------- Comment #2 from rguenth at gcc dot gnu dot org 2007-09-17 08:33 ------- I guess so. All the testcases rely on either availability of cexp() or sincos(). For the former via TARGET_C99_FUNCTIONS, for the latter via TARGET_HAS_SINCOS.
But it is interesting that -59 fails, as the transformation checked should be done via fold_builtin_sincos which only bails out if !TARGET_C99_FUNCTIONS or cepxi is not available (but that is a GCC builtin, so should be available all the times). I don't have time to investigate if we can somehow relax all the machinery around this transformation, but we basically try to avoid locking us up in a corner where we end up with the internal CEXPI but have no way to re-emit either sincos or cexp calls. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=31497
