[Bug target/82533] inefficient code generation for copy loop on falkor

[wilson at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82533

--- Comment #1 from Jim Wilson <wilson at gcc dot gnu.org> ---
A reviewer suggested using register_offset in the address cost structure to fix
this.  That doesn't prevent all reg+reg addressing modes from being generated,
though this might be fixable with some work.  However, a bigger problem, is
that it disables reg+reg addressing modes for both loads and stores, and
disabling reg+reg loads reduces performance on Falkor.  It is only the reg+reg
quad-word store that needs to be disabled for best performance.  To make this
happen, we need to modify the md file.