[gmx-users] icc

[Graham Smith]

I've investigated the new intel compiler a bit more, in particular
when it's used to compile the inner loops, to compare with gcc (2.96) 
and with the hand-coded assembler (because I have had to hack the inner
loop code generator for the hydrophobic potential I want, as Erik
suggested to me a month ago, so will have to use compiler-generated
code...)

The tests I've done are scarcely exhaustive and there are some
anomalous-seeming results, but I thought you might be interested to
see them anyway. I'm surprised that icc doesn't in general do better,
but, at least its best performance is where I'm going to need it...

There are 3 systems

2pol = E coli DNA Polymerase III beta subunit + water, 
55894 atom 17854 molecules, 500 x 2 fs steps, 
cutoffs 0.9 coul/1.4 vdw + PME, nstlist 5, const pT

5pti = BPTI + water, 
13905 atom 4503 molecules, 2000 x 2 fs steps
ES as 2pol 
(with perfect linear scaling of all algorithms and no cache
effects should take the same time as 2pol, within 1%) 

pcred = simple lipid model , no water, 3000 atoms 1000 molecules,
langevin dynamics, 10000 x 20 fs steps 
user-tabuled potential functions, cutoff 3.0, nstlist 1, const VT

for the lipid test I didn't use my hydrophobic potential but kept with
pairwise additive ones so I could still use the provided i386assembler. 

and there are two processors, 

p3 = 800 MHz PIII, 770Mb main mem, 256kb cache , Redhat 7.2 (2.4.9-31)
p4 = 2.0 GHz P4,   256Mb main mem, 512kb cache , Redhat 7.2 (2.4.9-31)

all code including fftw (though fftw's performance varied very
little) was recompiled on the p3 or p4 it was going to 
run on, after doing a make distclean first, with the flags described 
below

all runs are single-processor 

and there are 6 combinations of compilers and flags 

gcc_asm means just ./configure 
gcc_noasm means    ./configure --disable-x86-asm
icc_O2    means    setenv CC icc; setenv CFLAGS  -O2 
                   ./configure --disable-x86-asm
icc_opt   means    setenv CC icc; setenv CFLAGS " -O3 -axW -tpp7 " 
                   (p4 flags; " -O3 -axK " on p3)
                   ./configure --disable-x86-asm 
icc_asm   means    (same setenvs as icc_opt) ; ./configure 
icc_ifc   means    setenv CC icc; setenv CFLAGS  " -O3 -axW -tpp7 " 
                   setenv F77 ifc ; setenv FFLAGS  " -O3 -axW -tpp7 "         
                   (p4 flags; " -O3 -axK " on p3)
                   ./configure --disable-x86-asm --enable-fortran
(icc_ifc needed some hacking of the makefiles before it would compile)

I give the NODE time in seconds as reported in the .log files 

2pol

p3

gcc_asm        1000.8
gcc_noasm      1434.7
icc_asm         957.9
icc_ifp        1469.0
icc_opt        1863.5
icc_O2         1864.5

p4

gcc_asm         478.3
gcc_noasm       731.7
icc_asm         428.6
icc_ifp         706.2
icc_opt         678.2
icc_O2         1241.3

5pti 

p3

gcc_asm         759.6
gcc_noasm      1170.1
icc_asm         722.0
icc_ifp        1218.8
icc_opt        1528.2
icc_O2         1604.5

p4

gcc_asm         338.5
gcc_noasm       566.4
icc_asm         313.4
icc_ifp         574.0
icc_opt         531.7
icc_O2         1099.4

pcred

p3

gcc_asm         316.1
gcc_noasm       323.2
icc_asm         262.3
icc_ifp         270.1
icc_opt         268.5
icc_O2          276.8

p4

gcc_asm         157.7
gcc_noasm       181.8
icc_asm         104.9
icc_ifp          99.6
icc_opt          99.4
icc_O2          117.0


########################################################################

Dr. Graham R. Smith, 		       Department of Biological Sciences,  
Biomolecular Modelling Laboratory,     Biochemistry Building,              	
Cancer Research UK, 		       Imperial College of Science,        	
44 Lincoln's Inn Fields, 	        Technology & Medicine,             	
London WC2A 3PX,		       London SW7 2AZ,                     	
U.K. 				       U.K.                                	
Tel: +44-(0)20 7269 3348 	       Tel: +44-(0)20 7594 5737            	
email: graham.smith at cancer.org.uk	
URL: http://www.bmm.icnet.uk/~smithgr