Comment 3 for bug 1717257

I've found several bug reports and am sure that the problem stems from clang-3.8 being built with libstdc++-7-dev and not properly supporting __float128
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=678033

Google patched gcc by hiding _GLIBCXX_USE_FLOAT128 from clang:
https://gcc.gnu.org/ml/gcc-patches/2012-05/msg00672.html

There also seems to be a simple typedef workaround:
https://stackoverflow.com/questions/13525774/clang-and-float128-bug-error

I had to patch /usr/include/c++/7.2.0/bits/std_abs.h
in order to compile cupy from source with CUDA8.
#if !defined(__STRICT_ANSI__) && defined(_GLIBCXX_USE_FLOAT128) && !defined(__CUDACC__)