I wrote an performance testing function about in parameter modifer with large struct:
[Test, Performance, Category("Performance/Large Struct")]
public void In_ImmutableLargeStruct_WithInOrNot()
{
var unit = SampleUnit.Microsecond;
var withoutIn = new MeasurementSettings(100, 100, 10000, "Immutable struct without in", unit);
var withIn = new MeasurementSettings(100, 100, 10000, "Immutable struct with in ", unit);
var immutableStruct = new ImmutableStruct(1.1, 2.2);
Measure.Method(() => { StructInBenchmarkTest.Add(immutableStruct); }, withoutIn).Run();
Measure.Method(() => { StructInBenchmarkTest.AddWithIn(immutableStruct); }, withIn).Run();
}
ImmutableStruct is a 88-bytes readonly struct.Looks like:
public readonly struct ImmutableStruct
{
public double x { get; }
public double y { get; }
public double z { get; }
private readonly double m_A;
// ....
And the Add function looks like:
public static double Add(ImmutableStruct s) { return s.x + s.y; }
public static double AddWithIn(in ImmutableStruct s) { return s.x + s.y; }
While in Unity Editor and android device with mono 32-bit backend, the performance testing result matchs expection that in modifer will lead to performance improvement with immutable structure.
However, while in android device with il2cpp 64-bit backend, there is no noticeable performance effect with in modifier or not:
I tested on Oculus, XiaoMi, and microsoft android subsytem with Unity 2020.3.25f1 / 2022.2.1f1. All of these test doesn’t give expected performance test result.