rework benchmarks: script and output format

this is an attempt at making benchmarks easier to work with

• benchmarks run will read benchmarks from NUON data and run them
• benchmarks plot will plot benchmark results
• the output format will be something
  • a directory whose name is the hash of the CPU spec and the Komodo commit hash
  • contains cpu.json with the CPU info
  • contains komodo.txt with the Komodo commit hash
  • contains NDJSON result files

Note

results will typically be uploaded to https://gitlab.isae-supaero.fr/dragoon/komodo-benchmark-results

Note

this MR goes alongside the <code data-sourcepos="16:32-16:63">komodo-benchmark-results@restart</code> branch

changelog

• bump Nushell to 0.101.0
  • parallel $in => remove useless let input = $in when possible, e.g. still required when using the function's $in in a for loop
  • group-by changed => group-by x --to-table will now produce a table with columns x and items instead of group and items as in 0.95.0
• add link to results repo
• the Nushell benchmarks lib
  • rename --force to --no-confirm (-y)
  • add --append
  • reject columns that GPLT will complain about, e.g. $.points.k for the FRI plots
  • add --save to the FRI plot
• move the "field" and "curve group" benchmarks from benchmarks/src/bin/operations/ to benchmarks/src/bin/
• remove benchmarks/params/fri.nu because it's been uniformized with the other methods
• rewrite the README
• add main function to benchmarks/ that runs the benchmarks from a NUON record specification
• simplify the output of FRI run

TODO

• fix "atomic operations" (done in 4f69a1d6)
• check that plotting still works

images

Field

Linear algebra

FEC

ZK

FRI