This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: Fix -fwhole-program on LTO -- Another SPEC2000 benchmakring round
On Thu, 8 Oct 2009, Diego Novillo wrote:
> On Thu, Oct 8, 2009 at 11:32, Steven Bosscher <stevenb.gcc@gmail.com> wrote:
>
> > Is this kind of slow-down expected (i.e. similar to LTO cost of other
> > compilers)?
>
> Yes, we compile everything twice since .o files have both final object
> code and gimple bytecode. Add to that the penalty of doing IPA over
> larger callgraphs. We have some large polynomial algorithms there.
> On a very large internal application we spend 30 CPU minutes making
> inlining decisions.
>
> There is also the penalty of streaming bytecode. -ftime-report breaks
> those down. I don't recall exact figures, but streaming was not a
> significant factor.
Btw, on SPEC 2006 I see about a factor of 4 to 6 increase in compile-time
(skewed by those that run out of memory - I have yet to get my hands
on a convenient machine with enough memory).
Richard.