Pure Call Overhead Benchmark for folly::coro::Task. Credit to original author @keshpr

call_overhead.cpp

#include <coroutine>

#include <folly/experimental/coro/BlockingWait.h>
#include <folly/experimental/coro/Task.h>
#include <folly/init/Init.h>

#include <benchmark/benchmark.h>
#include <gflags/gflags.h>

DEFINE_int32(numThreads, 1, "");

template <size_t ArgRetSize>
struct Val {
  char arr[ArgRetSize];
};

template <
    template <typename>
    typename TTask,
    size_t RetSize,
    size_t ArgSize,
    size_t Depth>
struct echo_awaitable_task {
  static TTask<Val<RetSize>> task(Val<ArgSize> arg) {
    co_return co_await echo_awaitable_task<TTask, RetSize, ArgSize, Depth - 1>::
        task(arg);
  }
};

template <template <typename> typename TTask, size_t RetSize, size_t ArgSize>
struct echo_awaitable_task<TTask, RetSize, ArgSize, 0> {
  static TTask<Val<RetSize>> task(Val<ArgSize> arg) {
    Val<RetSize> ret;
    co_return ret;
  }
};

template <size_t RetSize, size_t ArgSize, size_t Depth>
void BM_EchoFollyCoroTaskWithDepth(benchmark::State& state) {
  for (auto _ : state) {
    Val<ArgSize> arg;
    auto task =
        echo_awaitable_task<folly::coro::Task, RetSize, ArgSize, Depth>::task(
            arg);
    auto result = folly::coro::blockingWait(std::move(task));
    benchmark::DoNotOptimize(result);
  }
}

#define BENCH_(name)              \
  BENCHMARK(name)                 \
      ->Threads(FLAGS_numThreads) \
      ->MeasureProcessCPUTime()   \
      ->UseRealTime()

#define COMMA ,

#define BENCH_DEPTH_FOLLY_CORO_(ret_size, arg_size, depth) \
  BENCH_(BM_EchoFollyCoroTaskWithDepth<ret_size COMMA arg_size COMMA depth>);

#define BENCH_DEPTH_FOLLY_CORO(depth)    \
  BENCH_DEPTH_FOLLY_CORO_(0, 0, depth)   \
  BENCH_DEPTH_FOLLY_CORO_(0, 70, depth)  \
  BENCH_DEPTH_FOLLY_CORO_(70, 0, depth)  \
  BENCH_DEPTH_FOLLY_CORO_(70, 70, depth) \
  BENCH_DEPTH_FOLLY_CORO_(24, 0, depth)  \
  BENCH_DEPTH_FOLLY_CORO_(24, 32, depth) \
  BENCH_DEPTH_FOLLY_CORO_(128, 128, depth)

BENCH_DEPTH_FOLLY_CORO(0)
BENCH_DEPTH_FOLLY_CORO(20)
BENCH_DEPTH_FOLLY_CORO(30)
BENCH_DEPTH_FOLLY_CORO(40)

int main(int argc, char **argv) {
  benchmark::Initialize(&argc, argv);
  folly::init(&argc, &argv);
  benchmark::RunSpecifiedBenchmarks();
  benchmark::Shutdown();
  return 0;
}