https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114484
Jonathan Wakely <redi at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|INVALID |DUPLICATE
--- Comment #11 from Jonathan Wakely <redi at gcc dot gnu.org> ---
(In reply to vincenzo Innocente from comment #9)
> We observe that including xmmintrin.h the behaviour of some code,
> notably abs(x), when x is float or double changes.
Yeah, it *fixes* your code. Without <xmmintrin.h> your abs(f) code calls
abs(int) which is wrong.
e.g. https://godbolt.org/z/sxv69hv5K
The arm code uses abs(int) which is wrong. The problem here is not that
xmmintrin.h includes another header, it's that your code is not including the
headers it requires. The fact that xmmintrin.h includes stdlib.h means you get
away with it on one platform, but not on another. That portability problem
isn't great, but it's not actually a bug in xmmintrin.h.
> And this depends on the platform as xmmintrin.h is x86_64 specific.
Yes, but different headers having different transitive inclues on different
platforms is not actually a bug.
> Yes, is 20 years that is like that and people always wandered why abs(x) was
> behaving differently in different parts of the code and now asking why it
> behaves differently on x86_64 and ARM.
Righ, so *that's* the bug, not that xmmintrin.h includes another header, and
not that C++ code is "supposed to #include <cstdlib> not <stdlib.h>".
There is nothing wrong with xmmintrin.h including other headers.
> The workaround is obvious: use std::abs.
>
> I personally find very unconfortable that including (even through cascade)
> xmmintrin.h changes the behaviour of "abs(x)"
When Bug 89855 is fixed, it won't change it. <cmath> will put all overloads of
abs in the global namespace.
This is a dup of Bug 89855
*** This bug has been marked as a duplicate of bug 89855 ***
