source: benchmark/readyQ/cycle.cfa @ deda7e6

#include "rq_bench.hfa"

struct Partner {
        unsigned long long count;
        unsigned long long blocks;
        bench_sem self;
        bench_sem * next;
};

void ?{}( Partner & this ) {
        this.count = this.blocks = 0;
}

thread BThrd {
        Partner & partner;
};

void ?{}( BThrd & this, Partner * partner ) {
        ((thread&)this){ bench_cluster };
        &this.partner = partner;
}

void ^?{}( BThrd & mutex this ) {}

void main( BThrd & thrd ) with(thrd.partner) {
        count = 0;
        for() {
                blocks += wait( self );
                post( *next );
                count ++;
                if( clock_mode && stop) break;
                if(!clock_mode && count >= stop_count) break;
        }

        __atomic_fetch_add(&threads_left, -1, __ATOMIC_SEQ_CST);
}

int main(int argc, char * argv[]) {
        unsigned ring_size = 2;
        cfa_option opt[] = {
                BENCH_OPT,
                { 'r', "ringsize", "Number of threads in a cycle", ring_size }
        };
        BENCH_OPT_PARSE("cforall cycle benchmark");

        {
                unsigned long long global_counter = 0;
                unsigned long long global_blocks  = 0;
                unsigned tthreads = nthreads * ring_size;
                Time start, end;
                BenchCluster bc = { nprocs };
                {
                        threads_left = tthreads;
                        BThrd **  threads = alloc(tthreads);
                        Partner * thddata = alloc(tthreads);
                        for(i; tthreads) {
                                (thddata[i]){};
                                unsigned pi = (i + nthreads) % tthreads;
                                thddata[i].next = &thddata[pi].self;
                        }
                        for(int i = 0; i < tthreads; i++) {
                                threads[i] = malloc();
                                (*threads[i]){ &thddata[i] };
                        }
                        printf("Starting\n");

                        bool is_tty = isatty(STDOUT_FILENO);
                        start = timeHiRes();

                        for(i; nthreads) {
                                post( thddata[i].self );
                        }
                        wait(start, is_tty);

                        stop = true;
                        end = timeHiRes();
                        printf("\nDone\n");

                        for(i; tthreads) {
                                post( thddata[i].self );
                                Partner & partner = join( *threads[i] ).partner;
                                global_counter += partner.count;
                                global_blocks  += partner.blocks;
                                delete(threads[i]);
                        }
                        free(threads);
                        free(thddata);
                }

                printf("Duration (ms)        : %'lf\n", (end - start)`dms);
                printf("Number of processors : %'d\n", nprocs);
                printf("Number of threads    : %'d\n", tthreads);
                printf("Cycle size (# thrds) : %'d\n", ring_size);
                printf("Total Operations(ops): %'15llu\n", global_counter);
                printf("Total blocks         : %'15llu\n", global_blocks);
                printf("Ops per second       : %'18.2lf\n", ((double)global_counter) / (end - start)`ds);
                printf("ns per ops           : %'18.2lf\n", (end - start)`dns / global_counter);
                printf("Ops per threads      : %'15llu\n", global_counter / tthreads);
                printf("Ops per procs        : %'15llu\n", global_counter / nprocs);
                printf("Ops/sec/procs        : %'18.2lf\n", (((double)global_counter) / nprocs) / (end - start)`ds);
                printf("ns per ops/procs     : %'18.2lf\n", (end - start)`dns / (global_counter / nprocs));
                fflush(stdout);
        }

        return 0;
}