Knl
Notes on using Strumpack within CCTBX Here are the documented results of using Strumpack on a single node for a variety of data set sizes (StrumpackSolverMPI_1K,StrumpackSolverMPI_5K,StrumpackSolverMPI_10K). All tests were performed on dials.lbl.gov, and allow the tests to be repeated at the user’s discretion. Example matrices for a variety of different refinement parameters are listed in the given paths, and the times represent a single solution.
Setting up and running STRUMPACK To build STRUMPACK alongside a conda cctbx.
STRUMPACK vs EIGEN performance The goal of this notebook is to allow the documentation of STRUMPACK vs EIGEN performance to be maintained in a single accessible location. The environment within which this notebook is run follows the standard cctbx conda build instructions available here. For this instance, we are using the STRUMPACK-enabled build of cctbx located at ExaFEL:cctbx_project(str_merge). STRUMPACK is currently built using the installation script STRUMPACK_installer_shared.sh, and if the installation takes place within the same directory as moddules and build, the cctbx build process can make use of it as a backend.
STRUMPACK vs EIGEN performance The goal of this notebook is to allow the documentation of STRUMPACK vs EIGEN performance to be maintained in a single accessible location. The environment within which this notebook is run follows the standard cctbx conda build instructions available here. For this instance, we are using the STRUMPACK-enabled build of cctbx located at ExaFEL:cctbx_project(str_merge). STRUMPACK is currently built using the installation script STRUMPACK_installer_shared.sh, and if the installation takes place within the same directory as moddules and build, the cctbx build process can make use of it as a backend.
STRUMPACK vs EIGEN performance The goal of this notebook is to allow the documentation of STRUMPACK vs EIGEN performance to be maintained in a single accessible location. The environment within which this notebook is run follows the standard cctbx conda build instructions available here. For this instance, we are using the STRUMPACK-enabled build of cctbx located at ExaFEL:cctbx_project(str_merge). STRUMPACK is currently built using the installation script STRUMPACK_installer_shared.sh, and if the installation takes place within the same directory as moddules and build, the cctbx build process can make use of it as a backend.
Performance tests of STRUMPACK using OpenMP and MPI on Cori We make use of the linear solver backends within cctbx.xfel, and evaluate their performance on a variety of sample A matrices and b vectors for 1k, 5k, 10k, and 32k images each, and for a variety of different refinement parameters. The time to solve a single sparse system will offer insight into the time spent during the Levenberg-Marquardt minimisation process.
STRUMPACK vs EIGEN performance The goal of this notebook is to allow the documentation of STRUMPACK vs EIGEN performance to be maintained in a single accessible location. The environment within which this notebook is run follows the standard cctbx conda build instructions available here. For this instance, we are using the STRUMPACK-enabled build of cctbx located at ExaFEL:cctbx_project(str_merge). STRUMPACK is currently built using the installation script STRUMPACK_installer_shared.sh, and if the installation takes place within the same directory as moddules and build, the cctbx build process can make use of it as a backend.
Exascale FEL crystallography