Stage 1 Idea: Make __float128 DTRT...

[Ed Smith-Rowland]

I would like to use __float128 in special functions...
To do that we need to make the library safe...
1. Specialize numeric_limits
2. Expose basic math functions.
...
I also have functional forms of numeric_limits to account for the fact 
that __float128 has non-constexpr infinity and NaNs.
Also, I would like to pave the way for multi-precision types were 
epsilon() will depend on the instance and may be dynamic.

-------------- next part --------------
#ifndef FLOAT128_TCC
#define FLOAT128_TCC 1

#if !defined(__STRICT_ANSI__) && defined(_GLIBCXX_USE_FLOAT128)

#include <iostream>
#include <sstream>
#include <quadmath.h>

namespace std
{

  template<typename _CharT, typename _Traits = std::char_traits<_CharT>>
    std::basic_ostream<_CharT, _Traits>&
    operator<<(std::basic_ostream<_CharT, _Traits>& __os,
	       const __float128& __x)
    {
      auto __sci = __os.flags() & std::ios::scientific;
      auto __hex = __os.flags() & std::ios::fixed
		&& __os.flags() & std::ios::scientific;
      //auto __hex = __os.flags() & (std::ios::fixed | std::ios::scientific);
      auto __upper = __os.flags() & std::ios::uppercase;
      auto __width = __os.width();
      std::ostringstream __fmt;
      __fmt << '%';

      if (__os.flags() & std::ios::showpos)
	__fmt << '+';
      else
	__fmt << ' '; // Space instead of plus standard?

      if (__os.flags() & std::ios::left)
	__fmt << '-';

      __fmt << __os.width() << '.' << __os.precision() << 'Q';

      if (__hex)
	__fmt << (__upper ? 'A' : 'a');
      else if (__sci)
	__fmt << (__upper ? 'E' : 'e');
      else
	__fmt << (__upper ? 'G' : 'g');

      constexpr int __strlen = 1000;
      char __str[__strlen];
      quadmath_snprintf(__str, __strlen, __fmt.str().c_str(), __x) ;
      __os << __str;
      return __os;
    }

  template<typename _CharT, typename _Traits = std::char_traits<_CharT>>
    std::basic_istream<_CharT, _Traits>&
    operator>>(std::basic_istream<_CharT, _Traits>& __is, __float128& __x)
    {
      constexpr int __strlen = 160;
      char __str[__strlen];
      __is >> std::setw(__strlen) >> __str;
      __x = strtoflt128(__str, 0);
      return __is;
    }

} // namespace std

#endif // __STRICT_ANSI__ && _GLIBCXX_USE_FLOAT128

#endif // FLOAT128_TCC
-------------- next part --------------
A non-text attachment was scrubbed...
Name: float128.h
Type: text/x-chdr
Size: 17971 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20151210/918328b3/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: numeric_limits.h
Type: text/x-chdr
Size: 7147 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20151210/918328b3/attachment-0001.bin>