#include "OpenCLChiSquare.h"

double OpenCLChiSquare::ocl_sum(cl_mem data, int length) {

  
  int ierr;
  //calc number of thread sper workgroup and nr of work groups
  size_t work_size_sum = 128;
  size_t work_items = (size_t)length;
  if (length % work_size_sum > 0)
    work_items = (length / work_size_sum + 1) * work_size_sum;

  int work_groups = length / work_size_sum + 1;
  
  //create tmp array for partial sums
  cl_mem tmp_ptr;

  double *partial_sums = new double[work_groups];
  tmp_ptr = m_oclbase->ocl_allocateMemory(work_groups * sizeof(double), ierr);
  
  //execute sum kernel
  m_oclbase->ocl_createKernel("parallelReductionSum");
  m_oclbase->ocl_setKernelArg(0, sizeof(cl_mem), &data);
  m_oclbase->ocl_setKernelArg(1, sizeof(cl_mem), &tmp_ptr);
  m_oclbase->ocl_setKernelArg(2, work_size_sum*sizeof(double), NULL);
  m_oclbase->ocl_setKernelArg(3, sizeof(int), &length); 
  m_oclbase->ocl_executeKernel(1, &work_items, &work_size_sum);
  
  //read partial sums and free temp mempry
  m_oclbase->ocl_readData(tmp_ptr, partial_sums, sizeof(double)*work_groups);
  m_oclbase->ocl_freeMemory(tmp_ptr);
  
  //sumup partial sums on the host
  double result = 0;
  for (int i = 0; i < work_groups; i++)
    result += partial_sums[i];

  delete[] partial_sums;

  return result;

}

int OpenCLChiSquare::ocl_PHistoTFFcn(void *mem_data, void *mem_par, void *mem_result, 
				     double fTimeResolution, double fRebin,
				     int sensors, int length, int numpar,
				     double &result)
{

  //set number of work items and work group sizes for kernel execution
  size_t work_size = 128;  

  size_t work_items = (size_t)length * sensors;
  if (length % work_size > 0)
    work_items = (length / work_size + 1) * work_size;

  cl_mem data = (cl_mem)mem_data;
  cl_mem par = (cl_mem)mem_par;
  cl_mem chi = (cl_mem)mem_result;

  //load and execute PHistotFFcn kernel
  m_oclbase->ocl_createKernel("kernelPHistoTFFcn");
  m_oclbase->ocl_setKernelArg(0, sizeof(cl_mem), &data);
  m_oclbase->ocl_setKernelArg(1, sizeof(cl_mem), &par);
  m_oclbase->ocl_setKernelArg(2, sizeof(cl_mem), &chi);
  m_oclbase->ocl_setKernelArg(3, sizeof(double), &fTimeResolution);
  m_oclbase->ocl_setKernelArg(4, sizeof(double), &fRebin);
  m_oclbase->ocl_setKernelArg(5, sizeof(int), &length);
  m_oclbase->ocl_setKernelArg(6, sizeof(int), &sensors);
  m_oclbase->ocl_setKernelArg(7, sizeof(int), &numpar);
  m_oclbase->ocl_setKernelArg(8, sizeof(double)*numpar, NULL);
  m_oclbase->ocl_executeKernel(1, &work_items, &work_size);

  result = ocl_sum(chi, sensors*length);
  
  return DKS_SUCCESS;
}

int OpenCLChiSquare::ocl_singleGaussTF(void *mem_data, void *mem_t0, void *mem_par, void *mem_result,
				       double fTimeResolution, double fRebin, double fGoodBinOffset,
				       int sensors, int length, int numpar,
				       double &result)
{

  //set number of work items and work group sizes for kernel execution
  size_t work_size = 128;  
  size_t work_items = (size_t)length * sensors;
  if (length % work_size > 0)
    work_items = (length / work_size + 1) * work_size;

  cl_mem data = (cl_mem)mem_data;
  cl_mem t0 = (cl_mem)mem_t0;
  cl_mem par = (cl_mem)mem_par;
  cl_mem chi = (cl_mem)mem_result;

  //load and execute PHistotFFcn kernel
  m_oclbase->ocl_createKernel("kernelSingleGaussTF");
  m_oclbase->ocl_setKernelArg(0, sizeof(cl_mem), &data);
  m_oclbase->ocl_setKernelArg(1, sizeof(cl_mem), &t0);
  m_oclbase->ocl_setKernelArg(2, sizeof(cl_mem), &par);
  m_oclbase->ocl_setKernelArg(3, sizeof(cl_mem), &chi);
  m_oclbase->ocl_setKernelArg(4, sizeof(double), &fTimeResolution);
  m_oclbase->ocl_setKernelArg(5, sizeof(double), &fRebin);
  m_oclbase->ocl_setKernelArg(6, sizeof(double), &fGoodBinOffset);
  m_oclbase->ocl_setKernelArg(7, sizeof(int), &length);
  m_oclbase->ocl_setKernelArg(8, sizeof(int), &sensors);
  m_oclbase->ocl_setKernelArg(9, sizeof(int), &numpar);
  m_oclbase->ocl_setKernelArg(10, sizeof(double)*numpar, NULL);
  m_oclbase->ocl_executeKernel(1, &work_items, &work_size);

  result = ocl_sum(chi, length);
  
  return DKS_SUCCESS;

}


int OpenCLChiSquare::ocl_doubleLorentzTF(void *mem_data, void *mem_t0, void *mem_par, void *mem_result,
					 double fTimeResolution, double fRebin, double fGoodBinOffset,
					 int sensors, int length, int numpar,
					 double &result)
{

  //set number of work items and work group sizes for kernel execution
  size_t work_size = 128;  
  size_t work_items = (size_t)length * sensors;
  if (length % work_size > 0)
    work_items = (length / work_size + 1) * work_size;

  cl_mem data = (cl_mem)mem_data;
  cl_mem t0 = (cl_mem)mem_t0;
  cl_mem par = (cl_mem)mem_par;
  cl_mem chi = (cl_mem)mem_result;

  //load and execute PHistotFFcn kernel
  m_oclbase->ocl_createKernel("kernelDoubleLorentzTF");
  m_oclbase->ocl_setKernelArg(0, sizeof(cl_mem), &data);
  m_oclbase->ocl_setKernelArg(1, sizeof(cl_mem), &t0);
  m_oclbase->ocl_setKernelArg(2, sizeof(cl_mem), &par);
  m_oclbase->ocl_setKernelArg(3, sizeof(cl_mem), &chi);
  m_oclbase->ocl_setKernelArg(4, sizeof(double), &fTimeResolution);
  m_oclbase->ocl_setKernelArg(5, sizeof(double), &fRebin);
  m_oclbase->ocl_setKernelArg(6, sizeof(double), &fGoodBinOffset);
  m_oclbase->ocl_setKernelArg(7, sizeof(int), &length);
  m_oclbase->ocl_setKernelArg(8, sizeof(int), &sensors);
  m_oclbase->ocl_setKernelArg(9, sizeof(int), &numpar);
  m_oclbase->ocl_setKernelArg(10, sizeof(double)*numpar, NULL);
  m_oclbase->ocl_executeKernel(1, &work_items, &work_size);

  result = ocl_sum(chi, length);
  
  return DKS_SUCCESS;

}