Performance comparison of iteration methods over arrays & lists

Hi there, .NET enthusiasts who care about performance!

I wondered whether it's worth writing for loops for iterating over random-access collections (arrays, lists) in .NET nowadays or the JIT compiler has got so smart by now that we can just use foreach loops in such cases without significant perfomance penalty.

So I did some measurements (by the help of BenchmarkDotNet) and seeing the results, I decided my findings might be worth sharing.

I benchmarked the following 3 types of iteration methods:

1. Plain old foreach loop: foreach (var item in collection) { /*...*/ }
2. for loop with Length/Count evaluated in stop condition: for (int i = 0; i < collection.Count; i++) { /*...*/ }
3. for loop with Length/Count cached in variable: for (int i = 0, n = collection.Count; i < n; i++) { /*...*/ }

I run tests for both arrays and lists, for small (10) and bigger (1000) item counts and for platforms .NET 4.8, .NET Core 3.1 and .NET 5.

You can view the results here.

I drew the following conclusions:

• If you aim for maximum performance, use method 3 (for loop with Length/Count cached in variable). The only exception is direct access to arrays, in which case foreach seems a tiny bit faster. Looks like the JIT compiler optimizes the hell out of that.
• Avoid iterating over collections via interfaces if possible. The performance penalty is in the range of 4x-6x! Definitely avoid foreach over interfaces because that allocates too (as the enumerator is also obtained through the interface, thus, it gets boxed). In this case, for, at least, is still allocation-free.