What is the overhead for a method call in a loop?

Question

I’ve been working on a C# maze generator for a while which can generate mazes of like 128000x128000 pixels. All memory usage is optimized already so I’m currently looking at speeding the generation up.

A problem (well more off an interest point) I found was the following (just some example code to illustrate the problem):

This code runs in about 1.4 seconds on my machine when pixelChanged is null:

public void Go()
{
    for (int i = 0; i < bitjes.Length; i++)
    {
        BitArray curArray = bitjes[i];
        for (int y = 0; y < curArray.Length; y++)
        {
            curArray[y] = !curArray[y];
            GoDrawPixel(i, y, false);
        }
    }
}

public void GoDrawPixel(int i, int y, Boolean enabled)
{
    if (pixelChanged != null)
    {
        pixelChanged.Invoke(new PixelChangedEventArgs(i, y, enabled));
    }
}

Where the following code runs actually 0.4 seconds faster

public void Go()
{
    for (int i = 0; i < bitjes.Length; i++)
    {
        BitArray curArray = bitjes[i];
        for (int y = 0; y < curArray.Length; y++)
        {
            curArray[y] = !curArray[y];
            if (pixelChanged != null)
            {
                pixelChanged.Invoke(new PixelChangedEventArgs(i, y, false));
            }
        }
    }
}

It seems that when just calling an “empty” method takes about 20% of the cpu this algorithm uses. Isn’t this strange? I’ve tried to compile the solution in debug and release mode but haven’t found any noticeable differences.

What this means is that every method call I have in this loop will slow my code down by about 0.4 seconds. Since the maze generator code currently consist of a lot of seperate method calls that excecute different actions this starts to get a substantial ammount.

I've also checked google and other posts on Stack Overflow but haven't really found a solution yet.

Is it possible to automatically optimize code like this? (Maybe with things like project Roslyn???) or should I place everything together in one big method?

Edit: I'm also interested in maybe some analysis on the JIT/CLR code differences in these 2 cases. (So where this problem actually comes from)

Edit2: All code was compiled in release mode

Answer 1

It is a problem, JIT has an inline optimization for methods (where the whole method code is actually injected inside the calling parent code) but this only happens for methods that are compiled to 32 bytes or less. I have no idea why the 32 byte limitation exists and would also like to see an 'inline' keyword in C# like in C/C++ exactly for these issues.