Re: compiler_rt missing support for 80-bit floats?

[Aymeric Vincent <aymericvincent%free.fr@localhost>]

Martin Husemann <martin%duskware.de@localhost> writes:

> But the library better have a very good reason to use such a stupid,
> obviously non portable feature. This should be checked and (as it probably
> is not reasonable) be reported upstream.

If we have to choose a culprit between boost and libuhd, it is boost;
libuhd doesn't use anything else than doubles.

The __float128's are used in a constructor (related to the Lanczos
function, I'm not at all familiar but I don't think it's relevant).
The following (small!) example exhibits the problem on netbsd-7/amd64.

Regards,
 Aymeric

y$ cat foo.cc
#include <boost/math/special_functions.hpp>

int
main(void) {
    return 0;
}
y$ g++ -I /usr/pkg/include -o foo foo.cc
/var/tmp//ccI3XALz.o: In function `long double const boost::math::tools::make_big_value<long double>(__float128, char const*, mpl_::bool_<true> const&, mpl_::bool_<false> const&)':
foo.cc:(.text._ZN5boost4math5tools14make_big_valueIeEEKT_gPKcRKN4mpl_5bool_ILb1EEERKNS8_ILb0EEE[_ZN5boost4math5tools14make_big_valueIeEEKT_gPKcRKN4mpl_5bool_ILb1EEERKNS8_ILb0EEE]+0x1f): undefined reference to `__trunctfxf2'
y$