This repository contains artifacts and workflows to use GPTune (https://github.com/gptune/GPTune) with the Collective Knowledge (CK) (https://cknowledge.org/) technology.
- Python 3+
- Collective Knowledge Framework (CK) - https://cknowledge.org/
$ pip install ck --user
$ alias ck="python -m ck"
Once you have installed CK, you can install CK-GPTune using the following command:
$ ck pull repo --url=https://github.com/gptune/CK-GPTune
This will also install several required CK repositories/modules/tools automatically. The user can check $HOME/CK directory for the details.
There are also several patch files in CK-GPTune to supplement the original CK tools. Please use the following command:
$ cd $HOME/CK/CK-GPTune/
$ bash patch.sh
Before using any CK-based workflows in CK-GPTune, the user needs to detect the installed GPTune package with the following CK command.
$ ck detect soft:lib.gptune
This command will detect the GPTune software installation paths and prepare an executable environment by setting all the required environment variables, this is also called CK virtual environment.
You can check the detected software environment variables in $HOME/CK/local/env/.
CK-GPTune provides a number of reproducible GPTune tuning workflows with CK. We are currently offering four examples gptune-demo, ScaLAPACK-PDGEQRF, SuperLU_DIST-PDDRIVE, and Hypre-IJ. Please check out the program directory for the details.
The user can install (compile) and run these examples using simple command line interafaces. If there are software dependencies to compile/run, CK will detect the installation path and version of the required softwares. Defined software dependencies can be found at the meta.json file in the .cm directory of each program directory (e.g. $HOME/CK/CK-GPTune/program/ScaLAPACK-PDGEQRF/.cm/meta.json)
$ ck compile CK-GPTune:program:ScaLAPACK-PDGEQRF
CK-GPTune provides a command line interface to run CK-enabled GPTune workflows with the GPTune's MLA and history database features.
$ ck MLA gptune program:ScaLAPACK-PDGEQRF
The provided examples in CK-GPTune automatically creates tmp directory in their program directory. After finishing the run, the user may want to check stdout and stderr files. The generated history database files will be stored in the gptune.db directory in the tmp directory.
The user can also pass arguments as follows. If the argument values are not given by this command line, GPTune will use the default values defined in the program code.
$ ck MLA gptune --bench=ScaLAPACK-PDGEQRF --nruns=10
One nice thing of CK is software detection plugins which detect the location and the version of the software. The CK community already provides many useful software detection plugs through the ck-env repository ($HOME/CK/ck-env/soft). But, CK-GPTune also contains a number of software detection plugins that are not (yet) supported by the ck-env repository.
The user can check out soft directory in CK-GPTune.
We have tested CK-GPTune using standalone Linux machine, the Cori supercomputer at NERSC, and a Mac OS. Please let us know if you have any troubles with using CK-GPTune.
GPTune Copyright (c) 2019, The Regents of the University of California, through Lawrence Berkeley National Laboratory (subject to receipt of any required approvals from the U.S. Dept. of Energy) and the University of California, Berkeley. All rights reserved.
If you have questions about your rights to use or distribute this software, please contact Berkeley Lab's Intellectual Property Office at IPO@lbl.gov.
NOTICE. This Software was developed under funding from the U.S. Department of Energy and the U.S. Government consequently retains certain rights. As such, the U.S. Government has been granted for itself and others acting on its behalf a paid-up, nonexclusive, irrevocable, worldwide license in the Software to reproduce, distribute copies to the public, prepare derivative works, and perform publicly and display publicly, and to permit other to do so.