JIT doesn't elide bounds checking when iterating over a span and an array of known length

[itsBuggingMe]

Description

If you ensure that the length of an array and a span are the same, the JIT will still emit bounds checking on only the span when indexing.
This does not occur when iterating over a span and a span, or an array and an array.

public static int SumSpanArray(int[] arr, Span<int> span)
{
    int counter = 0;
    
    if(arr.Length != span.Length)
        return 0;
    
    //bounds check here
    for(int i = 0; i < arr.Length; i++)
        counter += arr[i] + span[i];
    
    return counter;
}

Sharplab example here

Regression?

I have only used Sharplab to test this, but it doesn't seem likely to be a regression.

Analysis

I don't know why, but curiously enough when checking the span length when iterating over a span + array, there is no bound check

public static int SumSpanArray2(int[] arr, Span<int> span)
{
    int counter = 0;
    
    if(arr.Length != span.Length)
        return 0;
    
    //no bound check
    for(int i = 0; i < span.Length; i++)
        counter += arr[i] + span[i];
    
    return counter;
}

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-memory
See info in area-owners.md if you want to be subscribed.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[EgorBo]

It works if both are arrays. Presumably it should work here as well once #82946 is implemented.

[tannergooding]

Notably as well it does work if you insert a span1 = span1.Slice(span2.Length) after a if (span1.Length != span2.Length) { return; } check.

So there is a relatively easy workaround, but ideally we'd still just handle correlating the array/span or span/span lengths directly.