[gmx-users] G5 & G4 compiler options

Tyler Luchko tluchko at ualberta.ca
Sat Jun 25 00:36:09 CEST 2005


Hello,

I have been trying to compile GROMACS on my new dual 2.7 GHz G5 and  
1.67 GHz PowerBook but have been unable to reproduce the benchmark  
results for similar systems.  I have not tried the IBM compilers but  
I have tried both gcc 3.3 and 4.0 supplied with OS X 10.4.  I have  
listed the benchmark scores obtained with both below.  I have used  
the default options for compiling both fftw and GROMACS which result  
in the following compiler flags for all cases

fftw:

-O3 -fomit-frame-pointer -fno -schedule-insns -fschedule-insns2 - 
fstrict-aliasing -mcpu=powerpc

GROMACS:

-O6 -fomit-frame-pointer -finline-functions -Wall -Wno-unused -no-cpp- 
precomp -finline-limit=1000 -faltivec -funroll-all-loops -mcpu=7450 - 
mtune=970

I have tried other flags but with little success.

2.7 GHz G5 gcc-3.3

d.villin:  11426
d.lzm-cut:  3149
d.lzm-pme:  1467
d.poly-ch2: 3553
d.dppc:      150

2.7 GHz G5 gcc-4.0

d.villin:   9159
d.lzm-cut:  2737
d.lzm-pme:  1344
d.poly-ch2: 3183
d.dppc:      135

1.67 GHz G4 gcc-3.3

d.villin:   6689
d.lzm-cut:  1948
d.lzm-pme:   936
d.poly-ch2: 2131
d.dppc:       94

1.67 GHz G4 gcc-4.0
d.villin:   5737
d.lzm-cut:  1601
d.lzm-pme:   839
d.poly-ch2: 1838
d.dppc:       83

Can anyone suggest what  compiler options or running conditions may  
account for such low performance?

Thank you,

Tyler



More information about the gromacs.org_gmx-users mailing list