attribute_benchmark.md

fresh benchmarks for attributes and attribute strings with some new optimizations on tf2attributes' side

(note that the optimizations only affect native attributes)

---

Running benchmark with 65536 iterations. (This benchmark was run multiple times, and the best time for each was taken.)

• GetGameTime took 0.000384s
• GetEntPropString took 0.010754s

The above are just baseline values on known 'simple' operations for reference

• TF2Attrib_HookValueFloat took 0.037156s

TF2Attrib_HookValueFloat iterates over all items on a player if given a player entity to hook

• TF2Attrib_GetStringValue took 0.116503s
• TF2Attrib_GetStringValue (missing) took 1.443081s

TF2Attrib_GetStringValue does naive lookups and doesn't take advantage of the attrib cache

• TF2Attrib_HookValueString took 0.081578s
• TF2Attrib_HookValueString (missing) took 0.079989s

TF2Attrib_HookValueString does caching and iterates over all player items

• TF2CustAttr_GetInt took 0.044222s
• TF2CustAttr_GetFloat took 0.048780s
• TF2CustAttr_GetString took 0.051782s

custattr doesn't iterate over all items on a player, so you have to make multiple calls for an attribute that combines values — multiply the int / float bench times by at least 4x (player + 3 weapons) if you want to match hookvalue

---

tl;dr:

• native int / float hooks are faster than iterated calls to custattr int / float (they should be the same-ish speed for single entities)
• TF2CustAttr_GetString is ~40% faster than TF2Attrib_HookValueString
• TF2Attrib_HookValueString is ~75% faster than TF2Attrib_GetStringValue in the optimal case where the attribute exists as a runtime attribute, and 1800% faster than TF2Attrib_GetStringValue in the worst case where the attribute doesn't exist on an entity (since it iterates over runtime / SOC / static). TF2Attrib_GetStringValue is literally dead on arrival, since it's not even on a public release branch yet!