Re: Results for g++ 3.3 application testing on i686-pc-linux-gnu

[Paolo Carlini <pcarlini at unitus dot it>]

Peter Schmid wrote:

> OOPACK Version 1.7:
> Max=100000 Matrix=1000 Complex=100000 Iterator=100000
Hi! Interesting numbers indeed.
I wonder which kind of x86 cpu do you have, since it looks like 
something is going wrong for P4 (mine is 2.4 GHz, 533Mhz FSB) in the 
Complex test:

-O2
                        Seconds       Mflops
Test       Iterations     C    OOP     C    OOP  Ratio
----       ----------  -----------  -----------  -----
Max           1000000    2.9   2.8  346.0 353.4    1.0
Matrix          10000    2.9   3.1  862.1 811.7    1.1
Complex       1000000    8.2 107.3  -72.4  -5.5   13.2
Iterator      1000000    2.5   2.7  816.3 740.7    1.1

Compare to Intel6 (-O2):

                        Seconds       Mflops
Test       Iterations     C    OOP     C    OOP  Ratio
----       ----------  -----------  -----------  -----
Max           1000000    2.2   2.2  450.5 454.5    1.0
Matrix          10000    2.8   2.8  905.8 892.9    1.0
Complex       1000000    5.7   5.9  -103.5 -99.3    1.0
Iterator      1000000    2.2   2.1  917.4 939.0    1.0

The same problem occurs also for 3.2 or 2.95.3: on my P4 the Complex - 
OOP is always much slower than Complex - C (for the old PII this was not 
the case, and I obtained figures very similar to yours).

Is anyone able to reproduce either mine or yours behaviour???

Ciao, Paolo.

P.S. I didn't reallt investigate it but it seems that the negative 
numbers for the Mflops are just an artifact of the output routines, the 
times in seconds are 100% OK.