From: Francis T. <fx...@ma...> - 2012-01-11 08:06:32
|
nice! thanks for sharing this info kevin. fxt On Jan 10, 2012, at 12:44 PM, Kevin Moore wrote: > Success! > > I just got MESA compiled and ran through inlist_first_thing_to_try and everything looks to be working fine! It took 59s of cpu time and 13s of wall time to run. > > Here's what I had to do to get it working on Triton if anyone else is interested: > > Add the following lines to my .bashrc file: > module purge > module load intel > module load openmpi_mx > export MX_RCACHE=0 > export LD_LIBRARY_PATH=/opt/intel/Compiler/11.1/072/lib/intel64:$LD_LIBRARY_PATH > > Comment out these lines in mesa/adipls/test/ck: > scan-amdl.d amdl.mesa >> .list > check_okay > > For some reason, the adipls test was working fine, but there was a weird error when the script called scan-amd.d and this was dumped to the diff.txt file: > forrtl: severe (47): write to READONLY file, unit 5, file stdin > Image PC Routine Line Source > > scan-amdl.d.x 000000000047B8ED Unknown Unknown Unknown > scan-amdl.d.x 000000000047A3F5 Unknown Unknown Unknown > scan-amdl.d.x 00000000004348C0 Unknown Unknown Unknown > scan-amdl.d.x 00000000004063BF Unknown Unknown Unknown > scan-amdl.d.x 0000000000405BF2 Unknown Unknown Unknown > scan-amdl.d.x 000000000042E755 Unknown Unknown Unknown > scan-amdl.d.x 0000000000402B6F Unknown Unknown Unknown > scan-amdl.d.x 0000000000402ACC Unknown Unknown Unknown > libc.so.6 00000038F5C1D994 Unknown Unknown Unknown > scan-amdl.d.x 00000000004029D9 Unknown Unknown Unknown > > I couldn't track down why this was happening, but commenting out those lines in the test script fixed things. > > Then just ./install in the mesa directory as usual (Triton thankfully has ifort 11.1). I haven't tried linking with Triton's MKL libraries yet so I just used the default MESA blas & lapack libraries. > > -Kevin > > On Jan 7, 2012, at 6:52 PM, Kevin Moore wrote: > >> Yes, the purpose of doing this was to do parameter surveys as well as have a centralized location where students in our stars class could run MESA. We get free time on a variety of computing centers so this is a simple solution for us. >> >> Anyway, just an update - I actually got the people at Triton to compile a newer glibc (2.14) for me and now the installation gets all the way to the mtx module (yay?), giving the error: >> /home/kmoore-ucsb/mesa_3794/utils/mesasdk/lib/liblapack.a(xerbla.o): In function `xerbla_': >> xerbla.f:(.text+0x66): undefined reference to `_gfortran_transfer_character_write' >> xerbla.f:(.text+0x76): undefined reference to `_gfortran_transfer_integer_write' >> collect2: ld returned 1 exit status >> make: *** [tester] Error 1 >> >> This is enough to convince me I should probably be compiling it from source directly so will try that and post again if it succeeds/fails. >> >> Thanks for all the advice! >> >> -Kevin >> >> On Jan 5, 2012, at 2:51 PM, Richard Townsend wrote: >> >>> Agreed - and in fact I've done these sorts of calculations myself, using Bill's earlier EZ code. >>> >>> But in this context (and based on Ben's remarks about the need for specialized compilers) I was assuming 'supercomputer' means something with specialized hardware, such as a cluster linked together with Infiniband interconnects. While such systems can certainly be used for parameter surveys, a simple Ethernet-networked computer cluster can offer the same performance at a fraction of the price. >>> >>> Cheers, >>> >>> Rich >>> >>> On Jan 5, 2012, at 7:55 AM, Francis Timmes <fx...@ma...> wrote: >>> >>>> hi rich, >>>> >>>> there is lots to be gained by running multiple jobs simultaneously >>>> on a cluster; mesa (or any other non-mpi code) in embarrassingly parallel mode. >>>> for example, a parameter survey involving a 100 stars. >>>> >>>> fxt >>>> >>>> >>>> >>>> >>>> On Jan 5, 2012, at 2:41 AM, Richard Townsend wrote: >>>> >>>>> I'm not even sure what is to be gained from running MESA on a supercomputer. Parallelism in MESA is only via OpenMP - there is no MPIA usage. As such, MESA can take advantage of shared memory architectures, but not clustered architectures. >>>>> >>>>> Cheers, >>>>> >>>>> Rich >>>>> >>>>> On Jan 5, 2012, at 7:31 AM, Francis Timmes <fx...@ma...> wrote: >>>>> >>>>>> hi ben, rich, kevin (and the mesa community), >>>>>> >>>>>> i had no luck compiling mesa under pgi on the nersc >>>>>> machines hopper and franklin this past summer. the issue >>>>>> boiled down to certain pointer structures used in mesa >>>>>> not being supported in that version of the pgi compiler. >>>>>> >>>>>> i would be interested in hearing from anyone about >>>>>> successes or failures on running mesa on any supercomputer >>>>>> or largish parallel cluster. >>>>>> >>>>>> fxt >>>>>> >>>>>> >>>>>> >>>>>> >>>>>> On Jan 4, 2012, at 3:56 PM, Ben Brown wrote: >>>>>> >>>>>>> Kevin, >>>>>>> To follow up on Rich's note: you are almost certainly not going to be >>>>>>> able to use the MESA SDK on a supercomputer. From my experiences with >>>>>>> those systems (using a very different code), they tend to require highly >>>>>>> specialized libraries and compilers. A general purpose tool like the SDK >>>>>>> won't work, because often the libraries being referenced need to be >>>>>>> compiled against very specific hardware. >>>>>>> >>>>>>> Put differently, the login nodes and the compute nodes tend to run very >>>>>>> different operating systems and codes targeting the compute nodes need to >>>>>>> be compiled using the specific compilers supported by UCSD for Triton. >>>>>>> Which means you'll need to come up with a specialized solution for this >>>>>>> system. I don't have access to Triton myself, so I can't be much more >>>>>>> specific than that. Looking over the Triton webpages it looks like both >>>>>>> PGI and Intel compilers are available on the system: >>>>>>> >>>>>>> http://tritonresource.sdsc.edu/compile_jobs.php >>>>>>> >>>>>>> If you run into problems, you might try contacting Theodore Sande >>>>>>> (formerly at MIT) for help in getting a version of MESA to compile, as he >>>>>>> has had the code running on large-scale systems in the past. >>>>>>> >>>>>>> Otherwise I could take a look at Ranger or Kraken (supercomputers at TACC >>>>>>> and NICS; both have Intel and PGI compilers) later this week or next week. >>>>>>> They won't be a perfect match, but they may get you started. >>>>>>> --Ben >>>>>>> >>>>>>> >>>>>>>> >>>>>>>> Hi Kevin -- >>>>>>>> >>>>>>>> This is a common problem with systems using 'older' (pre-2.11) versions of the GNU C library (glibc), which is a standard (and significant) part of all Linux distributions. At the moment, I have no plans to support pre-2.11 glibc versions, since 2.11 has been out for over two years. So, the only workaround would be to upgrade to 2.11 or later. >>>>>>>> >>>>>>>> cheers, >>>>>>>> >>>>>>>> Rich >>>>>>>> >>>>>>>> On Jan 4, 2012, at 7:23 PM, Kevin Moore wrote: >>>>>>>> >>>>>>>>> Hi all, >>>>>>>>> >>>>>>>>> I'm trying to install MESA on the Triton computing cluster at UCSD and am getting some weird errors. I know next to nothing about this so am just following the instructions on Rich's website for using the MESA SDK (which has worked fine for my macbook many times). uname -m returns "x86_64" so I'm using the 64-bit linux version of the SDK. All the setup steps seem to work fine (no errors anyway), but I run into problems during the ./install step. >>>>>>>>> >>>>>>>>> The first error I encounter is: >>>>>>>>> gfortran: /lib64/libc.so.6: version `GLIBC_2.11' not found (required by gfortran) >>>>>>>>> make: *** [const_def.o] Error 1 >>>>>>>>> >>>>>>>>> Perhaps tellingly, there is not a file named libc.so.6 in my mesa/utils/mesasdk/lib64. >>>>>>>>> >>>>>>>>> In addition, now whenever I try and use the man command I get the error: >>>>>>>>> /usr/bin/gtbl: error while loading shared libraries: /home/kmoore-ucsb/mesa_3794/utils/mesasdk/lib64/libstdc++.so.6: ELF file OS ABI invalid >>>>>>>>> >>>>>>>>> Any ideas what's going on from someone who may have done this before? >>>>>>>>> >>>>>>>>> Thanks, >>>>>>>>> Kevin >>>>>>>>> >>>>>>>>> >>>>>>>>> >>>>>>>>> ------------------------------------------------------------------------------ >>>>>>>>> Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex >>>>>>>>> infrastructure or vast IT resources to deliver seamless, secure access to >>>>>>>>> virtual desktops. With this all-in-one solution, easily deploy virtual >>>>>>>>> desktops for less than the cost of PCs and save 60% on VDI infrastructure >>>>>>>>> costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox >>>>>>>>> _______________________________________________ >>>>>>>>> mesa-users mailing list >>>>>>>>> mes...@li... >>>>>>>>> https://lists.sourceforge.net/lists/listinfo/mesa-users >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> ------------------------------ >>>>>>>> >>>>>>>> ------------------------------------------------------------------------------ >>>>>>>> Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex >>>>>>>> infrastructure or vast IT resources to deliver seamless, secure access to >>>>>>>> virtual desktops. With this all-in-one solution, easily deploy virtual >>>>>>>> desktops for less than the cost of PCs and save 60% on VDI infrastructure >>>>>>>> costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox >>>>>>>> >>>>>>>> ------------------------------ >>>>>>>> >>>>>>>> _______________________________________________ >>>>>>>> mesa-users mailing list >>>>>>>> mes...@li... >>>>>>>> https://lists.sourceforge.net/lists/listinfo/mesa-users >>>>>>>> >>>>>>>> >>>>>>>> End of mesa-users Digest, Vol 29, Issue 2 >>>>>>>> ***************************************** >>>>>>>> >>>>>>> >>>>>>> ------------------------------------------------------------------------------ >>>>>>> Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex >>>>>>> infrastructure or vast IT resources to deliver seamless, secure access to >>>>>>> virtual desktops. With this all-in-one solution, easily deploy virtual >>>>>>> desktops for less than the cost of PCs and save 60% on VDI infrastructure >>>>>>> costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox >>>>>>> _______________________________________________ >>>>>>> mesa-users mailing list >>>>>>> mes...@li... >>>>>>> https://lists.sourceforge.net/lists/listinfo/mesa-users >>>>>> >>>>>> >>>>>> ------------------------------------------------------------------------------ >>>>>> Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex >>>>>> infrastructure or vast IT resources to deliver seamless, secure access to >>>>>> virtual desktops. With this all-in-one solution, easily deploy virtual >>>>>> desktops for less than the cost of PCs and save 60% on VDI infrastructure >>>>>> costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox >>>>>> _______________________________________________ >>>>>> mesa-users mailing list >>>>>> mes...@li... >>>>>> https://lists.sourceforge.net/lists/listinfo/mesa-users >>> >>> ------------------------------------------------------------------------------ >>> Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex >>> infrastructure or vast IT resources to deliver seamless, secure access to >>> virtual desktops. With this all-in-one solution, easily deploy virtual >>> desktops for less than the cost of PCs and save 60% on VDI infrastructure >>> costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox >>> _______________________________________________ >>> mesa-users mailing list >>> mes...@li... >>> https://lists.sourceforge.net/lists/listinfo/mesa-users >> >> >> ------------------------------------------------------------------------------ >> Ridiculously easy VDI. With Citrix VDI-in-a-Box, you don't need a complex >> infrastructure or vast IT resources to deliver seamless, secure access to >> virtual desktops. With this all-in-one solution, easily deploy virtual >> desktops for less than the cost of PCs and save 60% on VDI infrastructure >> costs. Try it free! http://p.sf.net/sfu/Citrix-VDIinabox >> _______________________________________________ >> mesa-users mailing list >> mes...@li... >> https://lists.sourceforge.net/lists/listinfo/mesa-users > > > ------------------------------------------------------------------------------ > Write once. Port to many. > Get the SDK and tools to simplify cross-platform app development. Create > new or port existing apps to sell to consumers worldwide. Explore the > Intel AppUpSM program developer opportunity. appdeveloper.intel.com/join > http://p.sf.net/sfu/intel-appdev > _______________________________________________ > mesa-users mailing list > mes...@li... > https://lists.sourceforge.net/lists/listinfo/mesa-users |