#include "rq_bench.hfa"

thread Partner {
	Partner * partner;
	unsigned long long count;
};

void ?{}( Partner & this ) {
	((thread&)this){ bench_cluster };
}

void main( Partner & this ) {
	this.count = 0;
	for() {
		park();
		unpark( *this.partner );
		this.count ++;
		if( clock_mode && stop) break;
		if(!clock_mode && this.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 tthreads = nthreads * ring_size;
		Time start, end;
		BenchCluster bc = { nprocs };
		{
			threads_left = tthreads;
			Partner threads[tthreads];
			for(i; tthreads) {
				unsigned pi = (i + nthreads) % tthreads;
				threads[i].partner = &threads[pi];
			}
			printf("Starting\n");

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

			for(i; nthreads) {
				unpark( threads[i] );
			}
			wait(start, is_tty);

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

			for(i; tthreads) {
				global_counter += join( threads[i] ).count;
			}
		}

		printf("Took %'ld ms\n", (end - start)`ms);
		printf("Yields per second   : %'18.2lf\n", ((double)global_counter) / (end - start)`s);
		printf("ns per yields       : %'18.2lf\n", ((double)(end - start)`ns) / global_counter);
		printf("Total yields        : %'15llu\n", global_counter);
		printf("Yields per threads  : %'15llu\n", global_counter / tthreads);
		printf("Yields per procs    : %'15llu\n", global_counter / nprocs);
		printf("Yields/sec/procs    : %'18.2lf\n", (((double)global_counter) / nprocs) / (end - start)`s);
		printf("ns per yields/procs : %'18.2lf\n", ((double)(end - start)`ns) / (global_counter / nprocs));
		fflush(stdout);
	}

	return 0;
}