/* SPDX-License-Identifier: GPL-2.0-or-later
 * Copyright 2001-2002 NaN Holding BV. All rights reserved. */

/** \file
 * \ingroup bke
 */

#include "MEM_guardedalloc.h"

#include "atomic_ops.h"

#include "BLI_bitmap.h"
#include "BLI_boxpack_2d.h"
#include "BLI_linklist.h"
#include "BLI_listbase.h"
#include "BLI_map.hh"
#include "BLI_path_util.h"
#include "BLI_threads.h"
#include "BLI_utility_mixins.hh"

#include "DNA_image_types.h"
#include "DNA_space_types.h"
#include "DNA_userdef_types.h"

#include "IMB_colormanagement.h"
#include "IMB_imbuf.h"
#include "IMB_imbuf_types.h"

#include "BKE_global.h"
#include "BKE_image.h"
#include "BKE_image_partial_update.hh"
#include "BKE_main.h"

#include "GPU_capabilities.h"
#include "GPU_state.h"
#include "GPU_texture.h"

#include "PIL_time.h"

using namespace blender::bke::image::partial_update;

namespace blender::bke::image::gpu {

/* Used to get the correct gpu texture from an Image datablock. */
typedef enum eGPUTextureTarget {
  TEXTARGET_2D = 0,
  TEXTARGET_2D_ARRAY,
  TEXTARGET_TILE_MAPPING,
  TEXTARGET_COUNT,
} eGPUTextureTarget;

struct ImageGPUTextureStore {
  class Entry : NonCopyable {
   public:
    GPUTexture *gputextures[TEXTARGET_COUNT][2];
    struct {
      bool in_use_gc : 1;
      bool in_use_bmain : 1;
    } flags;

    Entry()
    {
      for (int target = 0; target < TEXTARGET_COUNT; target++) {
        for (int eye = 0; eye < 2; eye++) {
          gputextures[target][eye] = nullptr;
        }
      }
      flags.in_use_gc = false;
      flags.in_use_bmain = false;
    }

    Entry(Entry &&other)
    {
      for (int target = 0; target < TEXTARGET_COUNT; target++) {
        for (int eye = 0; eye < 2; eye++) {
          gputextures[target][eye] = other.gputextures[target][eye];
          other.gputextures[target][eye] = nullptr;
        }
      }
      flags = other.flags;
    };

    virtual ~Entry()
    {
      clear();
    }

    void set_mipmap(const bool mipmap)
    {
      for (int eye = 0; eye < 2; eye++) {
        if (gputextures[TEXTARGET_2D][eye]) {
          GPU_texture_mipmap_mode(gputextures[TEXTARGET_2D][eye], mipmap, true);
        }
        if (gputextures[TEXTARGET_2D_ARRAY][eye]) {
          GPU_texture_mipmap_mode(gputextures[TEXTARGET_2D_ARRAY][eye], mipmap, true);
        }
      }
    }

    void tag_used()
    {
      flags.in_use_gc = true;
    }

    void clear()
    {
      for (int target = 0; target < TEXTARGET_COUNT; target++) {
        for (int eye = 0; eye < 2; eye++) {
          GPU_TEXTURE_FREE_SAFE(gputextures[target][eye]);
        }
      }
    }
  };

  using Entries = Map<std::string, Entry>;

  Entries entries;

  void reset_usage()
  {
    for (Entry &entry : entries.values()) {
      entry.flags.in_use_gc = false;
    }
  }

  void remove_unused()
  {
    for (auto it : entries.items()) {
      if (it.value.flags.in_use_gc) {
        continue;
      }
      entries.remove(it.key);
    }
  }

  void remove_unused(const Main *bmain)
  {
    for (Entry &entry : entries.values()) {
      entry.flags.in_use_bmain = false;
    }

    LISTBASE_FOREACH (Image *, image, &bmain->images) {
      std::string key = create_key(*image);
      Entry *entry = entries.lookup_ptr(key);
      if (entry == nullptr) {
        continue;
      }
      entry->flags.in_use_bmain = true;
    }

    for (auto it : entries.items()) {
      if (it.value.flags.in_use_bmain) {
        continue;
      }
      entries.remove(it.key);
    }
  }

  void set_mipmap(const bool mipmap)
  {
    for (Entry &entry : entries.values()) {
      entry.set_mipmap(mipmap);
    }
  }

  void clear()
  {
    entries.clear();
  }

  /**
   * Create a lookup key for the given image.
   *
   * Equal keys will map to the same set of GPUTextures.
   */
  std::string create_key(Image &image) const
  {
    bool add_ptr = false;
    bool add_filepath = false;

    switch (image.source) {
      case IMA_SRC_FILE:
        if (BKE_image_is_dirty(&image) || BKE_image_has_packedfile(&image)) {
          add_ptr = true;
        }
        else {
          add_filepath = true;
        }
        break;
      case IMA_SRC_SEQUENCE:
        add_ptr = true;
        break;
      case IMA_SRC_MOVIE:
        add_ptr = true;
        break;
      case IMA_SRC_GENERATED:
        add_ptr = true;
        break;
      case IMA_SRC_VIEWER:
        add_ptr = true;
        break;
      case IMA_SRC_TILED:
        add_ptr = true;
        break;
    }

    std::stringstream result;
    if (add_ptr) {
      result << "PTR:" << &image << ",";
    }
    if (add_filepath) {
      char filepath[FILE_MAX];
      BLI_strncpy(filepath, image.filepath, sizeof(filepath));
      BLI_path_abs(filepath, ID_BLEND_PATH_FROM_GLOBAL(&image.id));
      result << "FILE:" << filepath << ",";
    }

    return result.str();
  }

  Entry &operator[](Image &image)
  {
    std::string key = create_key(image);
    Entry &entry = entries.lookup_or_add_default(key);
    entry.tag_used();
    return entry;
  }
};

static ImageGPUTextureStore g_texture_store;

}  // namespace blender::bke::image::gpu

extern "C" {

using namespace blender::bke::image::gpu;

/* Prototypes. */
static void image_update_gputexture_ex(Image *ima,
                                       ImageGPUTextureStore::Entry &entry,
                                       ImageTile *tile,
                                       ImBuf *ibuf,
                                       int x,
                                       int y,
                                       int w,
                                       int h);

bool BKE_image_has_gpu_texture_premultiplied_alpha(Image *image, ImBuf *ibuf)
{
  if (image) {
    /* Render result and compositor output are always premultiplied */
    if (ELEM(image->type, IMA_TYPE_R_RESULT, IMA_TYPE_COMPOSITE)) {
      return true;
    }
    /* Generated images use pre multiplied float buffer, but straight alpha for byte buffers. */
    if (image->type == IMA_TYPE_UV_TEST && ibuf) {
      return ibuf->rect_float != nullptr;
    }
  }
  if (ibuf) {
    if (ibuf->rect_float) {
      return image ? (image->alpha_mode != IMA_ALPHA_STRAIGHT) : false;
    }

    return image ? (image->alpha_mode == IMA_ALPHA_PREMUL) : true;
  }
  return false;
}

/* -------------------------------------------------------------------- */
/** \name UDIM GPU Texture
 * \{ */

static bool is_over_resolution_limit(int w, int h)
{
  return (w > GPU_texture_size_with_limit(w) || h > GPU_texture_size_with_limit(h));
}

static int smaller_power_of_2_limit(int num)
{
  return power_of_2_min_i(GPU_texture_size_with_limit(num));
}

static GPUTexture *gpu_texture_create_tile_mapping(Image *ima,
                                                   ImageGPUTextureStore::Entry &entry,
                                                   const int multiview_eye)
{
  GPUTexture *tilearray = entry.gputextures[TEXTARGET_2D_ARRAY][multiview_eye];

  if (tilearray == nullptr) {
    return nullptr;
  }

  float array_w = GPU_texture_width(tilearray);
  float array_h = GPU_texture_height(tilearray);

  /* Determine maximum tile number. */
  BKE_image_sort_tiles(ima);
  ImageTile *last_tile = (ImageTile *)ima->tiles.last;
  int max_tile = last_tile->tile_number - 1001;

  /* create image */
  int width = max_tile + 1;
  float *data = (float *)MEM_callocN(width * 8 * sizeof(float), __func__);
  for (int i = 0; i < width; i++) {
    data[4 * i] = -1.0f;
  }
  LISTBASE_FOREACH (ImageTile *, tile, &ima->tiles) {
    int i = tile->tile_number - 1001;
    ImageTile_Runtime *tile_runtime = &tile->runtime;
    data[4 * i] = tile_runtime->tilearray_layer;

    float *tile_info = &data[4 * width + 4 * i];
    tile_info[0] = tile_runtime->tilearray_offset[0] / array_w;
    tile_info[1] = tile_runtime->tilearray_offset[1] / array_h;
    tile_info[2] = tile_runtime->tilearray_size[0] / array_w;
    tile_info[3] = tile_runtime->tilearray_size[1] / array_h;
  }

  GPUTexture *tex = GPU_texture_create_1d_array(ima->id.name + 2, width, 2, 1, GPU_RGBA32F, data);
  GPU_texture_mipmap_mode(tex, false, false);

  MEM_freeN(data);

  return tex;
}

struct PackTile {
  FixedSizeBoxPack boxpack;
  ImageTile *tile;
  float pack_score;
};

static int compare_packtile(const void *a, const void *b)
{
  const PackTile *tile_a = (const PackTile *)a;
  const PackTile *tile_b = (const PackTile *)b;

  return tile_a->pack_score < tile_b->pack_score;
}

static GPUTexture *gpu_texture_create_tile_array(Image *ima, ImBuf *main_ibuf)
{
  int arraywidth = 0, arrayheight = 0;
  ListBase boxes = {nullptr};

  LISTBASE_FOREACH (ImageTile *, tile, &ima->tiles) {
    ImageUser iuser;
    BKE_imageuser_default(&iuser);
    iuser.tile = tile->tile_number;
    ImBuf *ibuf = BKE_image_acquire_ibuf(ima, &iuser, nullptr);

    if (ibuf) {
      PackTile *packtile = MEM_cnew<PackTile>(__func__);
      packtile->tile = tile;
      packtile->boxpack.w = ibuf->x;
      packtile->boxpack.h = ibuf->y;

      if (is_over_resolution_limit(packtile->boxpack.w, packtile->boxpack.h)) {
        packtile->boxpack.w = smaller_power_of_2_limit(packtile->boxpack.w);
        packtile->boxpack.h = smaller_power_of_2_limit(packtile->boxpack.h);
      }
      arraywidth = max_ii(arraywidth, packtile->boxpack.w);
      arrayheight = max_ii(arrayheight, packtile->boxpack.h);

      /* We sort the tiles by decreasing size, with an additional penalty term
       * for high aspect ratios. This improves packing efficiency. */
      float w = packtile->boxpack.w, h = packtile->boxpack.h;
      packtile->pack_score = max_ff(w, h) / min_ff(w, h) * w * h;

      BKE_image_release_ibuf(ima, ibuf, nullptr);
      BLI_addtail(&boxes, packtile);
    }
  }

  BLI_assert(arraywidth > 0 && arrayheight > 0);

  BLI_listbase_sort(&boxes, compare_packtile);
  int arraylayers = 0;
  /* Keep adding layers until all tiles are packed. */
  while (boxes.first != nullptr) {
    ListBase packed = {nullptr};
    BLI_box_pack_2d_fixedarea(&boxes, arraywidth, arrayheight, &packed);
    BLI_assert(packed.first != nullptr);

    LISTBASE_FOREACH (PackTile *, packtile, &packed) {
      ImageTile *tile = packtile->tile;
      ImageTile_Runtime *tile_runtime = &tile->runtime;
      int *tileoffset = tile_runtime->tilearray_offset;
      int *tilesize = tile_runtime->tilearray_size;

      tileoffset[0] = packtile->boxpack.x;
      tileoffset[1] = packtile->boxpack.y;
      tilesize[0] = packtile->boxpack.w;
      tilesize[1] = packtile->boxpack.h;
      tile_runtime->tilearray_layer = arraylayers;
    }

    BLI_freelistN(&packed);
    arraylayers++;
  }

  const bool use_high_bitdepth = (ima->flag & IMA_HIGH_BITDEPTH);
  /* Create Texture without content. */
  GPUTexture *tex = IMB_touch_gpu_texture(
      ima->id.name + 2, main_ibuf, arraywidth, arrayheight, arraylayers, use_high_bitdepth);

  /* Upload each tile one by one. */
  LISTBASE_FOREACH (ImageTile *, tile, &ima->tiles) {
    ImageTile_Runtime *tile_runtime = &tile->runtime;
    int tilelayer = tile_runtime->tilearray_layer;
    int *tileoffset = tile_runtime->tilearray_offset;
    int *tilesize = tile_runtime->tilearray_size;

    if (tilesize[0] == 0 || tilesize[1] == 0) {
      continue;
    }

    ImageUser iuser;
    BKE_imageuser_default(&iuser);
    iuser.tile = tile->tile_number;
    ImBuf *ibuf = BKE_image_acquire_ibuf(ima, &iuser, nullptr);

    if (ibuf) {
      const bool store_premultiplied = BKE_image_has_gpu_texture_premultiplied_alpha(ima, ibuf);
      IMB_update_gpu_texture_sub(tex,
                                 ibuf,
                                 UNPACK2(tileoffset),
                                 tilelayer,
                                 UNPACK2(tilesize),
                                 use_high_bitdepth,
                                 store_premultiplied);
    }

    BKE_image_release_ibuf(ima, ibuf, nullptr);
  }

  if (GPU_mipmap_enabled()) {
    GPU_texture_generate_mipmap(tex);
    GPU_texture_mipmap_mode(tex, true, true);
    if (ima) {
      ima->gpuflag |= IMA_GPU_MIPMAP_COMPLETE;
    }
  }
  else {
    GPU_texture_mipmap_mode(tex, false, true);
  }

  return tex;
}

/** \} */

/* -------------------------------------------------------------------- */
/** \name Regular gpu texture
 * \{ */

static GPUTexture **get_image_gpu_texture_ptr(ImageGPUTextureStore::Entry &entry,
                                              eGPUTextureTarget textarget,
                                              const int multiview_eye)
{
  const bool in_range = (textarget >= 0) && (textarget < TEXTARGET_COUNT);
  BLI_assert(in_range);
  BLI_assert(ELEM(multiview_eye, 0, 1));

  if (in_range) {
    return &entry.gputextures[textarget][multiview_eye];
  }
  return nullptr;
}

static GPUTexture *image_gpu_texture_error_create(eGPUTextureTarget textarget)
{
  fprintf(stderr, "GPUTexture: Blender Texture Not Loaded!\n");
  switch (textarget) {
    case TEXTARGET_2D_ARRAY:
      return GPU_texture_create_error(2, true);
    case TEXTARGET_TILE_MAPPING:
      return GPU_texture_create_error(1, true);
    case TEXTARGET_2D:
    default:
      return GPU_texture_create_error(2, false);
  }
}

static void image_gpu_texture_partial_update_changes_available(
    Image *image,
    ImageGPUTextureStore::Entry &entry,
    PartialUpdateChecker<ImageTileData>::CollectResult &changes)
{
  while (changes.get_next_change() == ePartialUpdateIterResult::ChangeAvailable) {
    /* Calculate the clipping region with the tile buffer.
     * TODO(jbakker): should become part of ImageTileData to deduplicate with image engine. */
    rcti buffer_rect;
    BLI_rcti_init(
        &buffer_rect, 0, changes.tile_data.tile_buffer->x, 0, changes.tile_data.tile_buffer->y);
    rcti clipped_update_region;
    const bool has_overlap = BLI_rcti_isect(
        &buffer_rect, &changes.changed_region.region, &clipped_update_region);
    if (!has_overlap) {
      continue;
    }

    image_update_gputexture_ex(image,
                               entry,
                               changes.tile_data.tile,
                               changes.tile_data.tile_buffer,
                               clipped_update_region.xmin,
                               clipped_update_region.ymin,
                               BLI_rcti_size_x(&clipped_update_region),
                               BLI_rcti_size_y(&clipped_update_region));
  }
}

static void image_gpu_texture_try_partial_update(Image *image, ImageUser *iuser)
{
  PartialUpdateChecker<ImageTileData> checker(image, iuser, image->runtime.partial_update_user);
  PartialUpdateChecker<ImageTileData>::CollectResult changes = checker.collect_changes();
  ImageGPUTextureStore::Entry &entry = g_texture_store[*image];
  switch (changes.get_result_code()) {
    case ePartialUpdateCollectResult::FullUpdateNeeded: {
      entry.clear();
      break;
    }

    case ePartialUpdateCollectResult::PartialChangesDetected: {
      image_gpu_texture_partial_update_changes_available(image, entry, changes);
      break;
    }

    case ePartialUpdateCollectResult::NoChangesDetected: {
      /* GPUTextures are up to date. */
      break;
    }
  }
}

static GPUTexture *image_get_gpu_texture(Image *ima,
                                         ImageUser *iuser,
                                         ImBuf *ibuf,
                                         eGPUTextureTarget textarget)
{
  if (ima == nullptr) {
    return nullptr;
  }

  /* Free GPU textures when requesting a different render pass/layer.
   * When `iuser` isn't set (texture painting single image mode) we assume that
   * the current `pass` and `layer` should be 0. */
  short requested_pass = iuser ? iuser->pass : 0;
  short requested_layer = iuser ? iuser->layer : 0;
  short requested_view = iuser ? iuser->multi_index : 0;
  /* There is room for 2 multiview textures. When a higher number is requested we should always
   * target the first view slot. This is fine as multi view images aren't used together. */
  if (requested_view < 2) {
    requested_view = 0;
  }
  if (ima->gpu_pass != requested_pass || ima->gpu_layer != requested_layer ||
      ima->gpu_view != requested_view) {
    ima->gpu_pass = requested_pass;
    ima->gpu_layer = requested_layer;
    ima->gpu_view = requested_view;
  }
#undef GPU_FLAGS_TO_CHECK

  if (ima->runtime.partial_update_user == nullptr) {
    ima->runtime.partial_update_user = BKE_image_partial_update_create(ima);
  }

  image_gpu_texture_try_partial_update(ima, iuser);

  /* Tag as in active use for garbage collector. */
  BKE_image_tag_time(ima);

  /* Test if we already have a texture. */
  int current_view = iuser ? iuser->multi_index : 0;
  if (current_view >= 2) {
    current_view = 0;
  }
  ImageGPUTextureStore::Entry &entry = g_texture_store[*ima];
  GPUTexture **tex = get_image_gpu_texture_ptr(entry, textarget, current_view);
  if (*tex) {
    return *tex;
  }

  /* Check if we have a valid image. If not, we return a dummy
   * texture with zero bind-code so we don't keep trying. */
  ImageTile *tile = BKE_image_get_tile(ima, 0);
  if (tile == nullptr) {
    *tex = image_gpu_texture_error_create(textarget);
    return *tex;
  }

  /* check if we have a valid image buffer */
  ImBuf *ibuf_intern = ibuf;
  if (ibuf_intern == nullptr) {
    ibuf_intern = BKE_image_acquire_ibuf(ima, iuser, nullptr);
    if (ibuf_intern == nullptr) {
      *tex = image_gpu_texture_error_create(textarget);
      return *tex;
    }
  }

  if (textarget == TEXTARGET_2D_ARRAY) {
    *tex = gpu_texture_create_tile_array(ima, ibuf_intern);
  }
  else if (textarget == TEXTARGET_TILE_MAPPING) {
    *tex = gpu_texture_create_tile_mapping(ima, entry, iuser ? iuser->multiview_eye : 0);
  }
  else {
    const bool use_high_bitdepth = (ima->flag & IMA_HIGH_BITDEPTH);
    const bool store_premultiplied = BKE_image_has_gpu_texture_premultiplied_alpha(ima,
                                                                                   ibuf_intern);

    *tex = IMB_create_gpu_texture(
        ima->id.name + 2, ibuf_intern, use_high_bitdepth, store_premultiplied);

    if (*tex) {
      GPU_texture_wrap_mode(*tex, true, false);

      if (GPU_mipmap_enabled()) {
        GPU_texture_generate_mipmap(*tex);
        if (ima) {
          ima->gpuflag |= IMA_GPU_MIPMAP_COMPLETE;
        }
        GPU_texture_mipmap_mode(*tex, true, true);
      }
      else {
        GPU_texture_mipmap_mode(*tex, false, true);
      }
    }
  }

  if (*tex) {
    GPU_texture_orig_size_set(*tex, ibuf_intern->x, ibuf_intern->y);
  }

  if (ibuf != ibuf_intern) {
    BKE_image_release_ibuf(ima, ibuf_intern, nullptr);
  }

  return *tex;
}

GPUTexture *BKE_image_get_gpu_texture(Image *image, ImageUser *iuser, ImBuf *ibuf)
{
  return image_get_gpu_texture(image, iuser, ibuf, TEXTARGET_2D);
}

GPUTexture *BKE_image_get_gpu_tiles(Image *image, ImageUser *iuser, ImBuf *ibuf)
{
  return image_get_gpu_texture(image, iuser, ibuf, TEXTARGET_2D_ARRAY);
}

GPUTexture *BKE_image_get_gpu_tilemap(Image *image, ImageUser *iuser, ImBuf *ibuf)
{
  return image_get_gpu_texture(image, iuser, ibuf, TEXTARGET_TILE_MAPPING);
}

/** \} */

/* -------------------------------------------------------------------- */
/** \name Delayed GPU texture free
 *
 * Image datablocks can be deleted by any thread, but there may not be any active OpenGL context.
 * In that case we push them into a queue and free the buffers later.
 * \{ */

static int32_t free_unused_gpu_textures = false;

static void gpu_free_unused_buffers(const Main *bmain)
{
  BLI_assert(BLI_thread_is_main());

  if (G.is_rendering) {
    return;
  }

  int32_t do_free = atomic_cas_int32(&free_unused_gpu_textures, true, false);
  if (!do_free) {
    return;
  }

  g_texture_store.remove_unused(bmain);
}

void BKE_image_free_unused_gpu_textures(const Main *bmain)
{
  if (BLI_thread_is_main()) {
    gpu_free_unused_buffers(bmain);
  }
}

/** \} */

/* -------------------------------------------------------------------- */
/** \name Deletion
 * \{ */

void BKE_image_free_gputextures(Image *UNUSED(ima))
{
  atomic_store_int32(&free_unused_gpu_textures, true);
}

void BKE_image_free_all_gputextures(Main *UNUSED(bmain))
{
  g_texture_store.clear();
}

void BKE_image_free_anim_gputextures(Main *bmain)
{
  if (bmain) {
    LISTBASE_FOREACH (Image *, ima, &bmain->images) {
      if (BKE_image_is_animated(ima)) {
        BKE_image_free_gputextures(ima);
      }
    }
  }
}

void BKE_image_free_old_gputextures()
{

  static int lasttime = 0;
  int ctime = (int)PIL_check_seconds_timer();

  /*
   * Run garbage collector once for every collecting period of time
   * if textimeout is 0, that's the option to NOT run the collector
   */
  if (U.textimeout == 0 || ctime % U.texcollectrate || ctime == lasttime) {
    return;
  }

  /* of course not! */
  if (G.is_rendering) {
    return;
  }

  lasttime = ctime;

  g_texture_store.remove_unused();
  g_texture_store.reset_usage();
}

/** \} */

/* -------------------------------------------------------------------- */
/** \name Paint Update
 * \{ */

static ImBuf *update_do_scale(uchar *rect,
                              float *rect_float,
                              int *x,
                              int *y,
                              int *w,
                              int *h,
                              int limit_w,
                              int limit_h,
                              int full_w,
                              int full_h)
{
  /* Partial update with scaling. */
  float xratio = limit_w / (float)full_w;
  float yratio = limit_h / (float)full_h;

  int part_w = *w, part_h = *h;

  /* Find sub coordinates in scaled image. Take ceiling because we will be
   * losing 1 pixel due to rounding errors in x,y. */
  *x *= xratio;
  *y *= yratio;
  *w = (int)ceil(xratio * (*w));
  *h = (int)ceil(yratio * (*h));

  /* ...but take back if we are over the limit! */
  if (*x + *w > limit_w) {
    (*w)--;
  }
  if (*y + *h > limit_h) {
    (*h)--;
  }

  /* Scale pixels. */
  ImBuf *ibuf = IMB_allocFromBuffer((uint *)rect, rect_float, part_w, part_h, 4);
  IMB_scaleImBuf(ibuf, *w, *h);

  return ibuf;
}

static void gpu_texture_update_scaled(GPUTexture *tex,
                                      uchar *rect,
                                      float *rect_float,
                                      int full_w,
                                      int full_h,
                                      int x,
                                      int y,
                                      int layer,
                                      const int *tile_offset,
                                      const int *tile_size,
                                      int w,
                                      int h)
{
  ImBuf *ibuf;
  if (layer > -1) {
    ibuf = update_do_scale(
        rect, rect_float, &x, &y, &w, &h, tile_size[0], tile_size[1], full_w, full_h);

    /* Shift to account for tile packing. */
    x += tile_offset[0];
    y += tile_offset[1];
  }
  else {
    /* Partial update with scaling. */
    int limit_w = GPU_texture_width(tex);
    int limit_h = GPU_texture_height(tex);

    ibuf = update_do_scale(rect, rect_float, &x, &y, &w, &h, limit_w, limit_h, full_w, full_h);
  }

  void *data = (ibuf->rect_float) ? (void *)(ibuf->rect_float) : (void *)(ibuf->rect);
  eGPUDataFormat data_format = (ibuf->rect_float) ? GPU_DATA_FLOAT : GPU_DATA_UBYTE;

  GPU_texture_update_sub(tex, data_format, data, x, y, layer, w, h, 1);

  IMB_freeImBuf(ibuf);
}

static void gpu_texture_update_unscaled(GPUTexture *tex,
                                        uchar *rect,
                                        float *rect_float,
                                        int x,
                                        int y,
                                        int layer,
                                        const int tile_offset[2],
                                        int w,
                                        int h,
                                        int tex_stride,
                                        int tex_offset)
{
  if (layer > -1) {
    /* Shift to account for tile packing. */
    x += tile_offset[0];
    y += tile_offset[1];
  }

  void *data = (rect_float) ? (void *)(rect_float + tex_offset) : (void *)(rect + tex_offset);
  eGPUDataFormat data_format = (rect_float) ? GPU_DATA_FLOAT : GPU_DATA_UBYTE;

  /* Partial update without scaling. Stride and offset are used to copy only a
   * subset of a possible larger buffer than what we are updating. */
  GPU_unpack_row_length_set(tex_stride);

  GPU_texture_update_sub(tex, data_format, data, x, y, layer, w, h, 1);
  /* Restore default. */
  GPU_unpack_row_length_set(0);
}

static void gpu_texture_update_from_ibuf(
    GPUTexture *tex, Image *ima, ImBuf *ibuf, ImageTile *tile, int x, int y, int w, int h)
{
  bool scaled;
  if (tile != nullptr) {
    ImageTile_Runtime *tile_runtime = &tile->runtime;
    int *tilesize = tile_runtime->tilearray_size;
    scaled = (ibuf->x != tilesize[0]) || (ibuf->y != tilesize[1]);
  }
  else {
    scaled = (GPU_texture_width(tex) != ibuf->x) || (GPU_texture_height(tex) != ibuf->y);
  }

  if (scaled) {
    /* Extra padding to account for bleed from neighboring pixels. */
    const int padding = 4;
    const int xmax = min_ii(x + w + padding, ibuf->x);
    const int ymax = min_ii(y + h + padding, ibuf->y);
    x = max_ii(x - padding, 0);
    y = max_ii(y - padding, 0);
    w = xmax - x;
    h = ymax - y;
  }

  /* Get texture data pointers. */
  float *rect_float = ibuf->rect_float;
  uchar *rect = (uchar *)ibuf->rect;
  int tex_stride = ibuf->x;
  int tex_offset = ibuf->channels * (y * ibuf->x + x);

  const bool store_premultiplied = BKE_image_has_gpu_texture_premultiplied_alpha(ima, ibuf);
  if (rect_float == nullptr) {
    /* Byte pixels. */
    if (!IMB_colormanagement_space_is_data(ibuf->rect_colorspace)) {
      const bool compress_as_srgb = !IMB_colormanagement_space_is_scene_linear(
          ibuf->rect_colorspace);

      rect = (uchar *)MEM_mallocN(sizeof(uchar[4]) * w * h, __func__);
      if (rect == nullptr) {
        return;
      }

      tex_stride = w;
      tex_offset = 0;

      /* Convert to scene linear with sRGB compression, and premultiplied for
       * correct texture interpolation. */
      IMB_colormanagement_imbuf_to_byte_texture(
          rect, x, y, w, h, ibuf, compress_as_srgb, store_premultiplied);
    }
  }
  else {
    /* Float pixels. */
    if (ibuf->channels != 4 || scaled || !store_premultiplied) {
      rect_float = (float *)MEM_mallocN(sizeof(float[4]) * w * h, __func__);
      if (rect_float == nullptr) {
        return;
      }

      tex_stride = w;
      tex_offset = 0;

      IMB_colormanagement_imbuf_to_float_texture(
          rect_float, x, y, w, h, ibuf, store_premultiplied);
    }
  }

  if (scaled) {
    /* Slower update where we first have to scale the input pixels. */
    if (tile != nullptr) {
      ImageTile_Runtime *tile_runtime = &tile->runtime;
      int *tileoffset = tile_runtime->tilearray_offset;
      int *tilesize = tile_runtime->tilearray_size;
      int tilelayer = tile_runtime->tilearray_layer;
      gpu_texture_update_scaled(
          tex, rect, rect_float, ibuf->x, ibuf->y, x, y, tilelayer, tileoffset, tilesize, w, h);
    }
    else {
      gpu_texture_update_scaled(
          tex, rect, rect_float, ibuf->x, ibuf->y, x, y, -1, nullptr, nullptr, w, h);
    }
  }
  else {
    /* Fast update at same resolution. */
    if (tile != nullptr) {
      ImageTile_Runtime *tile_runtime = &tile->runtime;
      int *tileoffset = tile_runtime->tilearray_offset;
      int tilelayer = tile_runtime->tilearray_layer;
      gpu_texture_update_unscaled(
          tex, rect, rect_float, x, y, tilelayer, tileoffset, w, h, tex_stride, tex_offset);
    }
    else {
      gpu_texture_update_unscaled(
          tex, rect, rect_float, x, y, -1, nullptr, w, h, tex_stride, tex_offset);
    }
  }

  /* Free buffers if needed. */
  if (rect && rect != (uchar *)ibuf->rect) {
    MEM_freeN(rect);
  }
  if (rect_float && rect_float != ibuf->rect_float) {
    MEM_freeN(rect_float);
  }

  if (GPU_mipmap_enabled()) {
    GPU_texture_generate_mipmap(tex);
  }
  else {
    ima->gpuflag &= ~IMA_GPU_MIPMAP_COMPLETE;
  }

  GPU_texture_unbind(tex);
}

static void image_update_gputexture_ex(Image *ima,
                                       ImageGPUTextureStore::Entry &entry,
                                       ImageTile *tile,
                                       ImBuf *ibuf,
                                       int x,
                                       int y,
                                       int w,
                                       int h)
{
  const int eye = 0;
  GPUTexture *tex = entry.gputextures[TEXTARGET_2D][eye];
  /* Check if we need to update the main gputexture. */
  if (tex != nullptr && tile == ima->tiles.first) {
    gpu_texture_update_from_ibuf(tex, ima, ibuf, nullptr, x, y, w, h);
  }

  /* Check if we need to update the array gputexture. */
  tex = entry.gputextures[TEXTARGET_2D_ARRAY][eye];
  if (tex != nullptr) {
    gpu_texture_update_from_ibuf(tex, ima, ibuf, tile, x, y, w, h);
  }
}

void BKE_image_update_gputexture(Image *ima, ImageUser *iuser, int x, int y, int w, int h)
{
  ImageTile *image_tile = BKE_image_get_tile_from_iuser(ima, iuser);
  ImBuf *ibuf = BKE_image_acquire_ibuf(ima, iuser, nullptr);
  BKE_image_update_gputexture_delayed(ima, image_tile, ibuf, x, y, w, h);
  BKE_image_release_ibuf(ima, ibuf, nullptr);
}

void BKE_image_update_gputexture_delayed(struct Image *ima,
                                         struct ImageTile *image_tile,
                                         struct ImBuf *ibuf,
                                         int x,
                                         int y,
                                         int w,
                                         int h)
{
  /* Check for full refresh. */
  if (ibuf != nullptr && ima->source != IMA_SRC_TILED && x == 0 && y == 0 && w == ibuf->x &&
      h == ibuf->y) {
    BKE_image_partial_update_mark_full_update(ima);
  }
  else {
    rcti dirty_region;
    BLI_rcti_init(&dirty_region, x, x + w, y, y + h);
    BKE_image_partial_update_mark_region(ima, image_tile, ibuf, &dirty_region);
  }
}

void BKE_image_paint_set_mipmap(bool mipmap)
{
  g_texture_store.set_mipmap(mipmap);
}

/** \} */
}