IL2CPP (AOT) performance using Expressions

[gustavopsantos]

The expression trees implementation used for AOT platforms is an interpreter, so it is pretty slow compared to a JIT compiler like Mono. We need to check if theres a good way (clean, readable and scalable) to implement a IL Reweaver.

References:

• https://forum.unity.com/threads/are-c-expression-trees-or-ilgenerator-allowed-on-ios.489498/page-2
• https://answers.unity.com/questions/1731418/il2cpp-compiled-expressions-performance.html
• Compiled Lambda Expressions and AOT dotnet/runtime#17973

[boraty14]

I have no idea what does this issue means, but according to links it looks like reaallly slow (like you mentioned). Should I care for this when I develop for IL2CPP ?

[gustavopsantos]

@boraty14 You should not worry about it, expression trees are currently being used only for mono (JIT) builds.
See peformance charts for both mono (JIT) and IL2CPP (AOT) in https://github.com/gustavopsantos/Reflex?tab=readme-ov-file#-performance

[gustavopsantos]

Closing this issue as I already investigated this a long time ago, and decided to not go for IL Weaving (at least for now) as I'm currently happy with AOT performance