[Bug c++/14809] performance depends on allocation methods

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-04-01 15:59 -------
I can't reproduce this either, but it took me a while until I understood 
Falk's comment. The point is: you call malloc/new only once, so the run 
time of these functions must necessarily be negligible and can't account 
for the 13 second slowdown you observe. So it must be the output of these 
functions (for example the alignment of the data region returned), which 
however is not under gcc control. 
 
Since none of us can reproduce this, I guess we should close this PR. 
 
For the record, here are the results for running the four testcases: 
g/x> /home/bangerth/bin/gcc-3.4-pre/bin/c++ -O2 a.cc -o a 
g/x> /home/bangerth/bin/gcc-3.4-pre/bin/c++ -O2 b.cc -o b 
g/x> /home/bangerth/bin/gcc-3.4-pre/bin/c++ -O2 c.cc -o c 
g/x> /home/bangerth/bin/gcc-3.4-pre/bin/c++ -O2 d.cc -o d 
g/x> time ./a ; time ./b ; time ./c ; time ./d 
0 
 
real    0m19.101s 
user    0m18.792s 
sys     0m0.016s 
0 
 
real    0m19.340s 
user    0m19.092s 
sys     0m0.019s 
0 
 
real    0m19.135s 
user    0m18.869s 
sys     0m0.014s 
0 
[1]+  Done                    emacs -fn 9x15 b.cc 
 
real    0m20.012s 
user    0m19.669s 
sys     0m0.164s 
 
W. 

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14809