[dsesclei]

smem_container is kept in sorted order, so building it with N items becomes O(N^2). This splits that array into 1024 buckets so we square a smaller number.

Total animation load time drops from 386ms to 78ms for a Cordon save on an i9-14900K.

If this looks good I'll do a PR for MT as well.

[erepb]

this looks halfway to hash map. did you try replacing vector with hashmap?

[dsesclei]

Yeah that's faster thanks (52ms), just switched to xr_unordered_map

[themrdemonized]

Please retarget this change to MT branch only for this reasons:

1. It already has other changes to underlying shared containers
2. I'm not sure if a change from continuous vector to a map will not cause any problems. I had issues before where i changed underlying data of LTX processing from vector to set and it lead to some strange behaviour on some PCs.

I'll test it on MT branch later when your new PR lands

Closing this one