Allow operator overloading with "out" for the result

[nitroxis]

Since it is now possible to pass the parameters for operator overloading as references using the "in" keyword, it would make sense to allow returning the result via an "out" parameter. This has a huge impact on operations on large structs, such as a 4x4 float matrix.

This works right now:

public static Matrix4x4 operator *(in Matrix4x4 left, in Matrix4x4 right)

However, this does not:

public static void operator *(in Matrix4x4 left, in Matrix4x4 right, out Matrix4x4 result)

At the moment, it is still necessary to use static methods with "in, in, out" for maximum performance. I've made a small benchmark that compares the static method variant with the "in" operator overload and found that the static method is still more than 3 times faster than the operator on my machine (i7-6700K).
This is the code that I tested (in Release mode, of course):

struct Matrix4x4
{
	public float M11, M12, M13, M14;
	public float M21, M22, M23, M24;
	public float M31, M32, M33, M34;
	public float M41, M42, M43, M44;

	public static void Add(in Matrix4x4 left, in Matrix4x4 right, out Matrix4x4 result)
	{
		result.M11 = left.M11 + right.M11; result.M12 = left.M12 + right.M12; result.M13 = left.M13 + right.M13; result.M14 = left.M14 + right.M14;
		result.M21 = left.M21 + right.M21; result.M22 = left.M22 + right.M22; result.M23 = left.M23 + right.M23; result.M24 = left.M24 + right.M24;
		result.M31 = left.M31 + right.M31; result.M32 = left.M32 + right.M32; result.M33 = left.M33 + right.M33; result.M34 = left.M34 + right.M34;
		result.M41 = left.M41 + right.M41; result.M42 = left.M42 + right.M42; result.M43 = left.M43 + right.M43; result.M44 = left.M44 + right.M44;
	}
	
	public static Matrix4x4 operator +(in Matrix4x4 left, in Matrix4x4 right)
	{
		Matrix4x4 result;
		result.M11 = left.M11 + right.M11; result.M12 = left.M12 + right.M12; result.M13 = left.M13 + right.M13; result.M14 = left.M14 + right.M14;
		result.M21 = left.M21 + right.M21; result.M22 = left.M22 + right.M22; result.M23 = left.M23 + right.M23; result.M24 = left.M24 + right.M24;
		result.M31 = left.M31 + right.M31; result.M32 = left.M32 + right.M32; result.M33 = left.M33 + right.M33; result.M34 = left.M34 + right.M34;
		result.M41 = left.M41 + right.M41; result.M42 = left.M42 + right.M42; result.M43 = left.M43 + right.M43; result.M44 = left.M44 + right.M44;
		return result;
	}
}

[jnm2]

I'd love to see a BenchmarkDotNet table for the extra statistics.

[nitroxis]

I made a small benchmark scenario that implements all combinations of static methods and operators with and without aggressive inlining and calls them a couple of times with different inputs. Here's the code: https://gist.github.com/nitroxis/c700ef0f4bbe28cbc3090108410d4034

I used different symbols for the operator overloads but they all do the same (adding two 4x4 matrices).

This yields the following results on my machine:

// * Summary *

BenchmarkDotNet=v0.10.14, OS=Windows 10.0.16299.251 (1709/FallCreatorsUpdate/Redstone3)
Intel Core i7-6700K CPU 4.00GHz (Skylake), 1 CPU, 8 logical and 4 physical cores
Frequency=3914062 Hz, Resolution=255.4891 ns, Timer=TSC
.NET Core SDK=2.1.104
  [Host]     : .NET Core 2.0.6 (CoreCLR 4.6.26212.01, CoreFX 4.6.26212.01), 64bit RyuJIT
  DefaultJob : .NET Core 2.0.6 (CoreCLR 4.6.26212.01, CoreFX 4.6.26212.01), 64bit RyuJIT


                                  Method |     Mean |     Error |    StdDev |
---------------------------------------- |---------:|----------:|----------:|
                            StaticMethod | 37.41 ns | 0.7519 ns | 0.8357 ns |
                      StaticMethodInline | 46.63 ns | 0.9515 ns | 1.0181 ns |
                        StandardOperator | 95.10 ns | 1.1078 ns | 1.0362 ns |
                              InOperator | 89.95 ns | 1.1705 ns | 1.0949 ns |
             InOperatorCallsStaticMethod | 66.80 ns | 1.3256 ns | 2.1780 ns |
       InOperatorInlineCallsStaticMethod | 67.24 ns | 1.3530 ns | 2.0662 ns |
       InOperatorCallsStaticMethodInline | 77.35 ns | 1.3127 ns | 1.2279 ns |
 InOperatorInlineCallsStaticMethodInline | 62.82 ns | 1.2246 ns | 1.4103 ns |

Not sure why calling the static method from within the operator is faster than directly having the code in the operator.