Skip to content
Snippets Groups Projects
  1. Aug 24, 2006
  2. Jun 25, 2006
  3. Jun 20, 2006
  4. Apr 13, 2006
  5. Dec 10, 2004
  6. Dec 09, 2004
  7. Dec 08, 2004
  8. Jul 19, 2004
  9. Jun 17, 2004
  10. Jun 16, 2004
  11. May 05, 2004
  12. Apr 15, 2004
  13. Mar 17, 2004
  14. Sep 15, 2003
  15. Aug 29, 2003
  16. Aug 23, 2003
  17. Aug 22, 2003
    • Rene Brun's avatar
      Use a cast in a printf statement to avoid a warning with gcc3.4 · 3f97a932
      Rene Brun authored
      git-svn-id: http://root.cern.ch/svn/root/trunk@7144 27541ba8-7e3a-0410-8455-c3a389f83636
      3f97a932
    • Rene Brun's avatar
      Very nice performance improvement in the ZIP system by Michal Kapala. · 3b2f05c8
      Rene Brun authored
      Extract from a mail from Michal:
      
      "I'm a summer student working for the openlab project (supervisor: Sverre
      Jarp). I've been optimizing ROOT on an Itanium platform for a few weeks
      and I've found a bug, which, if it's accepted and corrected, will
      influence all the ROOT benchmarks that have been done by now.
      
      What I've found is that all zip/*.c files use the variable 'level' which
      is declared in ZIP.h as a static (local) variable. The result is that
      calling R__zip(1, ...) (as in the 'stress' program) changes the 'level'
      variable only in the Bits.c file, but its value remains default ( == 6) in
      the others. The resulting compressed file is still correct, but the
      drawback is quite huge slow-down of the programs like 'stress' which use
      data compression intensively.
      
      Some facts: by putting the files: Tree.c, Bits.c and Deflate.c in a single
      C file (with some small changes), what makes the 'level' variable appear
      in only one copy, I've increased the result of the 'stress' benchmarking
      program from 584 RM to 701 RM -- 20% improvement. At the same time, the
      size of all compressed data increased by about 1.7%. (Itanium 1000 MHz
      machine, Intel compiler v8.0 with -O3 and profile-guided optimization).
      
      The bug is really easy to fix, but it will require all the benchmarks done
      in the past to be repeated or rescaled..."
      
      
      git-svn-id: http://root.cern.ch/svn/root/trunk@7132 27541ba8-7e3a-0410-8455-c3a389f83636
      3b2f05c8
  18. Dec 02, 2002
  19. Jul 18, 2002
  20. Feb 06, 2002
  21. Nov 19, 2001
  22. Nov 13, 2001
  23. May 16, 2000
Loading