NAS_NPB_OpenACC_2.5

Code repository for the benchmarks discussed in our paper "Exploring translation of OpenMP to OpenACC 2.5: Lessons Learned"

Citation Information. AsHES 2017

Sergio Pino*, Lori Pollock, and Sunita Chandrasekaran. Exploring translation of OpenMP to OpenACC 2.5: Lessons learned. Proceedings of the Seventh International Workshop on Accelerators and Hybrid Exascale Systems (AsHES). 2017.

• There is a pre-print version in the following link.
• To download the paper from IEEE link.

@inproceedings{pino2017exploring,
 title={Exploring translation of OpenMP to OpenACC 2.5: Lessons Learned},
 author={Pino, Sergio and Pollock, Lori and Chandrasekaran, Sunita},
 booktitle={Proceedings of the Seventh International Workshop on Accelerators and Hybrid Exascale Systems (AsHES)},
 year={2017},
 organization={IEEE Press}
}

Cloning the repo

We use git-submodules to checkout the the individual benchmark suites into the global project. For more info in what this means at Git-Tools-Submodules. Thus, to successfully checkout all the required files to compile and run the benchmarks you need to do:

• You can clone recursively the repo:

$ git clone --recursive https://github.com/spino327/NAS_SHOC_OpenACC_2.5

or

• You can clone the repo as always and execute a couple more git commands:

$ git clone https://github.com/spino327/NAS_SHOC_OpenACC_2.5
$ cd NAS_SHOC_OpenACC_2.5
$ git submodule init
$ git submodule update

Make individual benchmarks

You can pass to the make command the following variables. Alternatively, you can export the env.

• CC: compiler to use.
• DEFINES: string to be passed to make (assuming you have a DEFINES variable within your makefile).
• TA: architecture for OpenACC (as the -ta flag for pgi). E.g. TA=multicore or TA=nvidia,cc35.
• EXTRA_CFLAGS: extra flags to pass to the c compiler.
• EXTRA_CLINKFLAGS: extra flags to pass to linker.
• PXM: Program execution model to use. E.g. PXM=acc, PXM=omp

NAS

Compiling OpenACC for multicore: make CC=pgcc TA=multicore CLASS=A. Compiling OpenACC for GPU: make CC=pgcc CLASS=A. Compiling OpenMP: make CC=pgcc CLASS=A.

SHOC

Compiling the OpenACC version for the multicore: make -f Makefile.acc CC=pgc++ TA=multicore. Compiling the OpenACC version for the GPU: make -f Makefile.acc CC=pgc++. Compiling the OpenMP version: make -f Makefile.omp CC=pgc++.

NPB-CUDA

Compile with make

NPB-OMP-C

Compile with make CC=pgcc CLASS=A

Using the benchmark execution scripts (ES)

To use the benchmark execution script (ES) you need to have a configuration file that describes the experiment to execute. In addition, you need to run the entry script execExperiments.sh with that configuration file.

We have provide several configuration file examples that compile the benchmarks using PGI. It is easy to make it work with other compilers.

1. The sample configuration files are at: scripts/conf_files.
2. ES are at: scripts/execExperiments.sh and scripts/singleExperiment.sh.
3. There are also scripts to plot the results scripts/plot_bar.py, scripts/plot_boxplot.py, scripts/plot_indbar.py, and scripts/plot_scatter.py.
4. There are also scripts to process raw benchmarks results to make it easier to comply with the format of the plotting scripts scripts/processResults.sh.

To execute an experiment using the ES. For instance to run the NAS-BT benchmark using OpenACC on a multicore:

$ cd script
$ ./execExperiments.sh scripts/conf_files/acc_multicore/BT.conf

The results will be placed at results/BT/BT_....

Compiler definitions

We provide Makefile definitions that guide the compiling process. You can read and modify these build system. For NAS, these files are at config/make.def and sys/make.common. For SHOC, these files are at crpl_conf/make.def and crpl_conf/make.common.