leviathan-lang.com / docs / benchmarks

Benchmarks

How Leviathan's performance is measured, and where the measurements are published.

How the numbers are produced

Every figure on this page comes from the same method, so that it can be checked:

• Like for like. Each benchmark is the same algorithm on the same input in Leviathan and in C++ compiled with -O2, and both programs must print the same checksum before their times count.
• Both sides measured the same way. Each program times its own work, and both timers report at microsecond resolution.
• Best of three. Every figure is the best of three runs.
• Reproducible. The page records the version, commit, and date a set of tables was measured on, so a figure can be re-measured against the same build.
• Release builds only. Numbers are entered only after they have been re-measured on a shipped release binary, never copied from an earlier development run.

What is published

Both kinds of result are published: the benchmarks where Leviathan is at or ahead of C++, and the ones where it is not, with a short note on why. A table that lists only the flattering rows is not a benchmark report.

The tables below are filled in from the release measurement. Until a release has been measured, they list no rows.

Results

Numerics vs C++ -O2

No measurements have been published for this table yet.