https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116651
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|RESOLVED |ASSIGNED
Last reconfirmed| |2025-08-28
Ever confirmed|0 |1
Resolution|DUPLICATE |---
Component|middle-end |tree-optimization
Severity|normal |enhancement
Assignee|unassigned at gcc dot gnu.org |pinskia at gcc dot
gnu.org
--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
.
For the simplified version of:
```
#include<vector>
using namespace std;
using T = int;
bool test1(const std::vector<int>& in) {
return in == std::vector<int>{42};
}
```
We don't fold __builtin_memcmp until it is too late and then we don't optimize
way the extra load.
That is the strlen pass is way too late. I will look into this further.
```
bool test1(const std::vector<std::string_view>& in) {
return in == std::vector<std::string_view>{"*"};
}
```
is much more complex.
There is a missed optimization dealing:
```
_12 = _10 - _11;
if (_12 == 16)
goto <bb 10>; [34.00%]
else
goto <bb 9>; [66.00%]
...
<bb 10> [local count: 364780225]:
if (_10 != _11)
goto <bb 3>; [96.34%]
else
goto <bb 8>; [3.66%]
```
if `_10 - _11` is 16 then obviously `_10 != _11`. And then we somehow lose that
the loop back-edge:
```
# __first1_13 = PHI <__first1_45(6), _11(9)>
...
<bb 6> [local count: 74161852]:
__first1_45 = __first1_13 + 16;
__first2_46 = __first2_58 + 16;
if (_10 != __first1_45)
goto <bb 3>; [96.34%]
else
goto <bb 7>; [3.66%]
```
Only iterates once.
