-march=athlon-xp slower than -march=i686

[Dmitry Ivanov]

I have g77 3.2.1 (ALT Linux, build 3.2.1-alt2) as distributed 
with ALT Linux Master 2.2

CPU is Athlon XP 1700+
Program is 1000d.f (see attachment)

[dimss@bigiron ~/tmp]$ f77 -march=athlon-xp -static -O3 1000d.f 
[dimss@bigiron ~/tmp]$ strip a.out 
[dimss@bigiron ~/tmp]$ time ./a.out > /dev/null
5.98user 0.07system 0:06.08elapsed 99%CPU (0avgtext+0avgdata 
0maxresident)k
0inputs+0outputs (51major+1974minor)pagefaults 0swaps
[dimss@bigiron ~/tmp]$ f77 -march=i686 -static -O3 1000d.f 
[dimss@bigiron ~/tmp]$ strip a.out 
[dimss@bigiron ~/tmp]$ time ./a.out > /dev/null
5.29user 0.06system 0:05.43elapsed 98%CPU (0avgtext+0avgdata 
0maxresident)k
0inputs+0outputs (52major+1974minor)pagefaults 0swaps

[dimss@bigiron ~/tmp]$ f77 -march=athlon-xp -static -O0 1000d.f 
[dimss@bigiron ~/tmp]$ strip a.out 
[dimss@bigiron ~/tmp]$ time ./a.out > /dev/null
8.11user 0.05system 0:08.16elapsed 99%CPU (0avgtext+0avgdata 
0maxresident)k
0inputs+0outputs (53major+1974minor)pagefaults 0swaps
[dimss@bigiron ~/tmp]$ f77 -march=i686 -static -O0 1000d.f 
[dimss@bigiron ~/tmp]$ strip a.out 
[dimss@bigiron ~/tmp]$ time ./a.out > /dev/null
8.05user 0.05system 0:08.18elapsed 99%CPU (0avgtext+0avgdata 
0maxresident)k
0inputs+0outputs (50major+1974minor)pagefaults 0swaps
[dimss@bigiron ~/tmp]$ 


Athlon-optimised code runs slower with any -O. It seems that 
-mcpu=athlon-xp is useless.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 1000d.f
Type: text/x-fsrc
Size: 21245 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-bugs/attachments/20030409/d9482528/attachment.bin>