Decision_Support_for_GPU_Accelerator_dataset/README.md

The dataset used in "Design-Decision Support for GPU Acceleration by Predicting Energy Gain and Programming Effort" manuscript

final

The final datasets (csv files) used for building the estimation models in the Manuscript

synthetic-dataset

Generates a simple synthetic dataset

• generate.py: Usage python generate.py <num_of_programs> <output_directory> generates the datapoints (synthetic programs)
• run_dataset_script.py: Usage python run_dataset_script.py <num_of_programs> <input_directory> runs the datapoints
• measure_energy_tegra: Specific scripts for running the synthetic programs and creating energy dataset on Nvidia Tegra TX1.
  • remake_dataset_for_tegra.py: Small changes to the granerated programs to run on Tegra TX1
  • tegra_measure_new.py: Usage python tegra_measure_new.py <num_of_programs> <input_directory> creates the dataset

real-life-dataset

Method for extracting datapoints from Rodinia and Polybench benchmark suites.

• polybench: The Polybench benchmark suite
• rodinia_3.1: The Rodinia benchmark suite
• config.txt: A config file that defines the areas that will be measured (corresponding to CPU-hotspots and GPU-kernels)
• run_dataset: The scripts to build the datapoints
  • run_dataset.py: Usage python run_dataset.py <app_hotspot_name> <energy|time> The is retrieved from the config.txt. The second argument is energy for measuring energy and time for measuring time
  • measure_time_energy.py: Used for measuring execution time and energy consumption on Nvidia Tegra TX1