#include<starpu.h>
#include <stdio.h>
#include<stdlib.h>
#include <sys/time.h>
// platform independent data types:
#include <stdint.h>     

// Number of iterations
#define NUM_ITERATIONS		10
// The window height is equal to the number of streams
#define NUM_INPUT_STREAMS	WINDOW_HEIGHT
#define WINDOW_HEIGHT		4
// The window width is equal to number of tuples required to fill in a window
#define WINDOW_WIDTH		100

// The total number of elements is equal to the window_height times window_width
#define elements			WINDOW_HEIGHT*WINDOW_WIDTH

// measure time: 
struct timeval start[NUM_ITERATIONS], end[NUM_ITERATIONS];

extern void cpu_output(void *buffers[], void *_args);
extern void output_thread_aggregation(void *buffers[], void *_args);

static struct starpu_perfmodel perf_model = {
	.type = STARPU_HISTORY_BASED,
	.symbol = "main",
};

static struct starpu_codelet cl =
{
	/*CPU implementation of the codelet */
	.cpu_funcs = { cpu_output },
	.cpu_funcs_name = { "cpu_output" },
	#ifdef STARPU_USE_CUDA
	/* CUDA implementation of the codelet */
	.cuda_funcs = { output_thread_aggregation },
	#endif
	.nbuffers = 1,
	.modes = { STARPU_RW },
	.model = &perf_model
};


int main(int argc, char **argv)
{

	uint32_t iterations_id;
	
	// create input streams:
	for (iterations_id=0; iterations_id<NUM_ITERATIONS; iterations_id++) {

		// dynamic allocation of the memory needed for all the elements
		uint32_t *window;
		window = (uint32_t*)calloc(elements, sizeof(uint32_t));
		
		// check if there's enough space for the allocation
		if(!window){
			printf("Allocation error for window - aborting.\n");
			exit(1);
		}
		
		uint64_t ag_val = 0;	// test variable to check if the cuda sum is equal to the cpu sum
		
		// initialization - fill in the window with random numbers:
		for (int i = 0; i < elements; i++) {
			window[i] = (rand()%1000);
			ag_val += window[i];
		}
		printf("TEST %lu\n", ag_val);
		
		gettimeofday(&start[iterations_id], NULL); // start time for each iteration only for StarPU initialization and time to calculate aggregated value
		
		/* initialize StarPU */
		starpu_init(NULL);
		
		/* initialize performance model */
		starpu_perfmodel_init(&perf_model);
		
		/* Tell StaPU to associate the "window" vector with the "vector_handle" */
		starpu_data_handle_t vector_handle;
		starpu_vector_data_register(&vector_handle, STARPU_MAIN_RAM, (uintptr_t)window, elements, sizeof(window[0]));
		
		/* create a synchronous task: any call to starpu_task_submit will block
		 * until it is terminated */
		struct starpu_task *task = starpu_task_create();
		task->synchronous = 1;
		task->cl = &cl; /* Pointer to the codelet defined above */
		
		/* the codelet manipulates one buffer in RW mode */
		task->handles[0] = vector_handle;
		
		uint64_t aggregated_value = 0;
		
		/* an argument is passed to the codelet, beware that this is a
		 * READ-ONLY buffer and that the codelet may be given a pointer to a
		 * COPY of the argument */
		task->cl_arg = &aggregated_value;
		task->cl_arg_size = sizeof(aggregated_value);
		
		/* submit the task to StarPU */
		starpu_task_submit(task);
		
		/* StarPU does not need to manipulate the array anymore so we can stop monitoring it */
		starpu_data_unregister(vector_handle);
		
		/* terminate StarPU */
		starpu_shutdown();
		gettimeofday(&end[iterations_id], NULL);	// stop time for each iteration after aggregation value has been calculated and StarPU has been shutted down
		
		printf("iter: %d - aggregated value: %lu\n", iterations_id, aggregated_value);
		
		//free the memory allocated on the CPU
		free(window);
	}

	uint64_t time = 0;		// variable that holds the time
	
	// calculate the time required for the calculation of the aggregated value for all iterations
	for (iterations_id=0; iterations_id<NUM_ITERATIONS; iterations_id++) {
		time += ((end[iterations_id].tv_sec * 1000000 + end[iterations_id].tv_usec) - (start[iterations_id].tv_sec * 1000000 + start[iterations_id].tv_usec));
	}

	printf("usec: %ld\n", time);

    return 0;
}