-0 Architect

[Daleth]

So I had some code that would return -0, so I used a checker similar to this:

if(result == 0)  result = 0;

However, after reading a bit from an assembly language tutorial, I found out that -0 is different by one bit. Does that mean that the following will work? And be way more efficient?

result ^= -0;

I can't test this on my code that produces -0 because there's a whole lot of introduced bugs I still need to fix. Plus I don't really know how to write another piece of code that would result in -0. Anyway, I just need to know if this alternative would be a great optimizer.

[greenleaf800073]

What is ^=

[greenleaf800073]

Why don't you do

if(result == -0) result = 0;

[Daleth]

If I'm not mistaken, ^= is the compound assignment operator for XOR?

The check I have currently works fine. I just want to know if the alternative will make the program more efficient. The code that produces the -0 is really slow with large amounts of data, but I'm going to optimize it after I finish fixing the bugs.

[LB]

^ and ^= are much slower than checking for equality and assigning.

-0 and 0 are identical unless you're dealing with floating point types, but it shouldn't matter anyway since you're smart enough not to use direct == comparison.

Why do you think -0 is a problem? For integers, there is no -0, it is the same as 0.

[Daleth]

Oh I should have mentioned that the result will be part of a printable vector. That's why I can't have "-0" showing up as in <1,2,-0,5>.

But thanks for answer! Here I was biased into thinking bitwise operators were always more efficient. :F

[LB]

Why would -0 show up? Integers cannot be -0

[Daleth]

Here's more of the code (result is actually double).

		//Perform Cross Product
	vector<double> toreturn;
	unsigned column;
	int sign;
		//Create outside for loop according to number of components
	for(column=0, sign = 1; column < vectorlist.begin()->size(); column++, sign *= -1){
			//Explicity check for a result of 0 and push back 0 to avoid the -0 architect
			//Declare temporary variable to avoid calling Determinant twice and avoid overhead cost
		double coefficient = ( catalyst.Splice(column) ).Determinant(componentnumber-1);
		if(coefficient == 0)	toreturn.push_back(0);
		else	toreturn.push_back(sign*coefficient);
	}
	return toreturn;

When I first tested it out (the code worked back then and without the 0 checker) with input that should have resulted in <0,0,0>, I always got <0,-0,0>.

[LB]

You cannot use == for float or double types.

[Daleth]

It worked correctly for me? It got rid of the -0, and when I compared the result of my code with what I did by hand, the answers were the same.

I hope I'm not missing some gigantic chunk of information about doubles.

Edit:
Reading articles 'bout epsilon and ways to compare floats. No need to answer anymore.