Skip to content

README_accelergy.md

Latest commit

 

History

History
56 lines (36 loc) · 2.25 KB

README_accelergy.md

File metadata and controls

56 lines (36 loc) · 2.25 KB

SCALE-Sim + Accerlegy Integration for Enabling Energy and Power Estimation of Systolic CNN-Accelerator

Requirement

Install SCALE-Sim

Please install SCALE-Sim following the instructions in the main branch: https://github.com/scalesim-project/scale-sim-v3/tree/main

Install Accelergy

Please install Accelergy following the instructions in their github repo https://github.com/Accelergy-Project/accelergy

For more accurate estimation, please install accelergy-plug-ins for 3rd party, technology-based estimators CACTI - https://github.com/Accelergy-Project/accelergy-cacti-plug-in.git Aladdin - https://github.com/Accelergy-Project/accelergy-aladdin-plug-in.git

Run

Run the following command:

$ cd rundir-accelergy

$ ./run_all.sh -c <path_to_config_file> -t <path_to_topology_file> -p <path_to_scale-sim_log_dir> -o <path_to_output_dir>

For Example:

$ ./run_all.sh -c ../configs/scale.cfg -t ../topologies/conv_nets/test.csv -p ../test_runs/ -o ./output

Tool Inputs

Configuration File

The configuration file is used to specify the architecture and run parameters for the simulations.

Built based upon SCALE-Sim, we add additional parameters for action count extraction.

  • SRAM_row_size: the size of the row buffer (block) that each SRAM access loads
  • SRAM_bank_size: temporal capacity for the memory to keep previously-accessed data (with more than one row buffer per bank)
  • DRAM_row_size, _bank_size: same concept as SRAM

The rest of the parameters are the same as (https://github.com/scalesim-project/scale-sim-v3).

Additional compound components for Accelergy

If any other kinds fo compound componetns are to be included, please add them to ./accelergy_input/components For more details, please refer to (http://accelergy.mit.edu/)

Result

Files

In the output directory,

  • scale_sim_out_<run_name> contains performance results in .csv format. Summary in COMPUTE_REPORT.csv
  • accelergy_out_<run_name> contains energy estimation results in .yaml format. Summary in energy_estimation.yaml More details can be found from each framework's github link.

Visualization

$ jupyter-notebook gen_plot.ipynb will also help to anaylize the result. Currently set to the example case (see the Single Run section).