Consider clang_ext_vector

[zalo]

Alex Evans (of Dreams-fame) writes that Clang has a special vector extension that automagically vectorizes well to SSE/AVX/NEON and has nice GLSL/HLSL-style syntax:
https://x.com/mmalex/status/1250449900514480128?s=46

Might be interesting to consider as a “drop-in” replacement for glm::vec3?

[elalish]

My understanding is that GLM has those same vectorization optimizations. Have you tried any simple performance comparisons?

[pca006132]

I feel that it would be more beneficial to work on bottlenecks such as the collider and decimator. For the collider, I think the state-of-the-art BVH are using things like multi-way trees that gives better cache performance, and the decimator is the major bottleneck for complex workload as it is sequential.