[Bug target/34529] [4.1/4.2/4.3 Regression] Wrong code with altivec stores and offsets

[dje at gcc dot gnu dot org]

------- Comment #10 from dje at gcc dot gnu dot org  2008-01-21 18:18 -------
The secondary_reload target hook does not provide the full MEM RTL as an
argument, only the address RTL, which prevents the hook implementation from
distinguishing ALTIVEC, which only allows indexed addresses, from GPR and FPR,
which allow offset addresses.

If the target hook can add a scratch register, reload should be able to add a
scratch register itself.

If reload cannot fix this address correctly, rs6000_legitimate_offset_address
needs to reject offset addresses for Altivec modes.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34529