source/src/engine/gl/image.cpp

/*
 * Copyright (C) 2023  Christopher J. Howard
 *
 * This file is part of Antkeeper source code.
 *
 * Antkeeper source code is free software: you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation, either version 3 of the License, or
 * (at your option) any later version.
 *
 * Antkeeper source code 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 Antkeeper source code.  If not, see <http://www.gnu.org/licenses/>.
 */

#include <engine/gl/image.hpp>
#include <engine/gl/cube-map.hpp>
#include <engine/gl/opengl/gl-format-lut.hpp>
#include <engine/resources/resource-loader.hpp>
#include <engine/resources/deserialize-error.hpp>
#include <engine/resources/deserializer.hpp>
#include <engine/debug/log.hpp>
#include <cmath>
#include <stdexcept>
#include <glad/gl.h>
#include <stb/stb_image.h>
#include <tinyexr.h>

namespace gl {

image::image
(
	std::uint8_t dimensionality,
	gl::format format,
	std::uint32_t width,
	std::uint32_t height,
	std::uint32_t depth,
	std::uint32_t mip_levels,
	std::uint32_t array_layers,
	std::uint32_t flags
)
{
	const auto format_index = std::to_underlying(format);
	const auto gl_internal_format = gl_format_lut[format_index][0];
	const auto gl_type = gl_format_lut[format_index][2];
	
	if (gl_internal_format == 0 || gl_type == 0)
	{
		throw std::invalid_argument("Image construction used unsupported format.");
	}
	
	if (!width || !height || !depth)
	{
		throw std::invalid_argument("Image dimensions must be nonzero.");
	}
	
	if (!mip_levels)
	{
		throw std::invalid_argument("Image mip levels must be nonzero.");
	}
	
	if (mip_levels > static_cast<std::uint32_t>(std::bit_width(std::max(std::max(width, height), depth))))
	{
		throw std::out_of_range("Image mip levels exceed `1 + log2(max(width, height, depth))`.");
	}
	
	if (!array_layers)
	{
		throw std::invalid_argument("Image array layers must be nonzero.");
	}
	
	if (dimensionality == 1)
	{
		if (height > 1 || depth > 1)
		{
			throw std::invalid_argument("1D image must have a height and depth of `1`.");
		}
	}
	else if (dimensionality == 2)
	{
		if (depth > 1)
		{
			throw std::invalid_argument("2D image must have a depth of `1`.");
		}
	}
	else if (dimensionality == 3)
	{
		if (array_layers > 1)
		{
			throw std::invalid_argument("3D image arrays not supported.");
		}
	}
	
	if (flags & std::to_underlying(image_flag::cube_compatible))
	{
		if (dimensionality != 2)
		{
			throw std::invalid_argument("Cube compatible image must be 2D.");
		}
		
		if (width != height)
		{
			throw std::invalid_argument("Cube compatible image width and height must be equal.");
		}
		
		if (array_layers % 6 != 0)
		{
			throw std::invalid_argument("Cube compatible image array layers must be a multiple of 6.");
		}
	}
	
	m_dimensionality = dimensionality;
	m_format = format;
	m_dimensions = {width, height, depth};
	m_mip_levels = mip_levels;
	m_array_layers = array_layers;
	m_flags = flags;
	
	if (m_array_layers == 1)
	{
		switch (m_dimensionality)
		{
			case 1:
				m_gl_texture_target = GL_TEXTURE_1D;
				glCreateTextures(m_gl_texture_target, 1, &m_gl_texture_name);
				glTextureStorage1D
				(
					m_gl_texture_name,
					static_cast<GLsizei>(m_mip_levels),
					gl_internal_format,
					static_cast<GLsizei>(m_dimensions[0])
				);
				break;
			
			case 2:
				m_gl_texture_target = GL_TEXTURE_2D;
				glCreateTextures(m_gl_texture_target, 1, &m_gl_texture_name);
				glTextureStorage2D
				(
					m_gl_texture_name,
					static_cast<GLsizei>(m_mip_levels),
					gl_internal_format,
					static_cast<GLsizei>(m_dimensions[0]),
					static_cast<GLsizei>(m_dimensions[1])
				);
				break;
			
			case 3:
				m_gl_texture_target = GL_TEXTURE_3D;
				glCreateTextures(m_gl_texture_target, 1, &m_gl_texture_name);
				glTextureStorage3D
				(
					m_gl_texture_name,
					static_cast<GLsizei>(m_mip_levels),
					gl_internal_format,
					static_cast<GLsizei>(m_dimensions[0]),
					static_cast<GLsizei>(m_dimensions[1]),
					static_cast<GLsizei>(m_dimensions[2])
				);
				break;
			
			default:
				break;
		}
	}
	else
	{
		switch (m_dimensionality)
		{
			case 1:
				m_gl_texture_target = GL_TEXTURE_1D_ARRAY;
				glCreateTextures(m_gl_texture_target, 1, &m_gl_texture_name);
				glTextureStorage2D
				(
					m_gl_texture_name,
					static_cast<GLsizei>(m_mip_levels),
					gl_internal_format,
					static_cast<GLsizei>(m_dimensions[0]),
					static_cast<GLsizei>(m_array_layers)
				);
				break;
			
			case 2:
				if (is_cube_compatible())
				{
					if (m_array_layers == 6)
					{
						m_gl_texture_target = GL_TEXTURE_CUBE_MAP;
						glCreateTextures(m_gl_texture_target, 1, &m_gl_texture_name);
						glTextureStorage2D
						(
							m_gl_texture_name,
							static_cast<GLsizei>(m_mip_levels),
							gl_internal_format,
							static_cast<GLsizei>(m_dimensions[0]),
							static_cast<GLsizei>(m_dimensions[1])
						);
					}
					else
					{
						m_gl_texture_target = GL_TEXTURE_CUBE_MAP_ARRAY;
						glCreateTextures(m_gl_texture_target, 1, &m_gl_texture_name);
						glTextureStorage3D
						(
							m_gl_texture_name,
							static_cast<GLsizei>(m_mip_levels),
							gl_internal_format,
							static_cast<GLsizei>(m_dimensions[0]),
							static_cast<GLsizei>(m_dimensions[1]),
							static_cast<GLsizei>(m_array_layers)
						);
					}
				}
				else
				{
					m_gl_texture_target = GL_TEXTURE_2D_ARRAY;
					glCreateTextures(m_gl_texture_target, 1, &m_gl_texture_name);
					glTextureStorage3D
					(
						m_gl_texture_name,
						static_cast<GLsizei>(m_mip_levels),
						gl_internal_format,
						static_cast<GLsizei>(m_dimensions[0]),
						static_cast<GLsizei>(m_dimensions[1]),
						static_cast<GLsizei>(m_array_layers)
					);
				}
				break;
			
			default:
				break;
		}
	}
}

image::~image()
{
	glDeleteTextures(1, &m_gl_texture_name);
}

void image::read
(
	std::uint32_t mip_level,
	std::uint32_t offset_x,
	std::uint32_t offset_y,
	std::uint32_t offset_z,
	std::uint32_t width,
	std::uint32_t height,
	std::uint32_t depth,
	gl::format format,
	std::span<std::byte> data
) const
{
	if (mip_level >= m_mip_levels)
	{
		throw std::out_of_range("Image read operation mip level out of range.");
	}
	
	const auto format_index = std::to_underlying(format);
	const auto gl_base_format = gl_format_lut[format_index][1];
	const auto gl_type = gl_format_lut[format_index][2];
	
	if (gl_base_format == 0 || gl_type == 0)
	{
		throw std::invalid_argument("Image read operation used unsupported format.");
	}
	
	glGetTextureSubImage
	(
		m_gl_texture_name,
		static_cast<GLint>(mip_level),
		static_cast<GLint>(offset_x),
		static_cast<GLint>(offset_y),
		static_cast<GLint>(offset_z),
		static_cast<GLsizei>(width),
		static_cast<GLsizei>(height),
		static_cast<GLsizei>(depth),
		gl_base_format,
		gl_type,
		static_cast<GLsizei>(data.size()),
		data.data()
	);
}

void image::write
(
	std::uint32_t mip_level,
	std::uint32_t offset_x,
	std::uint32_t offset_y,
	std::uint32_t offset_z,
	std::uint32_t width,
	std::uint32_t height,
	std::uint32_t depth,
	gl::format format,
	std::span<const std::byte> data
)
{
	if (mip_level >= m_mip_levels)
	{
		throw std::out_of_range("Image write operation mip level out of range.");
	}
	
	const auto format_index = std::to_underlying(format);
	const auto gl_base_format = gl_format_lut[format_index][1];
	const auto gl_type = gl_format_lut[format_index][2];
	
	if (gl_base_format == 0 || gl_type == 0)
	{
		throw std::invalid_argument("Image write operation used unsupported format.");
	}
	
	if (m_array_layers == 1)
	{
		if ((offset_x + width  > std::max<std::uint32_t>(1, m_dimensions[0] >> mip_level)) ||
			(offset_y + height > std::max<std::uint32_t>(1, m_dimensions[1] >> mip_level)) ||
			(offset_z + depth  > std::max<std::uint32_t>(1, m_dimensions[2] >> mip_level)))
		{
			throw std::out_of_range("Image write operation exceeded image bounds.");
		}
		
		switch (m_dimensionality)
		{
			case 1:
				glTextureSubImage1D
				(
					m_gl_texture_name,
					static_cast<GLint>(mip_level),
					static_cast<GLint>(offset_x),
					static_cast<GLsizei>(width),
					gl_base_format,
					gl_type,
					data.data()
				);
				break;
			
			case 2:
				glTextureSubImage2D
				(
					m_gl_texture_name,
					static_cast<GLint>(mip_level),
					static_cast<GLint>(offset_x),
					static_cast<GLint>(offset_y),
					static_cast<GLsizei>(width),
					static_cast<GLsizei>(height),
					gl_base_format,
					gl_type,
					data.data()
				);
				break;
			
			case 3:
				glTextureSubImage3D
				(
					m_gl_texture_name,
					static_cast<GLint>(mip_level),
					static_cast<GLint>(offset_x),
					static_cast<GLint>(offset_y),
					static_cast<GLint>(offset_z),
					static_cast<GLsizei>(width),
					static_cast<GLsizei>(height),
					static_cast<GLsizei>(depth),
					gl_base_format,
					gl_type,
					data.data()
				);
				break;
			
			default:
				break;
		}
	}
	else
	{
		switch (m_dimensionality)
		{
			case 1:
				if ((offset_x + width  > std::max<std::uint32_t>(1, m_dimensions[0] >> mip_level)) ||
					(offset_y + height > m_array_layers) ||
					(offset_z + depth  > 1))
				{
					throw std::out_of_range("Image write operation exceeded image dimensions.");
				}
				
				glTextureSubImage2D
				(
					m_gl_texture_name,
					static_cast<GLint>(mip_level),
					static_cast<GLint>(offset_x),
					static_cast<GLint>(offset_y),
					static_cast<GLsizei>(width),
					static_cast<GLsizei>(height),
					gl_base_format,
					gl_type,
					data.data()
				);
				break;
			
			case 2:
				if ((offset_x + width  > std::max<std::uint32_t>(1, m_dimensions[0] >> mip_level)) ||
					(offset_y + height > std::max<std::uint32_t>(1, m_dimensions[1] >> mip_level)) ||
					(offset_z + depth  > m_array_layers))
				{
					throw std::out_of_range("Image write operation exceeded image bounds.");
				}
				
				glTextureSubImage3D
				(
					m_gl_texture_name,
					static_cast<GLint>(mip_level),
					static_cast<GLint>(offset_x),
					static_cast<GLint>(offset_y),
					static_cast<GLint>(offset_z),
					static_cast<GLsizei>(width),
					static_cast<GLsizei>(height),
					static_cast<GLsizei>(depth),
					gl_base_format,
					gl_type,
					data.data()
				);
				break;
			
			default:
				break;
		}
	}
}

void image::copy
(
	std::uint32_t src_mip_level,
	std::uint32_t src_x,
	std::uint32_t src_y,
	std::uint32_t src_z,
	image& dst_image,
	std::uint32_t dst_mip_level,
	std::uint32_t dst_x,
	std::uint32_t dst_y,
	std::uint32_t dst_z,
	std::uint32_t width,
	std::uint32_t height,
	std::uint32_t depth
) const
{
	glCopyImageSubData
	(
		m_gl_texture_name,
		m_gl_texture_target,
		static_cast<GLint>(src_mip_level),
		static_cast<GLint>(src_x),
		static_cast<GLint>(src_y),
		static_cast<GLint>(src_z),
		dst_image.m_gl_texture_name,
		dst_image.m_gl_texture_target,
		static_cast<GLint>(dst_mip_level),
		static_cast<GLint>(dst_x),
		static_cast<GLint>(dst_y),
		static_cast<GLint>(dst_z),
		static_cast<GLsizei>(width),
		static_cast<GLsizei>(height),
		static_cast<GLsizei>(depth)
	);
}

void image::generate_mipmaps()
{
	if (m_mip_levels > 1)
	{
		glGenerateTextureMipmap(m_gl_texture_name);
	}
}

image_1d::image_1d
(
	gl::format format,
	std::uint32_t width,
	std::uint32_t mip_levels,
	std::uint32_t array_layers,
	std::uint32_t flags
):
	image
	(
		1,
		format,
		width,
		1,
		1,
		mip_levels,
		array_layers,
		flags
	)
{}

image_2d::image_2d
(
	gl::format format,
	std::uint32_t width,
	std::uint32_t height,
	std::uint32_t mip_levels,
	std::uint32_t array_layers,
	std::uint32_t flags
):
	image
	(
		2,
		format,
		width,
		height,
		1,
		mip_levels,
		array_layers,
		flags
	)
{}

image_3d::image_3d
(
	gl::format format,
	std::uint32_t width,
	std::uint32_t height,
	std::uint32_t depth,
	std::uint32_t mip_levels,
	std::uint32_t flags
):
	image
	(
		3,
		format,
		width,
		height,
		depth,
		mip_levels,
		1,
		flags
	)
{}

image_cube::image_cube
(
	gl::format format,
	std::uint32_t width,
	std::uint32_t mip_levels,
	std::uint32_t array_layers
):
	image_2d
	(
		format,
		width,
		width,
		mip_levels,
		array_layers,
		std::to_underlying(image_flag::cube_compatible)
	)
{}

} // namespace gl

namespace {
	
	int stb_io_read(void* user, char* data, int size)
	{
		deserialize_context& ctx = *static_cast<deserialize_context*>(user);
		return static_cast<int>(ctx.read8(reinterpret_cast<std::byte*>(data), static_cast<std::size_t>(size)));
	}
	
	void stb_io_skip(void* user, int n)
	{
		deserialize_context& ctx = *static_cast<deserialize_context*>(user);
		ctx.seek(ctx.tell() + n);
	}
	
	int stb_io_eof(void* user)
	{
		deserialize_context& ctx = *static_cast<deserialize_context*>(user);
		return static_cast<int>(ctx.eof());
	}
	
	struct stb_image_deleter
	{
		void operator()(void* p) const
		{
			stbi_image_free(p);
		}
	};
	
	[[nodiscard]] std::unique_ptr<gl::image> load_image_stb_image(deserialize_context& ctx, std::uint8_t dimensionality, std::uint32_t mip_levels)
	{
		// Setup IO callbacks
		const stbi_io_callbacks io_callbacks
		{
			&stb_io_read,
			&stb_io_skip,
			&stb_io_eof
		};
		
		// Determine image bit depth
		std::size_t component_size = stbi_is_16_bit_from_callbacks(&io_callbacks, &ctx) ? sizeof(std::uint16_t) : sizeof(std::uint8_t);
		ctx.seek(0);
		
		// Set vertical flip on load in order to correctly upload pixel data to OpenGL
		stbi_set_flip_vertically_on_load(true);
		
		// Load image data
		std::unique_ptr<void, stb_image_deleter> data;
		int width;
		int height;
		int components;
		gl::format format;
		if (component_size == sizeof(std::uint16_t))
		{
			// Load 16-bit image data
			data = std::unique_ptr<void, stb_image_deleter>(stbi_load_16_from_callbacks(&io_callbacks, &ctx, &width, &height, &components, 0));
			
			// Determine 16-bit image format
			format = [components]()
			{
				switch (components)
				{
					case 1:
						return gl::format::r16_unorm;
					case 2:
						return gl::format::r16g16_unorm;
					case 3:
						return gl::format::r16g16b16_unorm;
					case 4:
						return gl::format::r16g16b16a16_unorm;
					default:
						return gl::format::undefined;
				}
			}();
		}
		else
		{
			// Load 8-bit image data
			data = std::unique_ptr<void, stb_image_deleter>(stbi_load_from_callbacks(&io_callbacks, &ctx, &width, &height, &components, 0));
			
			// Determine 8-bit image format
			format = [components]()
			{
				switch (components)
				{
					case 1:
						return gl::format::r8_unorm;
					case 2:
						return gl::format::r8g8_unorm;
					case 3:
						return gl::format::r8g8b8_unorm;
					case 4:
						return gl::format::r8g8b8a8_unorm;
					default:
						return gl::format::undefined;
				}
			}();
		}
		
		// Check if image data was loaded
		if (!data)
		{
			throw deserialize_error(stbi_failure_reason());
		}
		
		// Determine number mip levels
		if (!mip_levels)
		{
			mip_levels = static_cast<std::uint32_t>(std::bit_width(static_cast<std::uint32_t>(std::max(width, height))));
		}
		
		// Allocate image
		std::unique_ptr<gl::image> image;
		switch (dimensionality)
		{
			case 1:
				image = std::make_unique<gl::image_1d>
				(
					format,
					static_cast<std::uint32_t>(std::max(width, height)),
					mip_levels
				);
				break; 
			
			case 2:
				image = std::make_unique<gl::image_2d>
				(
					format,
					static_cast<std::uint32_t>(width),
					static_cast<std::uint32_t>(height),
					mip_levels
				);
				break;
			
			case 3:
				image = std::make_unique<gl::image_3d>
				(
					format,
					static_cast<std::uint32_t>(width),
					static_cast<std::uint32_t>(height),
					1,
					mip_levels
				);
				break;
			
			default:
				break;
		}
		
		// Upload image data to image
		image->write
		(
			0,
			0,
			0,
			0,
			image->get_dimensions()[0],
			image->get_dimensions()[1],
			image->get_dimensions()[2],
			format,
			{
				reinterpret_cast<const std::byte*>(data.get()),
				image->get_dimensions()[0] *
					image->get_dimensions()[1] *
					image->get_dimensions()[2] *
					static_cast<std::size_t>(components) *
					component_size
			}
		);
		
		// Generate mipmaps
		image->generate_mipmaps();
		
		return image;
	}
	
	[[nodiscard]] std::unique_ptr<gl::image> load_image_tinyexr(deserialize_context& ctx, std::uint8_t dimensionality, std::uint32_t mip_levels)
	{
		const char* error = nullptr;
		auto tinyexr_error = [&error]()
		{
			const std::string error_message(error);
			FreeEXRErrorMessage(error);
			throw deserialize_error(error_message);
		};
		
		// Read data into file buffer
		std::vector<unsigned char> file_buffer(ctx.size());
		ctx.read8(reinterpret_cast<std::byte*>(file_buffer.data()), file_buffer.size());
		
		// Read EXR version
		EXRVersion exr_version;
		if (ParseEXRVersionFromMemory(&exr_version, file_buffer.data(), file_buffer.size()) != TINYEXR_SUCCESS)
		{
			tinyexr_error();
		}
		
		// Check if image is multipart
		if (exr_version.multipart)
		{
			throw deserialize_error("OpenEXR multipart images not supported.");
		}
		
		// Load image header
		EXRHeader exr_header;
		InitEXRHeader(&exr_header);
		if (ParseEXRHeaderFromMemory(&exr_header, &exr_version, file_buffer.data(), file_buffer.size(), &error) != TINYEXR_SUCCESS)
		{
			tinyexr_error();
		}
		
		// Check if image is tiled
		if (exr_header.tiled)
		{
			FreeEXRHeader(&exr_header);
			throw deserialize_error("OpenEXR tiled images not supported.");
		}
		
		// Check if image has a supported number of channels
		if (exr_header.num_channels < 1 || exr_header.num_channels > 4)
		{
			FreeEXRHeader(&exr_header);
			throw deserialize_error("OpenEXR images must have 1-4 channels.");
		}
		
		// Check if all channels have the same format
		for (int i = 1; i < exr_header.num_channels; ++i)
		{
			if (exr_header.pixel_types[i] != exr_header.pixel_types[i - 1])
			{
				FreeEXRHeader(&exr_header);
				throw deserialize_error("OpenEXR images must have the same pixel type per channel.");
			}
		}
		
		// Load image data
		EXRImage exr_image;
		InitEXRImage(&exr_image);
		if (LoadEXRImageFromMemory(&exr_image, &exr_header, file_buffer.data(), file_buffer.size(), &error) != TINYEXR_SUCCESS)
		{
			FreeEXRHeader(&exr_header);
			tinyexr_error();
		}
		
		// Free file buffer
		file_buffer.clear();
		
		// Determine image format
		constexpr gl::format uint_formats[4] =
		{
			gl::format::r32_uint,
			gl::format::r32g32_uint,
			gl::format::r32g32b32_uint,
			gl::format::r32g32b32a32_uint
		};
		constexpr gl::format half_formats[4] =
		{
			gl::format::r16_sfloat,
			gl::format::r16g16_sfloat,
			gl::format::r16g16b16_sfloat,
			gl::format::r16g16b16a16_sfloat
		};
		constexpr gl::format float_formats[4] =
		{
			gl::format::r32_sfloat,
			gl::format::r32g32_sfloat,
			gl::format::r32g32b32_sfloat,
			gl::format::r32g32b32a32_sfloat
		};
		gl::format format;
		int component_size;
		switch (exr_header.pixel_types[0])
		{
			case TINYEXR_PIXELTYPE_UINT:
				format = uint_formats[exr_header.num_channels - 1];
				component_size = static_cast<int>(sizeof(std::uint32_t));
				break;
				
			case TINYEXR_PIXELTYPE_HALF:
				format = half_formats[exr_header.num_channels - 1];
				component_size = static_cast<int>(sizeof(std::uint16_t));//sizeof(float16_t)
				break;
				
			case TINYEXR_PIXELTYPE_FLOAT:
				format = float_formats[exr_header.num_channels - 1];
				component_size = static_cast<int>(sizeof(float));//sizeof(float32_t)
				break;
			
			default:
				format = gl::format::undefined;
				component_size = 0;
				break;
		}
		
		// Allocate interleaved image data
		std::vector<std::byte> data(static_cast<std::size_t>(exr_image.width * exr_image.height * exr_header.num_channels * component_size));
		
		// Interleave image data from layers
		std::byte* component = data.data();
		for (auto y = exr_image.height - 1; y >= 0; --y)
		{
			const auto row_offset = y * exr_image.width;
			
			for (auto x = 0; x < exr_image.width; ++x)
			{
				const auto byte_offset = (row_offset + x) * component_size;
				
				for (auto c = exr_image.num_channels - 1; c >= 0; --c)
				{
					std::memcpy(component, exr_image.images[c] + byte_offset, static_cast<std::size_t>(component_size));
					component += component_size;
				}
			}
		}
		
		// Store image dimensions
		const auto width = static_cast<std::uint32_t>(exr_image.width);
		const auto height = static_cast<std::uint32_t>(exr_image.height);
		
		// Free loaded image data and image header
		FreeEXRImage(&exr_image);
		FreeEXRHeader(&exr_header);
		
		// Determine number mip levels
		if (!mip_levels)
		{
			mip_levels = static_cast<std::uint32_t>(std::bit_width(std::max(width, height)));
		}
		
		// Allocate image
		std::unique_ptr<gl::image> image;
		switch (dimensionality)
		{
			case 1:
				image = std::make_unique<gl::image_1d>
				(
					format,
					std::max(width, height),
					mip_levels
				);
				break; 
			
			case 2:
				image = std::make_unique<gl::image_2d>
				(
					format,
					width,
					height,
					mip_levels
				);
				break;
			
			case 3:
				image = std::make_unique<gl::image_3d>
				(
					format,
					width,
					height,
					1,
					mip_levels
				);
				break;
			
			default:
				break;
		}
		
		// Upload interleaved image data to image
		image->write
		(
			0,
			0,
			0,
			0,
			image->get_dimensions()[0],
			image->get_dimensions()[1],
			image->get_dimensions()[2],
			format,
			data
		);
		
		// Generate mipmaps
		image->generate_mipmaps();
		
		return image;
	}
	
	[[nodiscard]] std::unique_ptr<gl::image> load_image(deserialize_context& ctx, std::uint8_t dimensionality, std::uint32_t mip_levels)
	{
		// Select loader according to file extension
		if (ctx.path().extension() == ".exr")
		{
			// Load EXR images with TinyEXR
			return load_image_tinyexr(ctx, dimensionality, mip_levels);
		}
		else
		{
			// Load other image formats with stb_image
			return load_image_stb_image(ctx, dimensionality, mip_levels);
		}
	}
}

template <>
std::unique_ptr<gl::image_1d> resource_loader<gl::image_1d>::load(::resource_manager& resource_manager, deserialize_context& ctx)
{
	return std::unique_ptr<gl::image_1d>(static_cast<gl::image_1d*>(load_image(ctx, 1, 0).release()));
}

template <>
std::unique_ptr<gl::image_2d> resource_loader<gl::image_2d>::load(::resource_manager& resource_manager, deserialize_context& ctx)
{
	return std::unique_ptr<gl::image_2d>(static_cast<gl::image_2d*>(load_image(ctx, 2, 0).release()));
}

template <>
std::unique_ptr<gl::image_3d> resource_loader<gl::image_3d>::load(::resource_manager& resource_manager, deserialize_context& ctx)
{
	return std::unique_ptr<gl::image_3d>(static_cast<gl::image_3d*>(load_image(ctx, 3, 0).release()));
}

template <>
std::unique_ptr<gl::image_cube> resource_loader<gl::image_cube>::load(::resource_manager& resource_manager, deserialize_context& ctx)
{
	// Load cube map
	auto cube_map = std::unique_ptr<gl::image_2d>(static_cast<gl::image_2d*>(load_image(ctx, 2, 1).release()));
	
	// Determine cube map layout
	const auto layout = gl::infer_cube_map_layout(cube_map->get_dimensions()[0], cube_map->get_dimensions()[1]);
	if (layout == gl::cube_map_layout::unknown)
	{
		throw deserialize_error("Failed to load cube image from cube map with unknown layout.");
	}
	else if (layout == gl::cube_map_layout::equirectangular || layout == gl::cube_map_layout::spherical)
	{
		throw deserialize_error("Failed to load cube image from cube map with unsupported layout.");
	}
	
	// Determine cube map face width
	const auto face_width = gl::infer_cube_map_face_width(cube_map->get_dimensions()[0], cube_map->get_dimensions()[1], layout);
	
	// Allocate cube image
	auto image = std::make_unique<gl::image_cube>
	(
		cube_map->get_format(),
		face_width,
		static_cast<std::uint32_t>(std::bit_width(face_width))
	);
	
	// Vertical cross layout face offsets
	constexpr std::uint32_t vcross_offsets[6][2] =
	{
		{2, 2}, {0, 2}, // -x, +x
		{1, 3}, {1, 1}, // -y, +y
		{1, 0}, {1, 2}  // -z, +z
	};
	
	// Horizontal cross layout face offsets
	constexpr std::uint32_t hcross_offsets[6][2] =
	{
		{2, 1}, {0, 1}, // -x, +x
		{1, 2}, {1, 0}, // -y, +y
		{3, 1}, {1, 1}  // -z, +z
	};
	
	// Copy cube map faces to cube image
	switch (layout)
	{
		case gl::cube_map_layout::column:
			for (std::uint32_t i = 0; i < 6; ++i)
			{
				cube_map->copy(0, 0, face_width * i, 0, *image, 0, 0, 0, i, face_width, face_width, 1);
			}
			break;
		
		case gl::cube_map_layout::row:
			for (std::uint32_t i = 0; i < 6; ++i)
			{
				cube_map->copy(0, face_width * i, 0, 0, *image, 0, 0, 0, i, face_width, face_width, 1);
			}
			break;
		
		case gl::cube_map_layout::vertical_cross:
			for (std::uint32_t i = 0; i < 6; ++i)
			{
				cube_map->copy(0, face_width * vcross_offsets[i][0], face_width * vcross_offsets[i][1], 0, *image, 0, 0, 0, i, face_width, face_width, 1);
			}
			break;
		
		case gl::cube_map_layout::horizontal_cross:
			for (std::uint32_t i = 0; i < 6; ++i)
			{
				cube_map->copy(0, face_width * hcross_offsets[i][0], face_width * hcross_offsets[i][1], 0, *image, 0, 0, 0, i, face_width, face_width, 1);
			}
			break;
		
		default:
			break;
	}
	
	// Generate mipmaps
	image->generate_mipmaps();
	
	return image;
}