The type-generic builtins apply the default variadic promotions to their arguments before handing them off to the middle-end for processing. This is bad because e.g. __builtin_isnan(f), where f is a float, gets turned into __builtin_isnan((double)f).
In most cases, the cast to double merely becomes a pessimization. However if we use a variadic builtin to implement e.g. isnormal, then casting a float argument to double can take a subnormal float like FLT_MIN/2 and make it a "normal" value because of the extended range of double. See: http://gcc.gnu.org/ml/gcc/2007-07/msg00221.html -- Summary: The type-generic builtins apply default promotions Product: gcc Version: 4.1.2 Status: UNCONFIRMED Keywords: wrong-code Severity: normal Priority: P3 Component: middle-end AssignedTo: unassigned at gcc dot gnu dot org ReportedBy: ghazi at gcc dot gnu dot org http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32668
