This repository contains regression tests for assessing the performance of FFTW, OpenBLAS, etc.. Currently, only FFTW and OpenBLAS have been added.
The regression tests contained in this repo can be run on bare metal, in a container, or in a container on AWS via OpenShift.
To run the OpenBLAS regression tests on bare metal, use the script run_benchmarks.sh provided in the OpenBLAS folder. Make sure to compile everything using the compile scripts.
To run the FFTW regression tests on bare metal, use the script run_benchmarks.sh provided in the FFTW folder. Make sure to compile everything using the compile scripts.
To run the OpenBLAS or FFTW regression tests in containers, use the provided Dockerfiles under OpenBLAS/Dockerfiles (for OpenBLAS) or FFTW/Dockerfiles (for FFTW). See the README.md files under the FFTW and OpenBLAS folders to see how to create the containers.
To run the regression tests on OpenShift in AWS, view the OpenBLAS/OpenShift folder (for OpenBLAS), FFTW/OpenShift (for FFTW), or TensorFlow/OpenShift (for TensorFlow). Everything is automated, and there is even a script under helper_scripts/OpenShift for creating a new node (MachineSet) in OpenShift in the event you want a specific instance type (e.g., m4.4xlarge, c5.large, etc.).
To use/enable NFD (Node Feature Discovery), use the script helper_scripts/OpenShift/nfd_setup.sh:
$ cd helper_scripts/OpenShift
$ sh nfd_setup.sh
To use/enable GPUs, use the script helper_scripts/OpenShift/gpu_setup.sh. Make sure you run the nfd_setup.sh script as well. For GPU setup,
$ cd helper_scripts/OpenShift
$ sh gpu_setup.sh
To use/enable CPU manager on a specific node for managing CPU resources, use the script helper_scripts/OpenShift/enable_cpumanager.sh. To see how it's used,
$ cd helper_scripts/OpenShift
$ sh enable_cpumanager.sh -h
Under the misc folder are miscellaneous files for doing various things. Currently, there are a few playbooks for installing gcc and glibc, along with their dependencies. Running these playbooks is optional, but doing so enables you to install whatever version of gcc (and glibc) you'd like, which may be useful if your current version of TensorFlow, etc. does not support your existing version of gcc (e.g., more recent versions of TensorFlow do not support gcc-4.8.x if trying to build with AVX-512 instructions).
The secrets folder contains secrets for various things. For example, you will need a secret to pull images from the redhat.io registry if you wish to use RHEL 8 s2i images.
You can read more information on this by reading the secrets/README.md file. Instructions on where/how to obtain these secrets files are listed in the README.
Note that there are no actual "secrets" files contained in this folder, only sample files to show the format of such files.