quadmath.h not found on clang

[Nathan Ridge]

When trying to compile:

#include <boost/math/special_functions/sign.hpp>

with clang, I get the following compiler error:

In file included from test.cpp:1:
In file included from boost/math/special_functions/sign.hpp:16:
boost/math/tools/config.hpp:215:10: fatal error: 'quadmath.h' file not found
#include <quadmath.h>
         ^

[John Maddock]

(In [83356]) Remove dependency on quadmath.h. Fixes #8265.

[Nathan Ridge]

Thanks for the fix. Unfortunately, the example given in the description still doesn't compile with clang. This time the error is:

In file included from test.cpp:1:
In file included from boost/math/special_functions/sign.hpp:17:
In file included from boost/math/special_functions/math_fwd.hpp:28:
boost/math/policies/policy.hpp:819:18: error: use of undeclared identifier '__float128'
struct precision<__float128, Policy>
                 ^

[John Maddock]

OK some more questions:

• Do you know if it's possible to detect when clang supports __float128? Recent SVN Trunk versions certainly do so?
• Presumably your version can't handle libstdc++ as that also refers to __float128 in <type_traits> ?
• Which clang version are you using also?

Thanks!

[Nathan Ridge]

I am using a very recent clang (version 3.3, trunk revision 176405). I am using it with libstdc++, which it handles just fine. The libstdc++ version it uses is from a recent GCC build (4.8.0 20130224).

[Nathan Ridge]

Regarding <type_traits>, I did a bit of digging and discovered the following:

• <type_traits> can only be used in C++11 mode (obviously)
• There are two flags that get clang to operate in C++11 mode: -std=c++11 and -std=gnu++11. If the -std=c++11 flag is used, the symbol __STRICT_ANSI__ gets defined. If the -std=gnu++11 flag is used, that symbol does not get defined.
• The use of __float128 in <type_traits> is guarded by !defined(__STRICT_ANSI__)
• I usually compile my code with the -std=c++11 flag. In this configuration, <type_traits> compiles just fine. If I change the flag to -std=gnu++11, I get the same error about __float128 in <type_traits> that I get with Boost.Math.

[Nathan Ridge]

A quick google search suggests that clang doesn't have any support for __float128 yet.

[John Maddock]

You're right, I found this open bug report: ​http://llvm.org/bugs/show_bug.cgi?id=13530 on this very issue. So I'll unconditionally disable float128 support when the compiler is clang for now.

[John Maddock]

(In [83369]) Restrict float128 support to GCC only for now. Fixes #8265.

[anonymous]

I have the same problem, but with float128, can you help me, please?

/usr/local/include/boost/multiprecision/float128.hpp:40:10: fatal error:

'quadmath.h' file not found

#include <quadmath.h>

^~

[John Maddock]

clang has no support for float128 and libquadmath, so you're out of luck: that header is supported by gcc and Intel compilers only.