I found another solution to this problem.

add:

{
    "C_Cpp.codeAnalysis.clangTidy.args": [ "${default}", "-mlong-double-64" ],
}

to settings.json file. This will set the long-double size to 64 and so 128 won't be included.

I get a similar error when I compile from the command line manually without this option. But there it is about missing a stream << operator for the __float128 type. :-(