bgfx/examples/common/cube_atlas.cpp

/*
 * Copyright 2013 Jeremie Roy. All rights reserved.
 * License: http://www.opensource.org/licenses/BSD-2-Clause
 */

#include "cube_atlas.h"
#include <bx/bx.h>
#include <bgfx.h>
#include <vector>

//********** Rectangle packer implementation ************
class RectanglePacker
{
public:
RectanglePacker();
RectanglePacker(uint32_t _width, uint32_t _height);

/// non constructor initialization
void init(uint32_t _width, uint32_t _height);
/// find a suitable position for the given rectangle
/// @return true if the rectangle can be added, false otherwise
bool addRectangle(uint16_t _width, uint16_t _height, uint16_t& _outX, uint16_t& _outY);
/// return the used surface in squared unit
uint32_t getUsedSurface()
{
	return m_usedSpace;
}
/// return the total available surface in squared unit
uint32_t getTotalSurface()
{
	return m_width * m_height;
}
/// return the usage ratio of the available surface [0:1]
float getUsageRatio();
/// reset to initial state
void clear();

private:
int32_t fit(uint32_t _skylineNodeIndex, uint16_t _width, uint16_t _height);
/// Merges all skyline nodes that are at the same level.
void merge();

struct Node
{
	Node(int16_t _x, int16_t _y, int16_t _width) : m_x(_x), m_y(_y), m_width(_width)
	{
	}

	/// The starting x-coordinate (leftmost).
	int16_t m_x;
	/// The y-coordinate of the skyline level line.
	int16_t m_y;
	/// The line _width. The ending coordinate (inclusive) will be x+width-1.
	int32_t m_width;         //32bit to avoid padding
};

/// width (in pixels) of the underlying texture
uint32_t m_width;
/// height (in pixels) of the underlying texture
uint32_t m_height;
/// Surface used in squared pixel
uint32_t m_usedSpace;
/// node of the skyline algorithm
std::vector<Node> m_skyline;
};

RectanglePacker::RectanglePacker() : m_width(0), m_height(0), m_usedSpace(0)
{
}

RectanglePacker::RectanglePacker(uint32_t _width, uint32_t _height) : m_width(_width), m_height(_height), m_usedSpace(0)
{
	// We want a one pixel border around the whole atlas to avoid any artefact when
	// sampling texture
	m_skyline.push_back(Node(1, 1, _width - 2) );
}

void RectanglePacker::init(uint32_t _width, uint32_t _height)
{
	BX_CHECK(_width > 2, "_width must be > 2");
	BX_CHECK(_height > 2, "_height must be > 2");
	m_width = _width;
	m_height = _height;
	m_usedSpace = 0;

	m_skyline.clear();
	// We want a one pixel border around the whole atlas to avoid any artifact when
	// sampling texture
	m_skyline.push_back(Node(1, 1, _width - 2) );
}

bool RectanglePacker::addRectangle(uint16_t _width, uint16_t _height, uint16_t& _outX, uint16_t& _outY)
{
	int y, best_height, best_index;
	int32_t best_width;
	Node* node;
	Node* prev;
	_outX = 0;
	_outY = 0;

	uint32_t ii;

	best_height = INT_MAX;
	best_index = -1;
	best_width = INT_MAX;
	for (ii = 0; ii < m_skyline.size(); ++ii)
	{
		y = fit(ii, _width, _height);
		if (y >= 0)
		{
			node = &m_skyline[ii];
			if ( ( (y + _height) < best_height)
			   || ( ( (y + _height) == best_height)
			      && (node->m_width < best_width) ) )
			{
				best_height = y + _height;
				best_index = ii;
				best_width = node->m_width;
				_outX = node->m_x;
				_outY = y;
			}
		}
	}

	if (best_index == -1)
	{
		return false;
	}

	Node newNode(_outX, _outY + _height, _width);
	m_skyline.insert(m_skyline.begin() + best_index, newNode);

	for (ii = best_index + 1; ii < m_skyline.size(); ++ii)
	{
		node = &m_skyline[ii];
		prev = &m_skyline[ii - 1];
		if (node->m_x < (prev->m_x + prev->m_width) )
		{
			int shrink = prev->m_x + prev->m_width - node->m_x;
			node->m_x += shrink;
			node->m_width -= shrink;
			if (node->m_width <= 0)
			{
				m_skyline.erase(m_skyline.begin() + ii);
				--ii;
			}
			else
			{
				break;
			}
		}
		else
		{
			break;
		}
	}

	merge();
	m_usedSpace += _width * _height;
	return true;
}

float RectanglePacker::getUsageRatio()
{
	uint32_t total = m_width * m_height;
	if (total > 0)
	{
		return (float) m_usedSpace / (float) total;
	}
	else
	{
		return 0.0f;
	}
}

void RectanglePacker::clear()
{
	m_skyline.clear();
	m_usedSpace = 0;

	// We want a one pixel border around the whole atlas to avoid any artefact when
	// sampling texture
	m_skyline.push_back(Node(1, 1, m_width - 2) );
}

int32_t RectanglePacker::fit(uint32_t _skylineNodeIndex, uint16_t _width, uint16_t _height)
{
	int32_t width = _width;
	int32_t height = _height;

	const Node& baseNode = m_skyline[_skylineNodeIndex];

	int32_t x = baseNode.m_x, y;
	int32_t _width_left = width;
	int32_t i = _skylineNodeIndex;

	if ( (x + width) > (int32_t)(m_width - 1) )
	{
		return -1;
	}

	y = baseNode.m_y;
	while (_width_left > 0)
	{
		const Node& node = m_skyline[i];
		if (node.m_y > y)
		{
			y = node.m_y;
		}

		if ( (y + height) > (int32_t)(m_height - 1) )
		{
			return -1;
		}

		_width_left -= node.m_width;
		++i;
	}

	return y;
}

void RectanglePacker::merge()
{
	Node* node;
	Node* next;
	uint32_t ii;

	for (ii = 0; ii < m_skyline.size() - 1; ++ii)
	{
		node = (Node*) &m_skyline[ii];
		next = (Node*) &m_skyline[ii + 1];
		if (node->m_y == next->m_y)
		{
			node->m_width += next->m_width;
			m_skyline.erase(m_skyline.begin() + ii + 1);
			--ii;
		}
	}
}

//********** Cube Atlas implementation ************

struct Atlas::PackedLayer
{
	RectanglePacker packer;
	AtlasRegion faceRegion;
};

Atlas::Atlas(uint16_t _textureSize, uint16_t _maxRegionsCount)
{
	BX_CHECK(_textureSize >= 64
	        && _textureSize <= 4096, "suspicious texture size");
	BX_CHECK(_maxRegionsCount >= 64
	        && _maxRegionsCount <= 32000, "suspicious _regions count");
	m_layers = new PackedLayer[24];
	for (int ii = 0; ii < 24; ++ii)
	{
		m_layers[ii].packer.init(_textureSize, _textureSize);
	}

	m_usedLayers = 0;
	m_usedFaces = 0;

	m_textureSize = _textureSize;
	m_regionCount = 0;
	m_maxRegionCount = _maxRegionsCount;
	m_regions = new AtlasRegion[_maxRegionsCount];
	m_textureBuffer = new uint8_t[ _textureSize * _textureSize * 6 * 4 ];
	memset(m_textureBuffer, 0, _textureSize * _textureSize * 6 * 4);
	//BGFX_TEXTURE_MIN_POINT|BGFX_TEXTURE_MAG_POINT|BGFX_TEXTURE_MIP_POINT;
	//BGFX_TEXTURE_MIN_ANISOTROPIC|BGFX_TEXTURE_MAG_ANISOTROPIC|BGFX_TEXTURE_MIP_POINT
	//BGFX_TEXTURE_U_CLAMP|BGFX_TEXTURE_V_CLAMP
	uint32_t flags = 0; // BGFX_TEXTURE_MIN_ANISOTROPIC|BGFX_TEXTURE_MAG_ANISOTROPIC|BGFX_TEXTURE_MIP_POINT;

	//Uncomment this to debug atlas
	//const bgfx::Memory* mem = bgfx::alloc(textureSize*textureSize * 6 * 4);
	//memset(mem->data, 255, mem->size);
	const bgfx::Memory* mem = NULL;
	m_textureHandle = bgfx::createTextureCube(6
	                                         , _textureSize
	                                         , 1
	                                         , bgfx::TextureFormat::BGRA8
	                                         , flags
	                                         , mem
	                                         );
}

Atlas::Atlas(uint16_t _textureSize, const uint8_t* _textureBuffer, uint16_t _regionCount, const uint8_t* _regionBuffer, uint16_t _maxRegionsCount)
{
	BX_CHECK(_regionCount <= 64
	        && _maxRegionsCount <= 4096, "suspicious initialization");
	//layers are frozen
	m_usedLayers = 24;
	m_usedFaces = 6;

	m_textureSize = _textureSize;
	m_regionCount = _regionCount;
	//regions are frozen
	if (_regionCount < _maxRegionsCount)
	{
		m_maxRegionCount = _regionCount;
	}
	else
	{
		m_maxRegionCount = _maxRegionsCount;
	}

	m_regions = new AtlasRegion[_regionCount];
	m_textureBuffer = new uint8_t[getTextureBufferSize()];

	//BGFX_TEXTURE_MIN_POINT|BGFX_TEXTURE_MAG_POINT|BGFX_TEXTURE_MIP_POINT;
	//BGFX_TEXTURE_MIN_ANISOTROPIC|BGFX_TEXTURE_MAG_ANISOTROPIC|BGFX_TEXTURE_MIP_POINT
	//BGFX_TEXTURE_U_CLAMP|BGFX_TEXTURE_V_CLAMP
	uint32_t flags = 0; //BGFX_TEXTURE_MIN_ANISOTROPIC|BGFX_TEXTURE_MAG_ANISOTROPIC|BGFX_TEXTURE_MIP_POINT;
	memcpy(m_regions, _regionBuffer, _regionCount * sizeof(AtlasRegion) );
	memcpy(m_textureBuffer, _textureBuffer, getTextureBufferSize() );

	m_textureHandle = bgfx::createTextureCube(6
	                                         , _textureSize
	                                         , 1
	                                         , bgfx::TextureFormat::BGRA8
	                                         , flags
	                                         , bgfx::makeRef(m_textureBuffer, getTextureBufferSize() )
	                                         );
}

Atlas::~Atlas()
{
	delete[] m_layers;
	delete[] m_regions;
	delete[] m_textureBuffer;
}

uint16_t Atlas::addRegion(uint16_t _width, uint16_t _height, const uint8_t* _bitmapBuffer, AtlasRegion::Type _type, uint16_t outline)
{
	if (m_regionCount >= m_maxRegionCount)
	{
		return UINT16_MAX;
	}

	uint16_t x = 0, y = 0;
	// We want each bitmap to be separated by at least one black pixel
	// TODO manage mipmaps
	uint32_t idx = 0;
	while (idx < m_usedLayers)
	{
		if (m_layers[idx].faceRegion.getType() == _type)
		{
			if (m_layers[idx].packer.addRectangle(_width + 1, _height + 1, x, y) )
			{
				break;
			}
		}

		idx++;
	}

	if (idx >= m_usedLayers)
	{
		//do we have still room to add layers ?
		if ( (idx + _type) > 24
		   || m_usedFaces >= 6)
		{
			return UINT16_MAX;
		}

		//create new layers
		for (int ii = 0; ii < _type; ++ii)
		{
			m_layers[idx + ii].faceRegion.setMask(_type, m_usedFaces, ii);
		}

		m_usedLayers += _type;
		m_usedFaces++;

		//add it to the created layer
		if (!m_layers[idx].packer.addRectangle(_width + 1, _height + 1, x, y) )
		{
			return UINT16_MAX;
		}
	}

	AtlasRegion& region = m_regions[m_regionCount];
	region.m_x = x;
	region.m_y = y;
	region.m_width = _width;
	region.m_height = _height;
	region.m_mask = m_layers[idx].faceRegion.m_mask;

	updateRegion(region, _bitmapBuffer);

	region.m_x += outline;
	region.m_y += outline;
	region.m_width -= (outline * 2);
	region.m_height -= (outline * 2);

	return m_regionCount++;
}

void Atlas::updateRegion(const AtlasRegion& _region, const uint8_t* _bitmapBuffer)
{
	const bgfx::Memory* mem = bgfx::alloc(_region.m_width * _region.m_height * 4);
	//BAD!
	memset(mem->data, 0, mem->size);
	if (_region.getType() == AtlasRegion::TYPE_BGRA8)
	{
		const uint8_t* inLineBuffer = _bitmapBuffer;
		uint8_t* outLineBuffer = m_textureBuffer + _region.getFaceIndex() * (m_textureSize * m_textureSize * 4) + ( ( (_region.m_y * m_textureSize) + _region.m_x) * 4);

		//update the cpu buffer
		for (int yy = 0; yy < _region.m_height; ++yy)
		{
			memcpy(outLineBuffer, inLineBuffer, _region.m_width * 4);
			inLineBuffer += _region.m_width * 4;
			outLineBuffer += m_textureSize * 4;
		}

		//update the GPU buffer
		memcpy(mem->data, _bitmapBuffer, mem->size);
	}
	else
	{
		uint32_t layer = _region.getComponentIndex();
		//uint32_t face = _region.getFaceIndex();
		const uint8_t* inLineBuffer = _bitmapBuffer;
		uint8_t* outLineBuffer = (m_textureBuffer + _region.getFaceIndex() * (m_textureSize * m_textureSize * 4) + ( ( (_region.m_y * m_textureSize) + _region.m_x) * 4) );

		//update the cpu buffer
		for (int yy = 0; yy < _region.m_height; ++yy)
		{
			for (int xx = 0; xx < _region.m_width; ++xx)
			{
				outLineBuffer[(xx * 4) + layer] = inLineBuffer[xx];
			}

			//update the GPU buffer
			memcpy(mem->data + yy * _region.m_width * 4, outLineBuffer, _region.m_width * 4);
			inLineBuffer += _region.m_width;
			outLineBuffer += m_textureSize * 4;
		}
	}

	bgfx::updateTextureCube(m_textureHandle, (uint8_t)_region.getFaceIndex(), 0, _region.m_x, _region.m_y, _region.m_width, _region.m_height, mem);
}

void Atlas::packFaceLayerUV(uint32_t _idx, uint8_t* _vertexBuffer, uint32_t _offset, uint32_t _stride)
{
	packUV(m_layers[_idx].faceRegion, _vertexBuffer, _offset, _stride);
}

void Atlas::packUV(uint16_t handle, uint8_t* _vertexBuffer, uint32_t _offset, uint32_t _stride)
{
	const AtlasRegion& region = m_regions[handle];
	packUV(region, _vertexBuffer, _offset, _stride);
}

void Atlas::packUV(const AtlasRegion& _region, uint8_t* _vertexBuffer, uint32_t _offset, uint32_t _stride)
{
	float texMult = 65535.0f / ( (float)(m_textureSize) );
	static const int16_t minVal = -32768;
	static const int16_t maxVal = 32767;

	int16_t x0 = (int16_t)(_region.m_x * texMult) - 32768;
	int16_t y0 = (int16_t)(_region.m_y * texMult) - 32768;
	int16_t x1 = (int16_t)( (_region.m_x + _region.m_width) * texMult) - 32768;
	int16_t y1 = (int16_t)( (_region.m_y + _region.m_height) * texMult) - 32768;
	int16_t w = (int16_t) ( (32767.0f / 4.0f) * _region.getComponentIndex() );

	_vertexBuffer += _offset;
	switch (_region.getFaceIndex() )
	{
	case 0: // +X
		x0 = -x0;
		x1 = -x1;
		y0 = -y0;
		y1 = -y1;
		writeUV(_vertexBuffer, maxVal, y0, x0, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, maxVal, y1, x0, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, maxVal, y1, x1, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, maxVal, y0, x1, w); _vertexBuffer += _stride;
		break;

	case 1: // -X
		y0 = -y0;
		y1 = -y1;
		writeUV(_vertexBuffer, minVal, y0, x0, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, minVal, y1, x0, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, minVal, y1, x1, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, minVal, y0, x1, w); _vertexBuffer += _stride;
		break;

	case 2: // +Y
		writeUV(_vertexBuffer, x0, maxVal, y0, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x0, maxVal, y1, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, maxVal, y1, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, maxVal, y0, w); _vertexBuffer += _stride;
		break;

	case 3: // -Y
		y0 = -y0;
		y1 = -y1;
		writeUV(_vertexBuffer, x0, minVal, y0, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x0, minVal, y1, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, minVal, y1, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, minVal, y0, w); _vertexBuffer += _stride;
		break;

	case 4: // +Z
		y0 = -y0;
		y1 = -y1;
		writeUV(_vertexBuffer, x0, y0, maxVal, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x0, y1, maxVal, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, y1, maxVal, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, y0, maxVal, w); _vertexBuffer += _stride;
		break;

	case 5: // -Z
		x0 = -x0;
		x1 = -x1;
		y0 = -y0;
		y1 = -y1;
		writeUV(_vertexBuffer, x0, y0, minVal, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x0, y1, minVal, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, y1, minVal, w); _vertexBuffer += _stride;
		writeUV(_vertexBuffer, x1, y0, minVal, w); _vertexBuffer += _stride;
		break;
	}
}