Re: [easybuild] next EasyBuild conf call: Wed 15 Feb 2023 at 16:00 UTC (17:00 CET)
Hi all, Notes of today's conf call are available at https://github.com/easybuilders/easybuild/wiki/Conference-call-notes-20230215 . Next call is scheduled for Wed 1 March 2023 at 09:00 UTC, see also https://github.com/easybuilders/easybuild/wiki/Conference-calls regards, Kenneth On 14/02/2023 20:42, Kenneth Hoste wrote: Dear EasyBuilders, The next EasyBuild conf call is planned for Wed 15 Feb 2023 at 16:00 UTC (17:00 CET). You can join the conf call via https://tiny.cc/eb_conf_call . Current agenda: * overview of recent developments + outlook to next EasyBuild release; * Q Suggestions for additional topics are always welcome! More information on the EasyBuild conf calls is available at https://github.com/easybuilders/easybuild/wiki/Conference-calls . regards, Kenneth
Re: [easybuild] imkl/2021.2.0-gompi-2021a fails with 'illegal instruction' on AMD
Hi Loris, I'am not an easybuild expert but based on some experiences, I would say that such an error could appear when used GCC was built on a host with different architecture (with different processor flags) than the current target architecture (AMD EPYC). I would therefore check that the machine on which this GCC was built is indeed an AMD EPYC : normally, this is the GCC coming from gompi-2021a, so installed with EasyBuild, therefore you may have a look at the 1st line of the .log file in the GCC installation, in the "easybuild" directory, to have the hostname (and then you will have the architecture) of the host where GCC was installed. Maybe il will be also interesting to inevestigate if the GCC used in this postprocessing step is really the GCC coming from gompi-2021a. But if everything is good on these sides, then the problem is probably elsewhere. Good luck, Dan -- Dan-Gabriel CALUGARU IR en Calcul Scientifique (CNRS) Dr de Mathématiques et Applications Laboratoire de Mécanique des Fluides et d'Acoustique UMR 5509 CNRS - ECL - UCBL - INSA Lyon - Univ. de Lyon Bâtiment I11 - bureau 11098 ECOLE CENTRALE de LYON 36, avenue Guy de Collongue 69134 ECULLY tel: +33 (0)4 72 18 61 73 Le 15/02/2023 à 15:01, Loris Bennett a écrit : Hi, I tried to build imkl/2021.2.0-gompi-2021a on a node with an AMD EPYC 7713 CPU and got the following error in the postprocessing step: == 2023-02-15 14:55:46,431 run.py:236 INFO running cmd: make -f makefile libintel64 compiler=gnu PRECISION=MKL_DOUBLE == 2023-02-15 14:55:46,578 build_log.py:171 ERROR EasyBuild crashed with an error (at easybuild/software/EasyBuild/4.7.0/lib/python3.6/site-packages/easybuild/base/exceptions.py:126 in __init__): cmd "make -f makefile libintel64 compiler=gnu PRECISION=MKL_DOUBLE" exited with exit code 2 and output: make -f makefile lib _IA=intel64 make[1]: Entering directory `/trinity/shared/easybuild/arch/x86_64/amd/zen3/software/imkl/2021.2.0-gompi-2021a/mkl/2021.2.0/interfaces/fftw2xc' mkdir -p /trinity/shared/easybuild/build/imkl/2021.2.0/gompi-2021a/tmpx4or7epu/obj_double_gnu gcc -m64 -Wall -Werror-c \ -DMKL_DOUBLE -I../../include -I../../include/fftw \ ../../interfaces/fftw2xc/wrappers/fftw.c -o /trinity/shared/easybuild/build/imkl/2021.2.0/gompi-2021a/tmpx4or7epu/obj_double_gnu/fftw.o Assembler messages: Internal error (Illegal instruction). Please report this bug. make[1]: *** [/trinity/shared/easybuild/build/imkl/2021.2.0/gompi-2021a/tmpx4or7epu/obj_double_gnu/fftw.o] Error 1 make[1]: Leaving directory `/trinity/shared/easybuild/arch/x86_64/amd/zen3/software/imkl/2021.2.0-gompi-2021a/mkl/2021.2.0/interfaces/fftw2xc' make: *** [libintel64] Error 2 (at easybuild/software/EasyBuild/4.7.0/lib/python3.6/site-packages/easybuild/tools/run.py:643 in parse_cmd_output) Should this work or is this the result of some incompatibility between Intel and AMD? Cheers, Loris
Re: [easybuild] Which toolchain to use on AMD EPYC 9004 "Genoa"?
On Wed, 15 Feb 2023 15:26:26 +0100 Ole Holm Nielsen wrote: > Is "foss" also the preferred toolchain on AMD Rome and Genoa? For now, yes. There's some work going on to create a toolchain around amd compilers but it's questionable how much you gain as all features these compilers and libs bring eventually end up in upstream llvm and gcc. So for Genoa look for upcoming foss toolchain built around gcc 13. For Rome and Milan existing ones with gcc 11 and 12 are already working fine. -- Jurij Pečar HPC Engineer, IT Operations, IT Services EMBL Heidelberg, Meyerhofstraße 1, 69117, Heidelberg, Germany Room 13-401
[easybuild] Which toolchain to use on AMD EPYC 9004 "Genoa"?
In an upcoming procurement we plan to use a GPAW EasyBuild module for benchmarking. On 4th Gen Intel Xeon Scalable processors we should presumably specify a recent "intel" toolchain because of support for the latest instructions. Question: Which toolchain to use on AMD EPYC 9004 "Genoa"? There are a number of sites out there with AMD clusters, and I wonder which optimal and reliable toolchains people use with AMD Rome and Genoa? We have often used the "foss" toolchain on Intel Xeon because it provides OpenMPI, OpenBLAS, ScaLAPACK, and FFTW modules for the GPAW code. Is "foss" also the preferred toolchain on AMD Rome and Genoa? Looking at the long toolchain list in https://docs.easybuild.io/version-specific/toolchains/ I'm not much wiser :-( None of these toolchains seems to use the AMD AOCC compiler, and I wonder why? Thanks for sharing any advice and experiences! Best regards, Ole -- Ole Holm Nielsen PhD, Senior HPC Officer Department of Physics, Technical University of Denmark
[easybuild] imkl/2021.2.0-gompi-2021a fails with 'illegal instruction' on AMD
Hi, I tried to build imkl/2021.2.0-gompi-2021a on a node with an AMD EPYC 7713 CPU and got the following error in the postprocessing step: == 2023-02-15 14:55:46,431 run.py:236 INFO running cmd: make -f makefile libintel64 compiler=gnu PRECISION=MKL_DOUBLE == 2023-02-15 14:55:46,578 build_log.py:171 ERROR EasyBuild crashed with an error (at easybuild/software/EasyBuild/4.7.0/lib/python3.6/site-packages/easybuild/base/exceptions.py:126 in __init__): cmd "make -f makefile libintel64 compiler=gnu PRECISION=MKL_DOUBLE" exited with exit code 2 and output: make -f makefile lib _IA=intel64 make[1]: Entering directory `/trinity/shared/easybuild/arch/x86_64/amd/zen3/software/imkl/2021.2.0-gompi-2021a/mkl/2021.2.0/interfaces/fftw2xc' mkdir -p /trinity/shared/easybuild/build/imkl/2021.2.0/gompi-2021a/tmpx4or7epu/obj_double_gnu gcc -m64 -Wall -Werror-c \ -DMKL_DOUBLE -I../../include -I../../include/fftw \ ../../interfaces/fftw2xc/wrappers/fftw.c -o /trinity/shared/easybuild/build/imkl/2021.2.0/gompi-2021a/tmpx4or7epu/obj_double_gnu/fftw.o Assembler messages: Internal error (Illegal instruction). Please report this bug. make[1]: *** [/trinity/shared/easybuild/build/imkl/2021.2.0/gompi-2021a/tmpx4or7epu/obj_double_gnu/fftw.o] Error 1 make[1]: Leaving directory `/trinity/shared/easybuild/arch/x86_64/amd/zen3/software/imkl/2021.2.0-gompi-2021a/mkl/2021.2.0/interfaces/fftw2xc' make: *** [libintel64] Error 2 (at easybuild/software/EasyBuild/4.7.0/lib/python3.6/site-packages/easybuild/tools/run.py:643 in parse_cmd_output) Should this work or is this the result of some incompatibility between Intel and AMD? Cheers, Loris -- Dr. Loris Bennett (Herr/Mr) ZEDAT, Freie Universität Berlin