All source code of "Loop2Recursion: Compiler-Assisted Wear Leveling for Non-Volatile Memory", including the Loop2Recursion LLVM Pass, the Intel Pin-based profiling tools, as well as data preprocessing scripts.

Requirements

• Unix-like system with the /tmp directory
• At least 8 GB of RAM
• clang+llvm-13.0.1
• pypy, with the following extensions:
  • intervaltree
  • more-itertools
  • numpy
• Jupyter Notebook, with a Python 3 kernel and the following extensions:
  • matplotlib
  • ipympl
  • numpy
  • pandas
  • pylatex
  • scipy

Replication Instructions

Build Tools

1. Download Intel Pin, extract, and move the extracted directory (e.g. pin-3.11-97998-g7ecce2dac-gcc-linux) containing the pin executable to EXPERIMENT_ROOT.
2. Edit parameters.sh, and change the value of PIN_ROOT. Edit the values of other variables as required.
3. Run build_loop2recursion.sh
4. Run build_libuwlalloc.sh
5. Run build_pin_tools.sh

Compile Executables

1. bash compile_baseline.sh
2. bash extract_loop_information.sh
3. bash selective_loop2recursion.sh

for recursion_depth_limit in 8 16 32 64 128 256
do
    bash run_unoptimized_loop2recursion.sh -t "unoptimized_loop2recursion-recursion_depth_limit-${recursion_depth_limit}" -p "-recursion-depth-limit ${recursion_depth_limit}"
done

for recursion_depth_limit in 8 16 32 64 128 256
do
    bash run_optimized_loop2recursion.sh -t "new_loop2recursion-recursion_depth_limit-${recursion_depth_limit}" -p "-recursion-depth-limit ${recursion_depth_limit}"
done

for loop_iterations in 8 16 32 64 128 256
do
    bash run_optimized_loop2recursion.sh -t "new_loop2recursion-loop_iterations-${loop_iterations}" -p "-loop-iterations ${loop_iterations}"
done

Run and Profile Executables

This process takes a long time (around a day).

bash profile.sh -t baseline

bash profile.sh -t selective_loop2recursion

bash profile.sh -t unoptimized_loop2recursion

bash profile.sh -t new_loop2recursion

for recursion_depth_limit in 8 16 32 64 128 256
do
    bash profile.sh -t "unoptimized_loop2recursion-recursion_depth_limit-${recursion_depth_limit}"
done

for recursion_depth_limit in 8 16 32 64 128 256
do
    bash profile.sh -t "new_loop2recursion-recursion_depth_limit-${recursion_depth_limit}"
done

for loop_iterations in 8 16 32 64
do
    bash profile.sh -t "new_loop2recursion-loop_iterations-${loop_iterations}"
done

bash profile_extras.sh -t unoptimized_loop2recursion

bash profile_extras.sh -t new_loop2recursion

bash profile_uwlalloc.sh

Data Analysis

Run the following Jupyter Notebooks:

1. average_maximum_stack_usages_and_average_writes_on_hottest_stack_address_under_different_recursion_depth_limits_and_loop_iterations.ipynb
2. baseline_loop2recursion_new_loop2recursion.ipynb
3. baseline_uwlalloc_recursion_depth_limit.ipynb
4. experiment_data_visualization.ipynb
5. loop_iterations_list.ipynb
6. recursion_depth_limit.ipynb
7. cache_analysis.ipynb
8. writes_on_stack_addresses_for_benchmark_dijkstra_on_cache_less_architecture.ipynb
9. writes_on_stack_addresses_on_cache_enabled_architecture.ipynb