This library is a fork of Codspeed's instrumentation hooks library and adapted to be used as a Zig integration for the platform.
Add the library to your Zig project.
zig fetch --save git+https://github.com/james-elicx/codspeed-zigThen, in your build.zig, add the library to your exe or lib target.
benchmark_exe.root_module.addImport(
"codspeed",
b.dependency("codspeed", .{ .target = target, .optimize = optimize }).module("codspeed"),
);const Codspeed = @import("codspeed");
pub fn main() !void {
const alloc = std.heap.c_allocator; // or std.heap.page_allocator
const codspeed = Codspeed.init(alloc, "src/benchmarks.zig");
defer codspeed.deinit();
try codspeed.bench("benchmark_name", benchmarkThisCode);
try codspeed.bench("benchmark_name_2", benchmarkOtherCode);
}
fn benchmarkThisCode() void {
// Code to benchmark here
}
fn benchmarkOtherCode() void {
// Other code to benchmark here
}You can manually start and stop the instrumentation around a code block for more granular control.
const Codspeed = @import("codspeed");
pub fn main() !void {
const alloc = std.heap.c_allocator; // or std.heap.page_allocator
const codspeed = Codspeed.init(alloc, "src/benchmarks.zig");
defer codspeed.deinit();
try codspeed.start("benchmark_name");
// Code to benchmark here
// ...
try codspeed.stop("benchmark_name");
}- Use
std.heap.c_allocatorfor benchmarks. It will provide more stable and consistent results. - Avoid benchmarking logic that involves system calls. They cannot be reliably instrumented and will often be excluded in Codspeed's dashboard. For instance, do file I/O operations outside of the benchmarked code.
The process ID cannot be fetched using std.os.linux.getpid() for the "perf" instrument as it appears to result in an invalid system call error. The valgrind instrument works fine.
