RTune enables automated live analysis and tuning of application and system parameters amidst the simulation process to meet user-defined objectives. The RTune API enables users to program computing and computational objectives in a parallel iterative program, and the RTune runtime takes care of the measurement, modeling and analysis according to the defined objective during the computation.
-
Clone this repo
git clone https://github.com/passlab/rtune.git
-
Build and install the RTune library. You can build and install in the directory you want. Commands in the following are for building and installing in the
build
andinstall
folder under the source tree.cd rtune mkdir build install cd build cmake -DCMAKE_INSTALL_PREFIX=../install .. make make install
cd ../examples/jacobi-num_threads
make
export LD_LIBRARY_PATH=../../install/lib
./jacobi-rtune_num_threads
cd ../LULESH-num_threads
make
export LD_LIBRARY_PATH=../../install/lib
./LULESH-rtune_num_threads
Jacobi for automatically tuning computational precision according to the number of computing iteration
cd ../examples/jacobi-precision
make
export LD_LIBRARY_PATH=../../install/lib
./jacobi-rtune_precision
cd ../LULESH-boundary
make
export LD_LIBRARY_PATH=../../install/lib
./LULESH-boundary
Funding for this research and development was provided by the National Science Foundation under award No. 2001580 and 2015254.
The work is published in the follwing paper: ISPASS-2024
Yonghong Yan, Kewei Yan, and Anjia Wang, RTune: Towards Automated and Coordinated Optimization of Computing and Computational Objectives of Parallel Iterative Applications, ISPASS-2024, 2024 IEEE International Symposium on Performance Analysis of Systems and Software May 5-7, 2024, Indianapolis, Indiana