     |
 |
 |
 | | | | |
 |
|||
 |
||||||
          |
 |
 |
||||
|
(Page 1/3)
Next 
To overcome these limitations, models that use a number of joints above the number directly supported by the hardware can be split up so every partition is processed by the graphics card in one pass. If the algorithm used to split the mesh results in inefficient partitioning this can have a severe impact on data throughput and force the artists to use less joints and/or simpler models (which means they have to redo much of their work, and/or limit the quality of animation). This article describes a preprocessing software implementation that can handle an unlimited number of joints and at the same time partitions the model to allow maximal performance. Multiple target platforms are supported by creating classes that simulate the architecture of the target hardware and a number of different optimization heuristics can be used to search for the best solution. We have found that the software works very well and nicely fits in our toolchain, allowing programmers to easily adjust the software for new hardware and optimization techniques while at the same time providing quick turnaround time for the artists when (pre)viewing models in the game engine. Skinned MeshesThe use of skinning when rendering meshes is nowadays widely used. The technique works as follows: For a model, a skeleton is defined by using a hierarchy of bones and the joints of the skeleton are attached to the vertices of the mesh (the “skin”).
To animate the mesh we only have to animate the skeleton because the vertices of the mesh are deformed based on the joints they are attached to. The vertices of the mesh can be attached to a number of joints. We need to attach a vertex to at least one joint (otherwise the vertex will remain fixed while the rest of the mesh is animated) and for present-day game animation of human-like characters, animators need at most four joint influences per vertex. If we use exactly one joint per vertex this is called ‘Rigid Skinning’. If we use multiple joints to influence a vertex (called ‘Smooth Skinning’) we specify in which amount every joint influences the vertex with a total influence that sums to one. Typical PS2 titles use rigid skinning because the hardware does not support smooth skinning directly. PSP titles use smooth skinning because it is supported on the hardware level. The total number of joints used in a character can vary according to the level of detail needed in a game. For a PS2 or PSP title, typically around 40-50 joints are used, but for a next-generation console title where fingers, hair and clothing is animated the number of joints can easily exceed 100 or 150. Let us assume we have a character that uses 100 joints. To render this skinned mesh we load all the joint matrices in memory. We can walk through all the vertices of the mesh and calculate the new vertex position, taking into account the joint matrices that influence the vertex. When all the vertices are correctly positioned we draw the complete mesh in one pass. This method is known as software skinning: we handle all the skin processing on the CPU. While it is an easy solution, the CPU cannot perform other tasks while it is skinning and it has to create new vertices which are correctly positioned before sending it to the Graphical Processing Unit (GPU). On the other hand, software skinning has no limit on the number of joints that can be used: whether the mesh uses 50 or 250 joints, software skinning can handle it just as easily. To improve performance and unload the CPU, many GPUs support hardware skinning. On the PC for instance, skinning can also be performed in the vertex shader. The most popular method of performing skinning in the vertex shader is called Matrix Palette Skinning1. Using this method, the joint matrices are stored in the constant table of the shader and every vertex has indices into the constant table to get the joint matrices. The graphics hardware defines the maximum number of constant registers that can be used in a vertex shader, thereby dictating the maximal number of joints that can be used in one render pass. We have to send new constants to the shader to use other matrices. A number of constants registers are normally reserved for general constants that the vertex shader needs, like the camera, lighting information, time, wind or fogging parameters, etc. The remaining constants can be used for joint matrices. A joint matrix is a 4 x 4 matrix, but because the last column is always [0 0 0 1]T, we only need to store a 4 x 3 matrix, thus enabling us to use 33% more joints per shader pass. 1. Skinning. Perform a google search on “Matrix Palette Skinning”. (Page 1/3)
Next
|
|||||||||||||||
|
|
|