Errors compiling float128.h on windows with icc 17.0 and MSVC 14.11 toolchain

[Metodi Nikolov <metodi.nikolov@…>]

The example code from the help here ​https://www.boost.org/doc/libs/1_66_0/libs/multiprecision/doc/html/boost_multiprecision/tut/floats/float128.html cannot be compiled on windows using intel compiler version 17.0.6 and MSVC v14.11 toolchain. The errors reported are:

1>G:\Libraries\boost_1_66_0\boost/multiprecision/float128.hpp(487): error : identifier "fmaq" is undefined
1>     result.value() = fmaq(a.value(), b.value(), c.value());
1>                      ^
1>
1>G:\Libraries\boost_1_66_0\boost/multiprecision/float128.hpp(492): error : the global scope has no "signbitq"
1>     return ::signbitq(arg.value());
1>              ^
1>
1>G:\Libraries\boost_1_66_0\boost/multiprecision/float128.hpp(556): error : the global scope has no "copysignq"
1>        return ::copysignq(a.backend().value(), b.backend().value());
1>                 ^
1>
1>G:\Libraries\boost_1_66_0\boost/multiprecision/float128.hpp(561): error : identifier "remainderq" is undefined
1>        result.value() = remainderq(a.value(), b.value());
1>                         ^
1>
1>G:\Libraries\boost_1_66_0\boost/multiprecision/float128.hpp(565): error : identifier "remquoq" is undefined
1>        result.value() = remquoq(a.value(), b.value(), pi);
1>                         ^

Please note that newer MSVC toolchain versions seem to break the intel compiler (and the default version for MS2017 is newer), while for icc 18.0 one needs to remove -Za or the /permissive- options as they also need to problems with the compilation.

[John Maddock]

Unfortunately, that configuration has dropped off our testing radar... can you tell me if the patch below fixes things?

 include/boost/multiprecision/float128.hpp | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/include/boost/multiprecision/float128.hpp b/include/boost/multiprecision/float128.hpp
index a729523..095634f 100644
--- a/include/boost/multiprecision/float128.hpp
+++ b/include/boost/multiprecision/float128.hpp
@@ -482,15 +482,16 @@ inline void eval_atan2(float128_backend& result, const float128_backend& a, cons
 {
    result.value() = atan2q(a.value(), b.value());
 }
+#ifndef BOOST_MP_USE_QUAD
 inline void eval_multiply_add(float128_backend& result, const float128_backend& a, const float128_backend& b, const float128_backend& c)
 {
    result.value() = fmaq(a.value(), b.value(), c.value());
 }
-
 inline int eval_signbit BOOST_PREVENT_MACRO_SUBSTITUTION(const float128_backend& arg)
 {
    return ::signbitq(arg.value());
 }
+#endif
 
 inline std::size_t hash_value(const float128_backend& val)
 {
@@ -550,6 +551,7 @@ inline std::size_t hash_value(const float128_backend& val)
       return log1pq(arg.backend().value());
    }
 
+#ifndef BOOST_MP_USE_QUAD
    template <multiprecision::expression_template_option ExpressionTemplates>
    inline boost::multiprecision::number<boost::multiprecision::backends::float128_backend, ExpressionTemplates> copysign BOOST_PREVENT_MACRO_SUBSTITUTION(const boost::multiprecision::number<boost::multiprecision::backends::float128_backend, ExpressionTemplates>& a, const boost::multiprecision::number<boost::multiprecision::backends::float128_backend, ExpressionTemplates>& b)
    {
@@ -564,6 +566,7 @@ inline std::size_t hash_value(const float128_backend& val)
    {
       result.value() = remquoq(a.value(), b.value(), pi);
    }
+#endif
 
 } // namespace multiprecision
 

[Metodi Nikolov <metodi.nikolov@…>]

Hi,

yes, I can confirm that the patch fixes the errors I was seeing. (side note, in the spirit of full transparency, I couldn't get git to apply the patch, almost sure it is because of copying from the web to a file and windows messing the line endings. The four inclusions were done by hand)

Thanks a lot.

Best,

[John Maddock]

Thanks, fixed in develop.