zeta0134 / dsgx-converter Goto Github PK

Especially for small objects, vertex-based animation is more efficient both in terms of data storage for the points and in terms of processing power on the NDS. Dealing with bones is expensive for all the matrix calculations, and if the bone only affects a tiny number of vertices, it really isn't worth it.

This will require some way to indicate in the DSGX file which set of data we're working with, but in practice it'll work similar to bone-based animation. Each vertex will have a list of offsets (where that point is used in the DSGX list) and then one copy of the data per frame of each animation. We'll need to update vertex and normal data per frame, but everything else can stay static, so there's no real need to re-bake the whole DSGX file for each chunk.

As written, the euclid library for working with vectors and matrix structures is outdated, hasn't been updated since 2011, and seems to be packaged strangely, which complicates its use in deterministic builds. We should research a more appropriate library.

Avoiding emitting MTX_4x4 commands where possible will shave off cycles for matrix operations; this becomes significant with any animated object, as there are lots of matrix operations to handle the bones.

When multiple meshes are used within a single file, the exporter fails to account for the offsets of each mesh and causes wrapping of points when viewed on the DS.

This will greatly reduce the file size and cycle count where possible, should speed up rendering a bit (less point data to process per model) and should help to reduce the workload done applying animations.

General strategy: build up an adjacency graph based on polygons who share an edge, then repeatedly traverse this graph searching for the longest strip. Lather rinse repeat until there are no more strips, then emit everything else as separate polygons. (Handle quads and triangles separately.) Should be striaghtforward enough.

This is a micro-optimization, but it does save 1 cycle per vertex emitted, which adds up for lots of objects.