Re: [AMBER] PHI Intel Compilation

From: Jacopo Sgrignani <sgrigna.gmail.com>
Date: Thu, 4 Sep 2014 13:08:05 +0200

Hi,
I finally compiled pmemd.mic_native.MPI but when I run a calculation
I get this error:

./pmemd.mic_native.MPI: symbol lookup error: ./pmemd.mic_native.MPI:
undefined symbol: __intel_new_feature_proc_init


any advice to solve the problem?


I also tried to compile the offload version but, there were problems with
netcdf configuring.


checking whether we are cross compiling... configure: error: in
`/home/jacopo/amber14/AmberTools/src/netcdf-4.3.0':

configure: error: cannot run C compiled programs.


Thanks


Jacopo


2014-09-02 19:34 GMT+02:00 Ross Walker <ross.rosswalker.co.uk>:

> Hi Jacopo,
>
> No you have to use Intel MPI - The Xeon Phi cards only work with Intel's
> compilers and Intel's MPI.
>
> All the best
> Ross
>
>
> On 9/2/14, 10:16 AM, "Jacopo Sgrignani" <sgrigna.gmail.com> wrote:
>
> >Dear Ross
> >Thanks for your reply.
> >Do I have to use intel-mpi or can I Also use openmpi ?
> >
> >Thanks
> >
> >Jacopo
> >
> >> Il giorno 02/set/2014, alle ore 19:11, Ross Walker
> >><ross.rosswalker.co.uk> ha scritto:
> >>
> >> Hi Jacopo,
> >>
> >> You are using the serial version of the code - this only runs on 1 out
> >>of
> >> the 61 cores on a Xeon Phi - hence why it is so slow.
> >>
> >> You need to also build the parallel (MPI) version and the run that with
> >> mpirun specifying the number of cores on the Xeon Phi model you have.
> >>
> >> All the best
> >> Ross
> >>
> >>
> >>
> >>> On 9/2/14, 6:40 AM, "Jacopo Sgrignani" <sgrigna.gmail.com> wrote:
> >>>
> >>> Dear all
> >>> I compiled pmemd for a PHI intel accelerator in the native mode.
> >>>
> >>> To run the calculation I use this command: ssh mic0 "export
> >>> LD_LIBRARY_PATH=/tmp/; cd /tmp; ./pmemd.mic_native -O -i md.in -p
> >>> letroamber2.prmtop -c letroamber2.inpcrd -o prova.out² &
> >>>
> >>> The simulation starts but it is extremely slow 0.05 ns/day.
> >>>
> >>> I¹m new in using this hardware, so I would like to ask some advices for
> >>> speeding up my simulation.
> >>> Do I need intel_mpi libraries also for running in native mode?
> >>> Is my command line correct?
> >>>
> >>> Thanks in advance
> >>>
> >>> Jacopo
> >>> _______________________________________________
> >>> AMBER mailing list
> >>> AMBER.ambermd.org
> >>> http://lists.ambermd.org/mailman/listinfo/amber
> >>
> >>
> >>
> >> _______________________________________________
> >> AMBER mailing list
> >> AMBER.ambermd.org
> >> http://lists.ambermd.org/mailman/listinfo/amber
> >
> >_______________________________________________
> >AMBER mailing list
> >AMBER.ambermd.org
> >http://lists.ambermd.org/mailman/listinfo/amber
>
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Sep 04 2014 - 04:30:03 PDT
Custom Search