Re: Subtract smaller or larger number from x?


Steve O'Hara-Smith wrote:
On 29 Mar 2006 05:39:29 -0800
upeksharuvani@xxxxxxxxx wrote:


eksamor@xxxxxxxxx wrote:
I have a number x. I would now like to find the difference between this
number x and another number y.

But y can be either larger or smaller than x. Currently I check if the
number y is greater or smaller than x:

int diff; // the difference between x and y

if (y > x){
diff = y-x;
}else{
diff = x-y;
}

is there not a more elegant way to do this when I would like the
difference to be a positive value??

Yes, of course
As my sis pointed out it would be more efficient if you take the
absolute value of the differenc, like this:

And then again it might not be more efficient - consider the
function call overhead and observe that the source code for abs looks like
this:

int
abs(int j)
{
return(j < 0 ? -j : j);
}

So in addition to doing the subtraction and a test it is doing a
function call and perhaps a negation - whereas the original is doing a test
and a subtraction. Seems pretty certain to be less efficient to me.

This version might be more efficient but I doubt it, it probably
compiles to pretty much the same code as the original. It is perhaps a
little prettier.

diff = (y > x) ? y - x : x - y;


Of course that's all totally implementation dependant. MSVC (with
-Ox), for example, generates:

; 7 : z = abs(x-y); (x, y and z are ints)
mov eax, DWORD PTR _x
mov edx, DWORD PTR _y
sub eax, edx
cdq
xor eax, edx
sub eax, edx
mov DWORD PTR _z, eax

abs() is quite commonly inlined, and on many processors there's a way
to do the "abs" function without any conditional code (and often it's a
lot prettier than the cdq/xor/sub sequence above).

.

Relevant Pages