[Bug c/123015] __builtin_isnan() should not trigger exception on sNaNs with -fsignaling-nans

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123015

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |wrong-code
                 CC|                            |jsm28 at gcc dot gnu.org
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2025-12-05
             Status|UNCONFIRMED                 |NEW

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
This is fold_builtin_interclass_mathfn doing

    CASE_FLT_FN (BUILT_IN_ISNAN):
    case BUILT_IN_ISNAND32:
    case BUILT_IN_ISNAND64:
    case BUILT_IN_ISNAND128:
      {
        /* In IBM extended NaN and Inf are encoded in the high-order double
           value only.  The low-order value is not significant.  */
        if (is_ibm_extended)
          arg = fold_build1_loc (loc, NOP_EXPR, double_type_node, arg);
        arg = builtin_save_expr (arg);
        tree type = TREE_TYPE (TREE_TYPE (fndecl));
        return fold_build2_loc (loc, UNORDERED_EXPR, type, arg, arg);

and for the isnan() case it's likely glibc "interfering".

Confirmed what the reporter says, not sure if that's correct.
isunordered() says it does not raise any flags.