measure.sh

#!/usr/bin/env bash
set -euo pipefail
cd "$(git rev-parse --show-toplevel)"

if [[ "$(uname)" == "Linux" ]]; then
    my_cores=$(nproc)
elif [[ "$(uname)" == "Darwin" ]]; then
    my_cores=$(sysctl -n hw.ncpu)
else
    echo "Unknown number of cores ¯\_(ツ)_/¯"
    exit 1
fi

out="plot.csv"
if [[ ! -f "$out" ]]; then
    echo "cores,area,chunk,start,stop" >"$out"
fi

for cores in 4 8 12 16 24 32 48 64; do
  if [ "$cores" -le "$my_cores" ]; then
    for area in 32 64 128 256 512; do
      for chunk in 2 4 8; do
        cabal clean
        ghc_flags="--ghc-options=-j$cores +RTS -A${area}m -n${chunk}m -RTS"
        start=$(date --iso=sec)
        cabal build --disable-tests -j "$ghc_flags" language
        stop=$(date --iso=sec)
        echo "$cores,$area,$chunk,$start,$stop" >>"$out"
      done
    done
  fi
done

plot.py

import sys
import pandas as pd
import matplotlib.pyplot as plt

# Excuse my terrible python skills

if len(sys.argv) < 2:
    print("Usage: python script_name.py filename.csv")
    sys.exit(1)

filename = sys.argv[1]
basename = filename.rsplit('.', 1)[0]
df = pd.read_csv(filename)

df['duration'] = (pd.to_datetime(df['stop']) - pd.to_datetime(df['start'])).dt.total_seconds()
areas = df['area'].unique()
areas.sort()
cores = df['cores'].unique()
cores.sort()

fig, axs = plt.subplots(len(areas), 1, figsize=(10, 6 * len(areas)), sharex=True, sharey=True)
if len(areas) == 1: axs = [axs]
for ax, area in zip(axs, areas):
    df_area = df[df['area'] == area]
    chunks = df_area['chunk'].unique()
    chunks.sort()
    for chunk in chunks:
        df_chunk = df_area[df_area['chunk'] == chunk]
        mean_durations = df_chunk.groupby('cores')['duration'].mean().reset_index().sort_values(by='cores')
        ax.plot(mean_durations['cores'], mean_durations['duration'], marker='o', linestyle='-', label=f'-n{chunk}m')
    ax.set_title(f'-A{area}m')
    ax.legend()
    ax.grid(True)
    ax.set_xticks(cores)
    ax.set_yticks(range(int(df['duration'].min()), int(df['duration'].max()) + 1, 1))
fig.supxlabel('Cores')
fig.supylabel('Duration (seconds)')
plt.tight_layout()
plt.savefig(f'{basename}.png')