lammps install problem

please always copy the mailing list on your replies.

thanks axel for the informations given in your message. so, in reality my
cygwin is 32bit version. i have also tried to use the windows binary (both
32 and 64 bit)version on my windows professional 7 and after the
installation i have tested the installed lammps with some potential and it
work fine but when i have tested the COMB3 potential the run stop just in
the beginning which is not the case for COMB potential which work nicely
so i don't know what i should do

the *very first* step would have been to report this issue here on the
mailing list and also provide the input, so that it can be tested.
when new features are added to LAMMPS, there always is the chance that
something goes wrong. not reporting any issues in a way that somebody
can do something about it, only makes certain that the problem will
not get fixed until somebody else (by accident) stumbles across it.

i note that i have installed previous lammps version in my previuos cygwin

that is a useless statement. *which* previous version? LAMMPS keeps
changing incrementally all the time, so the differences can be quite
large and significant.

version (it was the 1.5 or 1.6 version) many times without any problems
so if any one of lammps user's has a suggestion please let me know

i already gave you a good suggestion that you should have applied in
any way, since it is flat out impossible to run as large a system on
your machine with a 32-bit binary that it would require 64-bit
integers.

axel.

as a point of reference. i just tried the in.comb3 example with the
latest windows build from here
http://git.icms.temple.edu/rpm/windows/32bit/
and it worked just fine. so if you have an input that still doesn't
work, please post it here, so we can check it out.

thanks,
    axel.

thank you very much axel
yes the latest windows bin executable work just fine.
i have installed the lammps on my 32 bit cygwin 2.8 version with mpi and fftw2.15 and i have tested the in.comb3 example and it workthe run is only on 1 proc
compared with windows version or provided linus log my run take 5 to 7 times much slower
so i don’t know if it is necessary to reinstall the 64bit cygwin version or there is an other solution
i have tried to use user-omp but no results ( the installation was not completed savely ).
best regards

thank you very much axel
yes the latest windows bin executable work just fine.
i have installed the lammps on my 32 bit cygwin 2.8 version with mpi and
fftw2.15 and i have tested the in.comb3 example and it workthe run is only
on 1 proc
compared with windows version or provided linus log my run take 5 to 7 times
much slower

it is not clear what is 5-7x slower than what here. please be complete
and specific in your descriptions. remember that nobody can see over
your should and see what exactly you are referring to.

so i don't know if it is necessary to reinstall the 64bit cygwin version or
there is an other solution

64-bit doesn't automatically mean that things get faster. one x86 the
move from 16-bit to 32-bit resulted in faster execution not because of
the "bitness", but because of changing from a complex and inefficient
segmented memory model to a flat memory model. moving to 64-bit should
make most things about 20% _slower_ (due to all pointers requiring
twice as much storage). it only doesn't happen on x86_64, because at
the same time with the switch to 64-bit instructions, also the number
of general purpose registers was doubled from the 4 to 8.
floating-point and SSE/AVX support are the same in both cases, anyway.

i have tried to use user-omp but no results ( the installation was not
completed savely ).

i don't understand what this is supposed to mean either. the windows
installers that i provide have an MPI based executable and a serial
executable, but with support for OpenMP. however, since there
currently is no comb3/omp style, it won't help you to use OpenMP with
these calculations. however, MPI support should just run fine.

can you please exactly state where the problem is.

axel.

Dear axel
i think that my problem was not clear for lammps mailing list user’s
so to simplified the problem:
i have tested the in.comb3 lammps input file with the comb3 forcefield file uploaded in the distributed package (both source file and windows installer)
the test on my cygwin take 127.8 s as time for 100 iterations
with windows version lmp_serial it take 28s for the same input file
the provided log under linux test show 18s as execution time.
this let me to suppose that it there is some thing which don’t work fine in my cygwin installation.

for the omp acceleration subject i’m very sorry. yes there is no support for the comb3 but only for comb style so it willn’t work with the comb3
best regards

Dear axel
i think that my problem was not clear for lammps mailing list user's
so to simplified the problem:
i have tested the in.comb3 lammps input file with the comb3 forcefield file
uploaded in the distributed package (both source file and windows installer)
the test on my cygwin take 127.8 s as time for 100 iterations
with windows version lmp_serial it take 28s for the same input file
the provided log under linux test show 18s as execution time.

performance depends on a variety of factors. hardware, compiler and
compiler optimization.
the latter can be quite significant and independent of the OS. on my
desktop this input completes in 58s with a binary using minimal
optimization settings (-O) whereas it completes in 22s with aggressive
optimization (too many flags to list) using the very same compiler.

this let me to suppose that it there is some thing which don't work fine in
my cygwin installation.

well, there is little that anybody but you can do about that. you may
need to try cranking up the optimization.

axel.

thanks axel
i will try to compile the source code as well as fftw and mpich with others optimization setting and perform new test
best regards