C :: Possible To Make Compiler To Generate Smarter Code

I am not a C programmer (or a programmer at all). I am just a hobbyist who writes random code (i.e.: ignorant).C is not my area, and I my experience with it is quite LIMITED!Today, I found out that if I need to get "square root" in C I will be in trouble if I depend solely on the compiler..Is it possible to make the compiler to generate smarter code for this (quite trivial code):

Code:

void intprime_c(unsigned long* fprimes, unsigned long count)
{
unsigned long divc = 3;
unsigned long num = 2;
unsigned long sqr;
unsigned long pfound = 0;
}

[code]...

By "without ugly hacks" I mean using already existing code that came with GCC. Because the others did not need ugly hacks to perform better (it is all about being fair).To compute the "square root" in x86 one just need 3 instructions:

fild
fsqrt
fistp

But the compiler is generating a call to a "fsqrt" function and wasting a lot of time by doing this. If not, then why is it taking so long? Is it something wrong in my code (at least the output was verified to be the correct)?

Time taken to get the first 2500000 primes (x86_64):

Assembly = 72.400s (human written, unoptimized, coder level: newbie (i.e.: ignorant))
pascal = 75.200s (written by the same person)
C (fsqrt) = 83.600s (written by the same person)
C (sqrt) = 85.900s (written by the same person)

If it matters, I compiled the C code with this command: gcc -O2 -fno-omit-frame-pointer -fPIC -c ./prime.