[Bug target/108018] Wide immediate sequences not scheduled for POWER10 fusion

[segher at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108018

Segher Boessenkool <segher at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |segher at gcc dot gnu.org

--- Comment #1 from Segher Boessenkool <segher at gcc dot gnu.org> ---
The sequences generating the GPR3 and GPR9 values are identical.  This is not
ideal at all, irrespective of the other problems you point out here.  Even if
they were just related (say, a small non-zero integer difference) we can do
better than this.

For the problem as pointed out here, it should be done in genfusion.pl,
probably
in gen_addadd, which currently only does register "add+add" things, not addi
and addis things?

Also, are there other fusion things missing?