• Header-only
• Solely depends on a C++ 11 capable compiler

All benchmarks must be created and registered with the BENCHMARK(name) or the BENCHMARK_F(fixture,name) macro:

BENCHMARK(isPrimeSimple_101)
{
    while (ctxt.Running())
    {
        // put the code to benchmark here
        isPrimeSimple(101);
    }
}

These macros register a benchmark in a global list and obtain a zenbench::Context& ctxt parameter which iterates over the code to benchmark for a given amount of time.

To execute all benchmarks and print the results to stdout call zenbench::Benchmark::RunAllBenchmarks() (in main, e.g.).

int main(int argc, const char* argv[])
{
    zenbench::Benchmark::RunAllBenchmarks(argc, argv);   
}

or add ZENBENCH_MAIN to the global scope.

Fixture classes must derive from zenbench::Benchmark and may overwrite SetUpand TearDown

class Fixture : public zenbench::Benchmark
{
protected:
    virtual void SetUp() override
    {
        zenbench::Benchmark::SetUp();
    
        // put your initialization code here
    }

    virtual void TearDown() override
    {
        // put your tear down code here
    
        zenbench::Benchmark::TearDown();
    }
};

For some Benchmarks it is necessary to setup new input data in every iteration. To exclude this setup from the benchmark, wrap the included parts in an additional scope and declare a BenchmarkArea object at the top of this scope.

while (ctxt.Running())
{
    // the copy of vec to workVector is not included in the Benchmark
    std::vector<float> workVector(size);
    std::copy(vec.begin(),vec.end(),workVector.begin());
    
    {
        zenbench::BenchmarkArea benchArea(ctxt);

        qsort(workVector.data(),size,sizeof(float),cmp_float);
    }
}

Pass --zenbench_filter=*foo* to restrict the execution to benchmarks which names or fixtures contain foo.