https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117321
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org
Resolution|MOVED |---
Ever confirmed|0 |1
Last reconfirmed| |2024-10-28
Status|RESOLVED |ASSIGNED
--- Comment #5 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Created attachment 59473
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=59473&action=edit
gcc15-pr117321.patch
Lightly tested patch.
While std::__is_constant_evaluated() is always_inline, at -O0 that just means
we inline the setting of some temporary to 0, but it doesn't lead to optimizing
away code guarded by it. We need something that is folded to false already in
the FE or worst case during the gimplification, which I think both
if consteval and if (__builtin_is_constant_evaluated()) guarantee.
As GCC predefines __STDCPP_FLOATn_T__ etc. macros only for -std=c++23 or later,
I think just using if consteval is enough, the guard is just in case clang does
something weird in the future (currently it doesn't support the paper nor
predefine __STDCPP_FLOATn_T__).
