[Bug c++/66573] Unexpected change in static, branch-prediction cost from O1 to O2.

[jmcguiness at liquidcapital dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66573

--- Comment #2 from Jason McG <jmcguiness at liquidcapital dot com> ---
If I try with this code:

#define likely(x)       __builtin_expect((x),1)
#define unlikely(x)     __builtin_expect((x),0)

extern void bar1();
extern void bar2();

void foo(bool i) {
//  if (i)
  if (likely(i))
        bar1();
  else
    bar2();
}

Then the assembly output for gcc 4.8.2 & 5.10 (the ones I tested) for -O0, -O1,
-O2 & -O3 is the same:

foo(bool):
        testb   %dil, %dil
        je      .L2
        jmp     bar1()
.L2:
        jmp     bar2()

i.e. gcc now behaves as per icc and clang++. So using __builtin_expect seems to
be a work-around to get gcc to behave like other compilers.

I would sincerely hope that at least the documentation for these optimisation
options warns the programmer that gcc optimises if-else statements in the
reverse manner to other major compilers and even across it's own optimisation
levels (!), as this apparent change in behaviour is most unexpected...