[gmx-users] gromacs 4.6.3 and Intel compiiler 11.x

Szilárd Páll szilard.pall at cbr.su.se
Wed Sep 4 01:40:23 CEST 2013


On Tue, Sep 3, 2013 at 9:50 PM, Guanglei Cui
<amber.mail.archive at gmail.com> wrote:
> Hi Mark,
>
> I agree with you and Justin, but let's just say there are things that are
> out of my control ;-) I just tried SSE2 and NONE. Both failed the
> regression check.

That's alarming, with GMX_CPU_ACCELERATION=None only the plain C
kernels get compiled which should pass the regressiontests even with
icc 11 - although I have not tried myself.

Some things that don't take much time and may be useful to try:
- make sure that when GMX_CPU_ACCELERATION=None the resulting binary
does not get compiled with flags that instruct the compiler to
auto-generate SSE4.1 code (e.g. -msse4.1 or -xHOST);
- run with the GMX_NOOPTIMIZEDKERNELS environment variable set which
disables the architecture-specific kernels at runtime (regardless of
what mdrun was compiled with); the end result should be the same as
above: the plain C kernels should be used (although with
GMX_CPU_ACCELERATION != None the instructions of requested instruction
set will be generated by the compiler as an optimization).

--
Szilárd

> I think I've spent enough time on this, which justifies
> escalating this to someone with the control, but is failing regression
> check with no CPU/instruction optimization normal?
>
> Regards,
> Guanglei
>
>
> On Tue, Sep 3, 2013 at 3:35 PM, Mark Abraham <mark.j.abraham at gmail.com>wrote:
>
>> On Tue, Sep 3, 2013 at 7:47 PM, Guanglei Cui
>> <amber.mail.archive at gmail.com> wrote:
>> > Dear GMX users,
>> >
>> > I'm attempting to compile gromacs 4.6.3 with an older Intel compiler (ver
>> > 11.x). Here is how I compiled FFTW,
>> >
>> > ./configure CC=icc F77=ifort CFLAGS="-O3 -gcc"
>> > --prefix=/tmp/gromacs-4.6.3/fftw-3.3.3/build-intel-threads
>> --enable-threads
>> > --enable-sse2 --with-combined-threads --with-our-malloc16 --enable-float
>>
>> I can't imagine you'll benefit from threaded FFTW, but feel free to try...
>>
>> > And, here is how I invoked cmake,
>> >
>> > CC=icc CXX=icpc ../cmake-2.8.11/bin/cmake .. -DGMX_FFT_LIBRARY=fftw3
>> >
>> -DFFTWF_LIBRARY=/tmp/gromacs-4.6.3/fftw-3.3.3/build-intel-threads/lib/libfftw3f.a
>> >
>> -DFFTWF_INCLUDE_DIR=/tmp/gromacs-4.6.3/fftw-3.3.3/build-intel-threads/include
>> > -DBUILD_SHARED_LIBS=no
>> >
>> -DCMAKE_INSTALL_PREFIX=/home/gc603449/APPLICATIONS/gromac-4.6.3/thread-float-static
>> > -DREGRESSIONTEST_PATH=/tmp/gromacs-4.6.3/regressiontests-4.6.1
>> >
>> > When I ran 'make check', 39 out of 142 kernel tests failed, and 2 out of
>> 9
>> > free energy tests failed.
>> >
>> > This is my first time to compile gromacs, which I am not very familiar
>> > with. I wonder if anyone can kindly point out what has gone wrong, and
>> > where to look for hints. Any help is much appreciated.
>>
>> As Justin said, get an up-to-date compiler. gcc regularly outperforms
>> icc, anyway. You can use cmake -DGMX_CPU_ACCELERATION=SSE2 to get a
>> least-common-denominator build, but if you care about threaded FFTW
>> then you care enough to get a new compiler!
>>
>> Mark
>>
>> > Best regards,
>> > --
>> > Guanglei Cui
>> > PS: I am aware of a warning about using older Intel compilers with 4.6.3,
>> > but that's the only quality compiler I got. The gcc compiler I have
>> access
>> > to is too old (ver 4.1.x)
>> > --
>> > gmx-users mailing list    gmx-users at gromacs.org
>> > http://lists.gromacs.org/mailman/listinfo/gmx-users
>> > * Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>> > * Please don't post (un)subscribe requests to the list. Use the
>> > www interface or send it to gmx-users-request at gromacs.org.
>> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>> --
>> gmx-users mailing list    gmx-users at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> * Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>> * Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-users-request at gromacs.org.
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>
>
>
> --
> Guanglei Cui
> --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists



More information about the gromacs.org_gmx-users mailing list