Get 1, 0, -1 as positive, zero, or negative for an integer with math only

Question

I have a situation where I'm performing a calculate over a huge number of rows, and I can really increase the performance if I can eschew a conditional statement.

What I need is for a given positive, zero, or negative integer I want the result 1, 0, -1 respectively.

So if I do col/ABS(col), I will get 1 for a positive number, and -1 for a negative number, but of course if col equals 0 then I'll get an error. I can't get an error.

This seems simple enough, but I can't wrap my ahead around it.

Answer 1

Assuming either two's complement 32-bit integers, or one's complement with no negative-zero to worry about, then the following works well:

(x>>31) - (-x>>31);

Replace 31 with 63 for 64-bit integers, and so on.