Hi Bart, Recompiling GGCcore 11.3.0 did the trick.
Thanks! Loris Bart Oldeman <[email protected]> writes: > Hi Loris, > > there are some LAPACK tests that are very sensitive to floating point > rounding, in particular those involving complex numbers, > vectorization, and fused-multiply-addition (FMA). > Most of these have been addressed by GCC patches, but it depends a bit > on the compilation flags you are using for OpenBLAS, and if you use > `-march=native`, the CPU architecture you are using. > > If you haven't (re)compiled GCCcore 11.3.0 since October last year, > please do so first, and see if that fixes the issue. > Otherwise you can try compiling with > toolchainopts = {'vectorize': False} > in the easyconfig. > > Cheers, > Bart > > On Thu, 30 Mar 2023 at 05:30, Loris Bennett <[email protected]> > wrote: >> >> Hi, >> >> With OpenBLAS-0.3.20-GCC-11.3.0.eb I'm getting the following error: >> >> == FAILED: Installation ended unsuccessfully (build directory: >> /trinity/shared/easybuild/build/OpenBLAS/0.3.20/GCC-11.3.0): build >> failed (first 300 chars): Too many LAPACK tests failed due to numerical >> errors: 3471 (> 150) >> >> This error is mentioned here: >> >> https://github.com/easybuilders/easybuild-easyconfigs/pull/16961 >> >> However, I don't think I have an issue with hooks or with old >> easyconfigs in the working directory. >> >> Any ideas what else could be going wrong? >> >> Cheers, >> >> Loris >> >> -- >> Dr. Loris Bennett (Herr/Mr) >> ZEDAT, Freie Universität Berlin -- Dr. Loris Bennett (Herr/Mr) ZEDAT, Freie Universität Berlin
