On Thu, 3 May 2018, Jakub Jelinek wrote: > On Thu, May 03, 2018 at 06:35:50PM +0200, Jakub Jelinek wrote: > > That requires a machine_mode or tree, but I don't have either of those, > > nor the caller (fold_const_call_sss) has those. > > > > I could change it to: > > if (flag_signalling_nans > > && !flag_finite_math_only > > && format->has_nans > > so that it would better duplicate what HONOR_SNANS actually tests. > > Though, I think it is ok to punt if one of the operands is a signalling nan > > even if flag_signalling_nans. > > Though in theory the arg0's mode for which we have format pointer could > not have nans, but arg1's mode (for which we only have the REAL_VALUE_TYPE) > could have nans (or vice versa).
Hmm, indeed. OTOH what's wrong with just removing the flag_singalling_nans check, just keeping the actual checks for REAL_VALUE_ISSIGNALING_NAN? Thus the patch is ok with just those checks. The requested assert shouldn't be neccessary indeed. Thanks, Richard.
