path: root/source/blender/compositor/operations/COM_ScreenLensDistortionOperation.cc

/*
 * This program 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 2
 * of the License, or (at your option) any later version.
 *
 * This program 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 this program; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
 *
 * Copyright 2011, Blender Foundation.
 */

#include "COM_ScreenLensDistortionOperation.h"

#include "COM_ConstantOperation.h"

#include "BLI_rand.h"

#include "PIL_time.h"

namespace blender::compositor {

ScreenLensDistortionOperation::ScreenLensDistortionOperation()
{
  this->add_input_socket(DataType::Color);
  this->add_input_socket(DataType::Value);
  this->add_input_socket(DataType::Value);
  this->add_output_socket(DataType::Color);
  this->flags.complex = true;
  input_program_ = nullptr;
  distortion_ = 0.0f;
  dispersion_ = 0.0f;
  distortion_const_ = false;
  dispersion_const_ = false;
  variables_ready_ = false;
}

void ScreenLensDistortionOperation::set_distortion(float distortion)
{
  distortion_ = distortion;
  distortion_const_ = true;
}

void ScreenLensDistortionOperation::set_dispersion(float dispersion)
{
  dispersion_ = dispersion;
  dispersion_const_ = true;
}

void ScreenLensDistortionOperation::init_data()
{
  cx_ = 0.5f * (float)get_width();
  cy_ = 0.5f * (float)get_height();

  switch (execution_model_) {
    case eExecutionModel::FullFrame: {
      NodeOperation *distortion_op = get_input_operation(1);
      NodeOperation *dispersion_op = get_input_operation(2);
      if (!distortion_const_ && distortion_op->get_flags().is_constant_operation) {
        distortion_ = static_cast<ConstantOperation *>(distortion_op)->get_constant_elem()[0];
      }
      if (!dispersion_const_ && distortion_op->get_flags().is_constant_operation) {
        dispersion_ = static_cast<ConstantOperation *>(dispersion_op)->get_constant_elem()[0];
      }
      update_variables(distortion_, dispersion_);
      break;
    }
    case eExecutionModel::Tiled: {
      /* If both are constant, init variables once. */
      if (distortion_const_ && dispersion_const_) {
        update_variables(distortion_, dispersion_);
        variables_ready_ = true;
      }
      break;
    }
  }
}

void ScreenLensDistortionOperation::init_execution()
{
  input_program_ = this->get_input_socket_reader(0);
  this->init_mutex();

  uint rng_seed = (uint)(PIL_check_seconds_timer_i() & UINT_MAX);
  rng_seed ^= (uint)POINTER_AS_INT(input_program_);
  rng_ = BLI_rng_new(rng_seed);
}

void *ScreenLensDistortionOperation::initialize_tile_data(rcti * /*rect*/)
{
  void *buffer = input_program_->initialize_tile_data(nullptr);

  /* get distortion/dispersion values once, by reading inputs at (0,0)
   * XXX this assumes invariable values (no image inputs),
   * we don't have a nice generic system for that yet
   */
  if (!variables_ready_) {
    this->lock_mutex();

    if (!distortion_const_) {
      float result[4];
      get_input_socket_reader(1)->read_sampled(result, 0, 0, PixelSampler::Nearest);
      distortion_ = result[0];
    }
    if (!dispersion_const_) {
      float result[4];
      get_input_socket_reader(2)->read_sampled(result, 0, 0, PixelSampler::Nearest);
      dispersion_ = result[0];
    }

    update_variables(distortion_, dispersion_);
    variables_ready_ = true;

    this->unlock_mutex();
  }

  return buffer;
}

void ScreenLensDistortionOperation::get_uv(const float xy[2], float uv[2]) const
{
  uv[0] = sc_ * ((xy[0] + 0.5f) - cx_) / cx_;
  uv[1] = sc_ * ((xy[1] + 0.5f) - cy_) / cy_;
}

void ScreenLensDistortionOperation::distort_uv(const float uv[2], float t, float xy[2]) const
{
  float d = 1.0f / (1.0f + sqrtf(t));
  xy[0] = (uv[0] * d + 0.5f) * get_width() - 0.5f;
  xy[1] = (uv[1] * d + 0.5f) * get_height() - 0.5f;
}

bool ScreenLensDistortionOperation::get_delta(float r_sq,
                                              float k4,
                                              const float uv[2],
                                              float delta[2]) const
{
  float t = 1.0f - k4 * r_sq;
  if (t >= 0.0f) {
    distort_uv(uv, t, delta);
    return true;
  }

  return false;
}

void ScreenLensDistortionOperation::accumulate(const MemoryBuffer *buffer,
                                               int a,
                                               int b,
                                               float r_sq,
                                               const float uv[2],
                                               const float delta[3][2],
                                               float sum[4],
                                               int count[3]) const
{
  float color[4];

  float dsf = len_v2v2(delta[a], delta[b]) + 1.0f;
  int ds = jitter_ ? (dsf < 4.0f ? 2 : (int)sqrtf(dsf)) : (int)dsf;
  float sd = 1.0f / (float)ds;

  float k4 = k4_[a];
  float dk4 = dk4_[a];

  for (float z = 0; z < ds; z++) {
    float tz = (z + (jitter_ ? BLI_rng_get_float(rng_) : 0.5f)) * sd;
    float t = 1.0f - (k4 + tz * dk4) * r_sq;

    float xy[2];
    distort_uv(uv, t, xy);
    switch (execution_model_) {
      case eExecutionModel::Tiled:
        buffer->read_bilinear(color, xy[0], xy[1]);
        break;
      case eExecutionModel::FullFrame:
        buffer->read_elem_bilinear(xy[0], xy[1], color);
        break;
    }

    sum[a] += (1.0f - tz) * color[a];
    sum[b] += (tz)*color[b];
    count[a]++;
    count[b]++;
  }
}

void ScreenLensDistortionOperation::execute_pixel(float output[4], int x, int y, void *data)
{
  MemoryBuffer *buffer = (MemoryBuffer *)data;
  float xy[2] = {(float)x, (float)y};
  float uv[2];
  get_uv(xy, uv);
  float uv_dot = len_squared_v2(uv);

  int count[3] = {0, 0, 0};
  float delta[3][2];
  float sum[4] = {0, 0, 0, 0};

  bool valid_r = get_delta(uv_dot, k4_[0], uv, delta[0]);
  bool valid_g = get_delta(uv_dot, k4_[1], uv, delta[1]);
  bool valid_b = get_delta(uv_dot, k4_[2], uv, delta[2]);

  if (valid_r && valid_g && valid_b) {
    accumulate(buffer, 0, 1, uv_dot, uv, delta, sum, count);
    accumulate(buffer, 1, 2, uv_dot, uv, delta, sum, count);

    if (count[0]) {
      output[0] = 2.0f * sum[0] / (float)count[0];
    }
    if (count[1]) {
      output[1] = 2.0f * sum[1] / (float)count[1];
    }
    if (count[2]) {
      output[2] = 2.0f * sum[2] / (float)count[2];
    }

    /* set alpha */
    output[3] = 1.0f;
  }
  else {
    zero_v4(output);
  }
}

void ScreenLensDistortionOperation::deinit_execution()
{
  this->deinit_mutex();
  input_program_ = nullptr;
  BLI_rng_free(rng_);
}

void ScreenLensDistortionOperation::determineUV(float result[6], float x, float y) const
{
  const float xy[2] = {x, y};
  float uv[2];
  get_uv(xy, uv);
  float uv_dot = len_squared_v2(uv);

  copy_v2_v2(result + 0, xy);
  copy_v2_v2(result + 2, xy);
  copy_v2_v2(result + 4, xy);
  get_delta(uv_dot, k4_[0], uv, result + 0);
  get_delta(uv_dot, k4_[1], uv, result + 2);
  get_delta(uv_dot, k4_[2], uv, result + 4);
}

bool ScreenLensDistortionOperation::determine_depending_area_of_interest(
    rcti * /*input*/, ReadBufferOperation *read_operation, rcti *output)
{
  rcti new_input_value;
  new_input_value.xmin = 0;
  new_input_value.ymin = 0;
  new_input_value.xmax = 2;
  new_input_value.ymax = 2;

  NodeOperation *operation = get_input_operation(1);
  if (operation->determine_depending_area_of_interest(&new_input_value, read_operation, output)) {
    return true;
  }

  operation = get_input_operation(2);
  if (operation->determine_depending_area_of_interest(&new_input_value, read_operation, output)) {
    return true;
  }

  /* XXX the original method of estimating the area-of-interest does not work
   * it assumes a linear increase/decrease of mapped coordinates, which does not
   * yield correct results for the area and leaves uninitialized buffer areas.
   * So now just use the full image area, which may not be as efficient but works at least ...
   */
#if 1
  rcti image_input;

  operation = get_input_operation(0);
  image_input.xmax = operation->get_width();
  image_input.xmin = 0;
  image_input.ymax = operation->get_height();
  image_input.ymin = 0;

  if (operation->determine_depending_area_of_interest(&image_input, read_operation, output)) {
    return true;
  }
  return false;
#else
  rcti new_input;
  const float margin = 2;

  BLI_rcti_init_minmax(&new_input);

  if (dispersion_const_ && distortion_const_) {
    /* update from fixed distortion/dispersion */
#  define UPDATE_INPUT(x, y) \
    { \
      float coords[6]; \
      determineUV(coords, x, y); \
      new_input.xmin = min_ffff(new_input.xmin, coords[0], coords[2], coords[4]); \
      new_input.ymin = min_ffff(new_input.ymin, coords[1], coords[3], coords[5]); \
      new_input.xmax = max_ffff(new_input.xmax, coords[0], coords[2], coords[4]); \
      new_input.ymax = max_ffff(new_input.ymax, coords[1], coords[3], coords[5]); \
    } \
    (void)0

    UPDATE_INPUT(input->xmin, input->xmax);
    UPDATE_INPUT(input->xmin, input->ymax);
    UPDATE_INPUT(input->xmax, input->ymax);
    UPDATE_INPUT(input->xmax, input->ymin);

#  undef UPDATE_INPUT
  }
  else {
    /* use maximum dispersion 1.0 if not const */
    float dispersion = dispersion_const_ ? dispersion_ : 1.0f;

#  define UPDATE_INPUT(x, y, distortion) \
    { \
      float coords[6]; \
      update_variables(distortion, dispersion); \
      determineUV(coords, x, y); \
      new_input.xmin = min_ffff(new_input.xmin, coords[0], coords[2], coords[4]); \
      new_input.ymin = min_ffff(new_input.ymin, coords[1], coords[3], coords[5]); \
      new_input.xmax = max_ffff(new_input.xmax, coords[0], coords[2], coords[4]); \
      new_input.ymax = max_ffff(new_input.ymax, coords[1], coords[3], coords[5]); \
    } \
    (void)0

    if (distortion_const_) {
      /* update from fixed distortion */
      UPDATE_INPUT(input->xmin, input->xmax, distortion_);
      UPDATE_INPUT(input->xmin, input->ymax, distortion_);
      UPDATE_INPUT(input->xmax, input->ymax, distortion_);
      UPDATE_INPUT(input->xmax, input->ymin, distortion_);
    }
    else {
      /* update from min/max distortion (-1..1) */
      UPDATE_INPUT(input->xmin, input->xmax, -1.0f);
      UPDATE_INPUT(input->xmin, input->ymax, -1.0f);
      UPDATE_INPUT(input->xmax, input->ymax, -1.0f);
      UPDATE_INPUT(input->xmax, input->ymin, -1.0f);

      UPDATE_INPUT(input->xmin, input->xmax, 1.0f);
      UPDATE_INPUT(input->xmin, input->ymax, 1.0f);
      UPDATE_INPUT(input->xmax, input->ymax, 1.0f);
      UPDATE_INPUT(input->xmax, input->ymin, 1.0f);

#  undef UPDATE_INPUT
    }
  }

  new_input.xmin -= margin;
  new_input.ymin -= margin;
  new_input.xmax += margin;
  new_input.ymax += margin;

  operation = get_input_operation(0);
  if (operation->determine_depending_area_of_interest(&new_input, read_operation, output)) {
    return true;
  }
  return false;
#endif
}

void ScreenLensDistortionOperation::update_variables(float distortion, float dispersion)
{
  k_[1] = max_ff(min_ff(distortion, 1.0f), -0.999f);
  /* Smaller dispersion range for somewhat more control. */
  float d = 0.25f * max_ff(min_ff(dispersion, 1.0f), 0.0f);
  k_[0] = max_ff(min_ff((k_[1] + d), 1.0f), -0.999f);
  k_[2] = max_ff(min_ff((k_[1] - d), 1.0f), -0.999f);
  maxk_ = max_fff(k_[0], k_[1], k_[2]);
  sc_ = (fit_ && (maxk_ > 0.0f)) ? (1.0f / (1.0f + 2.0f * maxk_)) : (1.0f / (1.0f + maxk_));
  dk4_[0] = 4.0f * (k_[1] - k_[0]);
  dk4_[1] = 4.0f * (k_[2] - k_[1]);
  dk4_[2] = 0.0f; /* unused */

  mul_v3_v3fl(k4_, k_, 4.0f);
}

void ScreenLensDistortionOperation::determine_canvas(const rcti &preferred_area, rcti &r_area)
{
  switch (execution_model_) {
    case eExecutionModel::FullFrame: {
      set_determined_canvas_modifier([=](rcti &canvas) {
        /* Ensure screen space. */
        BLI_rcti_translate(&canvas, -canvas.xmin, -canvas.ymin);
      });
      break;
    }
    default:
      break;
  }

  NodeOperation::determine_canvas(preferred_area, r_area);
}

void ScreenLensDistortionOperation::get_area_of_interest(const int input_idx,
                                                         const rcti &UNUSED(output_area),
                                                         rcti &r_input_area)
{
  if (input_idx != 0) {
    /* Dispersion and distortion inputs are used as constants only. */
    r_input_area = COM_CONSTANT_INPUT_AREA_OF_INTEREST;
  }

  /* XXX the original method of estimating the area-of-interest does not work
   * it assumes a linear increase/decrease of mapped coordinates, which does not
   * yield correct results for the area and leaves uninitialized buffer areas.
   * So now just use the full image area, which may not be as efficient but works at least ...
   */
#if 1
  NodeOperation *image = get_input_operation(0);
  r_input_area = image->get_canvas();

#else /* Original method in tiled implementation. */
  rcti new_input;
  const float margin = 2;

  BLI_rcti_init_minmax(&new_input);

  if (dispersion_const_ && distortion_const_) {
    /* update from fixed distortion/dispersion */
#  define UPDATE_INPUT(x, y) \
    { \
      float coords[6]; \
      determineUV(coords, x, y); \
      new_input.xmin = min_ffff(new_input.xmin, coords[0], coords[2], coords[4]); \
      new_input.ymin = min_ffff(new_input.ymin, coords[1], coords[3], coords[5]); \
      new_input.xmax = max_ffff(new_input.xmax, coords[0], coords[2], coords[4]); \
      new_input.ymax = max_ffff(new_input.ymax, coords[1], coords[3], coords[5]); \
    } \
    (void)0

    UPDATE_INPUT(input->xmin, input->xmax);
    UPDATE_INPUT(input->xmin, input->ymax);
    UPDATE_INPUT(input->xmax, input->ymax);
    UPDATE_INPUT(input->xmax, input->ymin);

#  undef UPDATE_INPUT
  }
  else {
    /* use maximum dispersion 1.0 if not const */
    float dispersion = dispersion_const_ ? dispersion_ : 1.0f;

#  define UPDATE_INPUT(x, y, distortion) \
    { \
      float coords[6]; \
      update_variables(distortion, dispersion); \
      determineUV(coords, x, y); \
      new_input.xmin = min_ffff(new_input.xmin, coords[0], coords[2], coords[4]); \
      new_input.ymin = min_ffff(new_input.ymin, coords[1], coords[3], coords[5]); \
      new_input.xmax = max_ffff(new_input.xmax, coords[0], coords[2], coords[4]); \
      new_input.ymax = max_ffff(new_input.ymax, coords[1], coords[3], coords[5]); \
    } \
    (void)0

    if (distortion_const_) {
      /* update from fixed distortion */
      UPDATE_INPUT(input->xmin, input->xmax, distortion_);
      UPDATE_INPUT(input->xmin, input->ymax, distortion_);
      UPDATE_INPUT(input->xmax, input->ymax, distortion_);
      UPDATE_INPUT(input->xmax, input->ymin, distortion_);
    }
    else {
      /* update from min/max distortion (-1..1) */
      UPDATE_INPUT(input->xmin, input->xmax, -1.0f);
      UPDATE_INPUT(input->xmin, input->ymax, -1.0f);
      UPDATE_INPUT(input->xmax, input->ymax, -1.0f);
      UPDATE_INPUT(input->xmax, input->ymin, -1.0f);

      UPDATE_INPUT(input->xmin, input->xmax, 1.0f);
      UPDATE_INPUT(input->xmin, input->ymax, 1.0f);
      UPDATE_INPUT(input->xmax, input->ymax, 1.0f);
      UPDATE_INPUT(input->xmax, input->ymin, 1.0f);

#  undef UPDATE_INPUT
    }
  }

  new_input.xmin -= margin;
  new_input.ymin -= margin;
  new_input.xmax += margin;
  new_input.ymax += margin;

  operation = get_input_operation(0);
  if (operation->determine_depending_area_of_interest(&new_input, read_operation, output)) {
    return true;
  }
  return false;
#endif
}

void ScreenLensDistortionOperation::update_memory_buffer_partial(MemoryBuffer *output,
                                                                 const rcti &area,
                                                                 Span<MemoryBuffer *> inputs)
{
  const MemoryBuffer *input_image = inputs[0];
  for (BuffersIterator<float> it = output->iterate_with({}, area); !it.is_end(); ++it) {
    float xy[2] = {(float)it.x, (float)it.y};
    float uv[2];
    get_uv(xy, uv);
    const float uv_dot = len_squared_v2(uv);

    float delta[3][2];
    const bool valid_r = get_delta(uv_dot, k4_[0], uv, delta[0]);
    const bool valid_g = get_delta(uv_dot, k4_[1], uv, delta[1]);
    const bool valid_b = get_delta(uv_dot, k4_[2], uv, delta[2]);
    if (!(valid_r && valid_g && valid_b)) {
      zero_v4(it.out);
      continue;
    }

    int count[3] = {0, 0, 0};
    float sum[4] = {0, 0, 0, 0};
    accumulate(input_image, 0, 1, uv_dot, uv, delta, sum, count);
    accumulate(input_image, 1, 2, uv_dot, uv, delta, sum, count);

    if (count[0]) {
      it.out[0] = 2.0f * sum[0] / (float)count[0];
    }
    if (count[1]) {
      it.out[1] = 2.0f * sum[1] / (float)count[1];
    }
    if (count[2]) {
      it.out[2] = 2.0f * sum[2] / (float)count[2];
    }

    /* Set alpha. */
    it.out[3] = 1.0f;
  }
}

}  // namespace blender::compositor