https://gcc.gnu.org/bugzilla/show_bug.cgi?id=96862
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jason at gcc dot gnu.org
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
What fold-const.c has is:
/* Don't constant fold this floating point operation if the
result may dependent upon the run-time rounding mode and
flag_rounding_math is set, or if GCC's software emulation
is unable to accurately represent the result. */
if ((flag_rounding_math
|| (MODE_COMPOSITE_P (mode) && !flag_unsafe_math_optimizations))
&& (inexact || !real_identical (&result, &value)))
return NULL_TREE;
i.e. if the result is inexact, it punts with -frounding-math (or for IBM double
double without -ffast-math).
So are you suggesting that constexpr.c should temporarily disable
flag_rounding_math during !allow_non_constant mode (or rather
manifestly_const_eval || !allow_non_constant mode), perhaps directly in
cxx_eval_outermost_constant_expr, on the basis that constant evaluation happens
when the rounding mode couldn't be changed?
For constant evaluation not required by the standard we need to still punt
IMNSHO, as that is normally evaluated at runtime where it could be affected by
rounding mode.
