Assassyn: ISCA 2025 Artifact Evaluation

Folder Structure

1. assassyn-public: The repo of the assassyn language
2. Workloads & Evaluation:
   1. riscv-bmark: The workloads repo for the CPU evaluation
   2. MachSuite: The workloads for the HLS evaluation
   3. components: The scripts for the architectural component breakdown.
   4. gem5: The simulator to be compared with.
   5. sv-source: Some special verilog source for post-synthesis while some of those original design are very difficult to be synthesized.
   6. scripts: Automation scripts for generating and summarizing of experiment data.
      1. gem5_cpu: Scripts about using gem5.
      2. hls_assassyn: Scripts about using assassyn for hls benchmarks.
      3. hls_baseline: Scripts about using verilator and Bambu for the hls benchmarks baseline.
      4. minor_cpu: Scripts about simulating assassyn minor cpu with different version.
      5. sodor_cpu: Scripts about simulating 5 pipeline sodor core of chipyard.
      6. post_synthesis: Scripts about post synthesis.
      7. my_time: Scripts for generating preciser timer.
3. plot: The scripts for the figure ploting. Each fig-x.py is a script for the corresponding figure in the paper.
4. bambu-0.9.7.AppImage,open_eda.sif: Prerequisites package.

Prerequisites

1. Anaconda: For language Python support, and the VM for synthesis.
2. Docker: For gem5 simulation.
3. Rust: For the assassyn backend elaboration.

Replicating the Results

1. Elaborate the CPU and run all the RISCV benchmarks.

   • minor_cpu.py: Elaborate the CPU written in our language, run each workload, simulate both simulator and verilog, and finally export the results to /scripts/minor_cpu/minor_logs/minor.csv.

   • cpu_test.py: Elaborate different CPU versions, run each workload, get each cycles of them, and finally export the results to /scripts/minor_cpu/minor_logs/cpu_all_version.csv.

       ```bash
               sourse ./setup.sh
               cd ./scripts/my_time
               g++ time.cpp -o my_time
               cd ../..
               cd ./scripts/minor_cpu
               python ./minor_cpu.py
               python ./cpu_test.py
               cd ../..
     

   • sodor_cpu.py: Elaborate the Sodor CPU from chipyard, run each workload,simulate and finally export the results to /scripts/sodor_cpu/sodor_logs/sodor.csv.

       ```bash
               cd ./scripts/sodor_cpu
               python ./sodor_cpu.py
               cd ../..
     

   • machsuite_hls.py: Elaborate the HLS Baseline, run each benchmarks, synthesize, and finally export the results to /scripts/hls_baseline/machsuite_logs/hls_baseline.csv.

       ```bash
               cd ./scripts/hls_baseline
               python ./machsuite_hls.py
               cd ../..
     

   • ass_hls.py: Elaborate the HLS using Assassyn, run each benchmarks, synthesize, and finally export the results to /scripts/hls_baseline/ass_hls_logs/ass_hls.csv.

       ```bash
               cd ./scripts/hls_assassyn
               python ./ass_hls.py
               cd ../..
     

   • csv_combine.py: Combine distributed data together /plot/HLS.csv,/plot/cpu-bmarks.csv.

       ```bash
               cd ./scripts/gem5_cpu
               python ./gem5_cpu.py
               cd ../..   
     

   • gem5_cpu.py: Simulate the benchmarks using gem5 and finally export the results to /scripts/gem5_cpu/gem5_logs/gem5.csv.

2. Post Synthesize.

   • base_report.py: Post Synthesize all the baseline workloads, and gather the json format area result, export the results to /plot/reports.

   • assassyn_report.py: Post Synthesize all the assassyn workloads, and gather the json format area result, export the results to /plot/reports.

3. Plot the figures.

   • fig-x.py: Run each script to plot the corresponding figure in the paper.