source: benchmark/readyQ/locality.cfa @ 4db0140

#include "rq_bench.hfa"

struct Result {
        uint64_t count;
        uint64_t dmigs;
        uint64_t gmigs;
};

// ==================================================
thread __attribute__((aligned(128))) MyThread {
        struct MyData * volatile data;

        struct {
                struct MySpot ** ptr;
                size_t len;
        } spots;

        bench_sem sem;

        Result result;

        bool share;
        size_t cnt;
        processor * ttid;
        size_t id;
};

uint64_t moved(MyThread & this, processor * ttid) {
        if(this.ttid == ttid) {
                return 0;
        }
        this.ttid = ttid;
        return 1;
}

// ==================================================
struct __attribute__((aligned(128))) MyData {
        uint64_t _p1[16];  // padding
        uint64_t * data;
        size_t len;
        processor * ttid;
        size_t id;
        uint64_t _p2[16];  // padding
};

void ?{}(MyData & this, size_t id, size_t size) {
        this.len = size;
        this.data = alloc(this.len, 128`align);
        this.ttid = active_processor();
        this.id = id;

        for(i; this.len) {
                this.data[i] = 0;
        }
}

uint64_t moved(MyData & this, processor * ttid) {
        if(this.ttid == ttid) {
                return 0;
        }
        this.ttid = ttid;
        return 1;
}

__attribute__((noinline)) void access(MyData & this, size_t idx) {
        size_t l = this.len;
        this.data[idx % l] += 1;
}

// ==================================================
// Atomic object where a single thread can wait
// May exchanges data
struct __attribute__((aligned(128))) MySpot {
        MyThread * volatile ptr;
        size_t id;
        uint64_t _p1[16];  // padding
};

void ?{}(MySpot & this, size_t id) {
        this.ptr = 0p;
        this.id  = id;
}

// Main handshake of the code
// Single seat, first thread arriving waits
// Next threads unblocks current one and blocks in its place
// if share == true, exchange data in the process
bool put( MySpot & this, MyThread & ctx, MyData * data, bool share) {
        // Attempt to CAS our context into the seat
        for() {
                MyThread * expected = this.ptr;
                if (expected == 1p) { // Seat is closed, return
                        return true;
                }

                if (__atomic_compare_exchange_n(&this.ptr, &expected, &ctx, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) {
                        if(expected) {
                                if(share) {
                                        expected->data = data;
                                }
                                post( expected->sem );
                        }
                        break; // We got the seat
                }
        }

        // Block once on the seat
        wait(ctx.sem);

        // Someone woke us up, get the new data
        return false;
}

// Shutdown the spot
// Wake current thread and mark seat as closed
void release( MySpot & this ) {
        MyThread * val = __atomic_exchange_n(&this.ptr, 1p, __ATOMIC_SEQ_CST);
        if (!val) {
                return;
        }

        // Someone was there, release them
        post( val->sem );
}

// ==================================================
// Do some work by accessing 'cnt' cells in the array
__attribute__((noinline)) void work(MyData & data, size_t cnt_, uint64_t & state) {
        for (cnt_) {
                access(data, xorshift_13_7_17(state));
        }
}

void main(MyThread & this) {
        uint64_t state = prng();

        // Wait for start
        wait(this.sem);

        // Main loop
        for() {
                // Touch our current data, write to invalidate remote cache lines
                work(*this.data, this.cnt, state);

                // Wait on a random spot
                uint64_t idx = xorshift_13_7_17(state) % this.spots.len;
                bool closed = put(*this.spots.ptr[idx], this, this.data, this.share);

                // Check if the experiment is over
                if (closed) break;
                if ( clock_mode && stop) break;
                if (!clock_mode && this.result.count >= stop_count) break;

                // Check everything is consistent
                verify(this.data);

                // write down progress and check migrations
                processor * ttid = active_processor();
                this.result.count += 1;
                this.result.gmigs += moved(this, ttid);
                this.result.dmigs += moved(*this.data, ttid);
        }

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

void ?{}( MyThread & this, MyData * data, MySpot ** spots, size_t spot_len, size_t cnt, bool share, size_t id) {
        ((thread&)this){ bench_cluster };
        this.data = data;
        this.spots.ptr = spots;
        this.spots.len = spot_len;
        (this.sem){};
        this.result.count = 0;
        this.result.gmigs = 0;
        this.result.dmigs = 0;
        this.share = share;
        this.cnt = cnt;
        this.ttid = active_processor();
        this.id = id;
}

// ==================================================
int main(int argc, char * argv[]) {
        unsigned wsize = 2;
        unsigned wcnt  = 2;
        unsigned nspots = 0;
        bool share = false;
        cfa_option opt[] = {
                BENCH_OPT,
                { 'n', "nspots", "Number of spots where threads sleep (nthreads - nspots are active at the same time)", nspots},
                { 'w', "worksize", "Size of the array for each threads, in words (64bit)", wsize},
                { 'c', "workcnt" , "Number of words to touch when working (random pick, cells can be picked more than once)", wcnt },
                { 's', "share"   , "Pass the work data to the next thread when blocking", share, parse_truefalse }
        };
        BENCH_OPT_PARSE("cforall cycle benchmark");

        unsigned long long global_count = 0;
        unsigned long long global_gmigs = 0;
        unsigned long long global_dmigs = 0;

        if( nspots == 0 ) { nspots = nthreads - nprocs; }
        if( nspots == 0 ) {
                fprintf(stderr, "--nspots must be set or --nthreads set to something bigger than --nprocs\n");
                exit(EXIT_FAILURE);
        }

        Time start, end;
        {
                MyData * data_arrays[nthreads];
                for(i; nthreads) {
                        data_arrays[i] = malloc();
                        (*data_arrays[i]){ i, wsize };
                }

                MySpot * spots[nspots];
                for(i; nspots) {
                        spots[i] = malloc();
                        (*spots[i]){ i };
                }

                BenchCluster bc = { nprocs };
                threads_left = nprocs;
                {
                        MyThread ** threads = alloc(nthreads);
                        for(i; nthreads) {
                                threads[i] = malloc();
                                (*threads[i]){
                                        data_arrays[i],
                                        spots,
                                        nspots,
                                        wcnt,
                                        share,
                                        i
                                };
                        }

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

                        for(i; nthreads) {
                                post( threads[i]->sem );
                        }
                        wait(start, is_tty);

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

                        for(i; nthreads) {
                                post( threads[i]->sem );
                                MyThread & thrd = join( *threads[i] );
                                global_count += thrd.result.count;
                                global_gmigs += thrd.result.gmigs;
                                global_dmigs += thrd.result.dmigs;
                        }

                        for(i; nthreads) {
                                ^( *threads[i] ){};
                                free( threads[i] );
                        }
                        free( threads );
                }

                for(i; nthreads) {
                        ^( *data_arrays[i] ){};
                        free( data_arrays[i] );
                }

                for(i; nspots) {
                        ^( *spots[i] ){};
                        free( spots[i] );
                }
        }

        setlocale( LC_NUMERIC, getenv( "LANG" ) );
        printf("Duration (ms)          : %'lf\n", (end - start)`dms);
        printf("Number of processors   : %'d\n", nprocs);
        printf("Number of threads      : %'d\n", nthreads);
        printf("Total Operations(ops)  : %'15llu\n", global_count);
        printf("Work size (64bit words): %'15u\n", wsize);
        printf("Data sharing           : %s\n", share ? "On" : "Off");
        printf("Total Operations(ops)  : %'15llu\n", global_count);
        printf("Total G Migrations     : %'15llu\n", global_gmigs);
        printf("Total D Migrations     : %'15llu\n", global_dmigs);
        printf("Ops per second         : %'18.2lf\n", ((double)global_count) / (end - start)`ds);
        printf("ns per ops             : %'18.2lf\n", (end - start)`dns / global_count);
        printf("Ops per threads        : %'15llu\n", global_count / nthreads);
        printf("Ops per procs          : %'15llu\n", global_count / nprocs);
        printf("Ops/sec/procs          : %'18.2lf\n", (((double)global_count) / nprocs) / (end - start)`ds);
        printf("ns per ops/procs       : %'18.2lf\n", (end - start)`dns / (global_count / nprocs));
        fflush(stdout);
}