https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84100
--- Comment #7 from Chris Hall <gcc at gmch dot uk> ---
And here's a funny thing...
... if I compile "-O3 -falign-functions -falign-loops=32" I get the alignment I
ask for.
... if I compile "-O3 -falign-functions -falign-loops=32 -fno-tree-vectorize"
I get the alignment I ask for (and I avoid Bug 84101).
... if I compile "-O3 -falign-functions -falign-loops=32" but with:
_Pragma("GCC optimize(\"no-tree-vectorize\")")
extern int func_x(int num) ;
extern int func_y(int num) ;
extern uint64_pair_t pair(int num) ;
extern int func_z(int num) ;
I no longer get the alignment I asked for, I get the default alignment for -O3
!
And, for further amusement:
extern int func_x(int num) ;
extern int func_y(int num) ;
extern uint64_pair_t pair(int num)
__attribute__((optimize("no-tree-vectorize"))) ;
extern int func_z(int num) ;
has exactly the same effect !
FWIW, I have tried _Pragma("GCC push/pop/reset_options") -- the effect of which
is hard to determine; but I can say that pop/reset do not restore all
optimization option settings.