#ifndef NBODY_H_
#define NBODY_H_

#include <CL/cl.h>

#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#include <assert.h>
#include <string.h>
#include <cmath>
#include <malloc.h>
#include <time.h>

#include <GL/glut.h>

#define GROUP_SIZE 64
#define ITER 5

/**
 * NBody 
 * Class implements OpenCL  NBody sample
 *
 */

class NBody 
{
	public:
    cl_double  setupTime;			// Time taken to setup OpenCL resources and building kernel;
    cl_double  kernelTime;			// Time taken to run kernel and read result back;
    
    size_t maxWorkGroupSize;        // Max allowed work-items in a group;
    cl_uint maxDimensions;          // Max group dimensions allowed;
    size_t* maxWorkItemSizes;       // Max work-items sizes in each dimensions;
    cl_ulong totalLocalMemory;      // Max local memory allowed;
    cl_ulong usedLocalMemory;       // Used local memory;

    cl_float delT;                  // dT (timestep);
    cl_float espSqr;                // Softening Factor;
    cl_float* initPos;              // Initial position;
    cl_float* initVel;              // Initial velocity; 
    cl_float* vel;                  // Output velocity; 
    cl_float* refPos;               // Reference position; 
    cl_float* refVel;               // Reference velocity; 
    cl_context context;             // CL context; 
    cl_device_id *devices;          // CL device list; 
    cl_mem   updatedPos;            // Position of partciles; 
    cl_mem   updatedVel;            // Velocity of partciles; 
    cl_command_queue commandQueue;  // CL command queue; 
    cl_program program;             // CL program; 
    cl_kernel kernel;               // CL kernel; 

    cl_int  numParticles;			// Number of particles in the system;  
	const char * filename;			// The name of the file that contains the kernel code;


private:

    float random(float randMax, float randMin);
    int compareArray(const float* mat0, const float* mat1, unsigned int size);

public:
    /** 
     * Constructor 
     * Initialize member variables
     * @param name name of sample (string)
     */
    explicit NBody(std::string name)
    {
		  setupTime = 0;
		  kernelTime = 0;
		  delT = 0.005f;
		  espSqr = 50.0f;
		  initPos = NULL;
		  initVel = NULL;
		  vel = NULL;
		  refPos = NULL;
		  refVel = NULL;
		  devices = NULL;
		  maxWorkItemSizes = NULL;
		  filename = "NBody_Kernels.cl";
		  numParticles = 164;
    }

    /** 
     * Constructor 
     * Initialize member variables
     * @param name name of sample (const char*)
     */
    explicit NBody(const char* name)
    {
		  setupTime = 0;
		  kernelTime = 0;
		  delT = 0.005f;
		  espSqr = 50.0f;
		  initPos = NULL;
		  initVel = NULL;
		  vel = NULL;
		  refPos = NULL;
		  refVel = NULL;
		  devices = NULL;
		  maxWorkItemSizes = NULL;
		  filename = "NBody_Kernels.cl";
		  numParticles = 1564;
    }

    ~NBody();

    /**
     * Allocate and initialize host memory array with random values
     * @return 1 on success and 0 on failure
     */
    int setupNBody();

    /**
     * OpenCL related initialisations. 
     * Set up Context, Device list, Command Queue, Memory buffers
     * Build CL kernel program executable
     * @return 1 on success and 0 on failure
     */
    int setupCL();

    /**
     * Set values for kernels' arguments
     * @return 1 on success and 0 on failure
     */
    int setupCLKernels();

    /**
     * Enqueue calls to the kernels
     * on to the command queue, wait till end of kernel execution.
     * Get kernel start and end time if timing is enabled
     * @return 1 on success and 0 on failure
     */
    int runCLKernels();

    /**
     * Load a .cl source file as a char* and it will be used
	 * for as a parameter for creating a program
     * @return the char vector
     */
	char * load_program_source(const char *filename);

    /**
     * Reference CPU implementation of Binomial Option
     * for performance comparison
     */
    void nBodyCPUReference();

    /**
     * Override from SDKSample. Print sample stats.
     */
    void printStats();

    /**
     * Override from SDKSample, adjust width and height 
     * of execution domain, perform all sample setup
     */
    int setup();

    /**
     * Override from SDKSample
     * Run OpenCL NBody
     */
    int run();

    /**
     * Override from SDKSample
     * Cleanup memory allocations
     */
    int cleanup();
};

#endif // NBODY_H_