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 *)¢er);
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_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_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_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_overexposed_data_t *d = (dt_iop_overexposed_data_t *)piece->data;
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 = d->lower / 100.0f;
const float upper = d->upper / 100.0f;
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);
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;
}
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_colorchecker_data_t *d = (dt_iop_colorchecker_data_t *)piece->data;
dt_iop_colorchecker_global_data_t *gd = (dt_iop_colorchecker_global_data_t *)self->data;
const int devid = piece->pipe->devid;
const int width = roi_out->width;
const int height = roi_out->height;
const int num_patches = d->num_patches;
cl_int err = -999;
cl_mem dev_params = NULL;
const size_t params_size = (size_t)(4 * (2 * num_patches + 4)) * sizeof(float);
float *params = malloc(params_size);
float *idx = params;
// re-arrange data->source_Lab and data->coeff_{L,a,b} into float4
for(int n = 0; n < num_patches; n++, idx += 4)
{
idx[0] = d->source_Lab[3 * n];
idx[1] = d->source_Lab[3 * n + 1];
idx[2] = d->source_Lab[3 * n + 2];
idx[3] = 0.0f;
}
for(int n = 0; n < num_patches + 4; n++, idx += 4)
{
idx[0] = d->coeff_L[n];
idx[1] = d->coeff_a[n];
idx[2] = d->coeff_b[n];
idx[3] = 0.0f;
}
dev_params = dt_opencl_copy_host_to_device_constant(devid, params_size, params);
if(dev_params == NULL) goto error;
size_t sizes[3] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
dt_opencl_set_kernel_arg(devid, gd->kernel_colorchecker, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorchecker, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorchecker, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorchecker, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorchecker, 4, sizeof(int), (void *)&num_patches);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorchecker, 5, sizeof(cl_mem), (void *)&dev_params);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_colorchecker, sizes);
if(err != CL_SUCCESS) goto error;
dt_opencl_release_mem_object(dev_params);
free(params);
return TRUE;
error:
free(params);
if(dev_params != NULL) dt_opencl_release_mem_object(dev_params);
dt_print(DT_DEBUG_OPENCL, "[opencl_colorchecker] 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_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_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_channelmixer_data_t *data = (dt_iop_channelmixer_data_t *)piece->data;
dt_iop_channelmixer_global_data_t *gd = (dt_iop_channelmixer_global_data_t *)self->data;
cl_mem dev_red = NULL;
cl_mem dev_green = NULL;
cl_mem dev_blue = 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 gray_mix_mode = (data->red[CHANNEL_GRAY] != 0.0f || data->green[CHANNEL_GRAY] != 0.0f
|| data->blue[CHANNEL_GRAY] != 0.0f)
? TRUE
: FALSE;
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
dev_red = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * CHANNEL_SIZE, data->red);
if(dev_red == NULL) goto error;
dev_green = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * CHANNEL_SIZE, data->green);
if(dev_green == NULL) goto error;
dev_blue = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * CHANNEL_SIZE, data->blue);
if(dev_blue == NULL) goto error;
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 4, sizeof(int), (void *)&gray_mix_mode);
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 5, sizeof(cl_mem), (void *)&dev_red);
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 6, sizeof(cl_mem), (void *)&dev_green);
dt_opencl_set_kernel_arg(devid, gd->kernel_channelmixer, 7, sizeof(cl_mem), (void *)&dev_blue);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_channelmixer, sizes);
if(err != CL_SUCCESS) goto error;
dt_opencl_release_mem_object(dev_red);
dt_opencl_release_mem_object(dev_green);
dt_opencl_release_mem_object(dev_blue);
return TRUE;
error:
if(dev_red != NULL) dt_opencl_release_mem_object(dev_red);
if(dev_green != NULL) dt_opencl_release_mem_object(dev_green);
if(dev_blue != NULL) dt_opencl_release_mem_object(dev_blue);
dt_print(DT_DEBUG_OPENCL, "[opencl_channelmixer] 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;
}
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_colorbalance_data_t *d = (dt_iop_colorbalance_data_t *)piece->data;
dt_iop_colorbalance_global_data_t *gd = (dt_iop_colorbalance_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 lift[4] = { 2.0f - (d->lift[CHANNEL_RED] * d->lift[CHANNEL_FACTOR]),
2.0f - (d->lift[CHANNEL_GREEN] * d->lift[CHANNEL_FACTOR]),
2.0f - (d->lift[CHANNEL_BLUE] * d->lift[CHANNEL_FACTOR]), 0.0f },
gamma[4] = { d->gamma[CHANNEL_RED] * d->gamma[CHANNEL_FACTOR],
d->gamma[CHANNEL_GREEN] * d->gamma[CHANNEL_FACTOR],
d->gamma[CHANNEL_BLUE] * d->gamma[CHANNEL_FACTOR], 0.0f },
gamma_inv[4] = { (gamma[0] != 0.0f) ? 1.0f / gamma[0] : 1000000.0f,
(gamma[1] != 0.0f) ? 1.0f / gamma[1] : 1000000.0f,
(gamma[2] != 0.0f) ? 1.0f / gamma[2] : 1000000.0f, 0.0f },
gain[4] = { d->gain[CHANNEL_RED] * d->gain[CHANNEL_FACTOR],
d->gain[CHANNEL_GREEN] * d->gain[CHANNEL_FACTOR],
d->gain[CHANNEL_BLUE] * d->gain[CHANNEL_FACTOR], 0.0f };
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
dt_opencl_set_kernel_arg(devid, gd->kernel_colorbalance, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorbalance, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorbalance, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorbalance, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorbalance, 4, 4 * sizeof(float), (void *)&lift);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorbalance, 5, 4 * sizeof(float), (void *)&gain);
dt_opencl_set_kernel_arg(devid, gd->kernel_colorbalance, 6, 4 * sizeof(float), (void *)&gamma_inv);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_colorbalance, sizes);
if(err != CL_SUCCESS) goto error;
return TRUE;
error:
dt_print(DT_DEBUG_OPENCL, "[opencl_colorbalance] couldn't enqueue kernel! %d\n", err);
return FALSE;
}
int process_cl(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_levels_data_t *d = (dt_iop_levels_data_t *)piece->data;
dt_iop_levels_global_data_t *gd = (dt_iop_levels_global_data_t *)self->data;
if(d->mode == LEVELS_MODE_AUTOMATIC)
{
commit_params_late(self, piece);
}
cl_mem dev_lut = NULL;
cl_int err = -999;
const int devid = piece->pipe->devid;
const int width = roi_out->width;
const int height = roi_out->height;
dev_lut = dt_opencl_copy_host_to_device(devid, d->lut, 256, 256, sizeof(float));
if(dev_lut == NULL) goto error;
size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 0, sizeof(cl_mem), &dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 1, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 2, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 3, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 4, sizeof(cl_mem), &dev_lut);
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 5, sizeof(float), &d->levels[0]);
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 6, sizeof(float), &d->levels[2]);
dt_opencl_set_kernel_arg(devid, gd->kernel_levels, 7, sizeof(float), &d->in_inv_gamma);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_levels, sizes);
if(err != CL_SUCCESS) goto error;
dt_opencl_release_mem_object(dev_lut);
return TRUE;
error:
if(dev_lut != NULL) dt_opencl_release_mem_object(dev_lut);
dt_print(DT_DEBUG_OPENCL, "[opencl_levels] 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_borders_data_t *d = (dt_iop_borders_data_t *)piece->data;
dt_iop_borders_global_data_t *gd = (dt_iop_borders_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 int bw = (piece->buf_out.width - piece->buf_in.width ) * roi_in->scale;
const int bh = (piece->buf_out.height - piece->buf_in.height) * roi_in->scale;
const int bx = MAX(bw/2 - roi_out->x, 0);
const int by = MAX(bh/2 - roi_out->y, 0);
const float col[4] = {d->color[0], d->color[1], d->color[2], 1.0f};
size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 0, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 1, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 2, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 3, 4*sizeof(float), &col);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_borders_fill, sizes);
if(err != CL_SUCCESS) goto error;
size_t iorigin[] = { 0, 0, 0};
size_t oorigin[] = { bx, by, 0};
size_t region[] = { roi_in->width, roi_in->height, 1};
// copy original input from dev_in -> dev_out as starting point
err = dt_opencl_enqueue_copy_image(devid, dev_in, dev_out, iorigin, oorigin, region);
if(err != CL_SUCCESS) goto error;
return TRUE;
error:
dt_print(DT_DEBUG_OPENCL, "[opencl_borders] 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_lowlight_data_t *d = (dt_iop_lowlight_data_t *)piece->data;
dt_iop_lowlight_global_data_t *gd = (dt_iop_lowlight_global_data_t *)self->data;
cl_mem dev_m = NULL;
cl_int err = -999;
const int devid = piece->pipe->devid;
const int width = roi_out->width;
const int height = roi_out->height;
// scotopic white, blue saturated
float Lab_sw[3] = { 100.0f, 0.0f, -d->blueness };
float XYZ_sw[4];
dt_Lab_to_XYZ(Lab_sw, XYZ_sw);
dev_m = dt_opencl_copy_host_to_device(devid, d->lut, 256, 256, sizeof(float));
if(dev_m == NULL) goto error;
size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
dt_opencl_set_kernel_arg(devid, gd->kernel_lowlight, 0, sizeof(cl_mem), &dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_lowlight, 1, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_lowlight, 2, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_lowlight, 3, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_lowlight, 4, 4 * sizeof(float), &XYZ_sw);
dt_opencl_set_kernel_arg(devid, gd->kernel_lowlight, 5, sizeof(cl_mem), &dev_m);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_lowlight, sizes);
if(err != CL_SUCCESS) goto error;
dt_opencl_release_mem_object(dev_m);
return TRUE;
error:
dt_opencl_release_mem_object(dev_m);
dt_print(DT_DEBUG_OPENCL, "[opencl_lowlight] 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_splittoning_data_t *d = (dt_iop_splittoning_data_t *)piece->data;
dt_iop_splittoning_global_data_t *gd = (dt_iop_splittoning_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 compress = (d->compress/110.0)/2.0; // Dont allow 100% compression..
const float balance = d->balance;
const float shadow_hue = d->shadow_hue;
const float shadow_saturation = d->shadow_saturation;
const float highlight_hue = d->highlight_hue;
const float highlight_saturation = d->highlight_saturation;
size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 0, sizeof(cl_mem), &dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 1, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 2, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 3, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 4, sizeof(float), &compress);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 5, sizeof(float), &balance);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 6, sizeof(float), &shadow_hue);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 7, sizeof(float), &shadow_saturation);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 8, sizeof(float), &highlight_hue);
dt_opencl_set_kernel_arg(devid, gd->kernel_splittoning, 9, sizeof(float), &highlight_saturation);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_splittoning, sizes);
if(err != CL_SUCCESS) goto error;
return TRUE;
error:
dt_print(DT_DEBUG_OPENCL, "[opencl_splittoning] 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_borders_data_t *d = (dt_iop_borders_data_t *)piece->data;
dt_iop_borders_global_data_t *gd = (dt_iop_borders_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 int border_tot_width = (piece->buf_out.width - piece->buf_in.width ) * roi_in->scale;
const int border_tot_height = (piece->buf_out.height - piece->buf_in.height) * roi_in->scale;
const int border_size_t = border_tot_height*d->pos_v;
const int border_size_b = border_tot_height - border_size_t;
const int border_size_l = border_tot_width*d->pos_h;
const int border_size_r = border_tot_width - border_size_l;
const int border_in_x = MAX(border_size_l - roi_out->x, 0);
const int border_in_y = MAX(border_size_t - roi_out->y, 0);
// ----- Filling border
const float col[4] = {d->color[0], d->color[1], d->color[2], 1.0f};
size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
const int zero = 0;
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 0, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 1, sizeof(int), &zero);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 2, sizeof(int), &zero);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 3, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 4, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 5, 4*sizeof(float), &col);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_borders_fill, sizes);
if(err != CL_SUCCESS) goto error;
// ----- Frame line
const int border_min_size = MIN(MIN(border_size_t, border_size_b), MIN(border_size_l, border_size_r));
const int frame_size = border_min_size * d->frame_size;
if (frame_size != 0)
{
const float col_frame[4] = {d->frame_color[0], d->frame_color[1], d->frame_color[2], 1.0f};
const int image_lx = border_size_l - roi_out->x;
const int image_ty = border_size_t - roi_out->y;
const int frame_space = border_min_size - frame_size;
const int frame_offset = frame_space * d->frame_offset;
const int frame_tl_in_x = MAX(border_in_x - frame_offset, 0);
const int frame_tl_out_x = MAX(frame_tl_in_x - frame_size, 0);
const int frame_tl_in_y = MAX(border_in_y - frame_offset, 0);
const int frame_tl_out_y = MAX(frame_tl_in_y - frame_size, 0);
const int frame_in_width = floor((piece->buf_in.width * roi_in->scale) + frame_offset*2);
const int frame_in_height = floor((piece->buf_in.height * roi_in->scale) + frame_offset*2);
const int frame_out_width = frame_in_width + frame_size*2;
const int frame_out_height = frame_in_height + frame_size*2;
const int frame_br_in_x = CLAMP(image_lx - frame_offset + frame_in_width, 0, roi_out->width);
const int frame_br_in_y = CLAMP(image_ty - frame_offset + frame_in_height, 0, roi_out->height);
// ... if 100% frame_offset we ensure frame_line "stick" the out border
const int frame_br_out_x = (d->frame_offset == 1.0f)
? (roi_out->width)
: CLAMP(image_lx - frame_offset - frame_size + frame_out_width, 0, roi_out->width);
const int frame_br_out_y = (d->frame_offset == 1.0f)
? (roi_out->height)
: CLAMP(image_ty - frame_offset - frame_size + frame_out_height, 0, roi_out->height);
const int roi_frame_in_width = frame_br_in_x - frame_tl_in_x;
const int roi_frame_in_height = frame_br_in_y - frame_tl_in_y;
const int roi_frame_out_width = frame_br_out_x - frame_tl_out_x;
const int roi_frame_out_height = frame_br_out_y - frame_tl_out_y;
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 0, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 1, sizeof(int), &frame_tl_out_x);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 2, sizeof(int), &frame_tl_out_y);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 3, sizeof(int), &roi_frame_out_width);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 4, sizeof(int), &roi_frame_out_height);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 5, 4*sizeof(float), &col_frame);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_borders_fill, sizes);
if(err != CL_SUCCESS) goto error;
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 0, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 1, sizeof(int), &frame_tl_in_x);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 2, sizeof(int), &frame_tl_in_y);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 3, sizeof(int), &roi_frame_in_width);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 4, sizeof(int), &roi_frame_in_height);
dt_opencl_set_kernel_arg(devid, gd->kernel_borders_fill, 5, 4*sizeof(float), &col);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_borders_fill, sizes);
if(err != CL_SUCCESS) goto error;
}
size_t iorigin[] = { 0, 0, 0};
size_t oorigin[] = { border_in_x, border_in_y, 0};
size_t region[] = { roi_in->width, roi_in->height, 1};
// copy original input from dev_in -> dev_out as starting point
err = dt_opencl_enqueue_copy_image(devid, dev_in, dev_out, iorigin, oorigin, region);
if(err != CL_SUCCESS) goto error;
return TRUE;
error:
dt_print(DT_DEBUG_OPENCL, "[opencl_borders] 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_global_tonemap_data_t *d = (dt_iop_global_tonemap_data_t *)piece->data;
dt_iop_global_tonemap_global_data_t *gd = (dt_iop_global_tonemap_global_data_t *)self->data;
dt_iop_global_tonemap_gui_data_t *g = (dt_iop_global_tonemap_gui_data_t *)self->gui_data;
dt_bilateral_cl_t *b = NULL;
cl_int err = -999;
cl_mem dev_m = NULL;
cl_mem dev_r = NULL;
float *maximum = NULL;
const int devid = piece->pipe->devid;
int gtkernel = -1;
const int width = roi_out->width;
const int height = roi_out->height;
float parameters[4] = { 0.0f };
switch(d->operator)
{
case OPERATOR_REINHARD:
gtkernel = gd->kernel_global_tonemap_reinhard;
break;
case OPERATOR_DRAGO:
gtkernel = gd->kernel_global_tonemap_drago;
break;
case OPERATOR_FILMIC:
gtkernel = gd->kernel_global_tonemap_filmic;
break;
}
if(d->operator== OPERATOR_DRAGO)
{
const float eps = 0.0001f;
float tmp_lwmax = NAN;
// see comments in process() about lwmax value
if(self->dev->gui_attached && g && piece->pipe->type == DT_DEV_PIXELPIPE_FULL)
{
dt_pthread_mutex_lock(&g->lock);
const uint64_t hash = g->hash;
dt_pthread_mutex_unlock(&g->lock);
if(hash != 0 && !dt_dev_sync_pixelpipe_hash(self->dev, piece->pipe, 0, self->priority, &g->lock, &g->hash))
dt_control_log(_("inconsistent output"));
dt_pthread_mutex_lock(&g->lock);
tmp_lwmax = g->lwmax;
dt_pthread_mutex_unlock(&g->lock);
}
if(isnan(tmp_lwmax))
{
dt_opencl_local_buffer_t flocopt
= (dt_opencl_local_buffer_t){ .xoffset = 0, .xfactor = 1, .yoffset = 0, .yfactor = 1,
.cellsize = sizeof(float), .overhead = 0,
.sizex = 1 << 4, .sizey = 1 << 4 };
if(!dt_opencl_local_buffer_opt(devid, gd->kernel_pixelmax_first, &flocopt))
goto error;
const size_t bwidth = ROUNDUP(width, flocopt.sizex);
const size_t bheight = ROUNDUP(height, flocopt.sizey);
const int bufsize = (bwidth / flocopt.sizex) * (bheight / flocopt.sizey);
dt_opencl_local_buffer_t slocopt
= (dt_opencl_local_buffer_t){ .xoffset = 0, .xfactor = 1, .yoffset = 0, .yfactor = 1,
.cellsize = sizeof(float), .overhead = 0,
.sizex = 1 << 16, .sizey = 1 };
if(!dt_opencl_local_buffer_opt(devid, gd->kernel_pixelmax_second, &slocopt))
goto error;
const int reducesize = MIN(REDUCESIZE, ROUNDUP(bufsize, slocopt.sizex) / slocopt.sizex);
size_t sizes[3];
size_t local[3];
dev_m = dt_opencl_alloc_device_buffer(devid, (size_t)bufsize * sizeof(float));
if(dev_m == NULL) goto error;
dev_r = dt_opencl_alloc_device_buffer(devid, (size_t)reducesize * sizeof(float));
if(dev_r == NULL) goto error;
sizes[0] = bwidth;
sizes[1] = bheight;
sizes[2] = 1;
local[0] = flocopt.sizex;
local[1] = flocopt.sizey;
local[2] = 1;
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_first, 0, sizeof(cl_mem), &dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_first, 1, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_first, 2, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_first, 3, sizeof(cl_mem), &dev_m);
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_first, 4, flocopt.sizex * flocopt.sizey * sizeof(float), NULL);
err = dt_opencl_enqueue_kernel_2d_with_local(devid, gd->kernel_pixelmax_first, sizes, local);
if(err != CL_SUCCESS) goto error;
sizes[0] = reducesize * slocopt.sizex;
sizes[1] = 1;
sizes[2] = 1;
local[0] = slocopt.sizex;
local[1] = 1;
local[2] = 1;
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_second, 0, sizeof(cl_mem), &dev_m);
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_second, 1, sizeof(cl_mem), &dev_r);
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_second, 2, sizeof(int), &bufsize);
dt_opencl_set_kernel_arg(devid, gd->kernel_pixelmax_second, 3, slocopt.sizex * sizeof(float), NULL);
err = dt_opencl_enqueue_kernel_2d_with_local(devid, gd->kernel_pixelmax_second, sizes, local);
if(err != CL_SUCCESS) goto error;
maximum = dt_alloc_align(16, reducesize * sizeof(float));
err = dt_opencl_read_buffer_from_device(devid, (void *)maximum, dev_r, 0,
(size_t)reducesize * sizeof(float), CL_TRUE);
if(err != CL_SUCCESS) goto error;
dt_opencl_release_mem_object(dev_r);
dt_opencl_release_mem_object(dev_m);
dev_r = dev_m = NULL;
for(int k = 1; k < reducesize; k++)
{
float mine = maximum[0];
float other = maximum[k];
maximum[0] = (other > mine) ? other : mine;
}
tmp_lwmax = MAX(eps, (maximum[0] * 0.01f));
dt_free_align(maximum);
maximum = NULL;
}
const float lwmax = tmp_lwmax;
const float ldc = d->drago.max_light * 0.01f / log10f(lwmax + 1.0f);
const float bl = logf(MAX(eps, d->drago.bias)) / logf(0.5f);
parameters[0] = eps;
parameters[1] = ldc;
parameters[2] = bl;
parameters[3] = lwmax;
if(self->dev->gui_attached && g && piece->pipe->type == DT_DEV_PIXELPIPE_PREVIEW)
{
uint64_t hash = dt_dev_hash_plus(self->dev, piece->pipe, 0, self->priority);
dt_pthread_mutex_lock(&g->lock);
g->lwmax = lwmax;
g->hash = hash;
dt_pthread_mutex_unlock(&g->lock);
}
}
const float scale = piece->iscale / roi_in->scale;
const float sigma_r = 8.0f; // does not depend on scale
const float iw = piece->buf_in.width / scale;
const float ih = piece->buf_in.height / scale;
const float sigma_s = fminf(iw, ih) * 0.03f;
if(d->detail != 0.0f)
{
b = dt_bilateral_init_cl(devid, roi_in->width, roi_in->height, sigma_s, sigma_r);
if(!b) goto error;
// get detail from unchanged input buffer
err = dt_bilateral_splat_cl(b, dev_in);
if(err != CL_SUCCESS) goto error;
}
size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
dt_opencl_set_kernel_arg(devid, gtkernel, 0, sizeof(cl_mem), &dev_in);
dt_opencl_set_kernel_arg(devid, gtkernel, 1, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gtkernel, 2, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gtkernel, 3, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gtkernel, 4, 4 * sizeof(float), ¶meters);
err = dt_opencl_enqueue_kernel_2d(devid, gtkernel, sizes);
if(err != CL_SUCCESS) goto error;
if(d->detail != 0.0f)
{
err = dt_bilateral_blur_cl(b);
if(err != CL_SUCCESS) goto error;
// and apply it to output buffer after logscale
err = dt_bilateral_slice_to_output_cl(b, dev_in, dev_out, d->detail);
if(err != CL_SUCCESS) goto error;
dt_bilateral_free_cl(b);
}
return TRUE;
error:
if(b) dt_bilateral_free_cl(b);
dt_opencl_release_mem_object(dev_m);
dt_opencl_release_mem_object(dev_r);
dt_free_align(maximum);
dt_print(DT_DEBUG_OPENCL, "[opencl_global_tonemap] couldn't enqueue kernel! %d\n", err);
return FALSE;
}
#endif
void tiling_callback(struct dt_iop_module_t *self, struct dt_dev_pixelpipe_iop_t *piece,
const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out,
struct dt_develop_tiling_t *tiling)
{
dt_iop_global_tonemap_data_t *d = (dt_iop_global_tonemap_data_t *)piece->data;
const float scale = piece->iscale / roi_in->scale;
const float iw = piece->buf_in.width / scale;
const float ih = piece->buf_in.height / scale;
const float sigma_s = fminf(iw, ih) * 0.03f;
const float sigma_r = 8.0f;
const int detail = (d->detail != 0.0f);
const int width = roi_in->width;
const int height = roi_in->height;
const int channels = piece->colors;
const size_t basebuffer = width * height * channels * sizeof(float);
tiling->factor = 2.0f + (detail ? (float)dt_bilateral_memory_use2(width, height, sigma_s, sigma_r) / basebuffer : 0.0f);
tiling->maxbuf
= (detail ? MAX(1.0f, (float)dt_bilateral_singlebuffer_size2(width, height, sigma_s, sigma_r) / basebuffer) : 1.0f);
tiling->overhead = 0;
tiling->overlap = (detail ? ceilf(4 * sigma_s) : 0);
tiling->xalign = 1;
tiling->yalign = 1;
return;
}
void commit_params(struct dt_iop_module_t *self, dt_iop_params_t *p1, dt_dev_pixelpipe_t *pipe,
dt_dev_pixelpipe_iop_t *piece)
{
dt_iop_global_tonemap_params_t *p = (dt_iop_global_tonemap_params_t *)p1;
dt_iop_global_tonemap_data_t *d = (dt_iop_global_tonemap_data_t *)piece->data;
d->operator= p->operator;
d->drago.bias = p->drago.bias;
d->drago.max_light = p->drago.max_light;
d->detail = p->detail;
// drago needs the maximum L-value of the whole image so it must not use tiling
if(d->operator == OPERATOR_DRAGO) piece->process_tiling_ready = 0;
#ifdef HAVE_OPENCL
if(d->detail != 0.0f)
piece->process_cl_ready = (piece->process_cl_ready && !(darktable.opencl->avoid_atomics));
#endif
}
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;
cl_mem dev_xtrans = NULL;
const int devid = piece->pipe->devid;
const int width = roi_in->width;
const int height = roi_in->height;
const float clip = d->clip
* fminf(piece->pipe->dsc.processed_maximum[0],
fminf(piece->pipe->dsc.processed_maximum[1], piece->pipe->dsc.processed_maximum[2]));
const uint32_t filters = piece->pipe->dsc.filters;
if(!filters)
{
// non-raw images use dedicated kernel which just clips
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
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 if(d->mode == DT_IOP_HIGHLIGHTS_CLIP)
{
// raw images with clip mode (both bayer and xtrans)
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 4, sizeof(float), (void *)&clip);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 5, sizeof(int), (void *)&roi_out->x);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 6, sizeof(int), (void *)&roi_out->y);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 7, sizeof(int), (void *)&filters);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights_1f_clip, sizes);
if(err != CL_SUCCESS) goto error;
}
else if(d->mode == DT_IOP_HIGHLIGHTS_LCH && filters != 9u)
{
// bayer sensor raws with LCH mode
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 4, sizeof(float), (void *)&clip);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 5, sizeof(int), (void *)&roi_out->x);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 6, sizeof(int), (void *)&roi_out->y);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 7, sizeof(int), (void *)&filters);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights_1f_lch_bayer, sizes);
if(err != CL_SUCCESS) goto error;
}
else if(d->mode == DT_IOP_HIGHLIGHTS_LCH && filters == 9u)
{
// xtrans sensor raws with LCH mode
int blocksizex, blocksizey;
dt_opencl_local_buffer_t locopt
= (dt_opencl_local_buffer_t){ .xoffset = 2 * 2, .xfactor = 1, .yoffset = 2 * 2, .yfactor = 1,
.cellsize = sizeof(float), .overhead = 0,
.sizex = 1 << 8, .sizey = 1 << 8 };
if(dt_opencl_local_buffer_opt(devid, gd->kernel_highlights_1f_lch_xtrans, &locopt))
{
blocksizex = locopt.sizex;
blocksizey = locopt.sizey;
}
else
blocksizex = blocksizey = 1;
dev_xtrans
= dt_opencl_copy_host_to_device_constant(devid, sizeof(piece->pipe->dsc.xtrans), piece->pipe->dsc.xtrans);
if(dev_xtrans == NULL) goto error;
size_t sizes[] = { ROUNDUP(width, blocksizex), ROUNDUP(height, blocksizey), 1 };
size_t local[] = { blocksizex, blocksizey, 1 };
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 4, sizeof(float), (void *)&clip);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 5, sizeof(int), (void *)&roi_out->x);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 6, sizeof(int), (void *)&roi_out->y);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 7, sizeof(cl_mem), (void *)&dev_xtrans);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_xtrans, 8,
(blocksizex + 4) * (blocksizey + 4) * sizeof(float), NULL);
err = dt_opencl_enqueue_kernel_2d_with_local(devid, gd->kernel_highlights_1f_lch_xtrans, sizes, local);
if(err != CL_SUCCESS) goto error;
}
// update processed maximum
const float m = fmaxf(fmaxf(piece->pipe->dsc.processed_maximum[0], piece->pipe->dsc.processed_maximum[1]),
piece->pipe->dsc.processed_maximum[2]);
for(int k = 0; k < 3; k++) piece->pipe->dsc.processed_maximum[k] = m;
dt_opencl_release_mem_object(dev_xtrans);
return TRUE;
error:
dt_opencl_release_mem_object(dev_xtrans);
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_monochrome_data_t *d = (dt_iop_monochrome_data_t *)piece->data;
dt_iop_monochrome_global_data_t *gd = (dt_iop_monochrome_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 sigma2 = (d->size*128.0)*(d->size*128.0f);
// TODO: alloc new buffer, bilat filter, and go on with that
const float scale = piece->iscale/roi_in->scale;
const float sigma_r = 250.0f; // does not depend on scale
const float sigma_s = 20.0f / scale;
const float detail = -1.0f; // bilateral base layer
cl_mem dev_tmp = NULL;
dev_tmp = dt_opencl_alloc_device(devid, roi_in->width, roi_in->height, 4*sizeof(float));
dt_bilateral_cl_t *b = dt_bilateral_init_cl(devid, roi_in->width, roi_in->height, sigma_s, sigma_r);
if(!b) goto error;
size_t sizes[2] = { ROUNDUPWD(width), ROUNDUPHT(height) };
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome_filter, 0, sizeof(cl_mem), &dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome_filter, 1, sizeof(cl_mem), &dev_tmp);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome_filter, 2, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome_filter, 3, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome_filter, 4, sizeof(float), &d->a);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome_filter, 5, sizeof(float), &d->b);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome_filter, 6, sizeof(float), &sigma2);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_monochrome_filter, sizes);
if(err != CL_SUCCESS) goto error;
err = dt_bilateral_splat_cl(b, dev_tmp);
if (err != CL_SUCCESS) goto error;
err = dt_bilateral_blur_cl(b);
if (err != CL_SUCCESS) goto error;
err = dt_bilateral_slice_cl(b, dev_tmp, dev_tmp, detail);
if (err != CL_SUCCESS) goto error;
dt_bilateral_free_cl(b);
b = NULL; // make sure we don't do double cleanup in case the next few lines err out
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 0, sizeof(cl_mem), &dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 1, sizeof(cl_mem), &dev_tmp);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 2, sizeof(cl_mem), &dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 3, sizeof(int), &width);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 4, sizeof(int), &height);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 5, sizeof(float), &d->a);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 6, sizeof(float), &d->b);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 7, sizeof(float), &sigma2);
dt_opencl_set_kernel_arg(devid, gd->kernel_monochrome, 8, sizeof(float), &d->highlights);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_monochrome, sizes);
if(err != CL_SUCCESS) goto error;
if (dev_tmp != NULL) dt_opencl_release_mem_object(dev_tmp);
return TRUE;
error:
if (dev_tmp != NULL) dt_opencl_release_mem_object(dev_tmp);
dt_bilateral_free_cl(b);
dt_print(DT_DEBUG_OPENCL, "[opencl_monochrome] 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_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;
cl_mem dev_xtrans = NULL;
const int devid = piece->pipe->devid;
const int width = roi_in->width;
const int height = roi_in->height;
const float clip = d->clip
* fminf(piece->pipe->dsc.processed_maximum[0],
fminf(piece->pipe->dsc.processed_maximum[1], piece->pipe->dsc.processed_maximum[2]));
const uint32_t filters = piece->pipe->dsc.filters;
if(!filters)
{
// non-raw images use dedicated kernel which just clips
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
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 if(d->mode == DT_IOP_HIGHLIGHTS_CLIP)
{
// raw images with clip mode (both bayer and xtrans)
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 4, sizeof(float), (void *)&clip);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 5, sizeof(int), (void *)&roi_out->x);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 6, sizeof(int), (void *)&roi_out->y);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_clip, 7, sizeof(int), (void *)&filters);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights_1f_clip, sizes);
if(err != CL_SUCCESS) goto error;
}
else if(d->mode == DT_IOP_HIGHLIGHTS_LCH && filters != 9u)
{
// bayer sensor raws with LCH mode
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 0, sizeof(cl_mem), (void *)&dev_in);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 1, sizeof(cl_mem), (void *)&dev_out);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 2, sizeof(int), (void *)&width);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 3, sizeof(int), (void *)&height);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 4, sizeof(float), (void *)&clip);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 5, sizeof(int), (void *)&roi_out->x);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 6, sizeof(int), (void *)&roi_out->y);
dt_opencl_set_kernel_arg(devid, gd->kernel_highlights_1f_lch_bayer, 7, sizeof(int), (void *)&filters);
err = dt_opencl_enqueue_kernel_2d(devid, gd->kernel_highlights_1f_lch_bayer, sizes);
if(err != CL_SUCCESS) goto error;
}
else if(d->mode == DT_IOP_HIGHLIGHTS_LCH && filters == 9u)
{
// xtrans sensor raws with LCH mode
// we use local buffering for speed reasons; determine suited work group size
size_t maxsizes[3] = { 0 }; // the maximum dimensions for a work group
size_t workgroupsize = 0; // the maximum number of items in a work group
unsigned long localmemsize = 0; // the maximum amount of local memory we can use
size_t kernelworkgroupsize = 0; // the maximum amount of items in work group for this kernel
int blocksizex = 1 << 8;
int blocksizey = 1 << 8;
if(dt_opencl_get_work_group_limits(devid, maxsizes, &workgroupsize, &localmemsize) == CL_SUCCESS
&& dt_opencl_get_kernel_work_group_size(devid, gd->kernel_highlights_1f_lch_xtrans, &kernelworkgroupsize) == CL_SUCCESS)
{
while(maxsizes[0] < blocksizex || maxsizes[1] < blocksizey
|| localmemsize < (blocksizex + 4) * (blocksizey + 4) * sizeof(float)
|| workgroupsize < blocksizex * blocksizey || kernelworkgroupsize < blocksizex * blocksizey)
{
if(blocksizex == 1 && blocksizey == 1) break;
if(blocksizex > blocksizey)
blocksizex >>= 1;
else
blocksizey >>= 1;
}
}
int process_cl(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_rawprepare_data_t *d = (dt_iop_rawprepare_data_t *)piece->data;
dt_iop_rawprepare_global_data_t *gd = (dt_iop_rawprepare_global_data_t *)self->data;
const int devid = piece->pipe->devid;
cl_mem dev_sub = NULL;
cl_mem dev_div = NULL;
cl_int err = -999;
int kernel = -1;
if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && piece->pipe->filters)
{
kernel = gd->kernel_rawprepare_1f;
}
else
{
kernel = gd->kernel_rawprepare_4f;
}
const float scale = roi_in->scale / piece->iscale;
const int csx = (int)roundf((float)d->x * scale), csy = (int)roundf((float)d->y * scale);
dev_sub = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * 4, d->sub);
if(dev_sub == NULL) goto error;
dev_div = dt_opencl_copy_host_to_device_constant(devid, sizeof(float) * 4, d->div);
if(dev_div == NULL) goto error;
const int width = roi_out->width;
const int height = roi_out->height;
size_t sizes[] = { ROUNDUPWD(roi_in->width), ROUNDUPHT(roi_in->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(int), (void *)&csx);
dt_opencl_set_kernel_arg(devid, kernel, 5, sizeof(int), (void *)&csy);
dt_opencl_set_kernel_arg(devid, kernel, 6, sizeof(cl_mem), (void *)&dev_sub);
dt_opencl_set_kernel_arg(devid, kernel, 7, sizeof(cl_mem), (void *)&dev_div);
dt_opencl_set_kernel_arg(devid, kernel, 8, sizeof(uint32_t), (void *)&roi_out->x);
dt_opencl_set_kernel_arg(devid, kernel, 9, 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_sub);
dt_opencl_release_mem_object(dev_div);
if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && piece->pipe->filters)
{
piece->pipe->filters = dt_rawspeed_crop_dcraw_filters(self->dev->image_storage.filters, csx, csy);
adjust_xtrans_filters(piece->pipe, csx, csy);
}
return TRUE;
error:
if(dev_sub != NULL) dt_opencl_release_mem_object(dev_sub);
if(dev_div != NULL) dt_opencl_release_mem_object(dev_div);
dt_print(DT_DEBUG_OPENCL, "[opencl_rawprepare] 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_graduatednd_data_t *data = (dt_iop_graduatednd_data_t *)piece->data;
dt_iop_graduatednd_global_data_t *gd = (dt_iop_graduatednd_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 int ix = (roi_in->x);
const int iy = (roi_in->y);
const float iw = piece->buf_in.width * roi_out->scale;
const float ih = piece->buf_in.height * roi_out->scale;
const float hw = iw / 2.0;
const float hh = ih / 2.0;
const float hw_inv = 1.0 / hw;
const float hh_inv = 1.0 / hh;
const float v = (-data->rotation / 180) * M_PI;
const float sinv = sin(v);
const float cosv = cos(v);
const float filter_radie = sqrt((hh * hh) + (hw * hw)) / hh;
const float offset = data->offset / 100.0 * 2;
const float density = data->density;
#if 1
const float filter_compression = 1.0 / filter_radie
/ (1.0 - (0.5 + (data->compression / 100.0) * 0.9 / 2.0)) * 0.5;
#else
const float compression = data->compression / 100.0f;
const float t = 1.0f - .8f / (.8f + compression);
const float c = 1.0f + 1000.0f * powf(4.0, compression);
#endif
const float length_base = (sinv * (-1.0 + ix * hw_inv) - cosv * (-1.0 + iy * hh_inv) - 1.0 + offset)
* filter_compression;
const float length_inc_y = -cosv * hh_inv * filter_compression;
const float length_inc_x = sinv * hw_inv * filter_compression;
size_t sizes[] = { ROUNDUPWD(width), ROUNDUPHT(height), 1 };
int kernel = density > 0 ? gd->kernel_graduatedndp : gd->kernel_graduatedndm;
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, 4 * sizeof(float), (void *)data->color);
dt_opencl_set_kernel_arg(devid, kernel, 5, sizeof(float), (void *)&density);
dt_opencl_set_kernel_arg(devid, kernel, 6, sizeof(float), (void *)&length_base);
dt_opencl_set_kernel_arg(devid, kernel, 7, sizeof(float), (void *)&length_inc_x);
dt_opencl_set_kernel_arg(devid, kernel, 8, sizeof(float), (void *)&length_inc_y);
err = dt_opencl_enqueue_kernel_2d(devid, kernel, sizes);
if(err != CL_SUCCESS) goto error;
return TRUE;
error:
dt_print(DT_DEBUG_OPENCL, "[opencl_graduatednd] couldn't enqueue kernel! %d\n", err);
return FALSE;
}