/Speculative-Cache-Hierarchy

Extension of GhostMinion (https://dl.acm.org/doi/10.1145/3466752.3480074) for core with large size ROBs

Primary LanguageC++OtherNOASSERTION

Speculative-Cache-Hierarchy

Extension of GhostMinion (https://dl.acm.org/doi/10.1145/3466752.3480074) for core with large size ROBs

Testing

  1. Add traces to traces/ folder (.gitkeep has been added as placeholder)

  2. Run the following commands to simulate the traces on different models

# Simulation with 1M Instructions
bash list_data.sh  # Setup Input
python run_all.py all_traces.txt  # Will run all SPEC17 & ClientServer traces
# Data Analysis
bash list_data.sh  # Setup input
python extract.py results_1M_list.txt  # Extract results from ChampSim output
python summarize.py extracted_results.csv  # Summarize results for each trace
  1. Summary of results will be stored in summary/ folder (uploaded to https://docs.google.com/spreadsheets/d/1FTAI6n2S5PncKy_ky9HtW7kDRofKTGrvTM61a7VmJno/edit?usp=sharing)
  • all_spec_diff_size.csv contains results for different sizes of speculative cache considering all instructions as speculative
  • all_spec_diff_setup.csv contains results for different models of speculative cache considering all instructions as speculative

Results

1-1.5% performance slowdown

Inferences

  1. Changing the size of Speculative Cache has no visible effect on performance

  2. Removing Speculative Caches at any cache level without compromising on security has a drastic effect on performance

Plots

Inference 1 Inference 2