#include <algorithm>
#include <benchmark/benchmark.h>
#include <cstddef>
#include <iterator>

const size_t N = 1 << 25; // 32M

int v[N];                 // 32M * 4B = 128MB

void f1(int i) { std::ranges::fill(v, i); }

void f2(int i) { std::fill(std::rbegin(v), std::rend(v), i); }

template<auto f> static void test(benchmark::State &state)
{
    int i = 0;
    for (auto _ : state)
        f1(i++);
}

BENCHMARK(test<f1>)->Iterations(100);
BENCHMARK(test<f2>)->Iterations(100);

BENCHMARK_MAIN();