> >   Major rehash of the register allocator.  Fixes at least 
> one problem:
> >
> >     - clobbering: we shouldn't delete clobbered temps from the
> >       assignment, because they might be read by this instruction.
> 
> Ah yes. Another strange situation that doesn't happen in the beautiful
> PowerPC instruction set ;-).
> The code now looks nicer than before, but it fails when dealing with
> ccalls on PowerPC (the only place where clobbering ever happens around
> here). AFAICT, there are two things missing:
> 
> * The spill instructions returned by saveClobberedTemps are discarded.
> * The clobbered reg seems to remain InBoth until after the 
> instruction,
>    so now load is generated when the temporary is needed again.

ITYM "no load is generated..." (that had me confused for a while!).
Yes, rigth on both counts.  I'll commit fixes.

Cheers,
        Simon
_______________________________________________
Cvs-ghc mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/cvs-ghc

Reply via email to