[Bug c++/69564] [5/6 Regression] lto and/or C++ make scimark2 LU slower

[jason at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69564

--- Comment #8 from Jason Merrill <jason at gcc dot gnu.org> ---
Created attachment 37774
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=37774&action=edit
loop inversion sketch

This patch does loop inversion sufficient for scimark.  It will break
constexpr, but might be useful for looking at the LTO issue.