Stable Dev Build Status Coverage

AirspeedVelocity.jl strives to make it easy to benchmark Julia packages over their lifetime. It is inspired by asv.

This package allows you to:

  • Generate benchmarks directly from the terminal with an easy-to-use CLI.
  • Compare many commits/tags/branches at once.
  • Plot those benchmarks, automatically flattening your benchmark suite into a list of plots with generated titles.
  • Run as a GitHub action to create benchmark comparisons for every submitted PR (in a bot comment).

This package also freezes the benchmark script at a particular revision, so there is no worry about the old history overwriting the benchmark.


You can install the CLI with:

julia -e 'using Pkg; Pkg.add("AirspeedVelocity");"AirspeedVelocity")'

This will install two executables at ~/.julia/bin - make sure to have it on your PATH.


You may use the CLI to generate benchmarks for any package with, e.g.,


This will benchmark the package defined in the current directory at the current dirty state, against the default branch (i.e., main or master), over all benchmarks defined in benchmark/benchmarks.jl. It will then print a markdown table of the results while also saving the JSON results to the current directory.

You can configure all options with the CLI flags. For example, to benchmark the registered package Transducers.jl at the revisions v0.4.20, v0.4.70, and master, you can use:

benchpkg Transducers \
    --rev=v0.4.20,v0.4.70,master \

This will further use the benchmark script benchmark/benchmarks.jl as it was defined at v0.4.20, and then save the JSON results in the current directory.

We can explicitly view the results of the benchmark as a table with benchpkgtable:

benchpkgtable Transducers \

We can also generate plots of the revisions with:

benchpkgplot Transducers \
    --rev=v0.4.20,v0.4.70,master \
    --format=pdf \

which will generate a pdf file for each set of 5 plots, showing the change with each revision:

Screenshot 2023-04-03 at 10 36 16 AM

You can also provide a custom benchmark. For example, let's say you have a file script.jl, defining a benchmark for SymbolicRegression.jl (we always need to define the SUITE variable as a BenchmarkGroup):

using BenchmarkTools, SymbolicRegression
const SUITE = BenchmarkGroup()

# Create hierarchy of benchmarks:
SUITE["eval_tree_array"] = BenchmarkGroup()

options = Options(; binary_operators=[+, -, *], unary_operators=[cos])
tree = Node(; feature=1) + cos(3.2f0 * Node(; feature=2))

for n in [10, 20]
    SUITE["eval_tree_array"][n] = @benchmarkable(
        eval_tree_array($tree, X, $options),
        setup=(X=randn(Float32, 2, $n))

Inside this script, we will also have access to the PACKAGE_VERSION constant, to allow for different behavior depending on tag. We can run this benchmark over the history of SymbolicRegression.jl with:

benchpkg SymbolicRegression \
    -r v0.15.3,v0.16.2 \
    -s script.jl \
    -o results/ \
    --exeflags="--threads=4 -O3"

where we have also specified the output directory and extra flags to pass to the julia executable. We can also now visualize this:

benchpkgplot SymbolicRegression \
    -r v0.15.3,v0.16.2 \
    -i results/ \
    -o plots/

Using in CI

You can use this package in GitHub actions to benchmark every PR submitted to your package, by copying the example: .github/workflows/benchmark_pr.yml.

Every time a PR is submitted to your package, this workflow will run and generate plots of the performance of the PR against the default branch, as well as a markdown table, showing whether the PR improves or worsens performance:



For running benchmarks, you can use the benchpkg command, which is built into the ~/.julia/bin folder:

    benchpkg [package_name] [-r --rev <arg>]
                            [--url <arg>]
                            [--path <arg>]
                            [-o, --output-dir <arg>]
                            [-e, --exeflags <arg>]
                            [-a, --add <arg>]
                            [-s, --script <arg>]
                            [--bench-on <arg>]
                            [-f, --filter <arg>]
                            [--nsamples-load-time <arg>]

Benchmark a package over a set of revisions.

# Arguments

- `package_name`: Name of the package. If not given, the package is assumed to be
  the current directory.

# Options

- `-r, --rev <arg>`: Revisions to test (delimit by comma). Use `dirty` to
  benchmark the current state of the package at `path` (and not a git commit).
  The default is `{DEFAULT},dirty`, which will attempt to find the default branch
  of the package.
- `--url <arg>`: URL of the package.
- `--path <arg>`: Path of the package. The default is `.` if other arguments are not given.
- `-o, --output-dir <arg>`: Where to save the JSON results. The default is `.`.
- `-e, --exeflags <arg>`: CLI flags for Julia (default: none).
- `-a, --add <arg>`: Extra packages needed (delimit by comma).
- `-s, --script <arg>`: The benchmark script. Default: `benchmark/benchmarks.jl` downloaded from `stable`.
- `--bench-on <arg>`: If the script is not set, this specifies the revision at which
  to download `benchmark/benchmarks.jl` from the package.
- `-f, --filter <arg>`: Filter the benchmarks to run (delimit by comma).
- `--nsamples-load-time <arg>`: Number of samples to take when measuring load time of
    the package (default: 5). (This means starting a Julia process for each sample.)
- `--dont-print`: Don't print the table.

# Flags

- `--tune`: Whether to run benchmarks with tuning (default: false).

You can also just generate a table:

    benchpkgtable [package_name] [-r --rev <arg>]
                                 [-i --input-dir <arg>]
                                 [--mode <arg>]
                                 [--url <arg>]
                                 [--path <arg>]

Print a table of the benchmarks of a package as created with `benchpkg`.

# Arguments

- `package_name`: Name of the package.

# Options

- `-r, --rev <arg>`: Revisions to test (delimit by comma).
  The default is `{DEFAULT},dirty`, which will attempt to find the default branch
  of the package.
- `-i, --input-dir <arg>`: Where the JSON results were saved (default: ".").
- `--url <arg>`: URL of the package. Only used to get the package name.
- `--path <arg>`: Path of the package. The default is `.` if other arguments are not given.
   Only used to get the package name.

# Flags

- `--ratio`: Whether to include the ratio (default: false). Only applies when
    comparing two revisions.
- `--mode`: Table mode(s). Valid values are "time" (default), to print the
    benchmark time, or "memory", to print the allocation and memory usage.
    Both options can be passed, if delimited by comma.

For plotting, you can use the benchpkgplot function:

    benchpkgplot package_name [-r --rev <arg>]
                              [-i --input-dir <arg>]
                              [-o --output-dir <arg>]
                              [-n --npart <arg>]
                              [--format <arg>]

Plot the benchmarks of a package as created with `benchpkg`.

# Arguments

- `package_name`: Name of the package.

# Options

- `-r, --rev <arg>`: Revisions to test (delimit by comma).
- `-i, --input-dir <arg>`: Where the JSON results were saved (default: ".").
- `-o, --output-dir <arg>`: Where to save the plots results (default: ".").
- `-n, --npart <arg>`: Max number of plots per page (default: 10).
- `--format <arg>`: File type to save the plots as (default: "png").

If you prefer to use the Julia API, you can use the benchmark function for generating data. The API is given here.

Also be sure to check out PkgBenchmark.jl. PkgBenchmark.jl is a simple wrapper of BenchmarkTools.jl to interface it with Git, and is a good choice for building custom analysis workflows.

However, for me this wrapper is a bit too thin, which is why I created this package. AirspeedVelocity.jl tries to have more features and workflows readily-available. It also emphasizes a CLI (though there is a Julia API), as my subjective view is that this is more suitable for interacting side-by-side with git.