TriMesh GPU Support

I am using Trimesh to render multiple spherical harmonics overlaid on an image similar to [this link](https://www.mrtrix.org/images/frontpage/ssst-csd.jpg). An image contains an average of 128×128 glyphs, and each spherical glyph has 384 nodes, and the edges are computed using a Delaunay approximation. All the computations are performed in a GPU. However, the rendering gets slow since I have to move all the data to the CPU and to holoviews + datashader.

In the [performace description of datashader ](https://datashader.org/user_guide/Performance.html), I read that extending the GPU support for Trimesh could be possible.

Thanks in advance for your help.

Start by reviewing the TriMesh implementation and the linked Datashader performance description. The issue names no files, tests, or concrete GPU-support design, so first determine the relevant rendering entry points and define the supported GPU workflow and measurable completion criteria.

Written by the indexing model from the issue text.