GCC optimization bugs -- still there or a historic artifact?
Paul Seniura
pdseniura at techie.com
Tue Mar 16 07:40:33 PST 2004
Hi,
> [snip]
> > And as for PPC, I've compiled things as high as -O5,
> > which is its limit I believe. ;)
> [snip]
>
> hm, no docs for -On, n > 3 at http://xrl.us/brh2 . Looking at the
> code: http://xrl.us/brh5 (search for 'optimize >= 3'), I don't see any
> evidence that -O5 is different from -O3. Though apple gcc is somewhat
> changed from fsf gcc, so maybe apple gcc does something different.
(kinda o.t.)
Oh Yes I'm quite sure of it -- Apple and IBM are working on
gcc and feeding the changes back to the GNU teams, who are
probably swamped. ;) So I bet it won't be seen in gcc34
yet, tho. And this work may also be going into IBM's
'enterprise' compiler (the US$3K version ;) .
I believe the man page for gcc on Panther does show what
the levels do at -O4 and -O5. (My G4 is at home, TPTB
would have a hissy-fit if I brought it in. I have the
Darwin CD for i386 based on Panther, here at work; maybe
gcc is installed from it, and I can try posting its manpage
somehow.) I have a feeling the higher -O levels won't show
until/unless gcc itself is compiled for ppc platforms,
y'know #ifdefs in the src. Else the patches for the GNU
teams are in a long line for review.
They're doing the work because IBM *really* want the G5 to
be very-well supported. And not just for Apple. You see
Sony & M$ have announced their next game consoles will be
based on the G5. I'm hoping all gcc users will eventually
see the benefits of their work.
-- thx, Paul Seniura.
More information about the freebsd-hackers
mailing list