ahatanak added a comment. __builtin_fpclassify takes five int arguments followed by one last argument that is of floating point type. Do you know if there is a way to specify the last one argument is a floating point rather than using '.'? https://reviews.llvm.org/D25480