https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87886
Jakub Jelinek <jakub at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Target Milestone|7.4 |--- Summary|[7/8/9 Regression] ICE in |ICE in format_helper, at |format_helper, at |real.h:227 |real.h:227 | --- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Actually, can't find a compiler that doesn't ICE, even r90000 ICEs on this: pr87886.c: In function ‘foo’: pr87886.c:7: internal compiler error: output_operand: floating constant misused Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. The current ICE is that because apparently various spots in match.pd assume that a builtin call will have compatible arguments with the builtin, but in generic nothing actually checks that, and the FUNCTION_DECL doesn't even have a type with TYPE_ARG_TYPES at all; this gets reverted during pop_scope of the file scope, so what later gimple opts see is already proper function call.