Hello everyone here,

Hope you're all good recently! I have been working on adding animation support to my project (I'm totally a nookie in this domain ?) for a while, and I had a few questions which I can't figure out by myself:

1. All the 101 tutorials would just do a (sometimes recursive) iteration over the skeleton node hierarchy to calculate the final pose transformation each frame, and then upload to GPU memory for skinning in the shader. I thought everybody would agree it's a waste of cycles for real products and it's better to cache or even pre-compute the transformations. Should I be wrong on this? Is it common to calculate the final pose transformation on-the-fly (For example, a FPS which has a medium-level (5-10) number of pawns running and shooting each other at the same time)?

2. I tried to pre-compute the transformations first when import and convert the assets to the custom form, but I found the number of keys of each bone is different, the number of the translation, rotation and scale keys are different, and some bones won't have any keys since they won't move (my test asset was downloaded from [https://sketchfab.com/3d-models/wolf-with-animations-f3769a474a714ebbbaca0d97f9b0a5a0], since it's written there that it's compatible with Unity and Unreal, I think it's an okay enough test case). So an interpolation process is necessary in this case? Or in the DCC software artist could export the animation with an aligned count of the key for each bone (I tried Blender but didn't find anything related ?)?

3. I would prefer to trade time by space, so an X key * Y bones 2D array is acceptable for me, and I think it's friendly for GPU sampling. How do you deal with this in your case? What's your idea about?