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_flipped_filter(&piece->pipe->image); dt_iop_temperature_data_t *d = (dt_iop_temperature_data_t *)piece->data; if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && filters && piece->pipe->image.bpp != 4) { const float coeffsi[3] = {d->coeffs[0]/65535.0f, d->coeffs[1]/65535.0f, d->coeffs[2]/65535.0f}; #ifdef _OPENMP #pragma omp parallel for default(none) shared(roi_out, ivoid, ovoid, d) schedule(static) #endif for(int j=0; j<roi_out->height; j++) { int i=0; const uint16_t *in = ((uint16_t *)ivoid) + j*roi_out->width; float *out = ((float*)ovoid) + j*roi_out->width; // process unaligned pixels for ( ; i < ((4-(j*roi_out->width & 3)) & 3) ; i++,out++,in++) *out = *in * coeffsi[FC(j+roi_out->y, i+roi_out->x, filters)]; const __m128 coeffs = _mm_set_ps(coeffsi[FC(j+roi_out->y, roi_out->x+i+3, filters)], coeffsi[FC(j+roi_out->y, roi_out->x+i+2, filters)], coeffsi[FC(j+roi_out->y, roi_out->x+i+1, filters)], coeffsi[FC(j+roi_out->y, roi_out->x+i , filters)]); // process aligned pixels with SSE for( ; i < roi_out->width - 3 ; i+=4,out+=4,in+=4) { _mm_stream_ps(out,_mm_mul_ps(coeffs,_mm_set_ps(in[3],in[2],in[1],in[0]))); } // process the rest for( ; i<roi_out->width; i++,out++,in++) *out = *in * coeffsi[FC(j+roi_out->y, i+roi_out->x, filters)]; } _mm_sfence(); } else if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && filters && piece->pipe->image.bpp == 4) { #ifdef _OPENMP #pragma omp parallel for default(none) shared(roi_out, ivoid, ovoid, d) schedule(static) #endif for(int j=0; j<roi_out->height; j++) { const float *in = ((float *)ivoid) + j*roi_out->width; float *out = ((float*)ovoid) + j*roi_out->width; for(int i=0; i<roi_out->width; i++,out++,in++) *out = *in * d->coeffs[FC(j+roi_out->x, i+roi_out->y, filters)]; } } else { const int ch = piece->colors; #ifdef _OPENMP #pragma omp parallel for default(none) shared(roi_out, ivoid, ovoid, d) schedule(static) #endif for(int k=0; k<roi_out->height; k++) { const float *in = ((float*)ivoid) + ch*k*roi_out->width; float *out = ((float*)ovoid) + ch*k*roi_out->width; for (int j=0; j<roi_out->width; j++,in+=ch,out+=ch) for(int c=0; c<3; c++) out[c] = in[c]*d->coeffs[c]; } } for(int k=0; k<3; k++) piece->pipe->processed_maximum[k] = d->coeffs[k] * piece->pipe->processed_maximum[k]; }
static gboolean draw(GtkWidget *widget, cairo_t *cr, dt_iop_module_t *self) { if(darktable.gui->reset) return FALSE; if(self->picked_color_max[0] < 0.0f) return FALSE; if(self->request_color_pick == DT_REQUEST_COLORPICK_OFF) return FALSE; dt_iop_invert_gui_data_t *g = (dt_iop_invert_gui_data_t *)self->gui_data; dt_iop_invert_params_t *p = (dt_iop_invert_params_t *)self->params; if(fabsf(p->color[0] - self->picked_color[0]) < 0.0001f && fabsf(p->color[1] - self->picked_color[1]) < 0.0001f && fabsf(p->color[2] - self->picked_color[2]) < 0.0001f) { // interrupt infinite loops return FALSE; } p->color[0] = self->picked_color[0]; p->color[1] = self->picked_color[1]; p->color[2] = self->picked_color[2]; GdkRGBA color = (GdkRGBA){.red = p->color[0], .green = p->color[1], .blue = p->color[2], .alpha = 1.0 }; gtk_color_chooser_set_rgba(GTK_COLOR_CHOOSER(g->colorpicker), &color); dt_dev_add_history_item(darktable.develop, self, TRUE); return FALSE; } static void colorpicker_callback(GtkColorButton *widget, dt_iop_module_t *self) { if(self->dt->gui->reset) return; dt_iop_invert_gui_data_t *g = (dt_iop_invert_gui_data_t *)self->gui_data; dt_iop_invert_params_t *p = (dt_iop_invert_params_t *)self->params; // turn off the other color picker so that this tool actually works ... gtk_toggle_button_set_active(GTK_TOGGLE_BUTTON(g->picker), FALSE); GdkRGBA c; gtk_color_chooser_get_rgba(GTK_COLOR_CHOOSER(widget), &c); p->color[0] = c.red; p->color[1] = c.green; p->color[2] = c.blue; dt_dev_add_history_item(darktable.develop, self, TRUE); } static int FC(const int row, const int col, const unsigned int filters) { return filters >> (((row << 1 & 14) + (col & 1)) << 1) & 3; } static uint8_t FCxtrans(const int row, const int col, const dt_iop_roi_t *const roi, uint8_t (*const xtrans)[6]) { return xtrans[(row + roi->y) % 6][(col + roi->x) % 6]; } 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) { dt_iop_invert_data_t *d = (dt_iop_invert_data_t *)piece->data; const float *const m = piece->pipe->processed_maximum; const float film_rgb[3] = { d->color[0], d->color[1], d->color[2] }; const float film_rgb_f[3] = { d->color[0] * m[0], d->color[1] * m[1], d->color[2] * m[2] }; // FIXME: it could be wise to make this a NOP when picking colors. not sure about that though. // if(self->request_color_pick){ // do nothing // } const int filters = dt_image_filter(&piece->pipe->image); uint8_t (*const xtrans)[6] = self->dev->image_storage.xtrans; if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && (filters == 9u)) { // xtrans float mosaiced #ifdef _OPENMP #pragma omp parallel for default(none) shared(roi_out, ivoid, ovoid) schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { const float *in = ((float *)ivoid) + (size_t)j * roi_out->width; float *out = ((float *)ovoid) + (size_t)j * roi_out->width; for(int i = 0; i < roi_out->width; i++, out++, in++) *out = CLAMP(film_rgb_f[FCxtrans(j, i, roi_out, xtrans)] - *in, 0.0f, 1.0f); } for(int k = 0; k < 3; k++) piece->pipe->processed_maximum[k] = 1.0f; } else if(!dt_dev_pixelpipe_uses_downsampled_input(piece->pipe) && filters) { // bayer float mosaiced const __m128 val_min = _mm_setzero_ps(); const __m128 val_max = _mm_set1_ps(1.0f); #ifdef _OPENMP #pragma omp parallel for default(none) shared(roi_out, ivoid, ovoid) schedule(static) #endif for(int j = 0; j < roi_out->height; j++) { const float *in = ((float *)ivoid) + (size_t)j * roi_out->width; float *out = ((float *)ovoid) + (size_t)j * roi_out->width; int i = 0; int alignment = ((4 - (j * roi_out->width & (4 - 1))) & (4 - 1)); // process unaligned pixels for(; i < alignment; i++, out++, in++) *out = CLAMP(film_rgb_f[FC(j + roi_out->y, i + roi_out->x, filters)] - *in, 0.0f, 1.0f); const __m128 film = _mm_set_ps(film_rgb_f[FC(j + roi_out->y, roi_out->x + i + 3, filters)], film_rgb_f[FC(j + roi_out->y, roi_out->x + i + 2, filters)], film_rgb_f[FC(j + roi_out->y, roi_out->x + i + 1, filters)], film_rgb_f[FC(j + roi_out->y, roi_out->x + i, filters)]); // process aligned pixels with SSE for(; i < roi_out->width - (4 - 1); i += 4, in += 4, out += 4) { const __m128 input = _mm_load_ps(in); const __m128 subtracted = _mm_sub_ps(film, input); _mm_stream_ps(out, _mm_max_ps(_mm_min_ps(subtracted, val_max), val_min)); } // process the rest for(; i < roi_out->width; i++, out++, in++) *out = CLAMP(film_rgb_f[FC(j + roi_out->y, i + roi_out->x, filters)] - *in, 0.0f, 1.0f); } _mm_sfence(); for(int k = 0; k < 3; k++) piece->pipe->processed_maximum[k] = 1.0f; } else { // non-mosaiced const int ch = piece->colors; const __m128 film = _mm_set_ps(1.0f, film_rgb[2], film_rgb[1], film_rgb[0]); #ifdef _OPENMP #pragma omp parallel for default(none) shared(roi_out, ivoid, ovoid) schedule(static) #endif for(int k = 0; k < roi_out->height; k++) { const float *in = ((float *)ivoid) + (size_t)ch * k * roi_out->width; float *out = ((float *)ovoid) + (size_t)ch * k * roi_out->width; for(int j = 0; j < roi_out->width; j++, in += ch, out += ch) { const __m128 input = _mm_load_ps(in); const __m128 subtracted = _mm_sub_ps(film, input); _mm_stream_ps(out, subtracted); } } _mm_sfence(); if(piece->pipe->mask_display) dt_iop_alpha_copy(ivoid, ovoid, roi_out->width, roi_out->height); } }