[Bug libstdc++/43108] mixed complex<T> multiplication horribly inefficient

["rguenth at gcc dot gnu dot org" <gcc-bugzilla at gcc dot gnu dot org>]

------- Comment #6 from rguenth at gcc dot gnu dot org  2010-02-17 22:15 -------
(In reply to comment #4)
> (In reply to comment #3)
> > Well, but float * complex<float> isn't the same as complex<float> *
> > complex<float>.  But the library (and also C promotion if you write
> > that in literal C) presents us with complex<float> * complex<float>.
> 
> Where, exactly, the library presents complex * complex instead of complex *
> float? I really don't see it. I see things like:
> 
>       complex<float>&
>       operator*=(float __f)
>       {
>         _M_value *= __f;
>         return *this;
>       }
> 
> and I don't see where the library explicitly transforms __f to a complex, I see
> something like complex = complex * float. But really, we have been through this
> *already*, I'm absolutely sure, and I have zero problems with changing the
> above, if Gaby agrees and you all are sure something changed since the lat
> time, it's a 5 minutes work, but I'm still convinced that this is something
> much more generally useful to optimize than the library.

Well, also in C++ you get promotion to _M_value * complex <__f, 0.0> which
we cannot optimize in case we care for the sign of zeros (you get the
desired optimization with -fno-signed-zeros).


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=43108