source: CIVL/examples/translation/openclversion2.1/square.cvl@ c5b89ea

//Forget the program

#include "openCLshared.cvl"
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <civlc.h>

$input int NUM_DEVICES;
$input int MAX_NUM_DEVICES;
$assume 0 < NUM_DEVICES && NUM_DEVICES < MAX_NUM_DEVICES;

$input int DATA_SIZE;
$input int MAX_DATA_SIZE;
$assume 0 < DATA_SIZE && DATA_SIZE < MAX_NUM_DEVICES;

$input int LOCAL;
$input int MAX_LOCAL;
$assume 0 < LOCAL && LOCAL < MAX_LOCAL;
//this args struct will hold all the parameters of for the kernel function
typedef struct
{
  //Variables for kernels
  
  float * input;
  float * output;
  int count;
  
}args;

/*
  args * argument       - Takes in the struct, which is changed for every program using a different kernel
*/
cl_kernel clCreateKernel(args * argument)
{
  cl_kernel kernel;
  kernel.arguments = argument;
  
  return kernel;
}

/*
  This is the kernel that processes compute with
  int workgroup         - Gives the workgroup that a particular process came from, made by clEnqueueNDRangeKernel
  int global_id         - Gives the global_id that a particular process has, given by workfunc
  int local_id          - Gives the local_id that a particular process has, given by workfunc
  float* input          - Kernel argument
  float* output         - Kernel argument
  int count                     - Kernel argument
*/
void square(int workgroup, int global_id, int local_id, float* input, float* output, int count)
{
  //int i = get_global_id(0);
  int i = global_id;
  if (i < count)
  {
    output[i] = input[i] * input[i];
    //printf("output[%d] is %d\n", i, output[i]);
  }
}
/*
  workfunc assigns local and global ids, before calling the kernel. 
  Note: The function should be identical in all transformations except the calling of the kernel, which means that it cannot be in openCLshared.cvl
  size_t local          - The size of the workgroups, used to calculate blocks
  size_t global         - The total amount of work to be done
  cl_kernel param       - Holds the data for local_id, global_id, and the workgroup
  Use the print statement to get a better idea of what it means to split workgroups, local_ids, and global_ids
*/
void workfunc(size_t local, size_t global, cl_kernel param)
{
  for(int i = local * param.workgroup; i < local * param.workgroup + local; i++)
  {
    param.local_id = i % local;
    param.global_id = i;
    printf("My workgroup id is %d, my global id is %d, my local id is %d\n", param.workgroup, param.global_id, param.local_id);
    square(param.workgroup, param.global_id, param.local_id, ((args*)param.arguments)->input, ((args*)param.arguments)->output, ((args*)param.arguments)->count);
  }
}

/*
  Splits up and spawns processes based on global and local, using block
  TODO: remove cl_command_queue completely and put into a "just in case" file, currently not needed
  cl_command_queue commands     - Holds a queue of the order that devices are to be executed
  cl_kernel kernel                      - Holds all the arguments for the kernel, as well as local_id, global_id, and the workgroup
  size_t global                         - The total amount of work to be done
  size_t local                          - Number to split into workgroups by
*/
int clEnqueueNDRangeKernel(cl_command_queue commands, cl_kernel kernel, size_t global, size_t local)
{
  $assert(global % local == 0);
  int numworkgroups = global/local;
  cl_kernel param[numworkgroups];
  $proc procs[numworkgroups];
  //consider $parfor
  $domain(1) dom = {0 .. numworkgroups - 1};
  
  $for(int i: dom)
  {
    param[i] = kernel;
    param[i].workgroup = i;
  }
  $parfor(int i: dom)
  {
    workfunc(local, global, param[i]);
  }
  
  /*
  for(int i = 0; i < global/local; i++)
  {
    param[i] = kernel;
    param[i].workgroup = i;
    procs[i] = $spawn workfunc(local, global, param[i]);
  }
       
  //this part here is the new clFinish(commands);
  for(int i = 0; i < global/local; i++)
  {
    $wait(procs[i]);
  }
  */
  return CL_SUCCESS;
}


int main(int argc, char** argv)
{
  args * arguments;
  arguments = (args*)malloc(sizeof(args));

  float data[DATA_SIZE];              // original data set given to device
  float results[DATA_SIZE];           // results returned from device
  unsigned int correct;               // number of correct results returned
  
  size_t global;                      // global domain size for our calculation
  size_t local;                       // local domain size for our calculation
  
  cl_device_id device_id;             // compute device id 
  cl_context context;                 // compute context
  cl_command_queue commands;          // compute command queue
  //cl_program program;                 // compute program
  cl_kernel kernel;                   // compute kernel
  
  
  float * input;                       // device memory used for the input array
  float * output;                      // device memory used for the output array

  //Puts in data for input
  unsigned int count = DATA_SIZE;
  for(int i = 0; i < count; i++)
  {
    data[i] = i;
  }
  
  int err = clGetDeviceIDs(1, &device_id);
  
  //ignore clCreateContext for now, until we get an example that uses multiple ones
  
  //clCreateCommandQueue, could use context later
  commands = clCreateCommandQueue(device_id);
  
  //clCreateProgram is far different from the real version, this just stores parameters for the kernel
  //In order to make this clear, it is clCreateProgram and not something like clCreateProgramFromSource, which actually exists in openCL code
  //program = clCreateProgram(arguments);
  
  kernel = clCreateKernel(arguments);
  
  //replaces clCreateBuffer
  input = (float *) malloc(sizeof(float) * count);
  output = (float *) malloc(sizeof(float) * count);
  
  
  //replaces clEnqueueWriteBuffer, puts data into the input to be put into the kernel arguments
  memcpy(input, data, sizeof(float) * count);
  
  /*
  err = 0;
  err = clSetKernelArg(kernel, 0, sizeof(cl_mem), &input);
  err = clSetKernelArg(kernel, 1, sizeof(cl_mem), &output);
  err = clSetKernelArg(kernel, 2, sizeof(unsigned int), &count);
  */
  
  //use pointer instead of malloc + memcpy for global variables
  //((args*)kernel.arguments)->input = (float*)malloc(sizeof(float) * count);
  //memcpy(((args *)kernel.arguments)->input, input, sizeof(float) * count);
  ((args*)kernel.arguments)->input = input;
  
  //((args*)kernel.arguments)->output = (float*)malloc(sizeof(float) * count);
  //memcpy(((args*)kernel.arguments)->output, output, sizeof(float));
  ((args*)kernel.arguments)->output = output;

  ((args*)kernel.arguments)->count = count;
  //no malloc needed for non pointers
  
  //clGetKernelWorkGroupInfo would get a local size optimal for a device, but is not needed here
  local = LOCAL;
  
  global = count;
  /*
    commands holds the "order" of devices
    kernel holds program, which holds variables
    offset not implemented
  */
  err = clEnqueueNDRangeKernel(commands, kernel, global, local);
  
  //Replaces clEnqueueReadBuffer, which takes one of the saved variables and puts it out to another one
  memcpy(results, output, sizeof(float) * count);
  
  //verifies that all values in results are actually squared
  correct = 0;
  for(int i = 0; i < count; i++)
  {
    if(results[i] == data[i] * data[i])
    {
      correct++;
    }
  }
  printf("Computed '%d/%d' correct values!\n", correct, count);
  
  //TODO: Think of using void * array instead of regular arguments to make freeing easier
  
  free(((args*)kernel.arguments)->input);
  free(((args*)kernel.arguments)->output);
  /*
  free(input);
  free(output);
  */
  free(arguments);
  
  return 0;
}