http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39612
--- Comment #13 from Richard Guenther <rguenth at gcc dot gnu.org> 2012-01-17
10:01:04 UTC ---
(In reply to comment #12)
> (In reply to comment #11)
> > Re-confirmed on trunk with the testcase in comment #4 and -Os:
> >
> > > ./cc1 -quiet t.c -Os -Wall -fdump-tree-all
> > t.c: In function 'f2':
> > t.c:4:11: warning: 'inter[0]' is used uninitialized in this function
> > [-Wuninitialized]
> > t.c:4:11: warning: 'inter[1]' is used uninitialized in this function
> > [-Wuninitialized]
> >
> > There is also a store data-race introduced by LIM when the loop header
> > is not copied. I suppose the real fix is to disable SM if the loop
> > header is not copied.
>
> I do not think this is the right way. Disabling a valid (and fairly
> important)
> optimization in order to silence a warning seems a bit extreme. I am not sure
> what you mean by "a store data-race".
>
> Also, the warning is at least morally right. If R <= 1, the original code
> will
> pass inter to foo uninitialized, which probably is not intended. So, the
> right
> thing to do could be issuing "may be used uninitialized" warning instead of
> "is
> used uninitialized" one.
Yes, but the point is that without the loop header copy we insert the
loads and stores from/to inter in a path that is executed even when
R <= 1 and thus the loop is not executed at all, and we warn about
the inserted loads - not about the final one. Modified testcase:
void foo (int *);
void
f2 (int R)
{
int i, inter[3];
for (i = 1; i < R; i++)
{
inter[0] = 1;
inter[1] = 1;
inter[2] = 1;
}
foo (inter);
}
For the store data-race consider
int inter[3];
void
f2 (int R)
{
int i;
for (i = 1; i < R; i++)
{
inter[0] = 1;
inter[1] = 1;
inter[2] = 1;
}
}
where inter is protected by a mutex, but only if R > 1. We still
insert loads/stores on paths that are executed when the loop is not
entered:
f2 (int R)
{
int inter_I_lsm.5;
int inter_I_lsm.4;
int inter_I_lsm.3;
int i;
<bb 2>:
inter_I_lsm.3_10 = inter[0];
inter_I_lsm.4_11 = inter[1];
inter_I_lsm.5_12 = inter[2];
goto <bb 4>;
<bb 3>:
inter_I_lsm.3_13 = 1;
inter_I_lsm.4_14 = 1;
inter_I_lsm.5_15 = 1;
i_4 = i_1 + 1;
<bb 4>:
# i_1 = PHI <1(2), i_4(3)>
# inter_I_lsm.3_9 = PHI <inter_I_lsm.3_10(2), inter_I_lsm.3_13(3)>
# inter_I_lsm.4_8 = PHI <inter_I_lsm.4_11(2), inter_I_lsm.4_14(3)>
# inter_I_lsm.5_7 = PHI <inter_I_lsm.5_12(2), inter_I_lsm.5_15(3)>
if (i_1 < R_3(D))
goto <bb 3>;
else
goto <bb 5>;
<bb 5>:
# inter_I_lsm.5_19 = PHI <inter_I_lsm.5_7(4)>
# inter_I_lsm.4_20 = PHI <inter_I_lsm.4_8(4)>
# inter_I_lsm.3_21 = PHI <inter_I_lsm.3_9(4)>
inter[0] = inter_I_lsm.3_21;
inter[1] = inter_I_lsm.4_20;
inter[2] = inter_I_lsm.5_19;
return;
}
The issue is mitigated by loop-header copying because we there have a suitable
insertion place that is only executed when the loop body is executed (we seem
to cater for not inserting possibly trapping insns that way but do not
consider data-races possible for global memory).
