Should the loop limit be evaluated or stored?

Question

In C++, is it faster to store the limit of a loop in a variable than evaluating the value?

For example:

Is it a slower approach to use

for(int i=0; i<n*n+2*n; ++i)
{ .... }

than doing the following?

for(int i=0, limit=n*n+2*n; i<limit; ++i)
{ .... }

For clarity, assume that n is some given variable that remains unchanged during the course of the loop.

Answer 1

If n is a globally declared non-volatile variable, then the behaviour of

for (int i = 0; i < n * n + 2 * n; ++i)

is unspecified. A compiler is allowed to optimise n * n + 2 * n to be evaluated once even if another thread modifies n. Furthermore, if another thread is able to modify n, then you should take steps to avoid the potential for simultaneous read and writes of n (the behaviour of which is undefined). Consider using std::atomic<int> as the type for n.

So really it's appropriate to introduce limit anyway if you want the stopping condition to be dependent on the value of n observable when program control reaches the for loop, irrespective of any performance considerations. Consider therefore

for (int i = 0, limit = n * n + 2 * n; i < limit; ++i)

which has the advantage that the scope of limit doesn't leak into the surrounding statements.

But if you are able to, you could always run the loop backwards:

for (int i = n * n + 2 * n - 1; i >= 0; --i)

Be very careful though with unsigned types if you adopt that idea.