cargo-bench --- Execute benchmarks of a package
cargo bench
[options] [benchname] [--
bench-options]
Compile and execute benchmarks.
The benchmark filtering argument benchname and all the arguments following the two dashes (--
) are passed to the benchmark binaries and thus to libtest (rustc‘s built in unit-test and micro-benchmarking framework). If you are passing arguments to both Cargo and the binary, the ones after --
go to the binary, the ones before go to Cargo. For details about libtest’s arguments see the output of cargo bench -- --help
and check out the rustc book's chapter on how tests work at https://doc.rust-lang.org/rustc/tests/index.html.
As an example, this will run only the benchmark named foo
(and skip other similarly named benchmarks like foobar
):
cargo bench -- foo --exact
Benchmarks are built with the --test
option to rustc
which creates a special executable by linking your code with libtest. The executable automatically runs all functions annotated with the #[bench]
attribute. Cargo passes the --bench
flag to the test harness to tell it to run only benchmarks, regardless of whether the harness is libtest or a custom harness.
The libtest harness may be disabled by setting harness = false
in the target manifest settings, in which case your code will need to provide its own main
function to handle running benchmarks.
Note: The
#[bench]
attribute is currently unstable and only available on the nightly channel. There are some packages available on crates.io that may help with running benchmarks on the stable channel, such as Criterion.
By default, cargo bench
uses the bench
profile, which enables optimizations and disables debugging information. If you need to debug a benchmark, you can use the --profile=dev
command-line option to switch to the dev profile. You can then run the debug-enabled benchmark within a debugger.
The working directory of every benchmark is set to the root directory of the package the benchmark belongs to. Setting the working directory of benchmarks to the package‘s root directory makes it possible for benchmarks to reliably access the package’s files using relative paths, regardless from where cargo bench
was executed from.
By default, when no package selection options are given, the packages selected depend on the selected manifest file (based on the current working directory if --manifest-path
is not given). If the manifest is the root of a workspace then the workspaces default members are selected, otherwise only the package defined by the manifest will be selected.
The default members of a workspace can be set explicitly with the workspace.default-members
key in the root manifest. If this is not set, a virtual workspace will include all workspace members (equivalent to passing --workspace
), and a non-virtual workspace will include only the root crate itself.
When no target selection options are given, cargo bench
will build the following targets of the selected packages:
The default behavior can be changed by setting the bench
flag for the target in the manifest settings. Setting examples to bench = true
will build and run the example as a benchmark, replacing the example's main
function with the libtest harness.
Setting targets to bench = false
will stop them from being bencharmked by default. Target selection options that take a target by name (such as --example foo
) ignore the bench
flag and will always benchmark the given target.
See Configuring a target for more information on per-target settings.
Binary targets are automatically built if there is an integration test or benchmark being selected to benchmark. This allows an integration test to execute the binary to exercise and test its behavior. The CARGO_BIN_EXE_<name>
environment variable is set when the integration test is built so that it can use the env
macro to locate the executable.
Passing target selection flags will benchmark only the specified targets.
Note that --bin
, --example
, --test
and --bench
flags also support common Unix glob patterns like *
, ?
and []
. However, to avoid your shell accidentally expanding glob patterns before Cargo handles them, you must use single quotes or double quotes around each glob pattern.
The feature flags allow you to control which features are enabled. When no feature options are given, the default
feature is activated for every selected package.
See the features documentation for more details.
By default the Rust test harness hides output from benchmark execution to keep results readable. Benchmark output can be recovered (e.g., for debugging) by passing --nocapture
to the benchmark binaries:
cargo bench -- --nocapture
The --jobs
argument affects the building of the benchmark executable but does not affect how many threads are used when running the benchmarks. The Rust test harness runs benchmarks serially in a single thread.
While cargo bench
involves compilation, it does not provide a --keep-going
flag. Use --no-fail-fast
to run as many benchmarks as possible without stopping at the first failure. To “compile” as many benchmarks as possible, use --benches
to build benchmark binaries separately. For example:
cargo build --benches --release --keep-going cargo bench --no-fail-fast
See the reference for details on environment variables that Cargo reads.
0
: Cargo succeeded.101
: Cargo failed to complete.Build and execute all the benchmarks of the current package:
cargo bench
Run only a specific benchmark within a specific benchmark target:
cargo bench --bench bench_name -- modname::some_benchmark