#include <stdio.h>
#include <stdlib.h>
#include <string.h>


$input int NUM_DEVICES;
$input int MAX_NUM_DEVICES;
$assume NUM_DEVICES < MAX_NUM_DEVICES;
$gbarrier gbarrier = $gbarrier_create($here, NUM_DEVICES);
//struct goes here
typedef struct
{
  int device_id;
  //other variables
}process;

//kernel goes here

/*
Note that the original lines were "__kernel void square(                                                       \n" \
"   __global float* input,                                              \n" \
"   __global float* output,                                             \n" \
"   const unsigned int count)                                           \n" \

Any parser must take note of and don't input \n, "", or \ as is
__global float * input, __global float * output, int count;
*/


int main(int argc, char** argv)
{
  //get the number from clGetDeviceIDs 3rd parameter
  //int num_devices = 1;
  
  //variables from __kernel come here
  
  //from the code before
  
  //comes from clCreateBuffer
  
  //Possibly keep a list of variables, with a flag for whether they are init or not
  //Not init, malloc one from what is found in 
  //output = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * count, NULL, NULL);
  

  //came from clEnqueueWriteBuffer rather than the start of code
  
  
  //Phase after this is the definitions
  //put device_ids
  for(int i = 0; i < NUM_DEVICES; i++)
  {
    now[i].device_id = i;
    //other variables
  }
  
  //spawns processes according to devices, uses the struct for inputs
  $proc procs[NUM_DEVICES];
  for(int i = 0; i < NUM_DEVICES; i++)
  {
    procs[i] = $spawn //insert kernel function here
  }
  
  for(int i = 0; i < NUM_DEVICES; i++)
  {
    $wait(procs[i]);
  }
  
  //$barrier barrier = $barrier_create($here, gbarrier, now[i].device_id);
  //$barrier_call(barrier);
  //$barrier_destroy(barrier);
  //may use later, but not now
  
  
  
  $gbarrier_destroy(gbarrier);
  
  //use the information from clEnqueueReadBuffer
  //may have to alter later


  return 0;
}