https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121789
--- Comment #2 from Avi Kivity <avi at scylladb dot com> ---
(testq/jle sequence)
The jle itself is from libstdc++:
```
// Overload for pointers.
template<typename _Tp, typename _Up>
static _Up*
__uninit_copy(_Tp* __first, _Tp* __last, _Up* __result)
{
// Ensure that we don't successfully memcpy in cases that should be
// ill-formed because is_constructible<_Up, _Tp&> is false.
typedef __typeof__(static_cast<_Up>(*__first)) __check
__attribute__((__unused__));
const ptrdiff_t __n = __last - __first;
if (__builtin_expect(__n > 0, true))
{
__builtin_memcpy(__result, __first, __n * sizeof(_Tp));
__result += __n;
}
return __result;
}
};
```
ptrdiff_t is signed, and the compiler believes you so it has to check.
I believe it's undefined to have __last not be reachable by incrementing
__first, so != would be fine here. Or not comparing and calling memcpy()
directly (which I see first compares against a bunch of small values before
detecting zero, I can't make up my mind if it's better to check in libstdc++ or
not).
I'll file an issue for the peephole optimization. It's a little tricky since
salq doesn't set the flags like other instructions.
