Look for optimisations via profiling

[php1ic]

Start looking for any easy wins to speed things up.

A few seconds to read 25 separate 3-6k line files isn't too bad, but now we have the functionality set up, working and tested, let's see if we can make some optimisations.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.73%. Comparing base (ced0658) to head (84621ef).

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #26      +/-   ##
==========================================
- Coverage   99.73%   99.73%   -0.01%     
==========================================
  Files          12       12              
  Lines         751      749       -2     
==========================================
- Hits          749      747       -2     
  Misses          2        2              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[php1ic]

We will use this simple script to quantify progress

import timeit

from nuclearmasses.mass_table import MassTable

runs = 5
t = timeit.repeat("MassTable()", globals=globals(), number=1, repeat=runs)
print(f"{runs} runs: min={min(t):.6f}s, max={max(t):.6f}s, avg={sum(t)/runs:.6f}s")

With the HEAD of main (ced0658) before the branch:

5 runs: min=1.018965s, max=1.032672s, avg=1.026384s

[php1ic]

8535832

5 runs: min=0.772643s, max=0.810523s, avg=0.786041s

[php1ic]

Use pyinstrument for targeted optimisations.

Much simpler script

from nuclearmasses.mass_table import MassTable

MassTable()

Then run with

pyinstrument -r html ./my_profile.py

[php1ic]

Can't replicate above values, particularly their ratios. Probably related to what else my machine was doing at the time.

Wrapper script to checkout hash, run the timer and return to where we were for a list of commits. This should hopefully give a consistent set of outputs.

#!/usr/bin/env bash

set -euo pipefail

HASHES=(
  "ced0658"
  "8535832"
  "6524f56"
)

for h in "${HASHES[@]}"; do
  git checkout "${h}" >/dev/null 2>&1

  if output=$(python "${HOME}/tmp/nm/timer.py" 2>/dev/null); then
    # pyinstrument -r html "${HOME}/tmp/nm/my_profile.py"
    :
  else
    output="ERROR"
  fi

  git checkout - >/dev/null 2>&1

  echo "$h $output"
done

Can't have the timer or profile scripts with the repo (unless I commit them) as it causes issues with the checkout.

Running with the latest commit (6524f56)

ced0658 5 runs: min=1.094489s, max=1.226982s, avg=1.149681s
8535832 5 runs: min=0.876762s, max=0.946023s, avg=0.893084s
6524f56 5 runs: min=0.732890s, max=0.746327s, avg=0.738968s

[php1ic]

I think we have found all of the obvious optimisations, and made some good gains with an almost 30% speed-up.

Final comparison

ced0658 | 5 runs: min=0.994367s, max=1.030123s, avg=1.019532s
8535832 | 5 runs: min=0.814077s, max=0.857223s, avg=0.824173s
6524f56 | 5 runs: min=0.718199s, max=0.722187s, avg=0.720171s
84621ef | 5 runs: min=0.709170s, max=0.715174s, avg=0.712312s