most efficient way to test for multiple NaNs in C#

I am looking for the fastest way to determine if there is at least one NaN in a float array.

• Arrays can have from 0 to many NaN values.
• Arrays to not have NaNs values in 95%+ of the cases.
• When there is at least one NaN, the location is very random.

Let's say I have an array of floats:

var a = float[10000];

I can do this:

(let's ignore the loop itself, I know foreach it not the most efficient, it is just to illustrate)

bool hasNan(ref float[] floats)
{
    foreach (var f in floats)
    {
        if (float.IsNaN(f)) return True;
    }

    return False;
}

We know that in most cases, there are no NaNs, so I have to scan the whole array anyways. Instead of calling IsNaN for every element, would it make sense to:

• compute a sum; if there is at least one NaN, the result will be NaN.
• or, check for equality since if x = NaN, x is not equal to x, but it's probably what IsNaN does and it means one branch per check.
• or, check with each element if x & 0x7fffffff >= 0x7f800000, (that includes infinity which the array shouldn't contain anyways)

The array is quite large and this is happening in a fairly tight timeframe as I'm processing streams of financial data coming in realtime.

Edit:

Check the fiddle: https://dotnetfiddle.net/W3VgQg

and look at the IL code generated.

When doing the float.IsNan method, the code is not inlined. The sum method seems faster, but I don't know enough about IL to really understand the details.