Currently lovr uses FNV1a for hashing. The implementation in util.h is like 5 lines, but because it hashes the input 1 byte at a time, it's slow for large inputs. It might be better to switch to a faster hash function like xxh3 or rapidhash (previously wyhash). These are a lot faster because they hash multiple bytes at a time. I wonder if this could have a meaningful difference in performance anywhere. Hashing the gpu_pipeline_info structs might be noticeable, because that struct is a few hundred bytes.