#define _GNU_SOURCE
#define _POSIX_SOURCE 1 

#include <stdio.h>
#include <stdlib.h>
#include <pthread.h>
#include <stdint.h>
#include <unistd.h>
#include <errno.h>
#include <limits.h>
#include <sys/stat.h>
#include <string.h>
#include <dlfcn.h>
#include <sys/time.h>

/* Non standards includes */
#include <commtech.h>
#include <specific_comm.h>


#define MAX_BLOCK_ENTRIES (page_size / sizeof(void *))
#define toString(x) doStringification(x)
#define doStringification(x) #x
#define WORDS_PER_BUF (BUF_SIZE / sizeof(uintptr_t))


static long nb_bufs_sent = 0;
long nb_prod = 0;
static int (*init_calc)(int) = NULL;
static void **(*do_calc)(void) = NULL;
static int (*end_calc)(void) = NULL;
static int shared = 0; /* We are not shared by default */
pthread_cond_t cond_cons_has_finished = PTHREAD_COND_INITIALIZER;
pthread_mutex_t mutex_cons_has_finished = PTHREAD_MUTEX_INITIALIZER;
static int init_calc_arg = 0;
static int block_reception = 1;
static int page_size = 0;

void usage(char *argv[])
{
	char format[] = "-n <num_buf> -p <num_prod> [options]";
	char options[] = "Required options :\n"
		"-n nb_buffer_sent\t\tNumber of buffer to send to another core\n"
		"\t\t\t\tBuffer size is " toString(BUF_SIZE) " bytes\n"
		"Facultative options :\n"
		"b\t\t\t\tReceive the biggest amount of data available (The default)\n"
		"-c calculation_libname arg\tLibrary to use for calculation with its argument\n"
		"\t\t\t\tThis library must implement functions in calc.h\n"
		"\t\t\t\t(default to none)\n"
		"d\t\t\t\tReceive one piece of data\n"
		"-h\t\t\t\tPrint this help\n"
		"-s <level>\t\t\tShare the same L<level> cache or not\n"
		"\t\t\t\tIf level is:\n"
		"\t\t\t\t\t> 0, then the same L<level> must be shared\n"
		"\t\t\t\t\t< 0, then different L<level> must be used\n"
		"\t\t\t\t\t= 0, then no constraint is given, only main memory (RAM) is guaranteed to be shared\n";
	printf("Usage : %s %s\n", argv[0], format);
	printf("Options :\n");
	printf("%s\n", options);
}

int do_noinit(int unused)
{
	return 0;
}

void **do_nocalc(void)
{
	static int an_int, *an_int_ptr = &an_int;

	return (void **) &an_int_ptr;
}

int do_noend(void)
{
	return 0;
}

int analyse_options(int argc, char *argv[])
{
	int opt;

	opterr = 0;
	while ((opt = getopt(argc, argv, ":hs::c:n:p:")) != -1)
	{
		switch (opt)
		{
			case 'b' :
				block_reception = 1;
				break;
			case 'c' :
				{
					struct stat file_stat;
					void *dl_descriptor;

					if (stat(optarg, &file_stat))
					{
						fprintf(stderr, "%s: %s\n", optarg, strerror(errno));
						return -1;
					}
					dl_descriptor = dlopen(optarg, RTLD_LAZY | RTLD_LOCAL);
					if (dl_descriptor == NULL)
					{
						fprintf(stderr, "dlopen error: %s\n", dlerror());
						return -1;
					}
					init_calc = (int (*)(int)) dlsym(dl_descriptor, "init_calc");
					do_calc = (void ** (*)(void)) dlsym(dl_descriptor, "do_calc");
					end_calc = (int (*)(void)) dlsym(dl_descriptor, "end_calc");
					if ((init_calc == NULL) || (do_calc == NULL) || (end_calc == NULL))
					{
						fprintf(stderr, "A symbol cannot be loaded: %s\n", dlerror());
						return -1;
					}
					if ((optind == argc) || (*argv[optind] == '-'))
					{
						fprintf(stderr, "Missing argument for -c option\n");
						return -1;
					}
					{
						char *inval;
						init_calc_arg = strtol(argv[optind], &inval, 10);
						if ((*argv[optind] == '\0') || (*inval != '\0'))
						{
							fprintf(stderr, "Option '-c' needs also an integer argument\n");
							return -1;
						}
						if ((init_calc_arg <= 0) || ((init_calc_arg == LONG_MAX) && errno == ERANGE))
						{
							fprintf(stderr, "Number of useless loop to be done between 2 send must be"
								" between 1 and %ld, both inclusive\n", LONG_MAX);
							return -1;
						}
					}
					optind++;
				}
				break;
			case 'd' :
				block_reception = 0;
				break;
			case 'h' :
				usage(argv);
				exit(EXIT_SUCCESS);
			case 'n' :
				{
					char *inval;
					nb_bufs_sent = strtol(optarg, &inval, 10);
					if ((*optarg == '\0') || (*inval != '\0'))
					{
						fprintf(stderr, "Option '-n' needs an integer argument\n");
						return -1;
					}
					if ((nb_bufs_sent <= 0) || ((nb_bufs_sent == LONG_MAX) && errno == ERANGE))
					{
						fprintf(stderr, "Number of cache lines to be sent must be between 1 and %ld, both inclusive\n", LONG_MAX);
						return -1;
					}
				}
				break;
#if 0
			case 'p' :
				{
					char *inval;
					nb_prod = strtol(optarg, &inval, 10);
					if ((*optarg == '\0') || (*inval != '\0'))
					{
						fprintf(stderr, "Option '-p' needs an integer argument\n");
						return -1;
					}
					if ((nb_prod <= 0) || ((nb_prod == LONG_MAX) && errno == ERANGE))
					{
						fprintf(stderr, "Number of producers must be between 1 and %ld, both inclusive\n", LONG_MAX);
						return -1;
					}
				}
				break;
#endif
			case 's' :
				if ((optind != argc) && (*argv[optind] != '-'))
				{
					int share_level;
					char *inval;
					share_level = strtol(argv[optind], &inval, 10);
					if ((*argv[optind] == '\0') || (*inval != '\0'))
					{
						fprintf(stderr, "Option '-p' needs an integer argument\n");
						return -1;
					}
					if ((share_level == LONG_MIN) || ((share_level == LONG_MAX) && errno == ERANGE))
					{
						fprintf(stderr, "Shared memory level must be between %ld and %ld, both inclusive\n", LONG_MIN, LONG_MAX);
						return -1;
					}
					/* TODO: Real management of shared memory level */
					if (share_level <= 0) /* -x: We wan't level x not to be shared; 0 do as we want, only memory is guaranteed to be shared */
						shared = 0;
					else
						shared = 1;
					optind++;
				}
				break;
			case '?' :
				fprintf(stderr, "Option inconnue\n");
				return -1;
			case ':' :
				fprintf(stderr, "Option %s needs an argument\n", argv[optind]);
				return -1;
			default :
				fprintf(stderr, "Error while analysing command line options\n");
				return -1;
		}
	}
	if (!nb_bufs_sent)
	{
		fprintf(stderr, "You must give the number of cache lines to be sent\n");
		return -1;
	}
#if 0
	if (!nb_prod)
	{
		fprintf(stderr, "You must give the number of producers\n");
		return -1;
	}
#endif
	if (shared && (nb_prod > 1))
	{
		fprintf(stderr, "Too many producers to fit with the consumer in processors which share a same cache\n");
		return -1;
	}
	if (do_calc == NULL)
	{
		init_calc = do_noinit;
		do_calc = do_nocalc;
		end_calc = do_noend;
	}
	printf("buf size: %lu\n", WORDS_PER_BUF);
	return 0;
}

void *producer(void *channel)
{
	int i, j;
	struct timeval tv1, tv2, tv_result;

	if (init_producer_thread(channel))
	{
		fprintf(stderr, "Initialization of thread has failed\n");
		return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
	}
	if (shared)
	{
		pthread_t tid;
		cpu_set_t cpuset;

		tid = pthread_self();
		CPU_ZERO(&cpuset);
		CPU_SET(1, &cpuset);
		if (pthread_setaffinity_np(tid, sizeof(cpu_set_t), &cpuset))
		{
			perror("pthread_setaffinity_np");
			return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
		}
	}
	else
	{
		pthread_t tid;
		cpu_set_t cpuset;

		tid = pthread_self();
		CPU_ZERO(&cpuset);
		CPU_SET(2, &cpuset);
		if (pthread_setaffinity_np(tid, sizeof(cpu_set_t), &cpuset))
		{
			perror("pthread_setaffinity_np");
			return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
		}
	}
	if (init_calc(init_calc_arg))
	{
		fprintf(stderr, "Initialization of calculation has failed\n");
		return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
	}
	gettimeofday(&tv1, NULL);
	for(i = 0; i < nb_bufs_sent; i++) {
		//printf("[%p] Send %d new CACHE_LINE\n", (void *) pthread_self(), BUF_SIZE / CACHE_LINE_SIZE);
		for(j = 0; j < WORDS_PER_BUF; j++)
			send(do_calc());
	}
	//print_results(WORDS_PER_BUF, nb_bufs_sent);
	gettimeofday(&tv2, NULL);
	tv_result.tv_sec = tv2.tv_sec - tv1.tv_sec;
	if (tv2.tv_usec < tv1.tv_usec)
	{
		tv_result.tv_usec = tv1.tv_usec - tv2.tv_usec;
		tv_result.tv_sec--;
	}
	else
		tv_result.tv_usec = tv2.tv_usec - tv1.tv_usec;
	printf("total_time: %u.%06u\n", (unsigned) tv_result.tv_sec,
		(unsigned) tv_result.tv_usec);
	if (end_calc())
	{
		fprintf(stderr, "uninitialization of calculation has failed\n");
		return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
	}
	printf("[%p] Producer finished !\n", (void*) pthread_self());
	/* 
	 * When a producer end its thread-local storage vanished. Thus,
	 * producers must finish only after consumer has stopped using them
	 */
	if (finalize_producer_thread(channel))
	{
		fprintf(stderr, "Finalization of thread has failed\n");
		return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
	}
	return NULL;
}

void on_message(void *val)
{
	//printf("Receive value: %p\n", (void *) val);
}

void *consumer(void *channel)
{
	if (shared)
	{
		pthread_t tid;
		cpu_set_t cpuset;

		tid = pthread_self();
		CPU_ZERO(&cpuset);
		CPU_SET(0, &cpuset);
		if (pthread_setaffinity_np(tid, sizeof(cpu_set_t), &cpuset))
		{
			perror("pthread_setaffinity_np");
			return NULL;
		}
	}
	if (init_consumer_thread(channel))
	{
		fprintf(stderr, "Initialization of thread has failed\n");
		return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
	}
	if (block_reception)
	{
		long long total_data_received = 0;
		void *data_buf[MAX_BLOCK_ENTRIES];

		while (total_data_received < nb_bufs_sent * WORDS_PER_BUF)
		{
			int i;
			ssize_t nb_data_received;

			nb_data_received = recv_some_data(data_buf, MAX_BLOCK_ENTRIES);
			total_data_received += nb_data_received;
			for (i = 0; i < nb_data_received; i++)
				on_message(data_buf[i]);
			//printf("[%p] Just received %d word-sized data%s\n", (void *) pthread_self(), nb_data_received, nb_data_received ? "s" : "");
		}
	}
	else
	{
		int i, j;

		for(i = 0; i < nb_bufs_sent; i++) {
			//printf("[%p] About to receive %d new cache line%s\n", (void *) pthread_self(), BUF_SIZE / CACHE_LINE_SIZE, (BUF_SIZE / CACHE_LINE_SIZE > 1) ? "s" : "");
			for(j = 0; j < WORDS_PER_BUF; j++)
				on_message(recv_one_data());
		}
	}
	if (finalize_consumer_thread(channel))
	{
		fprintf(stderr, "Finalization of thread has failed\n");
		return &page_size; /* &page_size can't be NULL, whatever NULL is bound to */
	}
	return NULL;
}

int main(int argc, char *argv[])
{
	pthread_t tid[2];
	int return_value;
	void *pthread_return_value;
	void *channel;

	return_value = EXIT_SUCCESS;
	if (analyse_options(argc, argv))
		return EXIT_FAILURE;
	page_size = sysconf(_SC_PAGE_SIZE);
	if (page_size <= 0)
		return EXIT_FAILURE;
	if (init_library())
		return EXIT_FAILURE;
	channel = create_comm_channel();
	if (channel != NULL)
	{
		pthread_create(&tid[0], NULL, producer, channel);
		pthread_create(&tid[1], NULL, consumer, channel);
		pthread_join(tid[0], &pthread_return_value);
		if (pthread_return_value != NULL)
			return_value = EXIT_FAILURE;
		pthread_join(tid[1], &pthread_return_value);
		if (pthread_return_value != NULL)
			return_value = EXIT_FAILURE;
	}
	else
		return_value = EXIT_FAILURE;
	if (destroy_comm_channel(channel))
		return_value = EXIT_FAILURE;
	if (finalize_library())
		return EXIT_FAILURE;
	return return_value;
}