Hi Ross,
Thank you for your reply. At the moment we are waiting for a site/ group license so that we can download, install and make Amber 14 available to our users. Actually Amber is only used by one research group in the university and I (the cluster administrator) am waiting for the group PI to purchase the license. One of PhD students in the group had been given that development version so that he could do some preliminary calculations/benchmarks in advance of the production version. Technically, I guess I should not really be experimenting with his development copy (apologies if that is the case), and from what you say I might be best to wait for the license and the production version anyway. In this instance I was using a very recent version (v15) of the Intel compilers, however as you say I might need to consider updating my GNU tools.
We have some PHIs installed in our cluster and we are keen to evaluate how to Amber 14 performs on these cards. Potentially our Amber users would be keen to make use of the PHI cards if the performance of the package is good. In a similar respect they made excellent use of our GPU nodes. If you could please supply a development version, suitable for PHI cards, to be used on a strictly private basis then that would be greatly appreciated. Of course, I would also need to be able to run Amber 14 on our conventional x86_64 hardware by way of comparison.
Best regards -- David.
-----Original Message-----
From: Ross Walker [mailto:ross.rosswalker.co.uk]
Sent: Monday, October 13, 2014 7:31 PM
To: AMBER Mailing List
Subject: Re: [AMBER] Building Amber 14 for Intel PHIs
Hi David,
Who did you get this version of AMBER from? - There are separate branches for Xeon Phi development so I would be surprised if you have the 'correct'
development version. Can I suggest trying the original AMBER 14 release version with the various update patches applied. There will be an additional update for Xeon Phi coming in about a month but that is not complete yet so a development version of that will not work right now.
Also note that this error is coming from the linker and has nothing to do with AMBER - looks like a problem with GNU binutils to me. I would suggest updating your system fully (yum update) and then also making sure you have the very latest version of the Intel Composer Suite.
All the best
Ross
On 10/13/14, 2:10 AM, "Baker D.J." <D.J.Baker.soton.ac.uk> wrote:
>Hello,
>
>I've been a development version of Amber 14, amber14_071014.tar.gz, so
>that I can do some performance experiments on our PHI cards. I have no
>problems building this version of Amber for our x86_64 nodes, however
>the PHI "make install" fails with what appears to be a bug. This is
>what I
>see:
>
>./configure -mic_native -noX11 intel
>make install
>
>......
>x86_64-k1om-linux-ld:
>/scratch/djb1/phi/amber14_071014/lib/libnetcdf.a(libdispatch_la-dcopy.o)(.
>text+0xe9f): unresolvable R_X86_64_PLT32 relocation against symbol
>`free..GLIBC_2.14'
>x86_64-k1om-linux-ld:
>/scratch/djb1/phi/amber14_071014/lib/libnetcdf.a(libdispatch_la-derror.
>o)(
>.text+0x16): unresolvable R_X86_64_PLT32 relocation against symbol
>`strerror..GLIBC_2.14'
>x86_64-k1om-linux-ld: BFD (GNU Binutils) 2.22.52.20120302 internal
>error, aborting at
>/sandbox/build/tmp/tmp/work/x86_64-nativesdk-mpsssdk-linux/binutils-cro
>ss-
>canadian-k1om-2.22+mpss3.4-1/binutils-2.22+mpss3.4/bfd/elf64-x86-64.c
>line 3382 in elf_x86_64_relocate_section
>
>x86_64-k1om-linux-ld: Please report this bug.
>
>make[3]: *** [/scratch/djb1/phi/amber14_071014/bin/pmemd.mic_native]
>Error 1
>make[3]: Leaving directory
>`/scratch/djb1/phi/amber14_071014/src/pmemd/src'
>make[2]: *** [mic] Error 2
>make[2]: Leaving directory `/scratch/djb1/phi/amber14_071014/src/pmemd'
>make[1]: *** [mic] Error 2
>make[1]: Leaving directory `/scratch/djb1/phi/amber14_071014/src'
>make: *** [install] Error 2
>
>I'm working on a RHELS 6.4 cluster and using a recent version, v15, of
>the Intel compilers and MKL v11.0_4. Does anyone please have any ideas
>what might be wrong here? I understand, of course, that this is a
>development version of Amber 14 and so it may not come with the latest
>bug fixes.
>
>Best regards - David.
>_______________________________________________
>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 Tue Oct 14 2014 - 02:00:02 PDT
