Description of problem: Impossble to install NWChem on Fedora 27 due to missing dependencies. Version-Release number of selected component (if applicable): How reproducible: Try to install parallel NWChem Steps to Reproduce: sudo dnf install nwchem* Actual results: Erro: Problema 1: conflicting requests - nothing provides libga.so.0()(64bit)(mpich-x86_64) needed by nwchem-mpich-6.6.27746-32.fc27.x86_64 Problema 2: conflicting requests - nothing provides libga.so.0()(64bit)(openmpi-x86_64) needed by nwchem-openmpi-6.6.27746-32.fc27.x86_64
I'm having problems building a working version of nwchem on fedora 27. Neither 6.6 https://koji.fedoraproject.org/koji/taskinfo?taskID=23207023 or 6.8 work properly https://koji.fedoraproject.org/koji/taskinfo?taskID=23207119 I'm getting 0:0:ga_diag_std_seq: dsyev failed:: 0 on the simple output, for both openmpi and mpich module load mpi/openmpi-x86_64 echo -e "geometry\nH 0 0 0\nH 0 0 1\nend\nbasis\nH library STO-3G\nend\ntask dft energy" > test.nw && mpiexec -np 2 `which nwchem_openmpi` test.nw It may have something to do with the ga version 5.6.1 available in fedora 27. Maybe you want to investigate on nwchem forum, apparently a similar error has been reported in the past, and I can try to build nwchem with the bundled ga instead.
Here is a version built using internally bundled ga https://koji.fedoraproject.org/koji/taskinfo?taskID=23231588 - works for me.
Hi, Marcin, sorry for my delay. It is working for me too. Do you need any assistance, lets say, by contacting and reporting to upstream or performing more tests? Thank you for the quick fix.
There is a thing you could do: can you test whether scalapack features work properly with https://koji.fedoraproject.org/koji/taskinfo?taskID=23231588 This is a nwchem version built with the ga bundled by nwchem, but I'm not sure what and whether this ga has picked up any scalapack libraries during the build.
Created attachment 1356853 [details] Output
Created attachment 1356855 [details] Input
It looks like NWChem heavily uses Scalapack in the diagonalization part of PW calculations. Looks fine in my test. Please, take a look at the attachments.
Which part of the output proves scalapack is used?
You,re right, my mistake here (I was reading a part of a book that stated that scalapack was used in the PW calculations, but despite citing NWChem it does not affirm that ths is how NWChem does it). Let me contact Edoardo Apra
It would be better to search nwchem forum how to verify how to the scalapack functionality and ask there if no examples are available.
It is done. http://www.nwchem-sw.org/index.php/Special:AWCforum/st/id2488 Let's wait.
You can cross link bugzilla on the nwchem forum too.
Henrique/Marcin, What GA was used in this RPM that exhibits the eigensolver failure? If I read correctly, you are not using the bundle GA, but a something from a Fedora RPM, right?
(In reply to marcindulak from comment #2) > Here is a version built using internally bundled ga > https://koji.fedoraproject.org/koji/taskinfo?taskID=23231588 - works for me. Dear Edoardo, Marcin build a version that uses the bundled GA and that one is working, but he wants to test if Scalapack is working as expected too.
(In reply to Edoardo Apra from comment #13) > Henrique/Marcin, > What GA was used in this RPM that exhibits the eigensolver failure? > If I read correctly, you are not using the bundle GA, but a something from a > Fedora RPM, right? Since ga is not well maintained in Fedora/EPEL (see bug #1432661) I'm looking into using the ga bundled by nwchem, as https://bugzilla.redhat.com/show_bug.cgi?id=1514542#c4 describes. I don't know whether the failure to build a working nwchem executable against the ga 5.6.1 RPMS available currently in Fedora 27 is due to a mismatch between nwchem 6.6/6.8 and ga 5.6.1, or improper build of Fedora 27 ga 5.6.1 against scalapack. Since I would need to modify the build process of the bundled ga in nwchem.spec to use Fedora's scalapack/blas, the question is what is the set of nwchem tests that exercise scalapack. Here are the tests I'm executing using 2 cores during the nwchem RPM build process https://src.fedoraproject.org/cgit/rpms/nwchem.git/tree/nwchem.spec#n426 timeout --preserve-status --kill-after 10 1800 time ./doafewqmtests.mpi 2
Yes, some of the test included in doafewqmtests.mpi do stress scalapack. However, the following one uses large matrices (but it is not a long test) ./runtests.mpi.unix procs 2 dft_siosi3 By the way, the NWChem repository has just moved to github under https://github.com/nwchemgit/nwchem Cheers, Edo PS I have just updated my recipe to build NWChem 6.8 on Fedora 27 https://github.com/nwchemgit/nwchem/wiki/Compiling-NWChem#nwchem-68-on-centos-71fedora-27
I need a test that forcibly tests scalapack, and fails if scalapack is not available. Will nwchem perform any fallback if ga fails to discover scalapack during the build process?
Once the Global Arrays are built (in the src/tools directory), the NWChem build process checks if Global Arrays have detected the presence of Scalapack. If Global Arrays have failed to find usable a usable Scalapack library, NWChem uses the builtin Peigs library. Here is the line of src/config/makefile.h that checks for the Global Arrays detection of Scalapack _USE_SCALAPACK = $(shell cat ${NWCHEM_TOP}/src/tools/build/config.h | awk ' /HAVE_SCALAPACK\ 1/ {print "Y"}')
Is GA performing this switch dynamically, I mean I start nwchem and GA discovers there is a so missing/mismatch and switches to Peigs? Is there a way to see this happens somewhere in nwchem output?
You cannot see it in the output of a NWChem run at this time, only during the build phase. Would you like to have a print option that write this information at the beginning of a NWChem output?
Yes, and in addition, if scalapack is used some summary of blacs context.
Let me know if this is what you are looking for. (I am not sure I understand your question about BLACS, please provide more details about it) Job information --------------- hostname = lagrange program = /tmp/nwchem/bin/LINUX64/nwchem date = Mon Nov 27 19:44:56 2017 compiled = Mon_Nov_27_19:44:50_2017 source = /tmp/nwchem nwchem branch = Development nwchem revision = nwchem_on_git-7-gb15e696289a21bd0a3c2e2bda421a2539db39847 ga revision = nwchem_on_git-7-gb15e696 use scalapack = F input = nwchem.nw prefix = h2o. data base = ./h2o.db status = startup nproc = 1 time left = -1s
I mean nprow/npcol https://info.gwdg.de/wiki/doku.php?id=wiki:hpc:scalapack and also the block size http://www.netlib.org/utk/papers/scalapack/node19.html I imagine GA (?) will perform a heuristic choice of these based on the matrix size? But "use scalapack" is already good.
Here is the URL for the new nwchem-6.8 beta3 tarball. It contains the "use scalapack" print out showed earlier https://github.com/nwchemgit/nwchem/releases/tag/v6.8-beta.3
Marcin/Henrique NWChem 6.8 is now available for download https://github.com/nwchemgit/nwchem/releases/tag/v6.8-release
Still broken in f28 bug #1573918. Nwchem needs rebuilding.
The discussion about providing a flatpak has started in NWChem's GitHub. Maybe this is the best way to make the package easier to distribute for everyone.
Rebuilding the 6.6.27746 version results still in segfaults on Fedora 28 https://koji.fedoraproject.org/koji/taskinfo?taskID=27497158 and updating the spec to the latest nwchem 6.8 is blocked by https://bugzilla.redhat.com/show_bug.cgi?id=1432661
NWChem 6.6.27746: it's likely due that while the NWChem build adopts 32-bit integer for BLAS (BLAS_SIZE=4), the GA library from the Fedora28 rpm was compiled with BLAS_SIZE=8 (default value, confirmed after inspecting ga.spec in ga-5.6.1-2.fc28.src.rpm) Since the current NWChem 6.8 build is blocked by the lack of compatible GA RPMS, what prevents from using the bundles GA shipped with NWChem 6.8? By the way, there is a bunch of fixes released after NWChem 6.8 came available in the hotfix/release-6-8 branch (aka 6.8.1)
I've asked fedora-devel about the possibility of using external ga source for the nwchem build https://lists.fedoraproject.org/archives/list/devel@lists.fedoraproject.org/thread/KUYUMGYUM73TGXBCSYLBKDPN3IHXLLOW/#KUYUMGYUM73TGXBCSYLBKDPN3IHXLLOW
nwchem-6.8.1-4.fc28 has been submitted as an update to Fedora 28. https://bodhi.fedoraproject.org/updates/FEDORA-2018-dccd733c0a
nwchem-6.8.1-4.fc28 has been pushed to the Fedora 28 testing repository. If problems still persist, please make note of it in this bug report. See https://fedoraproject.org/wiki/QA:Updates_Testing for instructions on how to install test updates. You can provide feedback for this update here: https://bodhi.fedoraproject.org/updates/FEDORA-2018-dccd733c0a
nwchem-6.8.1-4.fc28 has been pushed to the Fedora 28 stable repository. If problems still persist, please make note of it in this bug report.