/* SPDX-License-Identifier: GPL-2.0-or-later
 * Copyright 2011 Blender Foundation. */

#include "BLI_jitter_2d.h"

#include "COM_VectorBlurOperation.h"

namespace blender::compositor {

/* Defined */
#define PASS_VECTOR_MAX 10000.0f

/* Forward declarations */
struct DrawBufPixel;
struct ZSpan;
void zbuf_accumulate_vecblur(NodeBlurData *nbd,
                             int xsize,
                             int ysize,
                             float *newrect,
                             const float *imgrect,
                             float *vecbufrect,
                             const float *zbufrect);
void zbuf_alloc_span(ZSpan *zspan, int rectx, int recty, float clipcrop);
void zbuf_free_span(ZSpan *zspan);
void antialias_tagbuf(int xsize, int ysize, char *rectmove);

/* VectorBlurOperation */

VectorBlurOperation::VectorBlurOperation()
{
  this->add_input_socket(DataType::Color);
  this->add_input_socket(DataType::Value); /* ZBUF */
  this->add_input_socket(DataType::Color); /* SPEED */
  this->add_output_socket(DataType::Color);
  settings_ = nullptr;
  cached_instance_ = nullptr;
  input_image_program_ = nullptr;
  input_speed_program_ = nullptr;
  input_zprogram_ = nullptr;
  flags_.complex = true;
  flags_.is_fullframe_operation = true;
}
void VectorBlurOperation::init_execution()
{
  init_mutex();
  input_image_program_ = get_input_socket_reader(0);
  input_zprogram_ = get_input_socket_reader(1);
  input_speed_program_ = get_input_socket_reader(2);
  cached_instance_ = nullptr;
  QualityStepHelper::init_execution(COM_QH_INCREASE);
}

void VectorBlurOperation::execute_pixel(float output[4], int x, int y, void *data)
{
  float *buffer = (float *)data;
  int index = (y * this->get_width() + x) * COM_DATA_TYPE_COLOR_CHANNELS;
  copy_v4_v4(output, &buffer[index]);
}

void VectorBlurOperation::deinit_execution()
{
  deinit_mutex();
  input_image_program_ = nullptr;
  input_speed_program_ = nullptr;
  input_zprogram_ = nullptr;
  if (cached_instance_) {
    MEM_freeN(cached_instance_);
    cached_instance_ = nullptr;
  }
}
void *VectorBlurOperation::initialize_tile_data(rcti *rect)
{
  if (cached_instance_) {
    return cached_instance_;
  }

  lock_mutex();
  if (cached_instance_ == nullptr) {
    MemoryBuffer *tile = (MemoryBuffer *)input_image_program_->initialize_tile_data(rect);
    MemoryBuffer *speed = (MemoryBuffer *)input_speed_program_->initialize_tile_data(rect);
    MemoryBuffer *z = (MemoryBuffer *)input_zprogram_->initialize_tile_data(rect);
    float *data = (float *)MEM_dupallocN(tile->get_buffer());
    this->generate_vector_blur(data, tile, speed, z);
    cached_instance_ = data;
  }
  unlock_mutex();
  return cached_instance_;
}

bool VectorBlurOperation::determine_depending_area_of_interest(rcti * /*input*/,
                                                               ReadBufferOperation *read_operation,
                                                               rcti *output)
{
  if (cached_instance_ == nullptr) {
    rcti new_input;
    new_input.xmax = this->get_width();
    new_input.xmin = 0;
    new_input.ymax = this->get_height();
    new_input.ymin = 0;
    return NodeOperation::determine_depending_area_of_interest(&new_input, read_operation, output);
  }

  return false;
}

void VectorBlurOperation::get_area_of_interest(const int UNUSED(input_idx),
                                               const rcti &UNUSED(output_area),
                                               rcti &r_input_area)
{
  r_input_area = this->get_canvas();
}

void VectorBlurOperation::update_memory_buffer(MemoryBuffer *output,
                                               const rcti &area,
                                               Span<MemoryBuffer *> inputs)
{
  /* TODO(manzanilla): once tiled implementation is removed, run multi-threaded where possible. */
  if (!cached_instance_) {
    MemoryBuffer *image = inputs[IMAGE_INPUT_INDEX];
    const bool is_image_inflated = image->is_a_single_elem();
    image = is_image_inflated ? image->inflate() : image;

    /* Must be a copy because it's modified in #generate_vector_blur. */
    MemoryBuffer *speed = inputs[SPEED_INPUT_INDEX];
    speed = speed->is_a_single_elem() ? speed->inflate() : new MemoryBuffer(*speed);

    MemoryBuffer *z = inputs[Z_INPUT_INDEX];
    const bool is_z_inflated = z->is_a_single_elem();
    z = is_z_inflated ? z->inflate() : z;

    cached_instance_ = (float *)MEM_dupallocN(image->get_buffer());
    this->generate_vector_blur(cached_instance_, image, speed, z);

    if (is_image_inflated) {
      delete image;
    }
    delete speed;
    if (is_z_inflated) {
      delete z;
    }
  }

  const int num_channels = COM_data_type_num_channels(get_output_socket()->get_data_type());
  MemoryBuffer buf(cached_instance_, num_channels, this->get_width(), this->get_height());
  output->copy_from(&buf, area);
}

void VectorBlurOperation::generate_vector_blur(float *data,
                                               MemoryBuffer *input_image,
                                               MemoryBuffer *input_speed,
                                               MemoryBuffer *inputZ)
{
  NodeBlurData blurdata;
  blurdata.samples = settings_->samples / QualityStepHelper::get_step();
  blurdata.maxspeed = settings_->maxspeed;
  blurdata.minspeed = settings_->minspeed;
  blurdata.curved = settings_->curved;
  blurdata.fac = settings_->fac;
  zbuf_accumulate_vecblur(&blurdata,
                          this->get_width(),
                          this->get_height(),
                          data,
                          input_image->get_buffer(),
                          input_speed->get_buffer(),
                          inputZ->get_buffer());
}

/* -------------------------------------------------------------------- */
/** \name Spans
 *
 * Duplicated logic from `zbuf.c`.
 * \{ */

/** Span fill in method, is also used to localize data for Z-buffering. */
struct ZSpan {
  /* range for clipping */
  int rectx, recty;

  /* actual filled in range */
  int miny1, maxy1, miny2, maxy2;
  /* vertex pointers detect min/max range in */
  const float *minp1, *maxp1, *minp2, *maxp2;
  float *span1, *span2;

  /* transform from hoco to zbuf co */
  float zmulx, zmuly, zofsx, zofsy;

  int *rectz;
  DrawBufPixel *rectdraw;
  float clipcrop;
};

/* each zbuffer has coordinates transformed to local rect coordinates, so we can simply clip */
void zbuf_alloc_span(ZSpan *zspan, int rectx, int recty, float clipcrop)
{
  memset(zspan, 0, sizeof(ZSpan));

  zspan->rectx = rectx;
  zspan->recty = recty;

  zspan->span1 = (float *)MEM_mallocN(recty * sizeof(float), "zspan");
  zspan->span2 = (float *)MEM_mallocN(recty * sizeof(float), "zspan");

  zspan->clipcrop = clipcrop;
}

void zbuf_free_span(ZSpan *zspan)
{
  if (zspan) {
    if (zspan->span1) {
      MEM_freeN(zspan->span1);
    }
    if (zspan->span2) {
      MEM_freeN(zspan->span2);
    }
    zspan->span1 = zspan->span2 = nullptr;
  }
}

/* reset range for clipping */
static void zbuf_init_span(ZSpan *zspan)
{
  zspan->miny1 = zspan->miny2 = zspan->recty + 1;
  zspan->maxy1 = zspan->maxy2 = -1;
  zspan->minp1 = zspan->maxp1 = zspan->minp2 = zspan->maxp2 = nullptr;
}

static void zbuf_add_to_span(ZSpan *zspan, const float v1[2], const float v2[2])
{
  const float *minv, *maxv;
  float *span;
  float xx1, dx0, xs0;
  int y, my0, my2;

  if (v1[1] < v2[1]) {
    minv = v1;
    maxv = v2;
  }
  else {
    minv = v2;
    maxv = v1;
  }

  my0 = ceil(minv[1]);
  my2 = floor(maxv[1]);

  if (my2 < 0 || my0 >= zspan->recty) {
    return;
  }

  /* clip top */
  if (my2 >= zspan->recty) {
    my2 = zspan->recty - 1;
  }
  /* clip bottom */
  if (my0 < 0) {
    my0 = 0;
  }

  if (my0 > my2) {
    return;
  }
  /* if (my0>my2) should still fill in, that way we get spans that skip nicely */

  xx1 = maxv[1] - minv[1];
  if (xx1 > FLT_EPSILON) {
    dx0 = (minv[0] - maxv[0]) / xx1;
    xs0 = dx0 * (minv[1] - my2) + minv[0];
  }
  else {
    dx0 = 0.0f;
    xs0 = min_ff(minv[0], maxv[0]);
  }

  /* empty span */
  if (zspan->maxp1 == nullptr) {
    span = zspan->span1;
  }
  else { /* does it complete left span? */
    if (maxv == zspan->minp1 || minv == zspan->maxp1) {
      span = zspan->span1;
    }
    else {
      span = zspan->span2;
    }
  }

  if (span == zspan->span1) {
    //      printf("left span my0 %d my2 %d\n", my0, my2);
    if (zspan->minp1 == nullptr || zspan->minp1[1] > minv[1]) {
      zspan->minp1 = minv;
    }
    if (zspan->maxp1 == nullptr || zspan->maxp1[1] < maxv[1]) {
      zspan->maxp1 = maxv;
    }
    if (my0 < zspan->miny1) {
      zspan->miny1 = my0;
    }
    if (my2 > zspan->maxy1) {
      zspan->maxy1 = my2;
    }
  }
  else {
    //      printf("right span my0 %d my2 %d\n", my0, my2);
    if (zspan->minp2 == nullptr || zspan->minp2[1] > minv[1]) {
      zspan->minp2 = minv;
    }
    if (zspan->maxp2 == nullptr || zspan->maxp2[1] < maxv[1]) {
      zspan->maxp2 = maxv;
    }
    if (my0 < zspan->miny2) {
      zspan->miny2 = my0;
    }
    if (my2 > zspan->maxy2) {
      zspan->maxy2 = my2;
    }
  }

  for (y = my2; y >= my0; y--, xs0 += dx0) {
    /* xs0 is the X-coordinate! */
    span[y] = xs0;
  }
}

/** \} */

/* ******************** VECBLUR ACCUM BUF ************************* */

struct DrawBufPixel {
  const float *colpoin;
  float alpha;
};

/**
 * \note Near duplicate of `zspan_scanconvert` in `zbuf.c` with some minor adjustments.
 */
static void zbuf_fill_in_rgba(
    ZSpan *zspan, DrawBufPixel *col, float *v1, float *v2, float *v3, float *v4)
{
  DrawBufPixel *rectpofs, *rp;
  double zxd, zyd, zy0, zverg;
  float x0, y0, z0;
  float x1, y1, z1, x2, y2, z2, xx1;
  const float *span1, *span2;
  float *rectzofs, *rz;
  int x, y;
  int sn1, sn2, rectx, my0, my2;

  /* init */
  zbuf_init_span(zspan);

  /* set spans */
  zbuf_add_to_span(zspan, v1, v2);
  zbuf_add_to_span(zspan, v2, v3);
  zbuf_add_to_span(zspan, v3, v4);
  zbuf_add_to_span(zspan, v4, v1);

  /* clipped */
  if (zspan->minp2 == nullptr || zspan->maxp2 == nullptr) {
    return;
  }

  my0 = max_ii(zspan->miny1, zspan->miny2);
  my2 = min_ii(zspan->maxy1, zspan->maxy2);

  //  printf("my %d %d\n", my0, my2);
  if (my2 < my0) {
    return;
  }

  /* ZBUF DX DY, in floats still */
  x1 = v1[0] - v2[0];
  x2 = v2[0] - v3[0];
  y1 = v1[1] - v2[1];
  y2 = v2[1] - v3[1];
  z1 = v1[2] - v2[2];
  z2 = v2[2] - v3[2];
  x0 = y1 * z2 - z1 * y2;
  y0 = z1 * x2 - x1 * z2;
  z0 = x1 * y2 - y1 * x2;

  if (z0 == 0.0f) {
    return;
  }

  xx1 = (x0 * v1[0] + y0 * v1[1]) / z0 + v1[2];

  zxd = -(double)x0 / (double)z0;
  zyd = -(double)y0 / (double)z0;
  zy0 = ((double)my2) * zyd + (double)xx1;

  /* start-offset in rect */
  rectx = zspan->rectx;
  rectzofs = (float *)(zspan->rectz + rectx * my2);
  rectpofs = ((DrawBufPixel *)zspan->rectdraw) + rectx * my2;

  /* correct span */
  sn1 = (my0 + my2) / 2;
  if (zspan->span1[sn1] < zspan->span2[sn1]) {
    span1 = zspan->span1 + my2;
    span2 = zspan->span2 + my2;
  }
  else {
    span1 = zspan->span2 + my2;
    span2 = zspan->span1 + my2;
  }

  for (y = my2; y >= my0; y--, span1--, span2--) {

    sn1 = floor(*span1);
    sn2 = floor(*span2);
    sn1++;

    if (sn2 >= rectx) {
      sn2 = rectx - 1;
    }
    if (sn1 < 0) {
      sn1 = 0;
    }

    if (sn2 >= sn1) {
      zverg = (double)sn1 * zxd + zy0;
      rz = rectzofs + sn1;
      rp = rectpofs + sn1;
      x = sn2 - sn1;

      while (x >= 0) {
        if (zverg < (double)*rz) {
          *rz = zverg;
          *rp = *col;
        }
        zverg += zxd;
        rz++;
        rp++;
        x--;
      }
    }

    zy0 -= zyd;
    rectzofs -= rectx;
    rectpofs -= rectx;
  }
}

/* char value==255 is filled in, rest should be zero */
/* returns alpha values,
 * but sets alpha to 1 for zero alpha pixels that have an alpha value as neighbor. */
void antialias_tagbuf(int xsize, int ysize, char *rectmove)
{
  char *row1, *row2, *row3;
  char prev, next;
  int a, x, y, step;

  /* 1: tag pixels to be candidate for AA */
  for (y = 2; y < ysize; y++) {
    /* setup rows */
    row1 = rectmove + (y - 2) * xsize;
    row2 = row1 + xsize;
    row3 = row2 + xsize;
    for (x = 2; x < xsize; x++, row1++, row2++, row3++) {
      if (row2[1]) {
        if (row2[0] == 0 || row2[2] == 0 || row1[1] == 0 || row3[1] == 0) {
          row2[1] = 128;
        }
      }
    }
  }

  /* 2: evaluate horizontal scan-lines and calculate alphas. */
  row1 = rectmove;
  for (y = 0; y < ysize; y++) {
    row1++;
    for (x = 1; x < xsize; x++, row1++) {
      if (row1[0] == 128 && row1[1] == 128) {
        /* find previous color and next color and amount of steps to blend */
        prev = row1[-1];
        step = 1;
        while (x + step < xsize && row1[step] == 128) {
          step++;
        }

        if (x + step != xsize) {
          /* now we can blend values */
          next = row1[step];

          /* NOTE: prev value can be next value, but we do this loop to clear 128 then. */
          for (a = 0; a < step; a++) {
            int fac, mfac;

            fac = ((a + 1) << 8) / (step + 1);
            mfac = 255 - fac;

            row1[a] = (prev * mfac + next * fac) >> 8;
          }
        }
      }
    }
  }

  /* 3: evaluate vertical scan-lines and calculate alphas */
  /*    use for reading a copy of the original tagged buffer */
  for (x = 0; x < xsize; x++) {
    row1 = rectmove + x + xsize;

    for (y = 1; y < ysize; y++, row1 += xsize) {
      if (row1[0] == 128 && row1[xsize] == 128) {
        /* find previous color and next color and amount of steps to blend */
        prev = row1[-xsize];
        step = 1;
        while (y + step < ysize && row1[step * xsize] == 128) {
          step++;
        }

        if (y + step != ysize) {
          /* now we can blend values */
          next = row1[step * xsize];
          /* NOTE: prev value can be next value, but we do this loop to clear 128 then. */
          for (a = 0; a < step; a++) {
            int fac, mfac;

            fac = ((a + 1) << 8) / (step + 1);
            mfac = 255 - fac;

            row1[a * xsize] = (prev * mfac + next * fac) >> 8;
          }
        }
      }
    }
  }

  /* last: pixels with 0 we fill in zbuffer, with 1 we skip for mask */
  for (y = 2; y < ysize; y++) {
    /* setup rows */
    row1 = rectmove + (y - 2) * xsize;
    row2 = row1 + xsize;
    row3 = row2 + xsize;
    for (x = 2; x < xsize; x++, row1++, row2++, row3++) {
      if (row2[1] == 0) {
        if (row2[0] > 1 || row2[2] > 1 || row1[1] > 1 || row3[1] > 1) {
          row2[1] = 1;
        }
      }
    }
  }
}

/* in: two vectors, first vector points from origin back in time, 2nd vector points to future */
/* we make this into 3 points, center point is (0, 0) */
/* and offset the center point just enough to make curve go through midpoint */

static void quad_bezier_2d(float *result, const float *v1, const float *v2, const float *ipodata)
{
  float p1[2], p2[2], p3[2];

  p3[0] = -v2[0];
  p3[1] = -v2[1];

  p1[0] = v1[0];
  p1[1] = v1[1];

  /* official formula 2*p2 - 0.5*p1 - 0.5*p3 */
  p2[0] = -0.5f * p1[0] - 0.5f * p3[0];
  p2[1] = -0.5f * p1[1] - 0.5f * p3[1];

  result[0] = ipodata[0] * p1[0] + ipodata[1] * p2[0] + ipodata[2] * p3[0];
  result[1] = ipodata[0] * p1[1] + ipodata[1] * p2[1] + ipodata[2] * p3[1];
}

static void set_quad_bezier_ipo(float fac, float *data)
{
  float mfac = (1.0f - fac);

  data[0] = mfac * mfac;
  data[1] = 2.0f * mfac * fac;
  data[2] = fac * fac;
}

void zbuf_accumulate_vecblur(NodeBlurData *nbd,
                             int xsize,
                             int ysize,
                             float *newrect,
                             const float *imgrect,
                             float *vecbufrect,
                             const float *zbufrect)
{
  ZSpan zspan;
  DrawBufPixel *rectdraw, *dr;
  static float jit[256][2];
  float v1[3], v2[3], v3[3], v4[3], fx, fy;
  const float *dimg, *dz, *ro;
  float *rectvz, *dvz, *dvec1, *dvec2, *dz1, *dz2, *rectz;
  float *minvecbufrect = nullptr, *rectweight, *rw, *rectmax, *rm;
  float maxspeedsq = (float)nbd->maxspeed * nbd->maxspeed;
  int y, x, step, maxspeed = nbd->maxspeed, samples = nbd->samples;
  int tsktsk = 0;
  static int firsttime = 1;
  char *rectmove, *dm;

  zbuf_alloc_span(&zspan, xsize, ysize, 1.0f);
  zspan.zmulx = ((float)xsize) / 2.0f;
  zspan.zmuly = ((float)ysize) / 2.0f;
  zspan.zofsx = 0.0f;
  zspan.zofsy = 0.0f;

  /* the buffers */
  rectz = (float *)MEM_callocN(sizeof(float) * xsize * ysize, "zbuf accum");
  zspan.rectz = (int *)rectz;

  rectmove = (char *)MEM_callocN(xsize * ysize, "rectmove");
  rectdraw = (DrawBufPixel *)MEM_callocN(sizeof(DrawBufPixel) * xsize * ysize, "rect draw");
  zspan.rectdraw = rectdraw;

  rectweight = (float *)MEM_callocN(sizeof(float) * xsize * ysize, "rect weight");
  rectmax = (float *)MEM_callocN(sizeof(float) * xsize * ysize, "rect max");

  /* debug... check if PASS_VECTOR_MAX still is in buffers */
  dvec1 = vecbufrect;
  for (x = 4 * xsize * ysize; x > 0; x--, dvec1++) {
    if (dvec1[0] == PASS_VECTOR_MAX) {
      dvec1[0] = 0.0f;
      tsktsk = 1;
    }
  }
  if (tsktsk) {
    printf("Found uninitialized speed in vector buffer... fixed.\n");
  }

  /* Min speed? then copy speed-buffer to recalculate speed vectors. */
  if (nbd->minspeed) {
    float minspeed = (float)nbd->minspeed;
    float minspeedsq = minspeed * minspeed;

    minvecbufrect = (float *)MEM_callocN(sizeof(float[4]) * xsize * ysize, "minspeed buf");

    dvec1 = vecbufrect;
    dvec2 = minvecbufrect;
    for (x = 2 * xsize * ysize; x > 0; x--, dvec1 += 2, dvec2 += 2) {
      if (dvec1[0] == 0.0f && dvec1[1] == 0.0f) {
        dvec2[0] = dvec1[0];
        dvec2[1] = dvec1[1];
      }
      else {
        float speedsq = dvec1[0] * dvec1[0] + dvec1[1] * dvec1[1];
        if (speedsq <= minspeedsq) {
          dvec2[0] = 0.0f;
          dvec2[1] = 0.0f;
        }
        else {
          speedsq = 1.0f - minspeed / sqrtf(speedsq);
          dvec2[0] = speedsq * dvec1[0];
          dvec2[1] = speedsq * dvec1[1];
        }
      }
    }
    SWAP(float *, minvecbufrect, vecbufrect);
  }

  /* Make vertex buffer with averaged speed and Z-values. */
  rectvz = (float *)MEM_callocN(sizeof(float[4]) * (xsize + 1) * (ysize + 1), "vertices");
  dvz = rectvz;
  for (y = 0; y <= ysize; y++) {

    if (y == 0) {
      dvec1 = vecbufrect + 4 * y * xsize;
    }
    else {
      dvec1 = vecbufrect + 4 * (y - 1) * xsize;
    }

    if (y == ysize) {
      dvec2 = vecbufrect + 4 * (y - 1) * xsize;
    }
    else {
      dvec2 = vecbufrect + 4 * y * xsize;
    }

    for (x = 0; x <= xsize; x++) {

      /* two vectors, so a step loop */
      for (step = 0; step < 2; step++, dvec1 += 2, dvec2 += 2, dvz += 2) {
        /* average on minimal speed */
        int div = 0;

        if (x != 0) {
          if (dvec1[-4] != 0.0f || dvec1[-3] != 0.0f) {
            dvz[0] = dvec1[-4];
            dvz[1] = dvec1[-3];
            div++;
          }
          if (dvec2[-4] != 0.0f || dvec2[-3] != 0.0f) {
            if (div == 0) {
              dvz[0] = dvec2[-4];
              dvz[1] = dvec2[-3];
              div++;
            }
            else if ((fabsf(dvec2[-4]) + fabsf(dvec2[-3])) < (fabsf(dvz[0]) + fabsf(dvz[1]))) {
              dvz[0] = dvec2[-4];
              dvz[1] = dvec2[-3];
            }
          }
        }

        if (x != xsize) {
          if (dvec1[0] != 0.0f || dvec1[1] != 0.0f) {
            if (div == 0) {
              dvz[0] = dvec1[0];
              dvz[1] = dvec1[1];
              div++;
            }
            else if ((fabsf(dvec1[0]) + fabsf(dvec1[1])) < (fabsf(dvz[0]) + fabsf(dvz[1]))) {
              dvz[0] = dvec1[0];
              dvz[1] = dvec1[1];
            }
          }
          if (dvec2[0] != 0.0f || dvec2[1] != 0.0f) {
            if (div == 0) {
              dvz[0] = dvec2[0];
              dvz[1] = dvec2[1];
            }
            else if ((fabsf(dvec2[0]) + fabsf(dvec2[1])) < (fabsf(dvz[0]) + fabsf(dvz[1]))) {
              dvz[0] = dvec2[0];
              dvz[1] = dvec2[1];
            }
          }
        }
        if (maxspeed) {
          float speedsq = dvz[0] * dvz[0] + dvz[1] * dvz[1];
          if (speedsq > maxspeedsq) {
            speedsq = (float)maxspeed / sqrtf(speedsq);
            dvz[0] *= speedsq;
            dvz[1] *= speedsq;
          }
        }
      }
    }
  }

  /* set border speeds to keep border speeds on border */
  dz1 = rectvz;
  dz2 = rectvz + 4 * (ysize) * (xsize + 1);
  for (x = 0; x <= xsize; x++, dz1 += 4, dz2 += 4) {
    dz1[1] = 0.0f;
    dz2[1] = 0.0f;
    dz1[3] = 0.0f;
    dz2[3] = 0.0f;
  }
  dz1 = rectvz;
  dz2 = rectvz + 4 * (xsize);
  for (y = 0; y <= ysize; y++, dz1 += 4 * (xsize + 1), dz2 += 4 * (xsize + 1)) {
    dz1[0] = 0.0f;
    dz2[0] = 0.0f;
    dz1[2] = 0.0f;
    dz2[2] = 0.0f;
  }

  /* tag moving pixels, only these faces we draw */
  dm = rectmove;
  dvec1 = vecbufrect;
  for (x = xsize * ysize; x > 0; x--, dm++, dvec1 += 4) {
    if ((dvec1[0] != 0.0f || dvec1[1] != 0.0f || dvec1[2] != 0.0f || dvec1[3] != 0.0f)) {
      *dm = 255;
    }
  }

  antialias_tagbuf(xsize, ysize, rectmove);

  /* Has to become static, the jitter initialization calls a random-seed,
   * screwing up texture noise node. */
  if (firsttime) {
    firsttime = 0;
    BLI_jitter_init(jit, 256);
  }

  memset(newrect, 0, sizeof(float) * xsize * ysize * 4);

  /* accumulate */
  samples /= 2;
  for (step = 1; step <= samples; step++) {
    float speedfac = 0.5f * nbd->fac * (float)step / (float)(samples + 1);
    int side;

    for (side = 0; side < 2; side++) {
      float blendfac, ipodata[4];

      /* clear zbuf, if we draw future we fill in not moving pixels */
      if (false) {
        for (x = xsize * ysize - 1; x >= 0; x--) {
          rectz[x] = 10e16;
        }
      }
      else {
        for (x = xsize * ysize - 1; x >= 0; x--) {
          if (rectmove[x] == 0) {
            rectz[x] = zbufrect[x];
          }
          else {
            rectz[x] = 10e16;
          }
        }
      }

      /* clear drawing buffer */
      for (x = xsize * ysize - 1; x >= 0; x--) {
        rectdraw[x].colpoin = nullptr;
      }

      dimg = imgrect;
      dm = rectmove;
      dz = zbufrect;
      dz1 = rectvz;
      dz2 = rectvz + 4 * (xsize + 1);

      if (side) {
        if (nbd->curved == 0) {
          dz1 += 2;
          dz2 += 2;
        }
        speedfac = -speedfac;
      }

      set_quad_bezier_ipo(0.5f + 0.5f * speedfac, ipodata);

      for (fy = -0.5f + jit[step & 255][0], y = 0; y < ysize; y++, fy += 1.0f) {
        for (fx = -0.5f + jit[step & 255][1], x = 0; x < xsize;
             x++, fx += 1.0f, dimg += 4, dz1 += 4, dz2 += 4, dm++, dz++) {
          if (*dm > 1) {
            float jfx = fx + 0.5f;
            float jfy = fy + 0.5f;
            DrawBufPixel col;

            /* make vertices */
            if (nbd->curved) { /* curved */
              quad_bezier_2d(v1, dz1, dz1 + 2, ipodata);
              v1[0] += jfx;
              v1[1] += jfy;
              v1[2] = *dz;

              quad_bezier_2d(v2, dz1 + 4, dz1 + 4 + 2, ipodata);
              v2[0] += jfx + 1.0f;
              v2[1] += jfy;
              v2[2] = *dz;

              quad_bezier_2d(v3, dz2 + 4, dz2 + 4 + 2, ipodata);
              v3[0] += jfx + 1.0f;
              v3[1] += jfy + 1.0f;
              v3[2] = *dz;

              quad_bezier_2d(v4, dz2, dz2 + 2, ipodata);
              v4[0] += jfx;
              v4[1] += jfy + 1.0f;
              v4[2] = *dz;
            }
            else {
              ARRAY_SET_ITEMS(v1, speedfac * dz1[0] + jfx, speedfac * dz1[1] + jfy, *dz);
              ARRAY_SET_ITEMS(v2, speedfac * dz1[4] + jfx + 1.0f, speedfac * dz1[5] + jfy, *dz);
              ARRAY_SET_ITEMS(
                  v3, speedfac * dz2[4] + jfx + 1.0f, speedfac * dz2[5] + jfy + 1.0f, *dz);
              ARRAY_SET_ITEMS(v4, speedfac * dz2[0] + jfx, speedfac * dz2[1] + jfy + 1.0f, *dz);
            }
            if (*dm == 255) {
              col.alpha = 1.0f;
            }
            else if (*dm < 2) {
              col.alpha = 0.0f;
            }
            else {
              col.alpha = ((float)*dm) / 255.0f;
            }
            col.colpoin = dimg;

            zbuf_fill_in_rgba(&zspan, &col, v1, v2, v3, v4);
          }
        }
        dz1 += 4;
        dz2 += 4;
      }

      /* blend with a falloff. this fixes the ugly effect you get with
       * a fast moving object. then it looks like a solid object overlaid
       * over a very transparent moving version of itself. in reality, the
       * whole object should become transparent if it is moving fast, be
       * we don't know what is behind it so we don't do that. this hack
       * overestimates the contribution of foreground pixels but looks a
       * bit better without a sudden cutoff. */
      blendfac = ((samples - step) / (float)samples);
      /* Smooth-step to make it look a bit nicer as well. */
      blendfac = 3.0f * pow(blendfac, 2.0f) - 2.0f * pow(blendfac, 3.0f);

      /* accum */
      rw = rectweight;
      rm = rectmax;
      for (dr = rectdraw, dz2 = newrect, x = xsize * ysize - 1; x >= 0;
           x--, dr++, dz2 += 4, rw++, rm++) {
        if (dr->colpoin) {
          float bfac = dr->alpha * blendfac;

          dz2[0] += bfac * dr->colpoin[0];
          dz2[1] += bfac * dr->colpoin[1];
          dz2[2] += bfac * dr->colpoin[2];
          dz2[3] += bfac * dr->colpoin[3];

          *rw += bfac;
          *rm = MAX2(*rm, bfac);
        }
      }
    }
  }

  /* blend between original images and accumulated image */
  rw = rectweight;
  rm = rectmax;
  ro = imgrect;
  dm = rectmove;
  for (dz2 = newrect, x = xsize * ysize - 1; x >= 0; x--, dz2 += 4, ro += 4, rw++, rm++, dm++) {
    float mfac = *rm;
    float fac = (*rw == 0.0f) ? 0.0f : mfac / (*rw);
    float nfac = 1.0f - mfac;

    dz2[0] = fac * dz2[0] + nfac * ro[0];
    dz2[1] = fac * dz2[1] + nfac * ro[1];
    dz2[2] = fac * dz2[2] + nfac * ro[2];
    dz2[3] = fac * dz2[3] + nfac * ro[3];
  }

  MEM_freeN(rectz);
  MEM_freeN(rectmove);
  MEM_freeN(rectdraw);
  MEM_freeN(rectvz);
  MEM_freeN(rectweight);
  MEM_freeN(rectmax);
  if (minvecbufrect) {
    MEM_freeN(vecbufrect); /* rects were swapped! */
  }
  zbuf_free_span(&zspan);
}

}  // namespace blender::compositor