[Bug middle-end/71315] missing strlen optimization on a POINTER_PLUS expression
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71315 Martin Sebor changed: What|Removed |Added Status|UNCONFIRMED |RESOLVED See Also||https://gcc.gnu.org/bugzill ||a/show_bug.cgi?id=79547 Resolution|--- |FIXED --- Comment #5 from Martin Sebor --- Resolving the original bug as fixed. I've raised bug 79547 for the problem pointed out in comment #3.
[Bug middle-end/71315] missing strlen optimization on a POINTER_PLUS expression
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71315 Jakub Jelinek changed: What|Removed |Added CC||jakub at gcc dot gnu.org --- Comment #4 from Jakub Jelinek --- This is alias analysis deficiency. It considers s escaped, because [0] is passed to a function (strlen), but __builtin_strlen really is not allowed to save the argument into some global var that e.g. the following call to f function could then use to modify it. The strlen pass just uses the alias oracle to query if it needs to invalidate remembered string lengths, in particular on the f () call. As the alias oracle considers s to be escaped, it returns that the f () call might change it (which it indeed could if instead of __builtin_strlen one used some function with the same prototype, but on which the compiler can't make any assumptions).
[Bug middle-end/71315] missing strlen optimization on a POINTER_PLUS expression
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71315
prathamesh3492 at gcc dot gnu.org changed:
What|Removed |Added
CC||prathamesh3492 at gcc dot
gnu.org
--- Comment #3 from prathamesh3492 at gcc dot gnu.org ---
Hi,
As of r238513, gcc folds call to strlen into constant.
void f(unsigned x)
void g()
{
char s[] = "hello";
f (__builtin_strlen (s + 1));
}
optimized dump shows:
;; Function g (g, funcdef_no=0, decl_uid=1753, cgraph_uid=0, symbol_order=0)
Optimizing: _1 = __builtin_strlen ([(void *) + 1B]);
into: _1 = 4;
g ()
{
char s[6];
long unsigned int _1;
unsigned int _2;
:
s = "hello";
_1 = 4;
_2 = (unsigned int) _1;
f (_2);
s ={v} {CLOBBER};
return;
}
However there seems to be another issue, if there are repeated
calls to strlen() with same arg, only the first call is folded.
For example consider following case:
void f (unsigned);
void g (void)
{
char s[] = "hello";
f (__builtin_strlen (s));
f (__builtin_strlen (s));
}
Only the first call to __builtin_strlen() is folded to 5,
the second one isn't.
Optimized dump shows:
;; Function g (g, funcdef_no=0, decl_uid=1753, cgraph_uid=0, symbol_order=0)
Optimizing: _1 = __builtin_strlen ();
into: _1 = 5;
g ()
{
char s[6];
long unsigned int _1;
unsigned int _2;
long unsigned int _3;
unsigned int _4;
:
s = "hello";
_1 = 5;
_2 = (unsigned int) _1;
f (_2);
_3 = __builtin_strlen ();
_4 = (unsigned int) _3;
f (_4);
s ={v} {CLOBBER};
return;
}
clang seems to fold both the calls to 5.
Thanks,
Prathamesh
[Bug middle-end/71315] missing strlen optimization on a POINTER_PLUS expression
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71315
--- Comment #2 from Martin Sebor ---
I pasted the wrong test case in comment #0. The correct test case is as
follows:
void f (unsigned);
void g (void)
{
char s[] = "1234";
f (__builtin_strlen (s + 1));
f (__builtin_strlen (s + 1));
}
[Bug middle-end/71315] missing strlen optimization on a POINTER_PLUS expression
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71315 Martin Sebor changed: What|Removed |Added Keywords||missed-optimization Known to fail||4.5.3, 4.8.3, 4.9.3, 5.3.0, ||6.1.0, 7.0 --- Comment #1 from Martin Sebor --- This optimization was never performed (going as far back as 4.5.3).
