[Bug target/67260] [sh] Register spill bug for sibcall+complex+softfloat

[olegendo at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67260

--- Comment #13 from Oleg Endo <olegendo at gcc dot gnu.org> ---
(In reply to Alexander Monakov from comment #12)
> 
> I mostly wonder why does sh.md change RTL representation of a sibcall that
> way, instead of simply generating the required relative address load
> upfront, during RTL expand. This way, RTL better matches final instruction
> stream, register allocation should be more natural, and the load of the
> relative address is subject to all RTL optimizations (so if you call a
> function in a loop, you can load the relative address once, out of the loop).

Good question indeed.  Kaz, maybe you remember anything?