floating point multiplication vs repeated addition

Question

Let N be an a compile time unsigned integer.

GCC can optimize

unsigned sum = 0;
for(unsigned i=0; i

        

Answer 1

There are some fundamental difference between

 float funct( int N, float sum )
 {
     float value = 10.0;
     for( i = 0; i < N ;i ++ ) {
         sum += value;
     }
     return sum;
 }

and

float funct( int N, float sum )
{
    float value = 10.0;
    sum += value * N;
    return sum;
}

When the sum approaches FLT_EPSILON * larger than value, the repeated sum tends towards a no-op. So any large value of N, would result in no change to sum for repeated addition. For the multiplication choice, the result (value * N) needs to be FLT_EPSILON * smaller than sum for the operation to have a no-op.

So the compiler can't make the optimization, because it can't tell if you wanted the exact behavior (where multiply is better), or the implemented behavior, where the scale of sum affects the result of the addition.