equalizer_eaw.h

Go to the documentation of this file.

/*
    This file is part of darktable,
    Copyright (C) 2009-2021 darktable developers.
 
    darktable is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 3 of the License, or
    (at your option) any later version.
 
    darktable is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.
 
    You should have received a copy of the GNU General Public License
    along with darktable.  If not, see <http://www.gnu.org/licenses/>.
*/
 
#pragma once
 
// edge-avoiding wavelet:
#define gweight(i, j, ii, jj)                                                                                \
  1.0 / (fabsf(weight_a[l][(size_t)wd * ((j) >> (l - 1)) + ((i) >> (l - 1))]                                 \
               - weight_a[l][(size_t)wd * ((jj) >> (l - 1)) + ((ii) >> (l - 1))]) + 1.e-5)
// #define gweight(i, j, ii, jj) 1.0/(powf(fabsf(weight_a[l][wd*((j)>>(l-1)) + ((i)>>(l-1))] -
// weight_a[l][wd*((jj)>>(l-1)) + ((ii)>>(l-1))]),0.8)+1.e-5)
// std cdf(2,2) wavelet:
// #define gweight(i, j, ii, jj) (wd ? 1.0 : 1.0) //1.0
#define gbuf(BUF, A, B) ((BUF)[4 * ((size_t)width * ((B)) + ((A))) + ch])
 
 
static void dt_iop_equalizer_wtf(float *const buf, float **weight_a, const int l, const int width, const int height)
{
  const int wd = (int)(1 + (width >> (l - 1))), ht = (int)(1 + (height >> (l - 1)));
  int ch = 0;
  // store weights for luma channel only, chroma uses same basis.
  for(int j = 0; j < ht - 1; j++)
  {
    for(int i = 0; i < wd - 1; i++) weight_a[l][(size_t)j * wd + i] = gbuf(buf, i << (l - 1), j << (l - 1));
    weight_a[l][j * wd + (wd - 1)] = 0.0f; // zero out right-most column
  }
  for(int i = 0; i < wd; i++) // zero out the bottom row
    weight_a[l][(ht-1) * wd + i] = 0.0f;
 
  const int step = 1 << l;
  const int st = step / 2;
 
  size_t scratch_size;
  float *const restrict tmp_width_buf = dt_alloc_perthread_float(width, &scratch_size);
  if(tmp_width_buf == NULL) return;
 
#ifdef _OPENMP
#pragma omp parallel for default(none) \
  dt_omp_firstprivate(height, l, st, step, tmp_width_buf, scratch_size, wd, width) \
  dt_omp_sharedconst(buf) \
  shared(weight_a) \
  private(ch) \
  schedule(static)
#endif
  for(int j = 0; j < height; j++)
  {
    // rows
    // precompute weights:
    float *tmp = dt_get_perthread(tmp_width_buf, scratch_size);
    for(int i = 0; i < width - st; i += st) tmp[i] = gweight(i, j, i + st, j);
    // predict, get detail
    int i = st;
    for(; i < width - st; i += step)
      for(ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) -= (tmp[i - st] * gbuf(buf, i - st, j) + tmp[i] * gbuf(buf, i + st, j))
                           / (tmp[i - st] + tmp[i]);
    if(i < width)
      for(ch = 0; ch < 3; ch++) gbuf(buf, i, j) -= gbuf(buf, i - st, j);
    // update coarse
    for(ch = 0; ch < 3; ch++) gbuf(buf, 0, j) += gbuf(buf, st, j) * 0.5f;
    for(i = step; i < width - st; i += step)
      for(ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) += (tmp[i - st] * gbuf(buf, i - st, j) + tmp[i] * gbuf(buf, i + st, j))
                           / (2.0 * (tmp[i - st] + tmp[i]));
    if(i < width)
      for(ch = 0; ch < 3; ch++) gbuf(buf, i, j) += gbuf(buf, i - st, j) * .5f;
  }
 
  dt_free_align(tmp_width_buf);
 
  float *const restrict tmp_height_buf = dt_alloc_perthread_float(height, &scratch_size);
  if(tmp_height_buf == NULL) return;
 
#ifdef _OPENMP
#pragma omp parallel for default(none) \
  dt_omp_firstprivate(height, l, st, step, tmp_height_buf, scratch_size, wd, width) \
  dt_omp_sharedconst(buf) \
  shared(weight_a) \
  private(ch) \
  schedule(static)
#endif
  for(int i = 0; i < width; i++)
  {
    // cols
    // precompute weights:
    float *const tmp = dt_get_perthread(tmp_height_buf, scratch_size);
    for(int j = 0; j < height - st; j += st) tmp[j] = gweight(i, j, i, j + st);
    int j = st;
    // predict, get detail
    for(; j < height - st; j += step)
      for(ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) -= (tmp[j - st] * gbuf(buf, i, j - st) + tmp[j] * gbuf(buf, i, j + st))
                           / (tmp[j - st] + tmp[j]);
    if(j < height)
      for(ch = 0; ch < 3; ch++) gbuf(buf, i, j) -= gbuf(buf, i, j - st);
    // update
    for(ch = 0; ch < 3; ch++) gbuf(buf, i, 0) += gbuf(buf, i, st) * 0.5;
    for(j = step; j < height - st; j += step)
      for(ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) += (tmp[j - st] * gbuf(buf, i, j - st) + tmp[j] * gbuf(buf, i, j + st))
                           / (2.0 * (tmp[j - st] + tmp[j]));
    if(j < height)
      for(ch = 0; ch < 3; ch++) gbuf(buf, i, j) += gbuf(buf, i, j - st) * .5f;
  }
 
  dt_free_align(tmp_height_buf);
}
 
static void dt_iop_equalizer_iwtf(float *buf, float **weight_a, const int l, const int width, const int height)
{
  const int step = 1 << l;
  const int st = step / 2;
  const int wd = (int)(1 + (width >> (l - 1)));
 
  size_t scratch_size;
  float *const restrict tmp_height_buf = dt_alloc_perthread_float(height, &scratch_size);
  if(tmp_height_buf == NULL) return;
 
#ifdef _OPENMP
#pragma omp parallel for default(none) \
  dt_omp_firstprivate(height, l, st, step, tmp_height_buf, scratch_size, wd, width) \
  shared(weight_a, buf) \
  schedule(static)
#endif
  for(int i = 0; i < width; i++)
  {
    // cols
    float *const restrict tmp = dt_get_perthread(tmp_height_buf, scratch_size);
    int j;
    for(j = 0; j < height - st; j += st) tmp[j] = gweight(i, j, i, j + st);
    // update coarse
    for(int ch = 0; ch < 3; ch++) gbuf(buf, i, 0) -= gbuf(buf, i, st) * 0.5f;
    for(j = step; j < height - st; j += step)
      for(int ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) -= (tmp[j - st] * gbuf(buf, i, j - st) + tmp[j] * gbuf(buf, i, j + st))
                           / (2.0 * (tmp[j - st] + tmp[j]));
    if(j < height)
      for(int ch = 0; ch < 3; ch++) gbuf(buf, i, j) -= gbuf(buf, i, j - st) * .5f;
    // predict
    for(j = st; j < height - st; j += step)
      for(int ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) += (tmp[j - st] * gbuf(buf, i, j - st) + tmp[j] * gbuf(buf, i, j + st))
                           / (tmp[j - st] + tmp[j]);
    if(j < height)
      for(int ch = 0; ch < 3; ch++) gbuf(buf, i, j) += gbuf(buf, i, j - st);
  }
 
  dt_free_align(tmp_height_buf);
 
  float *const restrict tmp_width_buf = dt_alloc_perthread_float(width, &scratch_size);
  if(tmp_width_buf == NULL) return;
  
#ifdef _OPENMP
#pragma omp parallel for default(none) \
  dt_omp_firstprivate(height, l, st, step, tmp_width_buf, scratch_size, wd, width) \
  shared(weight_a, buf) \
  schedule(static)
#endif
  for(int j = 0; j < height; j++)
  {
    // rows
    float *const restrict tmp = dt_get_perthread(tmp_width_buf, scratch_size);
    for(int i = 0; i < width - st; i += st) tmp[i] = gweight(i, j, i + st, j);
    // update
    for(int ch = 0; ch < 3; ch++) gbuf(buf, 0, j) -= gbuf(buf, st, j) * 0.5f;
    int i;
    for(i = step; i < width - st; i += step)
      for(int ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) -= (tmp[i - st] * gbuf(buf, i - st, j) + tmp[i] * gbuf(buf, i + st, j))
                           / (2.0 * (tmp[i - st] + tmp[i]));
    if(i < width)
      for(int ch = 0; ch < 3; ch++) gbuf(buf, i, j) -= gbuf(buf, i - st, j) * 0.5f;
    // predict
    for(i = st; i < width - st; i += step)
      for(int ch = 0; ch < 3; ch++)
        gbuf(buf, i, j) += (tmp[i - st] * gbuf(buf, i - st, j) + tmp[i] * gbuf(buf, i + st, j))
                           / (tmp[i - st] + tmp[i]);
    if(i < width)
      for(int ch = 0; ch < 3; ch++) gbuf(buf, i, j) += gbuf(buf, i - st, j);
  }
 
  dt_free_align(tmp_width_buf);
}
 
#undef gbuf
#undef gweight
// clang-format off
// modelines: These editor modelines have been set for all relevant files by tools/update_modelines.py
// vim: shiftwidth=2 expandtab tabstop=2 cindent
// kate: tab-indents: off; indent-width 2; replace-tabs on; indent-mode cstyle; remove-trailing-spaces modified;
// clang-format on