Question

I’m using gcc 4.6.3 and creating a large array of random shorts. I’m generating them with the following statements:

val = SHRT_MAX; //as defined by limits.h
while(array<end) {
    *array++ = rand() % val;
}

This is a considerably fast operation, and even for arrays as large as 5,000,000 elements is completed almost instantly. I was curious about my sorts efficiency with a smaller variation in numbers and changed that to:

val = 3;

This caused a considerable speed difference, it ran much slower than the original statements. What is it that is causing such a considerable speed difference?

Answer 1

SHRT_MAX is most likely greater than or equal to RAND_MAX. The statement:

*array++ = rand() % val;

can be optimized into:

int rand_value= rand();
if (rand_value==RAND_MAX) rand_value= 0;
*array++= rand_value;

which is faster because it replaces a modulus with a branch. The second version, where val is 3, cannot be optimized into a simpler version that runs without modulus.

% SHRT_MAX cannot be simplified into a bitwise operation. But combined with knowledge of how rand() is specified, the compiler can certainly optimize statements dealing with rand() and values greater than or equal to RAND_MAX.