https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92649
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org
--- Comment #7 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Or it could be handled by constexpr-like evaluation of the loops, where we'd
evaluate the side-effects at compile time both to PHIs after loop and written
memory, and then FRE-like propagate that to the second loop if there is some
cheap simple expression with which it can be replaced. Though, unsure if it
wouldn't handle only toy testcases like these.
