[OT] gcc 3 [was Re: [plug] Slow kernel compile after rebuild]

James Devenish devenish at guild.uwa.edu.au
Fri Oct 3 08:41:12 WST 2003


In message <3F7C370D.8020105 at postnewspapers.com.au>
on Thu, Oct 02, 2003 at 10:32:45PM +0800, Craig Ringer wrote:
> >>These gcc 3 speed issues you say are normal ... is this the designed
> >>behaviour - if so do you know why?
> >
> >Perhaps gcc 3 is so much of a rewrite that no one has had time to
> >tighten up its code, yet.
> 
> My understanding is that a large part of the slowdown is caused by gcc3 
> being much more pedantic about the code. This qualifies as a good thing 
> in my mind. OTOH, I'll bet if there was a '--disable-error-checking 
> (DON'T USE EXCEPT FOR TESTING)'

Yes, a --compile-c-within-a-reasonable-time option would indeed be a
welcome feature. OTOH I suppose there are Java developers that put up
with Sun's Java compiler. Still, during development, there's often no
need be doing optimisations, and most lines of code get re-compiled/
re-linked a vast number of times with no new issues. And for C ... is it
really a language where there's much benefit from a new compiler? I
would have thought C++ would be more of a target, but I'm not sure that
gcc 3's C++ performance has been great, either (though I haven't really
looked in to it for at least a year). I do recall compiling gcc 3 with
gcc 3 and seeing if gcc 3 would then compile other software faster :-)

Although it's fairly natural that software gets slower over time (just
because people take advantage of faster hardware) I find it hard to
believe that gcc 3 could be fundamentally so slow -- there "must" be
room for improvement through refactoring. For me, code compiled with gcc
3.0 and 3.1 ran slower or not sufficiently faster for any practical
benefit in the code that was most often being executed, and random
application crashes were occurring. It seems that there were some bugs
that caused invalid output for me, at least up to 3.2.0. And having
started using gcc 3 at version 3.0 (which was probably too soon), I'd
never had a revelation that made me sit up and think gcc 3 was ready for
widespread deployment on "production" or end-user systems.


_______________________________________________
plug mailing list
plug at plug.linux.org.au
http://mail.plug.linux.org.au/cgi-bin/mailman/listinfo/plug


More information about the plug mailing list