https://gcc.gnu.org/bugzilla/show_bug.cgi?id=26388
Jan Hubicka <hubicka at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |hubicka at gcc dot gnu.org
--- Comment #20 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
with __builtin_operator_new we could theoretically optimize the initial
testcase to empty function, but we don't, since we can not handle the
reallocation.
_60 = operator new (_59);
<bb 9> [local count: 166566348]:
_46 = (long unsigned int) _38;
_47 = _60 + _46;
*_47 = i_56;
if (_38 != 0)
goto <bb 10>; [41.48%]
else
goto <bb 11>; [58.52%]
<bb 10> [local count: 69091722]:
__builtin_memcpy (_60, v$_M_start_73, _46);
_51 = _46 + 4;
__new_finish_54 = _60 + _51;
goto <bb 12>; [100.00%]
so here we would need to be first able to handle the chain of memcpy copies and
*_47 stores probably in DSE and once we get rid of the stores and memcpy
theoretically dce can optimize out the allocation/eh and other noise.
