source: benchmark/readyQ/transfer.cpp @ 9dc0836

#include "rq_bench.hpp"
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wunused-parameter"
        #include <libfibre/fibre.h>
#pragma GCC diagnostic pop

#define PRINT(...)

__lehmer64_state_t lead_seed;
volatile unsigned leader;
volatile size_t lead_idx;

bool exhaust = false;

bench_sem the_main;

class __attribute__((aligned(128))) MyThread;

MyThread ** threads;

class __attribute__((aligned(128))) MyThread {
        unsigned id;
        volatile size_t idx;
        bench_sem sem;

public:
        size_t rechecks;

        MyThread(unsigned _id)
                : id(_id), idx(0), rechecks(0)
        {}

        void unpark() { sem.post(); }
        void park  () { sem.wait(); }

        void waitgroup() {
                uint64_t start = timeHiRes();
                for(size_t i = 0; i < nthreads; i++) {
                        PRINT( std::cout << "Waiting for : " << i << " (" << threads[i]->idx << ")" << std::endl; )
                        while( threads[i]->idx != lead_idx ) {
                                Pause();
                                if( to_miliseconds(timeHiRes() - start) > 5'000 ) {
                                        std::cerr << "Programs has been blocked for more than 5 secs" << std::endl;
                                        std::exit(1);
                                }
                        }
                }
                PRINT( std::cout | "Waiting done"; )
        }

        void wakegroup(unsigned me) {
                if(!exhaust) return;

                for(size_t i = 0; i < nthreads; i++) {
                        if(i!= me) threads[i]->sem.post();
                }
        }

        void lead() {
                this->idx = ++lead_idx;
                if(lead_idx > stop_count) {
                        PRINT( std::cout << "Leader " << this->id << " done" << std::endl; )
                        the_main.post();
                        return;
                }

                PRINT( sout << "Leader no " << this->idx << ": " << this->id << std::endl; )

                waitgroup();

                unsigned nleader = __lehmer64( lead_seed ) % nthreads;
                __atomic_store_n( &leader, nleader, __ATOMIC_SEQ_CST );

                wakegroup(this->id);
        }

        void wait() {
                fibre_yield();
                if(lead_idx == this->idx) {
                        this->rechecks++;
                        return;
                }

                assert( (lead_idx - 1) == this->idx );
                __atomic_add_fetch( &this->idx, 1, __ATOMIC_SEQ_CST );
                if(exhaust) this->sem.wait();
                else fibre_yield();
        }

        static void main(void * arg) {
                MyThread & self = *reinterpret_cast<MyThread*>(arg);
                self.park();

                unsigned me = self.id;

                for(;;) {
                        if(leader == me) {
                                self.lead();
                        }
                        else {
                                self.wait();
                        }
                        if(lead_idx > stop_count) break;
                }
        }
};

// ==================================================
int main(int argc, char * argv[]) {
        __lehmer64_state_t lead_seed = getpid();
        for(int i = 0; i < 10; i++) __lehmer64( lead_seed );
        unsigned nprocs = 2;

        option_t opt[] = {
                BENCH_OPT,
                { 'e', "exhaust", "Whether or not threads that have seen the new epoch should yield or park.", exhaust, parse_yesno}
        };
        BENCH_OPT_PARSE("cforall transition benchmark");

        std::cout.imbue(std::locale(""));
        setlocale(LC_ALL, "");

        if(clock_mode) {
                std::cerr << "This benchmark doesn't support duration mode" << std::endl;
                return 1;
        }

        if(nprocs < 2) {
                std::cerr << "Must have at least 2 processors" << std::endl;
                return 1;
        }

        lead_idx = 0;
        leader = __lehmer64( lead_seed ) % nthreads;

        size_t rechecks = 0;

        uint64_t start, end;
        {
                FibreInit(1, nprocs);
                {
                        Fibre ** handles = new Fibre*[nthreads];
                        threads = new MyThread*[nthreads];
                        for(size_t i = 0; i < nthreads; i++) {
                                threads[i] = new MyThread( i );
                                handles[i] = new Fibre( MyThread::main, threads[i] );
                        }

                        start = timeHiRes();
                        for(size_t i = 0; i < nthreads; i++) {
                                threads[i]->unpark();
                        }

                        the_main.wait();
                        end = timeHiRes();

                        for(size_t i = 0; i < nthreads; i++) {
                                threads[i]->unpark();
                        }

                        for(size_t i = 0; i < nthreads; i++) {
                                MyThread & thrd = *threads[i];
                                fibre_join( handles[i], nullptr );
                                PRINT( std::cout << i << " joined" << std::endl; )
                                rechecks += thrd.rechecks;
                                // delete( handles[i] );
                                delete( threads[i] );
                        }

                        delete[] (threads);
                        delete[] (handles);
                }
        }

        std::cout << "Duration (ms)           : " << to_miliseconds(end - start) << std::endl;
        std::cout << "Number of processors    : " << nprocs << std::endl;
        std::cout << "Number of threads       : " << nthreads << std::endl;
        std::cout << "Total Operations(ops)   : " << stop_count << std::endl;
        std::cout << "Threads parking on wait : " << (exhaust ? "yes" : "no") << std::endl;
        std::cout << "Rechecking              : " << rechecks << std::endl;


}