This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

compile times: Success with gcc300


Hi,

I have some numerical C++ code, which implements the temporary base class
idiom to avoid unecessary copying of data.
The renewing of the framework also brought things like the possibility to
assign names and to have type info of the objetcs.

However, g++ 2.95.x was quite unhappy about the code. While it compiled it
within a few seconds without -g, it went crazy with -O2. It compiled 10 min
(Athlon 700) and needed like 480 MB of RAM. gcc-3.0 snapshots were ever more
cray: They went up to 700MB before the machine killed cc1plus with because
of out of memory. (Compaq CXX BTW could also compile with optimization
without any problems.)

To my amazement, recent (20010418) gcc-3.0.0 snapshots finally fixed the
problem. The compiler needs a minute and uses 50MB.

GREAT JOB! I'd like to thank you for achieving this. The g++ is finally
useable again for me. Thanks!

I was a bit curious andlooked at the gcc maling list archives.
I guess the c++-inline-throttle stuff did it. Is this correct?
Is there a way to tune this? I noticed slighlty decreased performance for
some stuff, comparing 2.95.3 with 3.0.0. The inline throttling could be
responsible, of course ...

Please keep up your great job!

PS: If you answer, please Cc: me, I'm not subscribed.
-- 
Kurt Garloff                   <kurt@garloff.de>         [Eindhoven, NL]
Physics: Plasma simulations  <K.Garloff@Phys.TUE.NL>  [TU Eindhoven, NL]
Linux: SCSI, Security          <garloff@suse.de>   [SuSE Nuernberg, FRG]
 (See mail header or public key servers for PGP2 and GPG public keys.)

PGP signature


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]