#include #include #include #include #include #include #include std::atomic total{ 0 }; long fib(long n) __attribute__((noinline)); void run(size_t nitr, long) __attribute__((noinline)); long fib(long n) { return (n < 2) ? n : fib(n - 1) + fib(n - 2); } #define RUN_LABEL \ std::string{ std::string{ __FUNCTION__ } + "(" + std::to_string(n) + ") x " + \ std::to_string(nitr) } \ .c_str() void run(size_t nitr, long n) { omnitrace_user_stop_thread_trace(); omnitrace_user_push_region(RUN_LABEL); long local = 0; for(size_t i = 0; i < nitr; ++i) local += fib(n); total += local; omnitrace_user_pop_region(RUN_LABEL); omnitrace_user_start_thread_trace(); } int main(int argc, char** argv) { omnitrace_user_push_region(argv[0]); omnitrace_user_push_region("initialization"); size_t nthread = std::min(16, std::thread::hardware_concurrency()); size_t nitr = 50000; long nfib = 10; if(argc > 1) nfib = atol(argv[1]); if(argc > 2) nthread = atol(argv[2]); if(argc > 3) nitr = atol(argv[3]); omnitrace_user_pop_region("initialization"); printf("[%s] Threads: %zu\n[%s] Iterations: %zu\n[%s] fibonacci(%li)...\n", argv[0], nthread, argv[0], nitr, argv[0], nfib); omnitrace_user_push_region("thread_creation"); std::vector threads{}; threads.reserve(nthread); for(size_t i = 0; i < nthread; ++i) { size_t _nitr = ((i % 2) == 1) ? (nitr - (0.1 * nitr)) : (nitr + (0.1 * nitr)); threads.emplace_back(&run, _nitr, nfib); } omnitrace_user_pop_region("thread_creation"); run(nitr - 0.25 * nitr, nfib - 0.1 * nfib); omnitrace_user_push_region("thread_wait"); for(auto& itr : threads) itr.join(); omnitrace_user_pop_region("thread_wait"); printf("[%s] fibonacci(%li) x %lu = %li\n", argv[0], nfib, nthread, total.load()); omnitrace_user_pop_region(argv[0]); return 0; }