This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Yet another attempt to solve multiplication by 11 and many (most?) other constants problem
> >>>>> Jan Hubicka writes:
>
> Jan> Current patch is biassed to 2 address architecture (that also saves registers) and I've
> Jan> verified that it produces no worse results in IA-64.
>
> This is not okay. GCC cannot repeatedly switch back and forth
> between which type of architecture is preferred. GCC 4.1 and 4.2
> synthetic multiply need to benefit three address architectures or we have
> a regression.
IA-64 is three address architecture and by counting clock cycles, we
don't have regression in first 65536 numbers after applying my patch. I
didn't tested PPC, but I can give it a try.
Due to nature of the _factor algorithm, we never considered those
alternatives originally. Technically this is not regression, but I
would be happy to provide the hook for controlling the algorithm if this
seems important. I would preffer followup patch, but I can bundle it
here too if we can agree on my proposed interface, or something else.
Honza
>
> David
>