// File: crn_ktx_texture.cpp
#include "crn_core.h"
#include "crn_ktx_texture.h"
#include "crn_console.h"

// Set #if CRNLIB_KTX_PVRTEX_WORKAROUNDS to 1 to enable various workarounds for oddball KTX files written by PVRTexTool.
#define CRNLIB_KTX_PVRTEX_WORKAROUNDS 1

namespace crnlib {
const uint8 s_ktx_file_id[12] = {0xAB, 0x4B, 0x54, 0x58, 0x20, 0x31, 0x31, 0xBB, 0x0D, 0x0A, 0x1A, 0x0A};

bool is_packed_pixel_ogl_type(uint32 ogl_type) {
  switch (ogl_type) {
    case KTX_UNSIGNED_BYTE_3_3_2:
    case KTX_UNSIGNED_BYTE_2_3_3_REV:
    case KTX_UNSIGNED_SHORT_5_6_5:
    case KTX_UNSIGNED_SHORT_5_6_5_REV:
    case KTX_UNSIGNED_SHORT_4_4_4_4:
    case KTX_UNSIGNED_SHORT_4_4_4_4_REV:
    case KTX_UNSIGNED_SHORT_5_5_5_1:
    case KTX_UNSIGNED_SHORT_1_5_5_5_REV:
    case KTX_UNSIGNED_INT_8_8_8_8:
    case KTX_UNSIGNED_INT_8_8_8_8_REV:
    case KTX_UNSIGNED_INT_10_10_10_2:
    case KTX_UNSIGNED_INT_2_10_10_10_REV:
    case KTX_UNSIGNED_INT_24_8:
    case KTX_UNSIGNED_INT_10F_11F_11F_REV:
    case KTX_UNSIGNED_INT_5_9_9_9_REV:
      return true;
  }
  return false;
}

uint get_ogl_type_size(uint32 ogl_type) {
  switch (ogl_type) {
    case KTX_UNSIGNED_BYTE:
    case KTX_BYTE:
      return 1;
    case KTX_HALF_FLOAT:
    case KTX_UNSIGNED_SHORT:
    case KTX_SHORT:
      return 2;
    case KTX_FLOAT:
    case KTX_UNSIGNED_INT:
    case KTX_INT:
      return 4;
    case KTX_UNSIGNED_BYTE_3_3_2:
    case KTX_UNSIGNED_BYTE_2_3_3_REV:
      return 1;
    case KTX_UNSIGNED_SHORT_5_6_5:
    case KTX_UNSIGNED_SHORT_5_6_5_REV:
    case KTX_UNSIGNED_SHORT_4_4_4_4:
    case KTX_UNSIGNED_SHORT_4_4_4_4_REV:
    case KTX_UNSIGNED_SHORT_5_5_5_1:
    case KTX_UNSIGNED_SHORT_1_5_5_5_REV:
      return 2;
    case KTX_UNSIGNED_INT_8_8_8_8:
    case KTX_UNSIGNED_INT_8_8_8_8_REV:
    case KTX_UNSIGNED_INT_10_10_10_2:
    case KTX_UNSIGNED_INT_2_10_10_10_REV:
    case KTX_UNSIGNED_INT_24_8:
    case KTX_UNSIGNED_INT_10F_11F_11F_REV:
    case KTX_UNSIGNED_INT_5_9_9_9_REV:
      return 4;
  }
  return 0;
}

uint32 get_ogl_base_internal_fmt(uint32 ogl_fmt) {
  switch (ogl_fmt) {
    case KTX_ETC1_RGB8_OES:
    case KTX_COMPRESSED_RGB8_ETC2:
    case KTX_RGB_S3TC:
    case KTX_RGB4_S3TC:
    case KTX_COMPRESSED_RGB_S3TC_DXT1_EXT:
    case KTX_COMPRESSED_SRGB_S3TC_DXT1_EXT:
      return KTX_RGB;
    case KTX_COMPRESSED_RGBA8_ETC2_EAC:
    case KTX_COMPRESSED_RGBA_S3TC_DXT1_EXT:
    case KTX_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT:
    case KTX_RGBA_S3TC:
    case KTX_RGBA4_S3TC:
    case KTX_COMPRESSED_RGBA_S3TC_DXT3_EXT:
    case KTX_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT:
    case KTX_COMPRESSED_RGBA_S3TC_DXT5_EXT:
    case KTX_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
    case KTX_RGBA_DXT5_S3TC:
    case KTX_RGBA4_DXT5_S3TC:
      return KTX_RGBA;
    case 1:
    case KTX_RED:
    case KTX_RED_INTEGER:
    case KTX_GREEN:
    case KTX_GREEN_INTEGER:
    case KTX_BLUE:
    case KTX_BLUE_INTEGER:
    case KTX_R8:
    case KTX_R8UI:
    case KTX_LUMINANCE8:
    case KTX_ALPHA:
    case KTX_LUMINANCE:
    case KTX_COMPRESSED_RED_RGTC1_EXT:
    case KTX_COMPRESSED_SIGNED_RED_RGTC1_EXT:
    case KTX_COMPRESSED_LUMINANCE_LATC1_EXT:
    case KTX_COMPRESSED_SIGNED_LUMINANCE_LATC1_EXT:
      return KTX_RED;
    case 2:
    case KTX_RG:
    case KTX_RG8:
    case KTX_RG_INTEGER:
    case KTX_LUMINANCE_ALPHA:
    case KTX_COMPRESSED_RED_GREEN_RGTC2_EXT:
    case KTX_COMPRESSED_SIGNED_RED_GREEN_RGTC2_EXT:
    case KTX_COMPRESSED_LUMINANCE_ALPHA_LATC2_EXT:
    case KTX_COMPRESSED_SIGNED_LUMINANCE_ALPHA_LATC2_EXT:
      return KTX_RG;
    case 3:
    case KTX_SRGB:
    case KTX_RGB:
    case KTX_RGB_INTEGER:
    case KTX_BGR:
    case KTX_BGR_INTEGER:
    case KTX_RGB8:
    case KTX_SRGB8:
      return KTX_RGB;
    case 4:
    case KTX_RGBA:
    case KTX_BGRA:
    case KTX_RGBA_INTEGER:
    case KTX_BGRA_INTEGER:
    case KTX_SRGB_ALPHA:
    case KTX_SRGB8_ALPHA8:
    case KTX_RGBA8:
      return KTX_RGBA;
  }
  return 0;
}

bool get_ogl_fmt_desc(uint32 ogl_fmt, uint32 ogl_type, uint& block_dim, uint& bytes_per_block) {
  uint ogl_type_size = get_ogl_type_size(ogl_type);

  block_dim = 1;
  bytes_per_block = 0;

  switch (ogl_fmt) {
    case KTX_COMPRESSED_RED_RGTC1_EXT:
    case KTX_COMPRESSED_SIGNED_RED_RGTC1_EXT:
    case KTX_COMPRESSED_LUMINANCE_LATC1_EXT:
    case KTX_COMPRESSED_SIGNED_LUMINANCE_LATC1_EXT:
    case KTX_ETC1_RGB8_OES:
    case KTX_COMPRESSED_RGB8_ETC2:
    case KTX_RGB_S3TC:
    case KTX_RGB4_S3TC:
    case KTX_COMPRESSED_RGB_S3TC_DXT1_EXT:
    case KTX_COMPRESSED_RGBA_S3TC_DXT1_EXT:
    case KTX_COMPRESSED_SRGB_S3TC_DXT1_EXT:
    case KTX_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT: {
      block_dim = 4;
      bytes_per_block = 8;
      break;
    }
    case KTX_COMPRESSED_RGBA8_ETC2_EAC:
    case KTX_COMPRESSED_LUMINANCE_ALPHA_LATC2_EXT:
    case KTX_COMPRESSED_SIGNED_LUMINANCE_ALPHA_LATC2_EXT:
    case KTX_COMPRESSED_RED_GREEN_RGTC2_EXT:
    case KTX_COMPRESSED_SIGNED_RED_GREEN_RGTC2_EXT:
    case KTX_RGBA_S3TC:
    case KTX_RGBA4_S3TC:
    case KTX_COMPRESSED_RGBA_S3TC_DXT3_EXT:
    case KTX_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT:
    case KTX_COMPRESSED_RGBA_S3TC_DXT5_EXT:
    case KTX_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT:
    case KTX_RGBA_DXT5_S3TC:
    case KTX_RGBA4_DXT5_S3TC: {
      block_dim = 4;
      bytes_per_block = 16;
      break;
    }
    case 1:
    case KTX_ALPHA:
    case KTX_RED:
    case KTX_GREEN:
    case KTX_BLUE:
    case KTX_RED_INTEGER:
    case KTX_GREEN_INTEGER:
    case KTX_BLUE_INTEGER:
    case KTX_LUMINANCE: {
      bytes_per_block = ogl_type_size;
      break;
    }
    case KTX_R8:
    case KTX_R8UI:
    case KTX_ALPHA8:
    case KTX_LUMINANCE8: {
      bytes_per_block = 1;
      break;
    }
    case 2:
    case KTX_RG:
    case KTX_RG_INTEGER:
    case KTX_LUMINANCE_ALPHA: {
      bytes_per_block = 2 * ogl_type_size;
      break;
    }
    case KTX_RG8:
    case KTX_LUMINANCE8_ALPHA8: {
      bytes_per_block = 2;
      break;
    }
    case 3:
    case KTX_SRGB:
    case KTX_RGB:
    case KTX_BGR:
    case KTX_RGB_INTEGER:
    case KTX_BGR_INTEGER: {
      bytes_per_block = is_packed_pixel_ogl_type(ogl_type) ? ogl_type_size : (3 * ogl_type_size);
      break;
    }
    case KTX_RGB8:
    case KTX_SRGB8: {
      bytes_per_block = 3;
      break;
    }
    case 4:
    case KTX_RGBA:
    case KTX_BGRA:
    case KTX_RGBA_INTEGER:
    case KTX_BGRA_INTEGER:
    case KTX_SRGB_ALPHA: {
      bytes_per_block = is_packed_pixel_ogl_type(ogl_type) ? ogl_type_size : (4 * ogl_type_size);
      break;
    }
    case KTX_SRGB8_ALPHA8:
    case KTX_RGBA8: {
      bytes_per_block = 4;
      break;
    }
    default:
      return false;
  }
  return true;
}

bool ktx_texture::compute_pixel_info() {
  if ((!m_header.m_glType) || (!m_header.m_glFormat)) {
    if ((m_header.m_glType) || (m_header.m_glFormat))
      return false;

    // Must be a compressed format.
    if (!get_ogl_fmt_desc(m_header.m_glInternalFormat, m_header.m_glType, m_block_dim, m_bytes_per_block)) {
#if CRNLIB_KTX_PVRTEX_WORKAROUNDS
      if ((!m_header.m_glInternalFormat) && (!m_header.m_glType) && (!m_header.m_glTypeSize) && (!m_header.m_glBaseInternalFormat)) {
        // PVRTexTool writes bogus headers when outputting ETC1.
        console::warning("ktx_texture::compute_pixel_info: Header doesn't specify any format, assuming ETC1 and hoping for the best");
        m_header.m_glBaseInternalFormat = KTX_RGB;
        m_header.m_glInternalFormat = KTX_ETC1_RGB8_OES;
        m_header.m_glTypeSize = 1;
        m_block_dim = 4;
        m_bytes_per_block = 8;
        return true;
      }
#endif
      return false;
    }

    if (m_block_dim == 1)
      return false;
  } else {
    // Must be an uncompressed format.
    if (!get_ogl_fmt_desc(m_header.m_glFormat, m_header.m_glType, m_block_dim, m_bytes_per_block))
      return false;

    if (m_block_dim > 1)
      return false;
  }
  return true;
}

bool ktx_texture::read_from_stream(data_stream_serializer& serializer) {
  clear();

  // Read header
  if (serializer.read(&m_header, 1, sizeof(m_header)) != sizeof(ktx_header))
    return false;

  // Check header
  if (memcmp(s_ktx_file_id, m_header.m_identifier, sizeof(m_header.m_identifier)))
    return false;

  if ((m_header.m_endianness != KTX_OPPOSITE_ENDIAN) && (m_header.m_endianness != KTX_ENDIAN))
    return false;

  m_opposite_endianness = (m_header.m_endianness == KTX_OPPOSITE_ENDIAN);
  if (m_opposite_endianness) {
    m_header.endian_swap();

    if ((m_header.m_glTypeSize != sizeof(uint8)) && (m_header.m_glTypeSize != sizeof(uint16)) && (m_header.m_glTypeSize != sizeof(uint32)))
      return false;
  }

  if (!check_header())
    return false;

  if (!compute_pixel_info())
    return false;

  uint8 pad_bytes[3];

  // Read the key value entries
  uint num_key_value_bytes_remaining = m_header.m_bytesOfKeyValueData;
  while (num_key_value_bytes_remaining) {
    if (num_key_value_bytes_remaining < sizeof(uint32))
      return false;

    uint32 key_value_byte_size;
    if (serializer.read(&key_value_byte_size, 1, sizeof(uint32)) != sizeof(uint32))
      return false;

    num_key_value_bytes_remaining -= sizeof(uint32);

    if (m_opposite_endianness)
      key_value_byte_size = utils::swap32(key_value_byte_size);

    if (key_value_byte_size > num_key_value_bytes_remaining)
      return false;

    uint8_vec key_value_data;
    if (key_value_byte_size) {
      key_value_data.resize(key_value_byte_size);
      if (serializer.read(&key_value_data[0], 1, key_value_byte_size) != key_value_byte_size)
        return false;
    }

    m_key_values.push_back(key_value_data);

    uint padding = 3 - ((key_value_byte_size + 3) % 4);
    if (padding) {
      if (serializer.read(pad_bytes, 1, padding) != padding)
        return false;
    }

    num_key_value_bytes_remaining -= key_value_byte_size;
    if (num_key_value_bytes_remaining < padding)
      return false;
    num_key_value_bytes_remaining -= padding;
  }

  // Now read the mip levels
  uint total_faces = get_num_mips() * get_array_size() * get_num_faces() * get_depth();
  if ((!total_faces) || (total_faces > 65535))
    return false;

// See Section 2.8 of KTX file format: No rounding to block sizes should be applied for block compressed textures.
// OK, I'm going to break that rule otherwise KTX can only store a subset of textures that DDS can handle for no good reason.
#if 0
      const uint mip0_row_blocks = m_header.m_pixelWidth / m_block_dim;
      const uint mip0_col_blocks = CRNLIB_MAX(1, m_header.m_pixelHeight) / m_block_dim;
#else
  const uint mip0_row_blocks = (m_header.m_pixelWidth + m_block_dim - 1) / m_block_dim;
  const uint mip0_col_blocks = (CRNLIB_MAX(1, m_header.m_pixelHeight) + m_block_dim - 1) / m_block_dim;
#endif
  if ((!mip0_row_blocks) || (!mip0_col_blocks))
    return false;

  bool has_valid_image_size_fields = true;
  bool disable_mip_and_cubemap_padding = false;

#if CRNLIB_KTX_PVRTEX_WORKAROUNDS
  {
    // PVRTexTool has a bogus KTX writer that doesn't write any imageSize fields. Nice.
    size_t expected_bytes_remaining = 0;
    for (uint mip_level = 0; mip_level < get_num_mips(); mip_level++) {
      uint mip_width, mip_height, mip_depth;
      get_mip_dim(mip_level, mip_width, mip_height, mip_depth);

      const uint mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
      const uint mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
      if ((!mip_row_blocks) || (!mip_col_blocks))
        return false;

      expected_bytes_remaining += sizeof(uint32);

      if ((!m_header.m_numberOfArrayElements) && (get_num_faces() == 6)) {
        for (uint face = 0; face < get_num_faces(); face++) {
          uint slice_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
          expected_bytes_remaining += slice_size;

          uint num_cube_pad_bytes = 3 - ((slice_size + 3) % 4);
          expected_bytes_remaining += num_cube_pad_bytes;
        }
      } else {
        uint total_mip_size = 0;
        for (uint array_element = 0; array_element < get_array_size(); array_element++) {
          for (uint face = 0; face < get_num_faces(); face++) {
            for (uint zslice = 0; zslice < mip_depth; zslice++) {
              uint slice_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
              total_mip_size += slice_size;
            }
          }
        }
        expected_bytes_remaining += total_mip_size;

        uint num_mip_pad_bytes = 3 - ((total_mip_size + 3) % 4);
        expected_bytes_remaining += num_mip_pad_bytes;
      }
    }

    if (serializer.get_stream()->get_remaining() < expected_bytes_remaining) {
      has_valid_image_size_fields = false;
      disable_mip_and_cubemap_padding = true;
      console::warning("ktx_texture::read_from_stream: KTX file size is smaller than expected - trying to read anyway without imageSize fields");
    }
  }
#endif

  for (uint mip_level = 0; mip_level < get_num_mips(); mip_level++) {
    uint mip_width, mip_height, mip_depth;
    get_mip_dim(mip_level, mip_width, mip_height, mip_depth);

    const uint mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
    const uint mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
    if ((!mip_row_blocks) || (!mip_col_blocks))
      return false;

    uint32 image_size = 0;
    if (!has_valid_image_size_fields)
      image_size = mip_depth * mip_row_blocks * mip_col_blocks * m_bytes_per_block * get_array_size() * get_num_faces();
    else {
      if (serializer.read(&image_size, 1, sizeof(image_size)) != sizeof(image_size))
        return false;

      if (m_opposite_endianness)
        image_size = utils::swap32(image_size);
    }

    if (!image_size)
      return false;

    uint total_mip_size = 0;

    if ((!m_header.m_numberOfArrayElements) && (get_num_faces() == 6)) {
      // plain non-array cubemap
      for (uint face = 0; face < get_num_faces(); face++) {
        CRNLIB_ASSERT(m_image_data.size() == get_image_index(mip_level, 0, face, 0));

        m_image_data.push_back(uint8_vec());
        uint8_vec& image_data = m_image_data.back();

        image_data.resize(image_size);
        if (serializer.read(&image_data[0], 1, image_size) != image_size)
          return false;

        if (m_opposite_endianness)
          utils::endian_swap_mem(&image_data[0], image_size, m_header.m_glTypeSize);

        uint num_cube_pad_bytes = disable_mip_and_cubemap_padding ? 0 : (3 - ((image_size + 3) % 4));
        if (serializer.read(pad_bytes, 1, num_cube_pad_bytes) != num_cube_pad_bytes)
          return false;

        total_mip_size += image_size + num_cube_pad_bytes;
      }
    } else {
      // 1D, 2D, 3D (normal or array texture), or array cubemap
      uint num_image_bytes_remaining = image_size;

      for (uint array_element = 0; array_element < get_array_size(); array_element++) {
        for (uint face = 0; face < get_num_faces(); face++) {
          for (uint zslice = 0; zslice < mip_depth; zslice++) {
            CRNLIB_ASSERT(m_image_data.size() == get_image_index(mip_level, array_element, face, zslice));

            uint slice_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
            if ((!slice_size) || (slice_size > num_image_bytes_remaining))
              return false;

            m_image_data.push_back(uint8_vec());
            uint8_vec& image_data = m_image_data.back();

            image_data.resize(slice_size);
            if (serializer.read(&image_data[0], 1, slice_size) != slice_size)
              return false;

            if (m_opposite_endianness)
              utils::endian_swap_mem(&image_data[0], slice_size, m_header.m_glTypeSize);

            num_image_bytes_remaining -= slice_size;

            total_mip_size += slice_size;
          }
        }
      }

      if (num_image_bytes_remaining)
        return false;
    }

    uint num_mip_pad_bytes = disable_mip_and_cubemap_padding ? 0 : (3 - ((total_mip_size + 3) % 4));
    if (serializer.read(pad_bytes, 1, num_mip_pad_bytes) != num_mip_pad_bytes)
      return false;
  }
  return true;
}

bool ktx_texture::write_to_stream(data_stream_serializer& serializer, bool no_keyvalue_data) {
  if (!consistency_check()) {
    CRNLIB_ASSERT(0);
    return false;
  }

  memcpy(m_header.m_identifier, s_ktx_file_id, sizeof(m_header.m_identifier));
  m_header.m_endianness = m_opposite_endianness ? KTX_OPPOSITE_ENDIAN : KTX_ENDIAN;

  if (m_block_dim == 1) {
    m_header.m_glTypeSize = get_ogl_type_size(m_header.m_glType);
    m_header.m_glBaseInternalFormat = m_header.m_glFormat;
  } else {
    m_header.m_glBaseInternalFormat = get_ogl_base_internal_fmt(m_header.m_glInternalFormat);
  }

  m_header.m_bytesOfKeyValueData = 0;
  if (!no_keyvalue_data) {
    for (uint i = 0; i < m_key_values.size(); i++)
      m_header.m_bytesOfKeyValueData += sizeof(uint32) + ((m_key_values[i].size() + 3) & ~3);
  }

  if (m_opposite_endianness)
    m_header.endian_swap();

  bool success = (serializer.write(&m_header, sizeof(m_header), 1) == 1);

  if (m_opposite_endianness)
    m_header.endian_swap();

  if (!success)
    return success;

  uint total_key_value_bytes = 0;
  const uint8 padding[3] = {0, 0, 0};

  if (!no_keyvalue_data) {
    for (uint i = 0; i < m_key_values.size(); i++) {
      uint32 key_value_size = m_key_values[i].size();

      if (m_opposite_endianness)
        key_value_size = utils::swap32(key_value_size);

      success = (serializer.write(&key_value_size, sizeof(key_value_size), 1) == 1);
      total_key_value_bytes += sizeof(key_value_size);

      if (m_opposite_endianness)
        key_value_size = utils::swap32(key_value_size);

      if (!success)
        return false;

      if (key_value_size) {
        if (serializer.write(&m_key_values[i][0], key_value_size, 1) != 1)
          return false;
        total_key_value_bytes += key_value_size;

        uint num_padding = 3 - ((key_value_size + 3) % 4);
        if ((num_padding) && (serializer.write(padding, num_padding, 1) != 1))
          return false;
        total_key_value_bytes += num_padding;
      }
    }
    (void)total_key_value_bytes;
  }

  CRNLIB_ASSERT(total_key_value_bytes == m_header.m_bytesOfKeyValueData);

  for (uint mip_level = 0; mip_level < get_num_mips(); mip_level++) {
    uint mip_width, mip_height, mip_depth;
    get_mip_dim(mip_level, mip_width, mip_height, mip_depth);

    const uint mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
    const uint mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
    if ((!mip_row_blocks) || (!mip_col_blocks))
      return false;

    uint32 image_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
    if ((m_header.m_numberOfArrayElements) || (get_num_faces() == 1))
      image_size *= (get_array_size() * get_num_faces() * get_depth());

    if (!image_size)
      return false;

    if (m_opposite_endianness)
      image_size = utils::swap32(image_size);

    success = (serializer.write(&image_size, sizeof(image_size), 1) == 1);

    if (m_opposite_endianness)
      image_size = utils::swap32(image_size);

    if (!success)
      return false;

    uint total_mip_size = 0;

    if ((!m_header.m_numberOfArrayElements) && (get_num_faces() == 6)) {
      // plain non-array cubemap
      for (uint face = 0; face < get_num_faces(); face++) {
        const uint8_vec& image_data = get_image_data(get_image_index(mip_level, 0, face, 0));
        if ((!image_data.size()) || (image_data.size() != image_size))
          return false;

        if (m_opposite_endianness) {
          uint8_vec tmp_image_data(image_data);
          utils::endian_swap_mem(&tmp_image_data[0], tmp_image_data.size(), m_header.m_glTypeSize);
          if (serializer.write(&tmp_image_data[0], tmp_image_data.size(), 1) != 1)
            return false;
        } else if (serializer.write(&image_data[0], image_data.size(), 1) != 1)
          return false;

        uint num_cube_pad_bytes = 3 - ((image_data.size() + 3) % 4);
        if ((num_cube_pad_bytes) && (serializer.write(padding, num_cube_pad_bytes, 1) != 1))
          return false;

        total_mip_size += image_size + num_cube_pad_bytes;
      }
    } else {
      // 1D, 2D, 3D (normal or array texture), or array cubemap
      for (uint array_element = 0; array_element < get_array_size(); array_element++) {
        for (uint face = 0; face < get_num_faces(); face++) {
          for (uint zslice = 0; zslice < mip_depth; zslice++) {
            const uint8_vec& image_data = get_image_data(get_image_index(mip_level, array_element, face, zslice));
            if (!image_data.size())
              return false;

            if (m_opposite_endianness) {
              uint8_vec tmp_image_data(image_data);
              utils::endian_swap_mem(&tmp_image_data[0], tmp_image_data.size(), m_header.m_glTypeSize);
              if (serializer.write(&tmp_image_data[0], tmp_image_data.size(), 1) != 1)
                return false;
            } else if (serializer.write(&image_data[0], image_data.size(), 1) != 1)
              return false;

            total_mip_size += image_data.size();
          }
        }
      }

      uint num_mip_pad_bytes = 3 - ((total_mip_size + 3) % 4);
      if ((num_mip_pad_bytes) && (serializer.write(padding, num_mip_pad_bytes, 1) != 1))
        return false;
      total_mip_size += num_mip_pad_bytes;
    }
    CRNLIB_ASSERT((total_mip_size & 3) == 0);
  }

  return true;
}

bool ktx_texture::init_2D(uint width, uint height, uint num_mips, uint32 ogl_internal_fmt, uint32 ogl_fmt, uint32 ogl_type) {
  clear();

  m_header.m_pixelWidth = width;
  m_header.m_pixelHeight = height;
  m_header.m_numberOfMipmapLevels = num_mips;
  m_header.m_glInternalFormat = ogl_internal_fmt;
  m_header.m_glFormat = ogl_fmt;
  m_header.m_glType = ogl_type;
  m_header.m_numberOfFaces = 1;

  if (!compute_pixel_info())
    return false;

  return true;
}

bool ktx_texture::init_2D_array(uint width, uint height, uint num_mips, uint array_size, uint32 ogl_internal_fmt, uint32 ogl_fmt, uint32 ogl_type) {
  clear();

  m_header.m_pixelWidth = width;
  m_header.m_pixelHeight = height;
  m_header.m_numberOfMipmapLevels = num_mips;
  m_header.m_numberOfArrayElements = array_size;
  m_header.m_glInternalFormat = ogl_internal_fmt;
  m_header.m_glFormat = ogl_fmt;
  m_header.m_glType = ogl_type;
  m_header.m_numberOfFaces = 1;

  if (!compute_pixel_info())
    return false;

  return true;
}

bool ktx_texture::init_3D(uint width, uint height, uint depth, uint num_mips, uint32 ogl_internal_fmt, uint32 ogl_fmt, uint32 ogl_type) {
  clear();

  m_header.m_pixelWidth = width;
  m_header.m_pixelHeight = height;
  m_header.m_pixelDepth = depth;
  m_header.m_numberOfMipmapLevels = num_mips;
  m_header.m_glInternalFormat = ogl_internal_fmt;
  m_header.m_glFormat = ogl_fmt;
  m_header.m_glType = ogl_type;
  m_header.m_numberOfFaces = 1;

  if (!compute_pixel_info())
    return false;

  return true;
}

bool ktx_texture::init_cubemap(uint dim, uint num_mips, uint32 ogl_internal_fmt, uint32 ogl_fmt, uint32 ogl_type) {
  clear();

  m_header.m_pixelWidth = dim;
  m_header.m_pixelHeight = dim;
  m_header.m_numberOfMipmapLevels = num_mips;
  m_header.m_glInternalFormat = ogl_internal_fmt;
  m_header.m_glFormat = ogl_fmt;
  m_header.m_glType = ogl_type;
  m_header.m_numberOfFaces = 6;

  if (!compute_pixel_info())
    return false;

  return true;
}

bool ktx_texture::check_header() const {
  if (((get_num_faces() != 1) && (get_num_faces() != 6)) || (!m_header.m_pixelWidth))
    return false;

  if ((!m_header.m_pixelHeight) && (m_header.m_pixelDepth))
    return false;

  if ((get_num_faces() == 6) && ((m_header.m_pixelDepth) || (!m_header.m_pixelHeight)))
    return false;

  if (m_header.m_numberOfMipmapLevels) {
    const uint max_mipmap_dimension = 1U << (m_header.m_numberOfMipmapLevels - 1U);
    if (max_mipmap_dimension > (CRNLIB_MAX(CRNLIB_MAX(m_header.m_pixelWidth, m_header.m_pixelHeight), m_header.m_pixelDepth)))
      return false;
  }

  return true;
}

bool ktx_texture::consistency_check() const {
  if (!check_header())
    return false;

  uint block_dim = 0, bytes_per_block = 0;
  if ((!m_header.m_glType) || (!m_header.m_glFormat)) {
    if ((m_header.m_glType) || (m_header.m_glFormat))
      return false;
    if (!get_ogl_fmt_desc(m_header.m_glInternalFormat, m_header.m_glType, block_dim, bytes_per_block))
      return false;
    if (block_dim == 1)
      return false;
    //if ((get_width() % block_dim) || (get_height() % block_dim))
    //   return false;
  } else {
    if (!get_ogl_fmt_desc(m_header.m_glFormat, m_header.m_glType, block_dim, bytes_per_block))
      return false;
    if (block_dim > 1)
      return false;
  }
  if ((m_block_dim != block_dim) || (m_bytes_per_block != bytes_per_block))
    return false;

  if (m_image_data.size() != get_total_images())
    return false;

  for (uint mip_level = 0; mip_level < get_num_mips(); mip_level++) {
    uint mip_width, mip_height, mip_depth;
    get_mip_dim(mip_level, mip_width, mip_height, mip_depth);

    const uint mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
    const uint mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
    if ((!mip_row_blocks) || (!mip_col_blocks))
      return false;

    for (uint array_element = 0; array_element < get_array_size(); array_element++) {
      for (uint face = 0; face < get_num_faces(); face++) {
        for (uint zslice = 0; zslice < mip_depth; zslice++) {
          const uint8_vec& image_data = get_image_data(get_image_index(mip_level, array_element, face, zslice));

          uint expected_image_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
          if (image_data.size() != expected_image_size)
            return false;
        }
      }
    }
  }

  return true;
}

const uint8_vec* ktx_texture::find_key(const char* pKey) const {
  const size_t n = strlen(pKey) + 1;
  for (uint i = 0; i < m_key_values.size(); i++) {
    const uint8_vec& v = m_key_values[i];
    if ((v.size() >= n) && (!memcmp(&v[0], pKey, n)))
      return &v;
  }

  return NULL;
}

bool ktx_texture::get_key_value_as_string(const char* pKey, dynamic_string& str) const {
  const uint8_vec* p = find_key(pKey);
  if (!p) {
    str.clear();
    return false;
  }

  const uint ofs = (static_cast<uint>(strlen(pKey)) + 1);
  const uint8* pValue = p->get_ptr() + ofs;
  const uint n = p->size() - ofs;

  uint i;
  for (i = 0; i < n; i++)
    if (!pValue[i])
      break;

  str.set_from_buf(pValue, i);
  return true;
}

uint ktx_texture::add_key_value(const char* pKey, const void* pVal, uint val_size) {
  const uint idx = m_key_values.size();
  m_key_values.resize(idx + 1);
  uint8_vec& v = m_key_values.back();
  v.append(reinterpret_cast<const uint8*>(pKey), static_cast<uint>(strlen(pKey)) + 1);
  v.append(static_cast<const uint8*>(pVal), val_size);
  return idx;
}

}  // namespace crnlib