>> I bet the above workaround in generic code was added for a reason, it would >> surprise me if _Float128 worked at all without that hack. > > OK, I'll have a look at those nan failures soon.
By investigating the exposed NaN failures, I found it's due to that it wants to convert _Float128 type constant to long double type constant, it goes through function real_convert which clears the signalling bit in the context of !HONOR_SNANS (arg). if (r->cl == rvc_nan) r->signalling = 0; The test cases don't have the explicit option -fsignaling-nans, I'm inclined to believe it's intentional since there is only a sNaN generation. If so, we don't want this kind of conversion which is useless and can clear signalling bit unexpectedly, one shortcut is to just copy the corresponding REAL_VALUE_TYPE and rebuild with the given type if the modes are the same. ----- diff --git a/gcc/fold-const.cc b/gcc/fold-const.cc index e80be8049e1..d036b09dc6f 100644 --- a/gcc/fold-const.cc +++ b/gcc/fold-const.cc @@ -2178,6 +2178,14 @@ fold_convert_const_real_from_real (tree type, const_tree arg1) REAL_VALUE_TYPE value; tree t; + /* If the underlying modes are the same, just copy the + TREE_REAL_CST information and rebuild with the given type. */ + if (TYPE_MODE (type) == TYPE_MODE (TREE_TYPE (arg1))) + { + t = build_real (type, TREE_REAL_CST (arg1)); + return t; + } + /* Don't perform the operation if flag_signaling_nans is on and the operand is a signaling NaN. */ if (HONOR_SNANS (arg1) ----- The above diff can fix all exposed NaN failures. BR, Kewen