C++ dt_opencl_enqueue_kernel_2d函数代码示例

int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out,
               const dt_iop_roi_t *const roi_in, const dt_iop_roi_t *const roi_out)
{
  dt_iop_highlights_data_t *d = (dt_iop_highlights_data_t *)piece->data;
  dt_iop_highlights_global_data_t *gd = (dt_iop_highlights_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
  const float clip = d->clip
                     * fminf(piece->pipe->processed_maximum[0],
                             fminf(piece->pipe->processed_maximum[1], piece->pipe->processed_maximum[2]));
  const int filters = dt_image_filter(&piece->pipe->image);
  if(dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) || !filters)
  {
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f_clip, 0, sizeof(cl_mem), (void *)&dev_in);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f_clip, 1, sizeof(cl_mem), (void *)&dev_out);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f_clip, 2, sizeof(int), (void *)&width);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f_clip, 3, sizeof(int), (void *)&height);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f_clip, 4, sizeof(int), (void *)&d->mode);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f_clip, 5, sizeof(float), (void *)&clip);
    err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights_4f_clip, sizes);
    if(err != CL_SUCCESS) goto error;
  }
  else
  {
    const int kernel
        = (d->mode == DT_IOP_HIGHLIGHTS_LCH) ? gd->kernel_highlights_1f_lch : gd->kernel_highlights_1f_clip;

    dt_opencl_set_kernel_arg(devid, kernel, 0, sizeof(cl_mem), (void *)&dev_in);
    dt_opencl_set_kernel_arg(devid, kernel, 1, sizeof(cl_mem), (void *)&dev_out);
    dt_opencl_set_kernel_arg(devid, kernel, 2, sizeof(int), (void *)&width);
    dt_opencl_set_kernel_arg(devid, kernel, 3, sizeof(int), (void *)&height);
    dt_opencl_set_kernel_arg(devid, kernel, 4, sizeof(float), (void *)&clip);
    dt_opencl_set_kernel_arg(devid, kernel, 5, sizeof(int), (void *)&roi_out->x);
    dt_opencl_set_kernel_arg(devid, kernel, 6, sizeof(int), (void *)&roi_out->y);
    dt_opencl_set_kernel_arg(devid, kernel, 7, sizeof(int), (void *)&filters);
    err = dt_opencl_enqueue_kernel_2d(devid, kernel, sizes);
    if(err != CL_SUCCESS) goto error;
  }

  // update processed maximum
  const float m = fmaxf(fmaxf(
        piece->pipe->processed_maximum[0],
        piece->pipe->processed_maximum[1]),
      piece->pipe->processed_maximum[2]);
  for(int k=0;k<3;k++) piece->pipe->processed_maximum[k] = m;

  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_highlights] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_relight_data_t *data = (dt_iop_relight_data_t *)piece->data;
  dt_iop_relight_global_data_t *gd = (dt_iop_relight_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  const float center = data->center;
  const float wings = data->width;
  const float ev = data->ev;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1};

  dt_opencl_set_kernel_arg(devid, gd->kernel_relight, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_relight, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_relight, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_relight, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_relight, 4, sizeof(float), (void *)&center);
  dt_opencl_set_kernel_arg(devid, gd->kernel_relight, 5, sizeof(float), (void *)&wings);
  dt_opencl_set_kernel_arg(devid, gd->kernel_relight, 6, sizeof(float), (void *)&ev);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_relight, sizes);
  if(err != CL_SUCCESS) goto error;
  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_relight] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out,
               const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_colorcorrection_data_t *d = (dt_iop_colorcorrection_data_t *)piece->data;
  dt_iop_colorcorrection_global_data_t *gd = (dt_iop_colorcorrection_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;

  const int width = roi_out->width;
  const int height = roi_out->height;

  size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 0, sizeof(cl_mem), &dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 1, sizeof(cl_mem), &dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 2, sizeof(int), &width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 3, sizeof(int), &height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 4, sizeof(float), &d->saturation);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 5, sizeof(float), &d->a_scale);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 6, sizeof(float), &d->a_base);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 7, sizeof(float), &d->b_scale);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcorrection, 8, sizeof(float), &d->b_base);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_colorcorrection, sizes);
  if(err != CL_SUCCESS) goto error;

  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_colorcorrection] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_highlights_data_t *d = (dt_iop_highlights_data_t *)piece->data;
  dt_iop_highlights_global_data_t *gd = (dt_iop_highlights_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1};
  const float clip = d->clip * fminf(piece->pipe->processed_maximum[0], fminf(piece->pipe->processed_maximum[1], piece->pipe->processed_maximum[2]));
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 4, sizeof(int), (void *)&d->mode);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 5, sizeof(float), (void *)&clip);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 6, sizeof(float), (void *)&d->blendL);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 7, sizeof(float), (void *)&d->blendC);
  dt_opencl_set_kernel_arg(devid, gd->kernel_highlights, 8, sizeof(float), (void *)&d->blendh);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights, sizes);
  if(err != CL_SUCCESS) goto error;
  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_highlights] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out,
               const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_exposure_data_t *d = (dt_iop_exposure_data_t *)piece->data;
  dt_iop_exposure_global_data_t *gd = (dt_iop_exposure_global_data_t *)self->data;

  commit_params_late(self, piece);

  cl_int err = -999;
  const float black = d->black;
  const float white = exposure2white(d->exposure);
  const float scale = 1.0 / (white - black);
  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
  dt_opencl_set_kernel_arg(devid, gd->kernel_exposure, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_exposure, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_exposure, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_exposure, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_exposure, 4, sizeof(float), (void *)&black);
  dt_opencl_set_kernel_arg(devid, gd->kernel_exposure, 5, sizeof(float), (void *)&scale);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_exposure, sizes);
  if(err != CL_SUCCESS) goto error;
  for(int k = 0; k < 3; k++) piece->pipe->processed_maximum[k] *= scale;

  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_exposure] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_colorcontrast_data_t *data = (dt_iop_colorcontrast_data_t *)piece->data;
  dt_iop_colorcontrast_global_data_t *gd = (dt_iop_colorcontrast_global_data_t *)self->data;
  cl_int err = -999;

  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  float scale[4] =  { 1.0f, data->a_steepness, data->b_steepness, 1.0f };
  float offset[4] = { 0.0f, data->a_offset, data->b_offset, 0.0f };

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1};

  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcontrast, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcontrast, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcontrast, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcontrast, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcontrast, 4, 4*sizeof(float), (void *)&scale);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorcontrast, 5, 4*sizeof(float), (void *)&offset);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_colorcontrast, sizes);

  if(err != CL_SUCCESS) goto error;
  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_colorcontrast] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out,
               const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_colisa_data_t *d = (dt_iop_colisa_data_t *)piece->data;
  dt_iop_colisa_global_data_t *gd = (dt_iop_colisa_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;

  const int width = roi_in->width;
  const int height = roi_in->height;

  const float saturation = d->saturation;

  cl_mem dev_cm = NULL;
  cl_mem dev_ccoeffs = NULL;
  cl_mem dev_lm = NULL;
  cl_mem dev_lcoeffs = NULL;

  dev_cm = dt_opencl_copy_host_to_device(devid, d->ctable, 256, 256, sizeof(float));
  if(dev_cm == NULL) goto error;
  dev_ccoeffs = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * 3, d->cunbounded_coeffs);
  if(dev_ccoeffs == NULL) goto error;
  dev_lm = dt_opencl_copy_host_to_device(devid, d->ltable, 256, 256, sizeof(float));
  if(dev_lm == NULL) goto error;
  dev_lcoeffs = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * 3, d->lunbounded_coeffs);
  if(dev_lcoeffs == NULL) goto error;

  size_t sizes[3];
  sizes[0] = ROUNDUPWD(width);
  sizes[1] = ROUNDUPWD(height);
  sizes[2] = 1;
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 4, sizeof(float), (void *)&saturation);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 5, sizeof(cl_mem), (void *)&dev_cm);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 6, sizeof(cl_mem), (void *)&dev_ccoeffs);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 7, sizeof(cl_mem), (void *)&dev_lm);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colisa, 8, sizeof(cl_mem), (void *)&dev_lcoeffs);

  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_colisa, sizes);
  if(err != CL_SUCCESS) goto error;

  if(dev_lcoeffs != NULL) dt_opencl_release_mem_object(dev_lcoeffs);
  if(dev_lm != NULL) dt_opencl_release_mem_object(dev_lm);
  if(dev_ccoeffs != NULL) dt_opencl_release_mem_object(dev_ccoeffs);
  if(dev_cm != NULL) dt_opencl_release_mem_object(dev_cm);
  return TRUE;

error:
  if(dev_lcoeffs != NULL) dt_opencl_release_mem_object(dev_lcoeffs);
  if(dev_lm != NULL) dt_opencl_release_mem_object(dev_lm);
  if(dev_ccoeffs != NULL) dt_opencl_release_mem_object(dev_ccoeffs);
  if(dev_cm != NULL) dt_opencl_release_mem_object(dev_cm);
  dt_print(DT_DEBUG_OPENCL, "[opencl_colisa] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_highlights_data_t *d = (dt_iop_highlights_data_t *)piece->data;
  dt_iop_highlights_global_data_t *gd = (dt_iop_highlights_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1};
  const float clip = d->clip * fminf(piece->pipe->processed_maximum[0], fminf(piece->pipe->processed_maximum[1], piece->pipe->processed_maximum[2]));
  const int filters = dt_image_flipped_filter(&piece->pipe->image);
  if(piece->pipe->type == DT_DEV_PIXELPIPE_PREVIEW || !filters)
  {
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f, 0, sizeof(cl_mem), (void *)&dev_in);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f, 1, sizeof(cl_mem), (void *)&dev_out);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f, 2, sizeof(int), (void *)&width);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f, 3, sizeof(int), (void *)&height);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f, 4, sizeof(int), (void *)&d->mode);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_4f, 5, sizeof(float), (void *)&clip);
    err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights_4f, sizes);
    if(err != CL_SUCCESS) goto error;
  }
  else
  {
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 0, sizeof(cl_mem), (void *)&dev_in);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 1, sizeof(cl_mem), (void *)&dev_out);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 2, sizeof(int), (void *)&width);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 3, sizeof(int), (void *)&height);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 4, sizeof(int), (void *)&d->mode);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 5, sizeof(float), (void *)&clip);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 6, sizeof(int), (void *)&roi_out->x);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 7, sizeof(int), (void *)&roi_out->y);
    dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f, 8, sizeof(int), (void *)&filters);
    err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights_1f, sizes);
    if(err != CL_SUCCESS) goto error;
  }
  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_highlights] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
    dt_iop_temperature_data_t *d = (dt_iop_temperature_data_t *)piece->data;
    dt_iop_temperature_global_data_t *gd = (dt_iop_temperature_global_data_t *)self->data;

    const int devid = piece->pipe->devid;
    const int filters = dt_image_flipped_filter(&piece->pipe->image);
    float coeffs[3] = {d->coeffs[0], d->coeffs[1], d->coeffs[2]};
    cl_mem dev_coeffs = NULL;
    cl_int err = -999;
    int kernel = -1;

    if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && filters && piece->pipe->image.bpp != 4)
    {
        kernel = gd->kernel_whitebalance_1ui;
        for(int k=0; k<3; k++) coeffs[k] /= 65535.0f;
    }
    else if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && filters && piece->pipe->image.bpp == 4)
    {
        kernel = gd->kernel_whitebalance_1f;
    }
    else
    {
        kernel = gd->kernel_whitebalance_4f;
    }

    dev_coeffs = dt_opencl_copy_host_to_device_constant(devid, sizeof(float)*3, coeffs);
    if (dev_coeffs == NULL) goto error;

    const int width = roi_in->width;
    const int height = roi_in->height;

    size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1};
    dt_opencl_set_kernel_arg(devid, kernel, 0, sizeof(cl_mem), (void *)&dev_in);
    dt_opencl_set_kernel_arg(devid, kernel, 1, sizeof(cl_mem), (void *)&dev_out);
    dt_opencl_set_kernel_arg(devid, kernel, 2, sizeof(int), (void *)&width);
    dt_opencl_set_kernel_arg(devid, kernel, 3, sizeof(int), (void *)&height);
    dt_opencl_set_kernel_arg(devid, kernel, 4, sizeof(cl_mem), (void *)&dev_coeffs);
    dt_opencl_set_kernel_arg(devid, kernel, 5, sizeof(uint32_t), (void *)&filters);
    dt_opencl_set_kernel_arg(devid, kernel, 6, sizeof(uint32_t), (void *)&roi_out->x);
    dt_opencl_set_kernel_arg(devid, kernel, 7, sizeof(uint32_t), (void *)&roi_out->y);
    err = dt_opencl_enqueue_kernel_2d(devid, kernel, sizes);
    if(err != CL_SUCCESS) goto error;

    dt_opencl_release_mem_object(dev_coeffs);
    for(int k=0; k<3; k++)
        piece->pipe->processed_maximum[k] = d->coeffs[k] * piece->pipe->processed_maximum[k];
    return TRUE;

error:
    if (dev_coeffs != NULL) dt_opencl_release_mem_object(dev_coeffs);
    dt_print(DT_DEBUG_OPENCL, "[opencl_white_balance] couldn't enqueue kernel! %d\n", err);
    return FALSE;
}

int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out,
               const dt_iop_roi_t *const roi_in, const dt_iop_roi_t *const roi_out)
{
  dt_iop_zonesystem_data_t *data = (dt_iop_zonesystem_data_t *)piece->data;
  dt_iop_zonesystem_global_data_t *gd = (dt_iop_zonesystem_global_data_t *)self->data;
  cl_mem dev_zmo, dev_zms = NULL;
  cl_int err = -999;

  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  /* calculate zonemap */
  const int size = data->params.size;
  float zonemap[MAX_ZONE_SYSTEM_SIZE] = { -1 };
  float zonemap_offset[ROUNDUP(MAX_ZONE_SYSTEM_SIZE, 16)] = { -1 };
  float zonemap_scale[ROUNDUP(MAX_ZONE_SYSTEM_SIZE, 16)] = { -1 };

  _iop_zonesystem_calculate_zonemap(&(data->params), zonemap);

  /* precompute scale and offset */
  for(int k = 0; k < size - 1; k++) zonemap_scale[k] = (zonemap[k + 1] - zonemap[k]) * (size - 1);
  for(int k = 0; k < size - 1; k++) zonemap_offset[k] = 100.0f * ((k + 1) * zonemap[k] - k * zonemap[k + 1]);

  dev_zmo = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * ROUNDUP(MAX_ZONE_SYSTEM_SIZE, 16),
                                                   zonemap_offset);
  if(dev_zmo == NULL) goto error;
  dev_zms = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * ROUNDUP(MAX_ZONE_SYSTEM_SIZE, 16),
                                                   zonemap_scale);
  if(dev_zms == NULL) goto error;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };

  dt_opencl_set_kernel_arg(devid, gd->kernel_zonesystem, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_zonesystem, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_zonesystem, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_zonesystem, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_zonesystem, 4, sizeof(int), (void *)&size);
  dt_opencl_set_kernel_arg(devid, gd->kernel_zonesystem, 5, sizeof(cl_mem), (void *)&dev_zmo);
  dt_opencl_set_kernel_arg(devid, gd->kernel_zonesystem, 6, sizeof(cl_mem), (void *)&dev_zms);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_zonesystem, sizes);

  if(err != CL_SUCCESS) goto error;
  dt_opencl_release_mem_object(dev_zmo);
  dt_opencl_release_mem_object(dev_zms);
  return TRUE;

error:
  if(dev_zmo != NULL) dt_opencl_release_mem_object(dev_zmo);
  if(dev_zms != NULL) dt_opencl_release_mem_object(dev_zms);
  dt_print(DT_DEBUG_OPENCL, "[opencl_zonesystem] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out,
               const dt_iop_roi_t *const roi_in, const dt_iop_roi_t *const roi_out)
{
  dt_iop_invert_data_t *d = (dt_iop_invert_data_t *)piece->data;
  dt_iop_invert_global_data_t *gd = (dt_iop_invert_global_data_t *)self->data;

  const int devid = piece->pipe->devid;
  const uint32_t filters = piece->pipe->dsc.filters;
  cl_mem dev_color = NULL;
  cl_int err = -999;
  int kernel = -1;

  float film_rgb_f[4] = { d->color[0], d->color[1], d->color[2], d->color[3] };

  if(filters)
  {
    kernel = gd->kernel_invert_1f;

    const float *const m = piece->pipe->dsc.processed_maximum;
    for(int c = 0; c < 4; c++) film_rgb_f[c] *= m[c];
  }
  else
  {
    kernel = gd->kernel_invert_4f;
  }

  dev_color = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * 3, film_rgb_f);
  if(dev_color == NULL) goto error;

  const int width = roi_in->width;
  const int height = roi_in->height;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
  dt_opencl_set_kernel_arg(devid, kernel, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, kernel, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, kernel, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, kernel, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, kernel, 4, sizeof(cl_mem), (void *)&dev_color);
  dt_opencl_set_kernel_arg(devid, kernel, 5, sizeof(uint32_t), (void *)&filters);
  dt_opencl_set_kernel_arg(devid, kernel, 6, sizeof(uint32_t), (void *)&roi_out->x);
  dt_opencl_set_kernel_arg(devid, kernel, 7, sizeof(uint32_t), (void *)&roi_out->y);
  err = dt_opencl_enqueue_kernel_2d(devid, kernel, sizes);
  if(err != CL_SUCCESS) goto error;

  dt_opencl_release_mem_object(dev_color);
  for(int k = 0; k < 4; k++) piece->pipe->dsc.processed_maximum[k] = 1.0f;
  return TRUE;

error:
  dt_opencl_release_mem_object(dev_color);
  dt_print(DT_DEBUG_OPENCL, "[opencl_invert] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int process_cl(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out,
               const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_colorout_data_t *d = (dt_iop_colorout_data_t *)piece->data;
  dt_iop_colorout_global_data_t *gd = (dt_iop_colorout_global_data_t *)self->data;
  cl_mem dev_m = NULL, dev_r = NULL, dev_g = NULL, dev_b = NULL, dev_coeffs = NULL;

  cl_int err = -999;
  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };

  dev_m = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * 9, d->cmatrix);
  if(dev_m == NULL) goto error;
  dev_r = dt_opencl_copy_host_to_device(devid, d->lut[0], 256, 256, sizeof(float));
  if(dev_r == NULL) goto error;
  dev_g = dt_opencl_copy_host_to_device(devid, d->lut[1], 256, 256, sizeof(float));
  if(dev_g == NULL) goto error;
  dev_b = dt_opencl_copy_host_to_device(devid, d->lut[2], 256, 256, sizeof(float));
  if(dev_b == NULL) goto error;
  dev_coeffs
      = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * 3 * 3, (float *)d->unbounded_coeffs);
  if(dev_coeffs == NULL) goto error;
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 4, sizeof(cl_mem), (void *)&dev_m);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 5, sizeof(cl_mem), (void *)&dev_r);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 6, sizeof(cl_mem), (void *)&dev_g);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 7, sizeof(cl_mem), (void *)&dev_b);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorout, 8, sizeof(cl_mem), (void *)&dev_coeffs);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_colorout, sizes);
  if(err != CL_SUCCESS) goto error;
  dt_opencl_release_mem_object(dev_m);
  dt_opencl_release_mem_object(dev_r);
  dt_opencl_release_mem_object(dev_g);
  dt_opencl_release_mem_object(dev_b);
  dt_opencl_release_mem_object(dev_coeffs);
  return TRUE;

error:
  if(dev_m != NULL) dt_opencl_release_mem_object(dev_m);
  if(dev_r != NULL) dt_opencl_release_mem_object(dev_r);
  if(dev_g != NULL) dt_opencl_release_mem_object(dev_g);
  if(dev_b != NULL) dt_opencl_release_mem_object(dev_b);
  if(dev_coeffs != NULL) dt_opencl_release_mem_object(dev_coeffs);
  dt_print(DT_DEBUG_OPENCL, "[opencl_colorout] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_tonecurve_data_t *d = (dt_iop_tonecurve_data_t *)piece->data;
  dt_iop_tonecurve_global_data_t *gd = (dt_iop_tonecurve_global_data_t *)self->data;
  cl_mem dev_L, dev_a, dev_b = NULL;
  cl_mem dev_coeffs = NULL;
  cl_int err = -999;

  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;
  const int autoscale_ab = d->autoscale_ab;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1};
  dev_L = dt_opencl_copy_host_to_device(devid, d->table[ch_L], 256, 256, sizeof(float));
  dev_a = dt_opencl_copy_host_to_device(devid, d->table[ch_a], 256, 256, sizeof(float));
  dev_b = dt_opencl_copy_host_to_device(devid, d->table[ch_b], 256, 256, sizeof(float));
  if (dev_L == NULL || dev_a == NULL || dev_b == NULL) goto error;

  dev_coeffs = dt_opencl_copy_host_to_device_constant(devid, sizeof(float)*3, d->unbounded_coeffs);
  if (dev_coeffs == NULL) goto error;
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 4, sizeof(cl_mem), (void *)&dev_L);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 5, sizeof(cl_mem), (void *)&dev_a);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 6, sizeof(cl_mem), (void *)&dev_b);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 7, sizeof(int), (void *)&autoscale_ab);
  dt_opencl_set_kernel_arg(devid, gd->kernel_tonecurve, 8, sizeof(cl_mem), (void *)&dev_coeffs);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_tonecurve, sizes);

  if(err != CL_SUCCESS) goto error;
  dt_opencl_release_mem_object(dev_L);
  dt_opencl_release_mem_object(dev_a);
  dt_opencl_release_mem_object(dev_b);
  dt_opencl_release_mem_object(dev_coeffs);
  return TRUE;

error:
  if (dev_L != NULL) dt_opencl_release_mem_object(dev_L);
  if (dev_a != NULL) dt_opencl_release_mem_object(dev_a);
  if (dev_b != NULL) dt_opencl_release_mem_object(dev_b);
  if (dev_coeffs != NULL) dt_opencl_release_mem_object(dev_coeffs);
  dt_print(DT_DEBUG_OPENCL, "[opencl_tonecurve] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_iop_colorize_data_t *data = (dt_iop_colorize_data_t *)piece->data;
  dt_iop_colorize_global_data_t *gd = (dt_iop_colorize_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;
  const int width = roi_in->width;
  const int height = roi_in->height;

  /* create Lab */
  float rgb[3]={0}, XYZ[3]={0}, Lab[3]={0};
  hsl2rgb(rgb,data->hue, data->saturation, data->lightness/100.0);

  XYZ[0] = (rgb[0] * 0.5767309) + (rgb[1] * 0.1855540) + (rgb[2] * 0.1881852);
  XYZ[1] = (rgb[0] * 0.2973769) + (rgb[1] * 0.6273491) + (rgb[2] * 0.0752741);
  XYZ[2] = (rgb[0] * 0.0270343) + (rgb[1] * 0.0706872) + (rgb[2] * 0.9911085);
  
  dt_XYZ_to_Lab(XYZ,Lab);


  /* a/b components */
  const float L = Lab[0];
  const float a = Lab[1];
  const float b = Lab[2];
  const float mix = data->source_lightness_mix/100.0f;

  size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1};

  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 0, sizeof(cl_mem), (void *)&dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 1, sizeof(cl_mem), (void *)&dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 2, sizeof(int), (void *)&width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 3, sizeof(int), (void *)&height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 4, sizeof(float), (void *)&mix);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 5, sizeof(float), (void *)&L);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 6, sizeof(float), (void *)&a);
  dt_opencl_set_kernel_arg(devid, gd->kernel_colorize, 7, sizeof(float), (void *)&b);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_colorize, sizes);
  if(err != CL_SUCCESS) goto error;
  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_colorize] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}

int
process_cl (struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, cl_mem dev_in, cl_mem dev_out, const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
  dt_develop_t *dev = self->dev;
  dt_iop_overexposed_global_data_t *gd = (dt_iop_overexposed_global_data_t *)self->data;

  cl_int err = -999;
  const int devid = piece->pipe->devid;

  const int width = roi_out->width;
  const int height = roi_out->height;

  const float lower  = dev->overexposed.lower / 100.0f;
  const float upper  = dev->overexposed.upper / 100.0f;
  const int colorscheme = dev->overexposed.colorscheme;

  const float *upper_color  = dt_iop_overexposed_colors[colorscheme][0];
  const float *lower_color = dt_iop_overexposed_colors[colorscheme][1];

  if(!dev->overexposed.enabled || !dev->gui_attached)
  {
    size_t origin[] = { 0, 0, 0};
    size_t region[] = { width, height, 1};
    err = dt_opencl_enqueue_copy_image(devid, dev_in, dev_out, origin, origin, region);
    if (err != CL_SUCCESS) goto error;
    return TRUE;
  }

  size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 0, sizeof(cl_mem), &dev_in);
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 1, sizeof(cl_mem), &dev_out);
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 2, sizeof(int), &width);
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 3, sizeof(int), &height);
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 4, sizeof(float), &lower);
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 5, sizeof(float), &upper);
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 6, 4*sizeof(float), lower_color);
  dt_opencl_set_kernel_arg(devid, gd->kernel_overexposed, 7, 4*sizeof(float), upper_color);
  err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_overexposed, sizes);
  if(err != CL_SUCCESS) goto error;
  return TRUE;

error:
  dt_print(DT_DEBUG_OPENCL, "[opencl_overexposed] couldn't enqueue kernel! %d\n", err);
  return FALSE;
}