https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123160
--- Comment #4 from Andrew Macleod <amacleod at redhat dot com> ---
(In reply to Jakub Jelinek from comment #3)
> On the other side, one could argue that ECF_RET1 is a pure optimization
> thing and unless the compiler optimizes hard, when you use the return value
> of memcpy/memset etc. and call objsz on that what exact subobject it points
> to can be lost. Definitely older versions of gcc wouldn't be able to track
> that, I bet they'd return a don't know value in those cases (all ones or 0
> depending on mode).
We can avoid memset. if I rewrite the fail() function to be simply:
r = &a.a[4] + 2;
if (__builtin_object_size (r, 3) != sizeof (a.a) - 6)
__builtin_abort ();
it goes from (in early_objsz)
<bb 2> :
r_2 = &a.a[4] + 2;
_5 = __builtin_object_size (r_2, 3);
_1 = MAX_EXPR <_5, 4>;
if (_1 != 4)
goto <bb 3>; [INV]
else
goto <bb 4>; [INV]
<bb 3> :
__builtin_abort ();
<bb 4> :
return;
to (in ccp1)
<bb 2> :
__builtin_abort ();
}
via
Folding statement: r_2 = &a.a[4] + 2;
Queued stmt for removal. Folds to: &MEM <char> [(void *)&a + 6B]
Folding statement: _5 = __builtin_object_size (r_2, 3);
Queued stmt for removal. Folds to: 22
Folding statement: _1 = MAX_EXPR <_5, 4>;
Queued stmt for removal. Folds to: 22
Folding statement: if (_1 != 4)
gimple_simplified to if (1 != 0)
which is likely CONSTANT
Match-and-simplified 1 != 0 to 1
Folding predicate 1 != 0 to 1
Folded into: if (1 != 0)
