Reichelt - Performance regression with libstdc++-v3 (original) (raw)

This is the mail archive of the libstdc++@gcc.gnu.orgmailing list for the libstdc++ project.

• To: libstdc++ at gcc dot gnu dot org
• Subject: Performance regression with libstdc++-v3
• From: Reichelt
• Date: Wed, 16 May 2001 15:10:51 +0200

Hello!

I'm experiencing quite a performance problem with lists under libstdc++-v3. Consider the following simple program:

#include

int main () { std::list List; for ( int i=0; i<10000000; ++i ) List.push_back( int() ); return 0; }

Compiled with different versions of g++ (-O3 -funroll-loops -static) I get the following timings (on an i686 system, but it looks similar on an IRIX6.5 workstation):

• old libstdc++:

• gcc 2.95.2: 2,51s user 1,01s system 3,516 total
• gcc 20001106: 2,36s user 1,08s system 3,420 total

• new libstdc++-v3:

• gcc 20001113: 4,58s user 1,13s system 5,702 total
• gcc 20010507: 4,69s user 1,06s system 5,749 total

This is a regression in performance of about 65% that came with the introduction of libstdc++-v3! (I also tested several snapshots between December 2000 and May 2001, but the results are the same. Compiling with "-fno-exceptions" doesn't help either.)

That leads to a couple of questions:

• Is that a known problem?
• Any ideas what causes this performance breakdown (maybe some debugging code, or is it a problem of the optimizer)?
• Are there any compiler options that I can use to increase performance?
• Can somebody improve speed again, please?

Greetings, Volker Reichelt

email: reichelt@igpm.rwth-aachen.de http://www.igpm.rwth-aachen.de/reichelt

• Follow-Ups:
  • Re: Performance regression with libstdc++-v3
    * From: Loren James Rittle
  • Re: Performance regression with libstdc++-v3
    * From: Benjamin Kosnik