Lets say my character model has different armor or pants or gloves, etc...
Now when I animate the character in the vertex shader I need to send the bone transforms which is a lot of data. I may need to make multiple draw calls for all the different pieces and I want to avoid sending all bones every time if I only need to send maybe the fingers for the characters glove.
Or is it better to maybe combine the character into a single mesh? This may be a bit tough if the character is changing appearance a lot mid game. This will also be impossible if a portion of the character is using a different material and requires a separate draw call. I guess if its a transparent helmet visor I can make it be a rigid object that is only relative to the head bone and isnt affected by any skeletal animations. I could probably do that with things like armor as well.
skeletal animation on peices of a model
[...]
Now when I animate the character in the vertex shader I need to send the bone transforms which is a lot of data. I may need to make multiple draw calls for all the different pieces and I want to avoid sending all bones every time if I only need to send maybe the fingers for the characters glove.[...]
The skeleton is just a big list of matrices or quaternion-translation pairs that are usually stored in a constant buffer. "Sending the bone transforms" is just a matter of updating a GPU resource; this only needs to happen oncer per frame. To use that data you simply bind it to the GPU, this isn't free but it's generally very cheap as it doesn't involve moving data from the CPU to the GPU. After binding the resource the shaders evoked by subsequent draw calls will have access to the skeleton data.
Generally speaking you should pack your entire skeleton into a single resource and update it once per frame regardless of how many draw calls it will take to render the skinned mesh.
Author
Are you talking about packing the matrices in a texture rather than sending it as a uniform?
Also my render state may not be sorted by animation state, but by shaders and materials and meshes, although I'm considering sorting by animation state as well.
I'm also considering trying instancing at some point.
Are you talking about packing the matrices in a texture rather than sending it as a uniform?
Also my render state may not be sorted by animation state, but by shaders and materials and meshes, although I'm considering sorting by animation state as well.
I'm also considering trying instancing at some point.
Packing them into a texture is the preferred approach. The texture containing bones is only written to once and can then be read from as many meshes as needed.
It avoids running into shader constant limits. It also makes use of the texture fetch hardware which runs in parallel to the shader math which can yield faster execution than shader constant fetching (google "shader constant waterfalling" for an explanation on this)
Are you talking about packing the matrices in a texture rather than sending it as a uniform?
Not necessarily but as DigitalFragment pointed out there are many advantages to doing so; either way the concept is the same. If your hardward supports texture fetches in the vertex shader and floating point texture formats it's generally a good idea (such features are common these days).
Author
Sounds like it could work pretty well since I'm making a PC game with deferred shading.
I haven't tried writing data to textures CPU side every frame yet. It's not too expensive? I have to upload a matrix array somehow anyway so I imagine there's some overhead either way.
Sounds like it could work pretty well since I'm making a PC game with deferred shading.
I haven't tried writing data to textures CPU side every frame yet. It's not too expensive? I have to upload a matrix array somehow anyway so I imagine there's some overhead either way.
Slightly more so than shader constants, but not by a wide margin. The real problem is the GPU stall associated with a CPU lock, but so long as you have extra work that the GPU can be chewing on while you tinker with the matrix buffer/texture there's nothing to be worried about. If you're concerned with PCI express bandwidth, you can also use a more compact transform representation such as vector + quaternion instead of a full 3x4 matrix. You lose scaling, but there are plenty of engines out there that do without. It's possible to pack that into, say, the transform w component (or also the quaternion, but reconstruction is more complex and it's less of a straight win.)
This topic is closed to new replies.
Advertisement
Popular Topics
Advertisement