Hacker News Reader

Comment by gameswithgo | Hacker News Reader

No way to leverage popcount instruction for this?

If you insist, you could calculate the log2 as something like

    x |= x >>  1;
    x |= x >>  2;
    x |= x >>  4;
    x |= x >>  8;
    x |= x >> 16;
    return popcount(x-1);

but there isn't really much of an advantage to it. lzcnt/bsr are better.

lzcnt is the first instruction in his solution. Seems it's left as implicit in this post, but if you go back through the chain of posts it's there.

LZCNT (base 2) would get you to a range of possible decimal digit (base 10) lengths that would need further processing, but it's still more expensive than a table-lookup direct solution directly to the base you're interested in.

> LZCNT (base 2) would get you to a range of possible decimal digit (base 10) lengths that would need further processing, but it's still more expensive than a table-lookup direct solution directly to the base you're interested in.

Are you saying you've got a better solution than the blog post that does a direct table lookup? Can you share it?

Why did you quote what I said that has nothing to do with it?

Do you have a better solution?

log2 is generally implemented with some variant of the "count leading zeros" instruction, but the general popcount is irrelevant

int->float conversion, followed by extraction and unbiasing of the exponent, also works for log2 and is more portable.

I don't normally convert my ints to floats, but when I do, I *(float *)(&x). ;-) (or fild)

Reply on news.ycombinator.com