Open eta000 opened 5 years ago
@eta000 Any updates?
@wpoely86 Any idea on this?
I am investigating. There are some more issues related to optimization.
I am ping-ponging with CP2K developers on psmp issues.
Et.
Fra: Kenneth Hoste notifications@github.com Svar til: easybuilders/easybuild-easyblocks reply@reply.github.com Dato: onsdag 08. mai 2019 10:00 Til: easybuilders/easybuild-easyblocks easybuild-easyblocks@noreply.github.com Kopi: Espen Tangen espen.tangen@uit.no, Mention mention@noreply.github.com Emne: Re: [easybuilders/easybuild-easyblocks] cp2k easyblock (#1714)
@eta000https://github.com/eta000 Any updates?
@wpoely86https://github.com/wpoely86 Any idea on this?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/easybuilders/easybuild-easyblocks/issues/1714#issuecomment-490389175, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ADTDD5UFWWFZ4TCJ7D4IAPLPUKCBZANCNFSM4HLPKOUQ.
I also had a lot of issue with the psmp build. If I build it (with foss/2018b
toolchain) by hand using EB dependencies, all tests but 1 pass. With the intel
toolchain it's a disaster. Their site says intel 18.0.1 should not work but 18.0.3. In my experience it's the other way around. Currently I'm avoid icc
/ifort
. With GCC it works much better. I've resorted to the gimkl
toolchain (GCC+MKL+Intel MPI). Still not all tests pass but the number of fails is pretty limited. I will try to open a PR with my changes to the easyblock.
Disaster is the right word indeed.
Et.
Fra: wpoely86 notifications@github.com Svar til: easybuilders/easybuild-easyblocks reply@reply.github.com Dato: onsdag 08. mai 2019 10:19 Til: easybuilders/easybuild-easyblocks easybuild-easyblocks@noreply.github.com Kopi: Espen Tangen espen.tangen@uit.no, Mention mention@noreply.github.com Emne: Re: [easybuilders/easybuild-easyblocks] cp2k easyblock (#1714)
I also had a lot of issue with the psmp build. If I build it (with foss/2018b toolchain) by hand using EB dependencies, all tests but 1 pass. With the intel toolchain it's a disaster. Their site says intel 18.0.1 should not work but 18.0.3. In my experience it's the other way around. Currently I'm avoid icc/ifort. With GCC it works much better. I've resorted to the gimkl toolchain (GCC+MKL+Intel MPI). Still not all tests pass but the number of fails is pretty limited. I will try to open a PR with my changes to the easyblock.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/easybuilders/easybuild-easyblocks/issues/1714#issuecomment-490394952, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ADTDD5V7O76GAJO6WA7TMNTPUKEJJANCNFSM4HLPKOUQ.
Hi all. There is something wrong with the cp2k setup causing psmp not to work. The main reason seems to be an incorrect hardcoded test for allowing threading in the src/pw/fft/fft_lib.F file in cp2k source. A workaround is to compile with FFTW libs linked before the mkl-libs in "arch" file. I am trying to discover where to manipulate this in the cp2k.py block with no luck so far.