[Bug rtl-optimization/125852] [16/17 Regression] Failure to late-combine address generation with load/store

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125852

--- Comment #1 from Drea Pinski <pinskia at gcc dot gnu.org> ---
What might be happening with my patch is causing expand to go differently.
That is it might cause out of ssa choices to be different.
This is especially true with loops.
I still need to look into some way of improving out of ssa choices and/or move
some statements out of the loop so it does not produce overlapping live ranges.
The loop vs overlapping live ranges is on my road map but I was finishing up
some other stuff first.