surprising optimization difference (I'm not quite courageous enough to call it a compiler bug yet). Additionally compiling it with -ffast-math gives the same result via the incomprehensible sequence below gcc version info: Target: x86_64-linux-gnu gcc version 4.7.2 (Debian 4.7.2-5) The wrong result turns out to be dot / (sqrt(radius) * radius * …

build and run it..

jepler@babs:~$ gcc -Os rsq.c -lm && ./a.out
0.993271
jepler@babs:~$ gcc -Os -funsafe-math-optimizations rsq.c -lm && ./a.out
1.365475

gistfile1.txt

000000000040057c <fn>:
  40057c:       f2 0f 51 d1             sqrtsd %xmm1,%xmm2
  400580:       f2 0f 59 c9             mulsd  %xmm1,%xmm1
  400584:       f2 0f 59 ca             mulsd  %xmm2,%xmm1
  400588:       f2 0f 5e c1             divsd  %xmm1,%xmm0
  40058c:       f2 0f 11 05 34 04 20    movsd  %xmm0,0x200434(%rip)        # 6009c8 <val>
  400593:       00 
  400594:       c3                      retq   

rsq.c

double val;

void fn(double dot, double radius) {
	val = dot / (radius * radius);
}

#include <stdio.h>
int main() {
	fn(.278102, 0.529137);
	printf("%f\n", val);
}