#include "rq_bench.hfa"

#include <locks.hfa>

unsigned spot_cnt = 2;
semaphore * spots;

thread BThrd {
	unsigned long long count;
	unsigned long long blocks;
	bool skip;
};

void ?{}( BThrd & this ) {
	((thread&)this){ bench_cluster };
	this.count  = 0;
	this.blocks = 0;
	this.skip = false;
}

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

void main( BThrd & this ) with( this ) {
	park();
	for() {
		uint32_t r = prng(this);
		semaphore & sem = spots[r % spot_cnt];
		if(!skip) V( sem );
		blocks += P( sem );
		skip = false;

		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[]) {
	cfa_option opt[] = {
		BENCH_OPT,
		{ 's', "spots", "Number of spots in the system", spot_cnt }
	};
	BENCH_OPT_PARSE("cforall churn benchmark");

	{
		unsigned long long global_counter = 0;
		unsigned long long global_blocks  = 0;
		Time start, end;
		BenchCluster bc = { nprocs };
		{
			spots = aalloc(spot_cnt);
			for(i; spot_cnt) {
				(spots[i]){ 0 };
			}

			threads_left = nthreads;
			BThrd ** threads = alloc(nthreads);
			for(i; nthreads ) {
				BThrd & t = *(threads[i] = malloc());
				(t){};
				t.skip = i < spot_cnt;
			}
			printf("Starting\n");

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

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

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

			for(i; spot_cnt) {
				for(10000) V( spots[i] );
			}

			for(i; nthreads) {
				BThrd & thrd = join( *threads[i] );
				global_counter += thrd.count;
				global_blocks  += thrd.blocks;
				delete(threads[i]);
			}

			free(spots);
			free(threads);
		}

		printf("Duration (ms)        : %'lf\n", (end - start)`dms);
		printf("Number of processors : %'d\n", nprocs);
		printf("Number of threads    : %'d\n", nthreads);
		printf("Number of spots      : %'d\n", spot_cnt);
		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 / nthreads);
		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;
}