Ole Schütt
Ole Schütt
It would be nice to have support for [hipBLAS](https://github.com/ROCmSoftwarePlatform/hipBLAS) as this allows to test the hip code path on a Nvidia device. In return the support for rocBLAS could be...
The first half of #2202.
During yesterday's ELPA meeting it was suggested that we try its new redistribution feature. This could have several advantages: - we could remove our [own redistribution code](https://github.com/cp2k/cp2k/blob/master/src/fm/cp_fm_diag_utils.F) - hopefully resolve...
Currently, [m_max_memory](https://github.com/cp2k/cp2k/blob/master/src/base/machine.F#L93) is determined by taking the maximum across all calls to `m_memory()`. Therefore, it's noisy and we've got several calls to `m_memory()` [scatter](https://github.com/cp2k/cp2k/blob/20fe0099bce4444784cb4990292e1d9e993a51dd/src/mp2_ri_gpw.F#L417-L418) throughout the code base. Instead we...
When using Intel [oneAPI HPC kit](https://hub.docker.com/r/intel/oneapi-hpckit) I get [random timeouts](https://github.com/cp2k/cp2k/files/8651236/report.log) - most of which seem to happen during diagonalization: ``` /opt/cp2k/regtesting/TEST-local-psmp-2022-05-08_16-14-07/xTB/regtest-5/ef_stress4.inp.out forrtl: error (78): process killed (SIGTERM) Image PC Routine...
The toolchain currently builds [ELPA without OpenMP on Cray](https://github.com/cp2k/cp2k/blob/cdcb2793be4c6302478245ba22b1e8c038bd7787/tools/toolchain/scripts/stage5/install_elpa.sh#L84). While this is ok for the current XC40/50 generation, we'll eventually need OpenMP support for EPYC and ARM based machines. See...
I ran the [H2O-64.inp](https://github.com/cp2k/cp2k/blob/master/benchmarks/QS/H2O-64.inp) benchmarks using 4 MPI, 1 OpenMP, and 4 P100 GPUs on a dual socket EPYC node. Full timing report ``` ------------------------------------------------------------------------------- - - - T I...
Currently, the [cuda toolchain](https://github.com/cp2k/cp2k/blob/master/tools/toolchain/Dockerfile.cuda_mkl) does not include ELPA. I briefly tried to build ELPA with GPU support and failed. Anecdotal evidence suggests that it's not quite ready yet. Nevertheless, we...
I think there are currently two problems with the way SIRIUS reads pseudo potentials: 1. SIRIUS uses a custom json format while UPF files are [commonly](https://www.materialscloud.org/discover/sssp) distributed in the [XML...