void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, const void *const ivoid,
void *const ovoid, const dt_iop_roi_t *const roi_in, const dt_iop_roi_t *const roi_out)
{
const uint32_t filters = piece->pipe->dsc.filters;
dt_iop_highlights_data_t *data = (dt_iop_highlights_data_t *)piece->data;
const float clip
= data->clip * fminf(piece->pipe->dsc.processed_maximum[0],
fminf(piece->pipe->dsc.processed_maximum[1], piece->pipe->dsc.processed_maximum[2]));
// const int ch = piece->colors;
if(!filters)
{
process_clip(piece, ivoid, ovoid, roi_in, roi_out, clip);
for(int k=0;k<3;k++)
piece->pipe->dsc.processed_maximum[k]
= fminf(piece->pipe->dsc.processed_maximum[0],
fminf(piece->pipe->dsc.processed_maximum[1], piece->pipe->dsc.processed_maximum[2]));
return;
}
switch(data->mode)
{
case DT_IOP_HIGHLIGHTS_INPAINT: // a1ex's (magiclantern) idea of color inpainting:
{
const float clips[4] = { 0.987 * data->clip * piece->pipe->dsc.processed_maximum[0],
0.987 * data->clip * piece->pipe->dsc.processed_maximum[1],
0.987 * data->clip * piece->pipe->dsc.processed_maximum[2], clip };
if(filters == 9u)
{
const uint8_t(*const xtrans)[6] = (const uint8_t(*const)[6])piece->pipe->dsc.xtrans;
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none)
#endif
for(int j = 0; j < roi_out->height; j++)
{
interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 0, 1, j, clips, xtrans, 0);
interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 0, -1, j, clips, xtrans, 1);
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none)
#endif
for(int i = 0; i < roi_out->width; i++)
{
interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 1, 1, i, clips, xtrans, 2);
interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 1, -1, i, clips, xtrans, 3);
}
}
else
{
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(data, piece)
#endif
for(int j = 0; j < roi_out->height; j++)
{
interpolate_color(ivoid, ovoid, roi_out, 0, 1, j, clips, filters, 0);
interpolate_color(ivoid, ovoid, roi_out, 0, -1, j, clips, filters, 1);
}
// up/down directions
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(data, piece)
#endif
for(int i = 0; i < roi_out->width; i++)
{
interpolate_color(ivoid, ovoid, roi_out, 1, 1, i, clips, filters, 2);
interpolate_color(ivoid, ovoid, roi_out, 1, -1, i, clips, filters, 3);
}
}
break;
}
case DT_IOP_HIGHLIGHTS_LCH:
if(filters == 9u)
process_lch_xtrans(self, piece, ivoid, ovoid, roi_in, roi_out, clip);
else
process_lch_bayer(self, piece, ivoid, ovoid, roi_in, roi_out, clip);
break;
default:
case DT_IOP_HIGHLIGHTS_CLIP:
process_clip(piece, ivoid, ovoid, roi_in, roi_out, clip);
break;
}
// 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;
if(piece->pipe->mask_display & DT_DEV_PIXELPIPE_DISPLAY_MASK) dt_iop_alpha_copy(ivoid, ovoid, roi_out->width, roi_out->height);
}
void process(struct dt_iop_module_t *self, dt_dev_pixelpipe_iop_t *piece, void *ivoid, void *ovoid,
const dt_iop_roi_t *roi_in, const dt_iop_roi_t *roi_out)
{
const int filters = dt_image_filter(&piece->pipe->image);
dt_iop_highlights_data_t *data = (dt_iop_highlights_data_t *)piece->data;
const float clip = data->clip
* fminf(piece->pipe->processed_maximum[0],
fminf(piece->pipe->processed_maximum[1], piece->pipe->processed_maximum[2]));
// const int ch = piece->colors;
if(dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) || !filters)
{
const __m128 clipm = _mm_set1_ps(clip);
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(ovoid, ivoid, roi_in, roi_out, data, piece)
#endif
for(int j = 0; j < roi_out->height; j++)
{
float *out = (float *)ovoid + (size_t)4 * roi_out->width * j;
float *in = (float *)ivoid + (size_t)4 * roi_in->width * j;
for(int i = 0; i < roi_out->width; i++)
{
_mm_stream_ps(out, _mm_min_ps(clipm, _mm_set_ps(in[3], in[2], in[1], in[0])));
in += 4;
out += 4;
}
}
_mm_sfence();
return;
}
switch(data->mode)
{
case DT_IOP_HIGHLIGHTS_INPAINT: // a1ex's (magiclantern) idea of color inpainting:
{
const float clips[4] = { 0.987 * data->clip * piece->pipe->processed_maximum[0],
0.987 * data->clip * piece->pipe->processed_maximum[1],
0.987 * data->clip * piece->pipe->processed_maximum[2], clip };
if(filters == 9u)
{
const dt_image_t *img = &self->dev->image_storage;
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(ovoid, ivoid, roi_in, roi_out, img)
#endif
for(int j = 0; j < roi_out->height; j++)
{
_interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 0, 1, j, clips, img->xtrans, 0);
_interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 0, -1, j, clips, img->xtrans, 1);
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(ovoid, ivoid, roi_in, roi_out, img)
#endif
for(int i = 0; i < roi_out->width; i++)
{
_interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 1, 1, i, clips, img->xtrans, 2);
_interpolate_color_xtrans(ivoid, ovoid, roi_in, roi_out, 1, -1, i, clips, img->xtrans, 3);
}
break;
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(ovoid, ivoid, roi_in, roi_out, data, piece)
#endif
for(int j = 0; j < roi_out->height; j++)
{
_interpolate_color(ivoid, ovoid, roi_out, 0, 1, j, clips, filters, 0);
_interpolate_color(ivoid, ovoid, roi_out, 0, -1, j, clips, filters, 1);
}
// up/down directions
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(ovoid, ivoid, roi_in, roi_out, data, piece)
#endif
for(int i = 0; i < roi_out->width; i++)
{
_interpolate_color(ivoid, ovoid, roi_out, 1, 1, i, clips, filters, 2);
_interpolate_color(ivoid, ovoid, roi_out, 1, -1, i, clips, filters, 3);
}
break;
}
case DT_IOP_HIGHLIGHTS_LCH:
if(filters == 9u)
{
process_lch_xtrans(ivoid, ovoid, roi_out->width, roi_out->height, clip);
break;
}
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic) default(none) shared(ovoid, ivoid, roi_in, roi_out, data, piece)
#endif
for(int j = 0; j < roi_out->height; j++)
{
float *out = (float *)ovoid + (size_t)roi_out->width * j;
float *in = (float *)ivoid + (size_t)roi_out->width * j;
for(int i = 0; i < roi_out->width; i++)
{
if(i == 0 || i == roi_out->width - 1 || j == 0 || j == roi_out->height - 1)
{
// fast path for border
out[0] = in[0];
}
else
{
// analyse one bayer block to get same number of rggb pixels each time
const float near_clip = 0.96f * clip;
const float post_clip = 1.10f * clip;
float blend = 0.0f;
float mean = 0.0f;
for(int jj = 0; jj <= 1; jj++)
{
for(int ii = 0; ii <= 1; ii++)
{
const float val = in[(size_t)jj * roi_out->width + ii];
mean += val * 0.25f;
blend += (fminf(post_clip, val) - near_clip) / (post_clip - near_clip);
}
}
blend = CLAMP(blend, 0.0f, 1.0f);
if(blend > 0)
{
// recover:
out[0] = blend * mean + (1.f - blend) * in[0];
}
else
out[0] = in[0];
}
out++;
in++;
}
}
break;
default:
case DT_IOP_HIGHLIGHTS_CLIP:
{
const __m128 clipm = _mm_set1_ps(clip);
const size_t n = (size_t)roi_out->height * roi_out->width;
float *const out = (float *)ovoid;
float *const in = (float *)ivoid;
#ifdef _OPENMP
#pragma omp parallel for schedule(static) default(none)
#endif
for(size_t j = 0; j < (n & ~3u); j += 4) _mm_stream_ps(out + j, _mm_min_ps(clipm, _mm_load_ps(in + j)));
_mm_sfence();
// lets see if there's a non-multiple of four rest to process:
if(n & 3)
for(size_t j = n & ~3u; j < n; j++) out[j] = MIN(clip, in[j]);
break;
}
}
if(piece->pipe->mask_display) dt_iop_alpha_copy(ivoid, ovoid, roi_out->width, roi_out->height);
}
