------- Comment #11 from rguenther at suse dot de 2010-01-16 13:19 ------- Subject: Re: [4.5 Regression] -O2 and verify_stmts failed again
On Sat, 16 Jan 2010, rakdver at gcc dot gnu dot org wrote: > > > ------- Comment #10 from rakdver at gcc dot gnu dot org 2010-01-16 12:53 > ------- > /* Reject a tailcall if the type conversion might need > 285 additional code. */ > 286 if (gimple_assign_cast_p (stmt) > 287 && TYPE_MODE (TREE_TYPE (dest)) != TYPE_MODE > (TREE_TYPE (src_var))) > 288 return false; > > This is suspicious; replacing the weird TYPE_MODE check with > useless_type_conversion_p fixes the problem. Correct, but for useless_type_conversion_p we can reject any conversions in the IL since only !useless_type_conversion_p remain. I suppose the code was explicitly trying to allow sign-changing (but not precision changing as it does), or conversion to/from pointers. Thus, if you want to go the useless_type_conversion_p route simply reject all conversions and only allow plain SSA name copies... Richard. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42749