added sumOfLog2s

[9il]

Computes accurate sum of binary logarithms of input range.

[DmitryOlshansky]

So I see this just uses the formula - logarithm of products is a sum of logarithms to speed up the latter.
Can't really comment on FP precision checking, but otherwise the idea is nice.

[JakobOvrum]

ElementType!Range needs to be constrained

[9il]

@DmitryOlshansky, accuracy like log2(r1...rn), but wihtout exponent overflow

[9il]

With native frexp (PR #2508) it is 4.5 times faster then sum of logs when compiled with LDC.

[9il]

Rebased

[JakobOvrum]

While the new return type is an improvement, ElementType!Range still needs to be constrained. For example if(isInputRange!Range && isNumeric!(ElementType!Range)) (though by the looks of it, it won't work well with integers?).

[9il]

@JakobOvrum A have added isFloatingPoint!(ElementType!Range). It looks like all std.numeric without checks like isFloatingPoint. For example

ElementType!Range entropy(Range)(Range r) if (isInputRange!Range)
{
    Unqual!(typeof(return)) result = 0.0;
    foreach (e; r)
    {
        if (!e) continue;
        result -= e * log2(e);
    }
    return result;
}

[quickfur]

LGTM

[dnadlinger]

Shouldn't this be sumOfLog2s or something for consistency?

[9il]

All std.numeric uses only log2. But names without 2. I can rename if you ask to do this.

[dnadlinger]

@9il: I don't think std.numeric has any other functions with "log" in the name, does it? The only mentions in 2.066 seem to be in some comments, where the base is explicitly specified. std.math uses log2, hence the question.

[9il]

@klickverbot Yes ) I will change it =)

[9il]

Rebased

[quickfur]

You forgot to rename these calls. The autotester will probably report failure. :-)

[9il]

Thanks!

[9il]

Rebased again)

[quickfur]

LGTM

[quickfur]

Auto-merge toggled on