perf: Cache directories in background before parsing

[SimplisticMind]

Title.
Warm disk cache, current version:

[02:50:17.214] [36548] [I]   Parsing: 8021ms
[02:50:17.214] [36548] [I]   Adding mods: 87ms
[02:50:17.214] [36548] [I]   Adding legacy mods: 35ms
[02:50:17.214] [36548] [I]   Checking for problems: 0ms
[02:50:17.214] [36548] [I]   Total: 8145ms
[02:50:18.646] [36548] [I] Time spent creating replacement animations for Actors\Character\:
[02:50:18.646] [36548] [I]   Parsing: 1064ms
[02:50:18.646] [36548] [I]   Updating animations in submods: 111ms
[02:50:18.646] [36548] [I]   Initializing replacment animations: 0ms
[02:50:18.646] [36548] [I]   Total: 1175ms
[02:50:20.470] [36548] [I] Time spent creating replacement animations for Actors\Character\:
[02:50:20.470] [36548] [I]   Parsing: 1150ms
[02:50:20.470] [36548] [I]   Updating animations in submods: 212ms
[02:50:20.470] [36548] [I]   Initializing replacment animations: 0ms
[02:50:20.470] [36548] [I]   Total: 1363ms

(with a total of 10.6s spent in tracked parts)
Warm disk cache, this fork:

[03:29:04.502] [43044] [I] Time spent creating replacer mods:
[03:29:04.502] [43044] [I]   Parsing: 7ms
[03:29:04.502] [43044] [I]   Adding mods: 412ms
[03:29:04.502] [43044] [I]   Adding legacy mods: 44ms
[03:29:04.502] [43044] [I]   Checking for problems: 0ms
[03:29:04.502] [43044] [I]   Total: 465ms
[03:29:04.970] [43044] [I] Time spent creating replacement animations for Actors\Character\:
[03:29:04.970] [43044] [I]   Parsing: 123ms
[03:29:04.970] [43044] [I]   Updating animations in submods: 65ms
[03:29:04.970] [43044] [I]   Initializing replacment animations: 0ms
[03:29:04.970] [43044] [I]   Total: 189ms
[03:29:05.849] [43044] [I] Time spent creating replacement animations for Actors\Character\:
[03:29:05.849] [43044] [I]   Parsing: 175ms
[03:29:05.849] [43044] [I]   Updating animations in submods: 177ms
[03:29:05.849] [43044] [I]   Initializing replacment animations: 0ms
[03:29:05.849] [43044] [I]   Total: 353ms

(with a total of 1s spent in tracked parts)

Cold disk cache will present even more perf gain as the IO load is spread across the entire loading process instead of being concentrated in one place.

Caveats to be aware of:

• I wouldn't suggest caching more stuff than is already done as the time spent caching is getting close to saturated on a cold disk cache.
• Was not tested on Linux/Wine - it may handle the IO priority poorly.
• Be careful when optimizing the caching path to not create additional IO burden as it may choke other game loading stuff.
• Mutexes have been switched to std variants as CommonLibSSE locks are quite expensive (and provide, seemingly, no actual gain)

[ersh1]

That's fantastic. Thank you!

[Textolyte]

Thank you so much for this fork! Restart time has dropped by 7 seconds, from total 32 seconds to 25 seconds on the PCIe 5. OAR load 8 sec to 2.12 sec. ❤️