zBench is a benchmarking library for the Zig programming language. It is designed to provide easy-to-use functionality to measure and compare the performance of your code.
For installation instructions, please refer to the documentation.
Create a new benchmark function in your Zig code. This function should take a single argument of type std.mem.Allocator
. The function would run the code you wish to benchmark.
fn benchmarkMyFunction(allocator: std.mem.Allocator) void {
// Code to benchmark here
}
You can then run your benchmarks in a test:
test "bench test" {
var bench = zbench.Benchmark.init(std.testing.allocator, .{});
defer bench.deinit();
try bench.add("My Benchmark", myBenchmark, .{});
try bench.run(std.io.getStdOut().writer());
}
To customize your benchmark runs, zBench provides a Config
struct that allows you to specify several options:
pub const Config = struct {
iterations: u16 = 0,
max_iterations: u16 = 16384,
time_budget_ns: u64 = 2e9, // 2 seconds
hooks: Hooks = .{},
track_allocations: bool = false,
};
iterations
: The number of iterations the benchmark has been run. This field is usually managed by zBench itself.max_iterations
: Set the maximum number of iterations for a benchmark. Useful for controlling long-running benchmarks.time_budget_ns
: Define a time budget for the benchmark in nanoseconds. Helps in limiting the total execution time of the benchmark.hooks
: Setbefore_all
,after_all
,before_each
, andafter_each
hooks to function pointers.track_allocations
: Boolean to enable or disable tracking memory allocations during the benchmark.
Zig is in active development and the APIs can change frequently, making it challenging to support every dev build. This project currently aims to be compatible with stable, non-development builds to provide a consistent experience for the users.
Supported Version: As of now, zBench is tested and supported on Zig version 0.13.0.
Performance Note: It's important to acknowledge that a no-op time of 15 ns (or more) is expected and is not an issue with zBench itself. This is a normal occurrence and does not reflect any inefficiency in the benchmarking process.
Benchmark functions have the following signature:
fn(allocator: std.mem.Allocator) void
The function body contains the code you wish to benchmark.
zBench provides a comprehensive report for each benchmark run. It includes the total operations performed, the average, min, and max durations of operations, and the percentile distribution (p75, p99, p995) of operation durations.
benchmark runs time (avg ยฑ ฯ) (min ... max) p75 p99 p995
---------------------------------------------------------------------------------------------------------------
benchmarkMyFunction 1000 1200ms ยฑ 10ms (100ms ... 2000ms) 1100ms 1900ms 1950ms
This example report indicates that the benchmark "benchmarkMyFunction" ran with an average of 1200 ms per execution and a standard deviation of 10 ms. The minimum and maximum execution times were 100 ms and 2000 ms, respectively. The 75th, 99th and 99.5th percentiles of execution times were 1100 ms, 1900 ms, and 1950 ms, respectively.
You can run all example tests with the following command:
zig build examples
- If Zig doesn't detect changes in a dependency, clear the project's
zig-cache
folder and~/.cache/zig
. - Non-ASCII characters not printed correctly on Windows
The main purpose of this repository is to continue to evolve zBench, making it faster and more efficient. We are grateful to the community for contributing bugfixes and improvements. Read below to learn how you help improve zBench.
Read our contributing guide to learn about our development process, how to propose bugfixes and improvements, and how to build and test your changes to zBench.
zBench is MIT licensed.