[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Bug#102193: gcc-3.0: compiled code with gcc 3.0 is slow and big



Anthony DeRobertis writes:
> 
> On Friday, January 4, 2002, at 08:17 , Morten Brix Pedersen wrote:
> 
> > mbp:~$ g++ benchmark.cpp ; ls -l a.out ; time a.out ; g++-3.0 
> > benchmark.cpp
> > ; ls -l a.out ; time a.out
> >
> Well, first, take . out of your path! I get:
> 
> -rwxr-xr-x    1 anthony  anthony     42840 Jan  5 23:46 a.out
> 
> real    0m7.870s
> user    0m7.210s
> sys     0m0.490s
> -rwxr-xr-x    1 anthony  anthony     37976 Jan  5 23:47 a.out
> 
> real    0m13.835s
> user    0m13.240s
> sys     0m0.420s
> 
> Ouch! Turning on the optimizer does help some, though:
> 
> anthony@bohr:morten$ g++ -O3 -march=i686 benchmark.cpp ; ls -l 
> a.out ; time ./a.out ; g++-3.0 -O3 -march=i686 benchmark.cpp; ls 
> -l a.out; time ./a.out
> -rwxr-xr-x    1 anthony  anthony     19173 Jan  5 23:50 a.out
> 
> real    0m3.146s
> user    0m2.640s
> sys     0m0.470s
> -rwxr-xr-x    1 anthony  anthony     19109 Jan  5 23:51 a.out
> 
> real    0m4.617s
> user    0m4.130s
> sys     0m0.410s
> anthony@bohr:morten$
> 
> Now only 1/3 slower. Still not good.

updated results for g++-3.1:

                  no opt         -O2               -O3  
-----------------------------------------------------------------
gcc-2.95: real    0m9.825s       real    0m4.322s  real    0m4.327s
          user    0m8.840s       user    0m3.700s  user    0m3.690s
          sys     0m0.540s       sys     0m0.600s  sys     0m0.590s

gcc-3.0:  real    0m16.070s      real    0m6.147s  real    0m6.079s
          user    0m15.710s      user    0m5.660s  user    0m5.700s
          sys     0m0.310s       sys     0m0.400s  sys     0m0.360s

gcc-3.1:  real    0m9.332s       real    0m7.815s  real    0m6.184s
          user    0m8.890s       user    0m7.420s  user    0m5.770s
          sys     0m0.370s       sys     0m0.380s  sys     0m0.390s



Worth forwarding the report, or just close it, because gcc-3.x comes
with a conforming libstdc++, and 2.95 not?


-- 
To UNSUBSCRIBE, email to debian-gcc-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org



Reply to: