Expose Row Index in PixelRowOperation delegate

[PeteMichaud]

I've been working through Joe's answer to this question: https://stackoverflow.com/questions/71388492/right-way-to-parallelize-pixel-access-across-multiple-images-using-imagesharp/71388733#71388733

I am trying to take his recommendation seriously not to used the Advanced stuff if possible, but I really do need parallel processing as the images I'm processing regularly push dimensions in the range of 25-50k width*height, so I've been looking into the Vector4 api he mentioned, which is given here as an example:

// from https://docs.sixlabors.com/articles/imagesharp/pixelbuffers.html
image.Mutate(c => c.ProcessPixelRowsAsVector4(row =>
{
    for (int x = 0; x < row.Length; x++)
    {
        // We can apply any custom processing logic here
        row[x] = Vector4.SquareRoot(row[x]);
    }
}));

I can't use this because my calculations absolutely require the current y/rowIndex coordinate. I looked here and found that one version of the delegate takes a T value arg, but as far as I can tell, that can't get me the index (maybe I'm wrong here?)

My suggestion is simply to provide an overload that has an index like so:

image.Mutate(c => c.ProcessPixelRowsAsVector4((rowIndex, row) =>
{
    for (int x = 0; x < row.Length; x++)
    {
        // y = rowIndex
        // x = x
    }
}));

I assume it doesn't make sense to attach an index type property to the row Span object itself, but that's a different possibility if the Span is less generic than I'm assuming it is.

[antonfirsov]

Sounds reasonable to me, are you interested creating a PR?

[PeteMichaud]

The spirit is willing, but I have very little confidence that I can navigate this codebase within a reasonable timeframe in a way that ends up with an acceptable-to-the-team PR.

One thing I'll note is that I investigated the code around this idea, and found a bunch of overloads for ProcessPixelRowsAsVector4 here

And one in particular seemed promising:

ImageSharp/src/ImageSharp/Processing/Extensions/Effects/PixelRowDelegateExtensions.cs

Line 100 in e54219a

=> source.ApplyProcessor(new PositionAwarePixelRowDelegateProcessor(rowOperation, modifiers), rectangle);

Because it creates something called a PositionAwarePixelRowDelegateProcessor which sounds like the sort of thing we're after.

That class does provide an Invoke signature that includes a Point offset which is maybe equivalent to the functionality we're talking about here, and might be the right place to start to provide the more clear (index, row) delegate signature we're talking about.

[JimBobSquarePants]

As you've discovered there's an overload that passes the current position as a Point of which Y is the current row so you have access to that functionality without any code changes.

We won't be adding additional overloads as to do so actually involves a lot of work with no added value.