices for every vertex, i.e. 3 blend weights for vertex 1, 4 blend weights for vertex 2 without using VS 2.0 as described in "Skinned Mesh Character Animation with Direct3D 9.0c". Thank you for you help.

Aldenar

Dec 27, 2004 at 9:10 AM

Hi,

The WORLDMATRIXARRAY is the array of bone matrices.

Actually, from a vertex shader, the values you can access are the VertexShaderConstants (See IDirect3DDevice9::SetVertexShaderConstantF). In vertex shader assembler, those are the c[n] (4 floats vector) registers. You have at least 96 of them on a VS 1.1 card, but you can find out the real number using D3DCAPS9::MaxVertexShaderConst. In that example, using 26 4x3 matrices requires 78 (26 * 3) constant registers, so it is possible ;)

If the hardware doesn't support all the features you need, yes, you can create a mixed device and switch software processing on and off. Another thing is that you can do software skinning and output the results to a dynamic vertex buffer on which you can use HW TnL to render. The former method can be useful when you need to render the exact same mesh a few times per frame (eg with shadow volumes).

It is a bit hard to have a variable weight count since VS 1.1 does not support conditional branching. But you can have a vertex format for 4 weights and set the ones you don't use for a given vertex to 0. That way you'll be able to have from 1 to 4 weights per vertex. The drawback is that you'll always process 4 weights even if the last 3 are 0.

Topic Locked

This topic has been locked by a moderator. New replies are not allowed.