diff --git a/MacGLide/MacGLide b/MacGLide/MacGLide
index 5f17549..6751e1d 100644
Binary files a/MacGLide/MacGLide and b/MacGLide/MacGLide differ
diff --git a/MacGLide/OpenGLide/Framebuffer.cpp b/MacGLide/OpenGLide/Framebuffer.cpp
index 19cf29a..849267b 100644
--- a/MacGLide/OpenGLide/Framebuffer.cpp
+++ b/MacGLide/OpenGLide/Framebuffer.cpp
@@ -1 +1 @@
-//**************************************************************
//*            OpenGLide - Glide to OpenGL Wrapper
//*             http://openglide.sourceforge.net
//*
//*                  framebuffer emulation
//*
//*         OpenGLide is OpenSource under LGPL license
//*  Mac version and additional features by Jens-Olaf Hemprich
//**************************************************************

#include "Framebuffer.h"
#include "Glide.h"
#include "GlideApplication.h"
#include "GlideSettings.h"
#include "GLRender.h"
#include "GLRenderUpdateState.h"
#include "GLColorAlphaCombineEnvTables.h"

// check if tile needs to be displayed
#define CHECK_RENDER_TILE

// Display small dots at opposite corners of rendered framebuffer tiles
//#define DEBUG_TILE_RENDERING

Framebuffer::Framebuffer()
: m_x_step_start(0)
, m_y_step_start(0)
, m_x_step_start_opaque(0)
, m_y_step_start_opaque(0)
, m_width(0)
, m_height(0)
, m_framebuffer(NULL)
, m_texbuffer(NULL)
, m_origin(GR_ORIGIN_UPPER_LEFT)
, m_glInternalFormat(-1)
, m_glFormat(-1)
, m_glType(-1)
, m_glDepth(1.0f)
, m_format_valid(false)
, m_use_client_storage(false)
, m_useRectangleARB(false)
, m_must_clear_buffer(true)
, m_custom_tilesizes(NULL)
{
	// Don't set the checksum to 0, as this would cause white screen in Carmageddon
	// because fully black tile also have a 0 checksum and no texture data would be
	// download at all (and the tile would be rendered as if TEXTURE_2D was disabled) 
	memset(m_tileChecksums, 0xff, sizeof(vector unsigned long) * m_tileCount);
}

Framebuffer::~Framebuffer()
{
}

bool Framebuffer::initialise_buffers(BufferStruct* framebuffer, BufferStruct* texbuffer, FxU32 width, FxU32 height, const tilesize* tilesizetable)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "GlideFrameBuffer::initialise_buffers(---, ---, %d, %d, ---)\n", width, height);
	#endif

	m_custom_tilesizes = tilesizetable;
	return initialise_buffers(framebuffer, texbuffer, width, height, 0, 0);
}

bool Framebuffer::initialise_buffers(BufferStruct* framebuffer, BufferStruct* texbuffer, FxU32 width, FxU32 height, FxU32 x_tile, FxU32 y_tile)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::initialise_buffers(---, ---, %d, %d, %d, %d)\n", width, height, x_tile, y_tile);
	#endif

	m_framebuffer = framebuffer;
	m_texbuffer = texbuffer;
	m_framebuffer->WriteMode = m_texbuffer->WriteMode = GR_LFBWRITEMODE_UNUSED;
	m_width = width;
	m_height = height;
	// find out largest texture size
	GLint tile_size;
	glGetIntegerv(GL_MAX_TEXTURE_SIZE, &tile_size);
	m_x_step_start_opaque = tile_size;
	m_y_step_start_opaque = tile_size;
	m_x_step_start = min(tile_size, x_tile);
	m_y_step_start = min(tile_size, y_tile);
	m_x_step_start = max(16, m_x_step_start);
	m_y_step_start = max(16, m_y_step_start);
	// Disabled because this is a global setting now
	// m_use_client_storage = InternalConfig.EXT_Client_Storage;
	m_useRectangleARB = InternalConfig.ARB_texture_rectangle &&
	                    InternalConfig.EXT_compiled_vertex_array;
	// The texture priority is set to minimun because
	// frame buffer textures are never used a second time
	// @todo: This is not true anymore in all cases
	// because of the altivec checksum feature
	const GLfloat zero = 0.0f;
	glGenTextures(m_tileCount, &m_textureNames[0]);
	glPrioritizeTextures(m_tileCount, &m_textureNames[0], &zero);
	for(int i = 0; i < m_tileCount; i++)
	{
		const GLenum textureTarget = m_useRectangleARB ? GL_TEXTURE_RECTANGLE_ARB : GL_TEXTURE_2D;
		glBindTexture(textureTarget, m_textureNames[i]);
		if (m_useRectangleARB)
		{
			// This will not probably not work right now, because
			// we're using RGBA (the OS9 prefered texture format)
			// @todo: use ABGR??? in order to avoid byte swizzling
			// glTextureRangeAPPLE(GLenum target, GLsizei length, GLvoid *pointer);
			glTexParameteri(textureTarget,
			                GL_TEXTURE_STORAGE_HINT_APPLE,
			                GL_STORAGE_CACHED_APPLE);
			glReportError();
		}
		glTexParameteri(textureTarget, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
		glTexParameteri(textureTarget, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
		glTexParameteri(textureTarget, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
		glTexParameteri(textureTarget, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
	}
	// If a game has its own tilesize table, use
	// the largest tiles for opaque renderings
	GLint y_step = y_tile == 0 ? m_y_step_start_opaque : m_y_step_start;
	// init default/opaque tilesize table
	int w = 0;
	for(FxU32 y = 0; y < m_height && w < MaxTiles ; y += y_step, w++)
	{
		while (m_height - y < y_step)
		{
			y_step = y_step >> 1;
		}
		m_tilesizes[w].y = y_step;
		GLint x_step = x_tile == 0 ? m_x_step_start_opaque : m_x_step_start;
		int v = 0;
		for(FxU32 x = 0; x < m_width && v < MaxTiles; x += x_step, v++ )
		{
			while (m_width - x < x_step)
			{
				x_step = x_step >> 1;
			}
			m_tilesizes[w].x[v] = x_step;
		}
	}
	// Build compiled vertex arrays
	if (InternalConfig.EXT_compiled_vertex_array)
	{
		// Store various render buffers indices 
		m_tilesizesVertexArrayIndex = OGLRender.FrameBufferStartIndex;
		m_tilesizesCount = buildVertexArrays(&m_tilesizes[0], m_tilesizesVertexArrayIndex);
		if (m_custom_tilesizes)
		{
			m_customtilesizesVertexArrayIndex = m_tilesizesVertexArrayIndex + m_tilesizesCount * 2;
			m_customtilesizesCount = buildVertexArrays(m_custom_tilesizes, m_customtilesizesVertexArrayIndex);
		}
	}
	return m_width > 0 && m_height > 0 && m_x_step_start > 0 && m_y_step_start > 0 && m_format_valid;
}

void Framebuffer::free_buffers()
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::free_buffers()\n");
	#endif

	if (m_tilesizes) FreeObject(m_tilesizes);
	glDeleteTextures(m_tileCount, &m_textureNames[0]);
}

void Framebuffer::initialise_format(GrLfbWriteMode_t writemode)
{
	#if defined(OGL_PART_DONE) || defined(OGL_FRAMEBUFFER)
		GlideMsg("Framebuffer::initialise_format(0x%x)\n", writemode);
	#endif

	// Enlarge buffer?
	if (writemode >= GR_LFBWRITEMODE_888 &&
	    (m_framebuffer->WriteMode < GR_LFBWRITEMODE_888 || m_framebuffer->WriteMode == GR_LFBWRITEMODE_UNUSED) &&
	    m_framebuffer->Address)
	{
		// Delete existing buffer
		FreeFrameBuffer(m_framebuffer->Address);
		m_framebuffer->Address = NULL;
		m_texbuffer->Address = NULL;
	}
	// Allocate 32-bit buffer (16bit buffer has been allocated in grSstWinOpen()
	if (m_framebuffer->Address == NULL)
	{
		unsigned long openglpixels = OpenGL.WindowWidth * OpenGL.WindowHeight;
		// Framebuffer can be written to with 16bit or 32bit data
		unsigned long buffertypesize = (writemode >= GR_LFBWRITEMODE_888) ? sizeof(FxU32) : sizeof(FxU16);
		Glide.FrameBuffer.Address = (FxU16*) AllocFrameBuffer(Glide.WindowTotalPixels * buffertypesize + openglpixels * sizeof(FxU32), 1);
		Glide.TempBuffer.Address = &Glide.FrameBuffer.Address[Glide.WindowTotalPixels * buffertypesize >> 1];
		memset( Glide.FrameBuffer.Address, 0, Glide.WindowTotalPixels * buffertypesize);
		memset( Glide.TempBuffer.Address, 0, openglpixels * sizeof(FxU32));
	}
	m_framebuffer->WriteMode = writemode;
	m_glInternalFormat = 4;
	m_glFormat = GL_RGBA;
	m_glType = GL_UNSIGNED_BYTE;
	FxU16 chromakeyvalue;
	switch (writemode)
	{
	case GR_LFBWRITEMODE_565:
		chromakeyvalue = 	s_GlideApplication.GetType() == GlideApplication::Carmageddon
		                  ? 0x1f1f : 0x07ff;
		m_format_valid = true;
		break;
	case GR_LFBWRITEMODE_1555:
		chromakeyvalue = 0x03ff;
		m_format_valid = true;
		break;
	case GR_LFBWRITEMODE_888:
		chromakeyvalue = 0x7ffdfeff;
		m_format_valid = true;
		break;
	default:
		chromakeyvalue = 0x0;
		m_format_valid = false;
		break;
	}
	// When the chromakeyvalue changes, the buffer has to be cleared
	if (chromakeyvalue != m_ChromaKey.Scalar)
	{
		SetChromaKeyValue(chromakeyvalue);
		m_must_clear_buffer = true;
	}
}

bool Framebuffer::begin_write()
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg("Framebuffer::begin_write()\n");
#endif

	if (m_must_clear_buffer)
	{
		Clear();
		m_must_clear_buffer = false;
	}
	return true;
}

void Framebuffer::Clear()
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::Clear()\n");
	#endif

	const FxU16 chromakey = GetChromaKeyValue();
	const FxU32 count = m_width * m_height ;
	FxU16* framebuffer = m_framebuffer->Address;
	for ( int i = 0; i < count; i++)
	{
		framebuffer[i] = chromakey;
	}
}

bool Framebuffer::end_write(FxU32 alpha, GLfloat depth, bool pixelpipeline)
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg("Framebuffer::end_write(%d, %f, %d)\n", alpha, depth, pixelpipeline);
#endif

	m_glDepth = depth;
#ifdef __ALTIVEC__
	for(int i = 0; i < 4; i++)
	{
		(&m_glAlpha.Scalar)[i] = alpha;
	}
#else
	m_glAlpha.Scalar = alpha;
#endif
	// if all pixels are invisible, nothing must be rendered.
	// The pixel conversion functions assume alpha is != 0 in order
	// to determine if a tile contains any pixels to be rendered.
	if (m_glAlpha.Scalar == 0) return false;
	set_gl_state(pixelpipeline);
	if (InternalConfig.EXT_compiled_vertex_array)
	{
		if (m_custom_tilesizes)
		{
			drawCompiledVertexArrays(m_custom_tilesizes, m_customtilesizesVertexArrayIndex, m_customtilesizesCount, pixelpipeline);
		}
		else
		{
			drawCompiledVertexArrays(m_tilesizes, m_tilesizesVertexArrayIndex, m_tilesizesCount, pixelpipeline);
		}
	}
	else
	{
		const tilesize* tilesizes = m_custom_tilesizes ? m_custom_tilesizes : m_tilesizes;
		draw(tilesizes, pixelpipeline);
	}
	restore_gl_state(pixelpipeline);
	return true;
}

bool Framebuffer::end_write(FxU32 alpha)
{
#ifdef OGL_DONE
	GlideMsg("Framebuffer::end_write(%d)\n", alpha);
#endif

	// draw frame buffer
	// @todo: Depth should OpenGL.ZNear, but that breaks overlays in Myth
	FxBool result = end_write(alpha, 0.0, false);
	return result;
}

bool Framebuffer::end_write()
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg( "Framebuffer::end_write( )\n" );
#endif

	return end_write(0x000000ff);
}

bool Framebuffer::end_write_opaque()
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg("Framebuffer::end_write_opaque()\n");
#endif

	// @todo: Depth should OpenGL.ZNear, but that breaks overlays in Myth
	return end_write(0x000000ff, 0.0, false);
}

void Framebuffer::set_gl_state(bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::set_gl_state(%d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::set_gl_state");
	VERIFY_ACTIVE_TEXTURE_UNIT(OpenGL.ColorAlphaUnit1);
	// Disable the cull mode
	glDisable(GL_CULL_FACE);
	// Disable clip volume hint manually to avoid recursion
	if (InternalConfig.EXT_clip_volume_hint && OpenGL.ClipVerticesEnabledState)
	{
		glHint(GL_CLIP_VOLUME_CLIPPING_HINT_EXT, GL_FASTEST);
	}
	
	if (pixelpipeline)
	{
		if (OpenGL.ColorAlphaUnit2)
		{
			// Pixelpipeline support for env cobine based rendering:
			// Framebuffer pixels must be routed through the coloralpha unit
			// as if they were produced by the vertex iterators without an 
			// additional GL texture unit -> source must be changed accordingly
			m_bRestoreColorCombine = false;
			if (Glide.State.ColorCombineLocal == GR_COMBINE_LOCAL_ITERATED)
			{
				Glide.State.ColorCombineLocal = GR_COMBINE_LOCAL_PIXELPIPELINE;
				m_bRestoreColorCombine = true;
			}
			if (Glide.State.ColorCombineOther == GR_COMBINE_OTHER_ITERATED)
			{
				Glide.State.ColorCombineOther = GR_COMBINE_OTHER_PIXELPIPELINE;
				m_bRestoreColorCombine = true;
			}
			if (m_bRestoreColorCombine)	SetColorCombineState();
			m_bRestoreAlphaCombine = false;
			if (Glide.State.AlphaLocal == GR_COMBINE_LOCAL_ITERATED)
			{
				Glide.State.AlphaLocal = GR_COMBINE_LOCAL_PIXELPIPELINE;
				m_bRestoreAlphaCombine = true;
			}
			if (Glide.State.AlphaOther == GR_COMBINE_OTHER_ITERATED)
			{
				Glide.State.AlphaOther = GR_COMBINE_OTHER_PIXELPIPELINE;
				m_bRestoreAlphaCombine = true;
			}
			if (m_bRestoreAlphaCombine)	SetAlphaCombineState();
			// Update the opengl state for the pixel pipeline
			RenderUpdateState();
			// If the write mode doesn't provide alpha then m_glAlpha is used
			// as the constant alpha value, and we can use the alpha test
			// to mask out chromakey pixels
			switch (m_framebuffer->WriteMode)
			{
			case GR_LFBWRITEMODE_565:
			case GR_LFBWRITEMODE_888:
				glEnable(GL_ALPHA_TEST);
				const GLenum alphaTestFunction = GL_EQUAL;
				const GLfloat alphaTestReferenceValue= m_glAlpha.Scalar * D1OVER255;
				OpenGL.AlphaTestFunction = alphaTestFunction;
				OpenGL.AlphaReferenceValue = alphaTestReferenceValue;
				glAlphaFunc(alphaTestFunction, alphaTestReferenceValue);
				glReportError();
				break;
			}
		}
		// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D
		if (m_useRectangleARB)
		{
			const GLenum textureTarget = GL_TEXTURE_RECTANGLE_ARB;
			const	bool enableColoralphaTextureUnit1 = OpenGL.ColorAlphaUnitColorEnabledState[0] || OpenGL.ColorAlphaUnitAlphaEnabledState[0];
			if (enableColoralphaTextureUnit1)
			{
				glEnable(textureTarget);
				glReportError();
			}
			if (OpenGL.ColorAlphaUnit2)
			{
				const	bool enableColoralphaTextureUnit2 = OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1];
				if (enableColoralphaTextureUnit2)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glReportError();
					glEnable(textureTarget);
					glReportError();
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
					glReportError();
				}
			}
			// The client texture state is already setup correctly since we just
			// have to adjust the texture rectangle state
			// (the texture_2d state is not changed when the pixelpipeline mode is active)
		}
		// Set the origin with clipping
		glMatrixMode(GL_PROJECTION);
		glLoadIdentity();
		if (m_origin == GR_ORIGIN_LOWER_LEFT)
		{
			glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
			        Glide.State.ClipMinY, Glide.State.ClipMaxY, 
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
			           OpenGL.OriginY + OpenGL.ClipMinY, 
			           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
			           OpenGL.ClipMaxY - OpenGL.ClipMinY); 
		}
		else
		{
			glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
			        Glide.State.ClipMaxY, Glide.State.ClipMinY, 
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
			           OpenGL.OriginY + OpenGL.WindowHeight - OpenGL.ClipMaxY, 
			           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
			           OpenGL.ClipMaxY - OpenGL.ClipMinY); 
		}
		// The scissor rectangle is not reset, because scissor mode
		// is only enabled when clearing the buffer
		glMatrixMode(GL_MODELVIEW);
		glReportError();
	}
	else
	{
		// disable blend
		glDisable(GL_BLEND);
		// disable depth buffer
		glDepthMask(false);
		// Enable colormask
		glColorMask( true, true, true, false);
		// Needed for displaying in-game menus
		if (Glide.State.DepthBufferMode != GR_DEPTHBUFFER_DISABLE)
		{
			glDisable(GL_DEPTH_TEST);
		}
		glEnable(GL_ALPHA_TEST);
		// Update state as we're calling update triggers on restore
		const GLenum alphaTestFunction = GL_GREATER;
		const GLfloat alphaTestReferenceValue= 0.0;
		OpenGL.AlphaTestFunction = alphaTestFunction;
		OpenGL.AlphaReferenceValue = alphaTestReferenceValue;
		glAlphaFunc(alphaTestFunction, alphaTestReferenceValue);
		glReportError();
		if (InternalConfig.EXT_secondary_color)
		{
			glDisable(GL_COLOR_SUM_EXT);
			glReportError();
		}
		// Reset the clipping window
		// and set the origin
		glMatrixMode(GL_PROJECTION);
		glLoadIdentity();
		if (m_origin == GR_ORIGIN_LOWER_LEFT)
		{
			glOrtho(0, Glide.WindowWidth,
			        0, Glide.WindowHeight,
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX,
			           OpenGL.OriginY,
			           OpenGL.WindowWidth,
			           OpenGL.WindowHeight); 
		}
		else
		{
			glOrtho(0, Glide.WindowWidth,
			        Glide.WindowHeight, 0,
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX,
			           OpenGL.OriginY,
			           OpenGL.WindowWidth,
			           OpenGL.WindowHeight); 
		}
		// The scissor rectangle is not changed, because scissor mode
		// is only enabled when clearing the buffer
		glMatrixMode(GL_MODELVIEW);
		glReportError();
		
		// Disable fog
		bool disable_fog_texture_unit = OpenGL.FogTextureUnit;
		if (disable_fog_texture_unit)
		{
			glActiveTextureARB(OpenGL.FogTextureUnit);
			if (InternalConfig.EXT_compiled_vertex_array)
			{
				glClientActiveTextureARB(OpenGL.FogTextureUnit);
				glDisableClientState(GL_TEXTURE_COORD_ARRAY);
				glTexCoordPointer(4, GL_FLOAT, 0, NULL);
			}
			glDisable(GL_TEXTURE_2D);
		}
		if (OpenGL.Fog &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_None &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_EnvCombine)
		{
			glDisable(GL_FOG);
		}
		glReportError();
		
		// enable framebuffer texture unit
		if (OpenGL.ColorAlphaUnit2)
		{
			bool disable_coloralpha_texture_unit_2 = OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1];
			if (disable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit2);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
					// On MacOS9 (Classic?) the texcoord pointer needs to be reset
					// to the default value when glLockArrays/glUnlockArrays is used
					glTexCoordPointer( 4, GL_FLOAT, 0, NULL );
				}
				glDisable(GL_TEXTURE_2D);
			}
			if (disable_fog_texture_unit || disable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (!OpenGL.ColorAlphaUnitColorEnabledState[0] && !OpenGL.ColorAlphaUnitAlphaEnabledState[0])
			{
				glEnable(GL_TEXTURE_2D);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glEnableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer(4, GL_FLOAT, 0, &OGLRender.TTexture[0]);
				}
			}
		}
		else
		{
			if (disable_fog_texture_unit)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (OpenGL.Texture == false)
			{
				// GL_RECTANGLE_ARB overrides GL_RECTANGLE_2D
				if (!m_useRectangleARB) glEnable(GL_TEXTURE_2D);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glEnableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer(4, GL_FLOAT, 0, &OGLRender.TTexture[0]);
				}
			}
		}
		if (m_useRectangleARB)
		{
			glEnable(GL_TEXTURE_RECTANGLE_ARB);
			glReportError();
		}
		glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);
		glReportError();
	}
}

void Framebuffer::restore_gl_state(bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::restore_gl_state(%d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::restore_gl_state");
	// Restore the cull mode
	switch (Glide.State.CullMode)
	{
	case GR_CULL_DISABLE:
		break;
	case GR_CULL_NEGATIVE:
	case GR_CULL_POSITIVE:
		glEnable(GL_CULL_FACE);
		break;
	}
	if (InternalConfig.EXT_clip_volume_hint && OpenGL.ClipVerticesEnabledState)
	{
		glHint(GL_CLIP_VOLUME_CLIPPING_HINT_EXT, GL_NICEST);
	}
	
	// Restore the clipping window
	glMatrixMode(GL_PROJECTION);
	glLoadIdentity();
	if ( Glide.State.OriginInformation == GR_ORIGIN_LOWER_LEFT )
	{
		glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
							Glide.State.ClipMinY, Glide.State.ClipMaxY, 
							OpenGL.ZNear, OpenGL.ZFar);
		glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
		           OpenGL.OriginY + OpenGL.ClipMinY, 
		           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
		           OpenGL.ClipMaxY - OpenGL.ClipMinY); 
	}
	else
	{
		glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
							Glide.State.ClipMaxY, Glide.State.ClipMinY, 
							OpenGL.ZNear, OpenGL.ZFar);
		glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
		           OpenGL.OriginY + OpenGL.WindowHeight - OpenGL.ClipMaxY, 
		           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
		           OpenGL.ClipMaxY - OpenGL.ClipMinY);
	}
	// The scissor rectangle is not reset, because scissor mode
	// is only enabled when clearing the buffer
	glMatrixMode( GL_MODELVIEW );
	glReportError();
	
	if (pixelpipeline)
	{
		if (OpenGL.ColorAlphaUnit2)
		{
			// restore current values
			if (m_bRestoreColorCombine)
			{
				if (Glide.State.ColorCombineLocal == GR_COMBINE_LOCAL_PIXELPIPELINE)	Glide.State.ColorCombineLocal = GR_COMBINE_LOCAL_ITERATED;
				if (Glide.State.ColorCombineOther == GR_COMBINE_OTHER_PIXELPIPELINE)	Glide.State.ColorCombineOther = GR_COMBINE_OTHER_ITERATED;
				SetColorCombineState();
			}
			if(m_bRestoreAlphaCombine)
			{
				if (Glide.State.AlphaLocal == GR_COMBINE_LOCAL_PIXELPIPELINE)	Glide.State.AlphaLocal = GR_COMBINE_LOCAL_ITERATED;
				if (Glide.State.AlphaOther == GR_COMBINE_OTHER_PIXELPIPELINE)	Glide.State.AlphaOther = GR_COMBINE_LOCAL_ITERATED;
				SetAlphaCombineState();
			}
		}
		switch (m_framebuffer->WriteMode)
		{
		case GR_LFBWRITEMODE_565:
		case GR_LFBWRITEMODE_888:
			SetChromaKeyAndAlphaState();
			break;
		}
		// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D
		if (m_useRectangleARB)
		{
			const GLenum textureTarget = GL_TEXTURE_RECTANGLE_ARB;
			const	bool disableColoralphaTextureUnit1 = !(OpenGL.ColorAlphaUnitColorEnabledState[0] || OpenGL.ColorAlphaUnitAlphaEnabledState[0]);
			if (disableColoralphaTextureUnit1)
			{
				glDisable(textureTarget);
				glReportError();
			}
			if (OpenGL.ColorAlphaUnit2)
			{
				const	bool disableColoralphaTextureUnit2 = !(OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1]);
				if (disableColoralphaTextureUnit2)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glReportError();
					glDisable(textureTarget);
					glReportError();
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
					glReportError();
				}
			}
			// The client texture state is already setup correctly since we just
			// have to adjust the texture rectangle state to the texture_2d state
		}
	}
	else
	{
		// restore previous state
		if (OpenGL.DepthBufferWritting )
		{
			glDepthMask( true );
		}
		if (Glide.State.DepthBufferMode != GR_DEPTHBUFFER_DISABLE)
		{
			glEnable( GL_DEPTH_TEST );
		}
		// Restore colormask
		bool rgb = Glide.State.ColorMask;
		glColorMask(rgb, rgb, rgb, Glide.State.AlphaMask);

		if ( OpenGL.Blend )
		{
			glEnable( GL_BLEND );
		}
		if ( InternalConfig.EXT_secondary_color )
		{
			glEnable( GL_COLOR_SUM_EXT );
		}
		glReportError();
		// Enable fog?
		bool enable_fog_texture_unit = OpenGL.FogTextureUnit &&
		    ((OpenGL.Fog && InternalConfig.FogMode == OpenGLideFogEmulation_EnvCombine) ||
		     Glide.State.ColorCombineInvert ||
		     Glide.State.AlphaInvert);
		if (enable_fog_texture_unit)
		{
			glActiveTextureARB(OpenGL.FogTextureUnit);
			glEnable(GL_TEXTURE_2D);
			// We're not using glDrawArrays to render the frame buffer,
			// but without disabling the client state the next texture drawn
			// by RenderDrawTriangles would get the wrong coordinates.
			// Can be observed in Carmageddon: The sky texture is rendered "too high"
			if (InternalConfig.EXT_compiled_vertex_array)
			{
				glClientActiveTextureARB(OpenGL.FogTextureUnit);
				glEnableClientState(GL_TEXTURE_COORD_ARRAY);
				glTexCoordPointer(1, GL_FLOAT, 0, &OGLRender.TFog[0]);
			}
		}
		if (OpenGL.Fog &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_None &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_EnvCombine)
		{
			glEnable(GL_FOG);
		}
		glReportError();
		
		if (OpenGL.ColorAlphaUnit2)
		{
			bool enable_coloralpha_texture_unit_2 = OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1];
			if (enable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit2);
				glEnable(GL_TEXTURE_2D);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glEnableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer(4, GL_FLOAT, 0, &OGLRender.TTexture[0]);
				}
			}
			if (enable_fog_texture_unit || enable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (!OpenGL.ColorAlphaUnitColorEnabledState[0] && !OpenGL.ColorAlphaUnitAlphaEnabledState[0])
			{
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer( 4, GL_FLOAT, 0, NULL );
				}
				glDisable(GL_TEXTURE_2D);
			}
			// Restore the previous texture environment
			glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE_EXT);
		}
		else
		{
			if (enable_fog_texture_unit)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (OpenGL.Texture == false)
			{
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer( 4, GL_FLOAT, 0, NULL );
				}
				// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D and this
				// has already been skipped in setState		
				if (!m_useRectangleARB) glDisable(GL_TEXTURE_2D);
			}
			// Restore the previous texture environment
			SetColorCombineState();
		}
		// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D
		if (m_useRectangleARB)
		{
			glDisable(GL_TEXTURE_RECTANGLE_ARB);
			glReportError();
		}
		glReportError();
		// This must be a forced update because GlideState changes of ChromaKeyMode
		// that don't change the corresponding GL-state are filtered out
		ForceChromaKeyAndAlphaStateUpdate();
	}

	glReportError();
	VERIFY_ACTIVE_TEXTURE_UNIT(OpenGL.ColorAlphaUnit1);
	VERIFY_TEXTURE_ENABLED_STATE();
}

bool Framebuffer::draw(const tilesize* tilesizetable, bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::draw(---, %d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::draw()");
	bool init_second_textureunit = pixelpipeline && OpenGL.ColorAlphaUnit2;
	FxU32* texbuffer = reinterpret_cast<FxU32*>(m_texbuffer->Address);
	if (m_use_client_storage)
	{
		glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, true);
		glReportError();
	}
	// Render the tiles
	GLint n = 0;
	GLint x;
	GLint y = 0;
	GLint y_step;
	for(int w = 0; y < m_height && w < MaxTiles; w++, y += y_step)
	{
		y_step = tilesizetable[w].y;
		x = 0;
		GLint x_step;
		for(int v = 0; x < m_width && v < MaxTiles; v++, x += x_step)
		{
			x_step = tilesizetable[w].x[v];
			// Use unique (but always the same) name for each texture in order
			// to maintain the size and avoid vram memory reallocation
			GLint texturename = m_textureNames[n];
			const TileUpdateState updateState = createTextureData(texbuffer, x, y, x_step, y_step, n);
			if (updateState != TileUpdateState_TileEmpty)
			{
#ifdef DEBUG_TILE_RENDERING
				unsigned int color;
				if (updateState == TileUpdateState_TileDownloadToGPU)
					// edges of downloaded tiles are red
					color = 0xff0000ff;
				else
					// edges of rendered tiles are cyan
					color = 0x00ffffff;
				((long*) texbuffer)[0] =
				((long*) texbuffer)[1] =
				((long*) texbuffer)[x_step] =
				((long*) texbuffer)[x_step -1] =
				((long*) texbuffer)[x_step -2] =
				((long*) texbuffer)[2 * x_step -1] =
				((long*) texbuffer)[(x_step - 1) * y_step] =
				((long*) texbuffer)[(x_step - 1) * y_step + 1] =
				((long*) texbuffer)[(x_step - 2) * y_step] =
				((long*) texbuffer)[x_step * y_step - 1] =
				((long*) texbuffer)[x_step * y_step - 2] =
				((long*) texbuffer)[x_step * (y_step - 1) - 1] = color;
#endif
				glBindTexture(GL_TEXTURE_2D, texturename);
				if (init_second_textureunit)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glBindTexture(GL_TEXTURE_2D, texturename);
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
#ifndef DEBUG_TILE_RENDERING
				if (updateState == TileUpdateState_TileDownloadToGPU || InternalConfig.EXT_Client_Storage == false)
				{
					glTexImage2D(GL_TEXTURE_2D, 0, m_glInternalFormat, x_step, y_step, 0, m_glFormat, m_glType, texbuffer);
					glReportError();
				}
#endif
				static struct
				{
					const GLfloat bl[4];
					const GLfloat br[4];
					const GLfloat tr[4];
					const GLfloat tl[4];
				}
				texcoords =
				{
					{0.0, 0.0, 1.0, 1.0},
					{1.0, 0.0, 1.0, 1.0},
					{1.0, 1.0, 1.0, 1.0},
					{0.0, 1.0, 1.0, 1.0}
				};
				glBegin(GL_QUADS);
					// counter clockwise
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.bl[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.bl[0]);
					}
					glVertex3f(x, y, m_glDepth);
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.br[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.br[0]);
					}
					glVertex3f(x + x_step, y, m_glDepth);
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.tr[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.tr[0]);
					}
					glVertex3f(x + x_step, y + y_step, m_glDepth);
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.tl[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.tl[0]);
					}
					glVertex3f(x, y + y_step , m_glDepth);
				glEnd();
				glReportError();
				// Advance to the next texbuffer location
				texbuffer += x_step * y_step;
			}
			n ++;
		}
	}
	if (m_use_client_storage)
	{
 		glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE , false);
		glReportError();
 	}
	s_Framebuffer.SetRenderBufferChanged();
	return y == m_height && x == m_width; 
}

bool Framebuffer::drawCompiledVertexArrays(const tilesize* tilesizetable, int vertexarrayindex, int tilecount, bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::draw(---, %d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::drawCompiledVertexArrays()");

	if (m_use_client_storage)
	{
		glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, true);
		glReportError();
	}
	// Finish rendering
	RenderUnlockArrays();
	// Transfer coords to VRAM
	glLockArraysEXT(vertexarrayindex * 3, tilecount * 6);
	OGLRender.BufferLocked = true;
	const bool init_second_textureunit = pixelpipeline &&
	                                     (OpenGL.ColorAlphaUnitColorEnabledState[1] ||
	                                      OpenGL.ColorAlphaUnitAlphaEnabledState[1]);
	FxU32* texbuffer = reinterpret_cast<FxU32*>(m_texbuffer->Address);
	// Render the tiles
	GLint n = 0;
	GLint x;
	GLint y = 0;
	GLint y_step;
	const GLenum textureTarget = m_useRectangleARB ? GL_TEXTURE_RECTANGLE_ARB : GL_TEXTURE_2D;
	for(int w = 0; y < m_height && w < MaxTiles; w++, y += y_step)
	{
		y_step = tilesizetable[w].y;
		x = 0;
		GLint x_step;
		for(int v = 0; x < m_width && v < MaxTiles; v++, x += x_step)
		{
			x_step = tilesizetable[w].x[v];
			const TileUpdateState updateState = createTextureData(texbuffer, x, y, x_step, y_step, n);
			if (updateState != TileUpdateState_TileEmpty)
			{
#ifdef DEBUG_TILE_RENDERING
				unsigned int color;
				if (updateState == TileUpdateState_TileDownloadToGPU)
					// edges of downloaded tiles are red
					color = 0xff0000ff;
				else
					// edges of rendered tiles are cyan
					color = 0x00ffffff;
				((long*) texbuffer)[0] =
				((long*) texbuffer)[1] =
				((long*) texbuffer)[x_step] =
				
				((long*) texbuffer)[x_step -1] =
				((long*) texbuffer)[x_step -2] =
				((long*) texbuffer)[2 * x_step -1] =

				((long*) texbuffer)[x_step * (y_step - 2)] =
				((long*) texbuffer)[x_step * (y_step - 1)] =
				((long*) texbuffer)[x_step * (y_step - 1) + 1] =

				((long*) texbuffer)[x_step * y_step - 1] =
				((long*) texbuffer)[x_step * y_step - 2] =
				((long*) texbuffer)[x_step * (y_step - 1) - 1] = color;
#endif
				// Use unique (but always the same) name for each
				// texture in order to be able to reuse tile data
				const GLint texturename = m_textureNames[n];
				// The texture rectangle is better suited for video,
				// which is close to a framebuffer
				if (init_second_textureunit)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glReportError();
					glBindTexture(textureTarget, texturename);
					glReportError();
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
					glReportError();
				}
				glBindTexture(textureTarget, texturename);
				glReportError();
#ifndef DEBUG_TILE_RENDERING
				if (updateState == TileUpdateState_TileDownloadToGPU || InternalConfig.EXT_Client_Storage == false)
#endif
				{
					glTexImage2D(textureTarget, 0, m_glInternalFormat, x_step, y_step, 0, m_glFormat, m_glType, texbuffer);
					glReportError();
				}
				// Draw the tile
				glDrawArrays(GL_TRIANGLES, vertexarrayindex * 3 + n * 6,  6);
				// Advance to the next texbuffer location
				texbuffer += x_step * y_step;
				glReportError();
			}
			n++;
		}
	}
	if (m_use_client_storage)
	{
 		glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE , false);
		glReportError();
 	}
	s_Framebuffer.SetRenderBufferChanged();
	return y == m_height && x == m_width; 
}

int Framebuffer::buildVertexArrays(const tilesize* tilesizetable, int vertexarrayindex)
{
	// Compute coordinates for compiled vertex arrays
	TColorStruct* pC = &OGLRender.TColor[vertexarrayindex];
	TVertexStruct* pV = &OGLRender.TVertex[vertexarrayindex];
	TTextureStruct* pTS = &OGLRender.TTexture[vertexarrayindex];
	int n = 0;
	GLint y = 0;
	GLint y_step;
	for(int w = 0; y < m_height && w < MaxTiles; w++, y += y_step)
	{
		y_step = tilesizetable[w].y;
		GLint x = 0;
		GLint x_step;
		for(int v = 0; x < m_width && v < MaxTiles; v++, x += x_step)
		{
			x_step = tilesizetable[w].x[v];
			// Write coordinates counter clockwise into render buffers
			pC->ar = pC->ag = pC->ab =
			pC->br = pC->bg = pC->bb =
			pC->cr = pC->cg = pC->cb =
			pC->aa = pC->ba = pC->ca = 1.0f;
			pV->ax = x;	pV->ay = y;
			pV->bx = x + x_step; pV->by = y;
			pV->cx = x + x_step; pV->cy = y + y_step;
			pV->az = pV->bz = pV->cz = m_glDepth;

			pTS->as = 0.0;
			pTS->at = 0.0;
			pTS->bs = m_useRectangleARB ? x_step : 1.0f;
			pTS->bt = 0.0;
			pTS->cs = m_useRectangleARB ? x_step : 1.0f;
			pTS->ct = m_useRectangleARB ? y_step : 1.0f;

			pTS->aq = pTS->bq = pTS->cq = 0.0f;
			pTS->aoow = pTS->boow = pTS->coow = 1.0f;
			pC++; pV++; pTS++;

			pC->ar = pC->ag = pC->ab =
			pC->br = pC->bg = pC->bb =
			pC->cr = pC->cg = pC->cb =
			pC->aa = pC->ba = pC->ca = 1.0f;
			pV->ax = x + x_step; pV->ay = y + y_step;
			pV->bx = x;	pV->by = y + y_step;
			pV->cx = x;	pV->cy = y;
			pV->az = pV->bz = pV->cz = m_glDepth;

			pTS->as = m_useRectangleARB ? x_step : 1.0f;
			pTS->at = m_useRectangleARB ? y_step : 1.0f;
			pTS->bs = 0.0;
			pTS->bt = m_useRectangleARB ? y_step : 1.0f;
			pTS->cs = 0.0;
			pTS->ct = 0.0;
			
			pTS->aq = pTS->bq = pTS->cq = 0.0f;
			pTS->aoow = pTS->boow = pTS->coow = 1.0f;
			pC++; pV++; pTS++;
			
			n++;
		}
	}
	return n;
}

#ifdef __ALTIVEC__
// altivec code
inline Framebuffer::TileUpdateState Framebuffer::Convert565Kto8888_AV(FxU16* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride, int checksumIndex)
{
	const vector bool short chromakey_565_av = m_ChromaKey.Vector;
	const int width_av = width >> 3; // 8 16-bit words
	const int stride_av = stride >> 3; // 8 16-bit words
	const int jump_av = width_av + stride_av;
	vector bool short* src_av = (vector bool short*) buffer1;
	// Setup channel 0 for reading one row of 565 ushorts from src into the L1 cache
	// This isn't read again soon, just written back once so we can bypass L2 cache
	const int src_control = (((width_av >> 4) & 0x1f) << 3) + (1 << 8) + (stride_av << 16);
	vec_dstt(src_av, src_control, 0);
	int h = height;
	// loop through the src to check whether anything has to be copied at all
	vector bool short* stop_zero_av = &src_av[width_av];
	do
	{
		do
		{
			const vector bool short pixels_565_av = *src_av;
			if (!vec_all_eq(pixels_565_av, chromakey_565_av)) goto create_8888_texture_1_av;  // Test clear first before jumping to create_8888_texture_1_av
			src_av++;
		}	while (src_av != stop_zero_av);
		src_av += stride_av;
		// Update channel 0 to prefetch the next row into the L1 cache
		vec_dstt(src_av, src_control, 0);
		stop_zero_av += jump_av;
	}	while (--h);
	return TileUpdateState_TileEmpty;
 create_8888_texture_1_av:
	// Delete dst up to the last chromakey entry in src
	stop_zero_av = src_av;
	src_av = (vector bool short*) buffer1;
	vector unsigned long* dst_av = (vector unsigned long*) buffer2;
	const vector unsigned long null_av = vec_splat_u32(0);
	// We're just writing to dst, no reading must occur, and starting a prefetch is a bad idea
	h = height;
	vector bool short* stop_av = &src_av[width_av];
	do
	{
 		do
		{
			if (src_av == stop_zero_av) goto create_8888_texture_2_av;  // Test clear first
			// clear cacheline to prevent it from being read-in (32 bytes = 2 altivec writes)
			// we're just clearing the cache line since we're going to write zeros anyway
	 		__dcbz(dst_av, 0);
			dst_av += 2;
			src_av++;
		}	while (src_av != stop_av);
		src_av += stride_av;
		stop_av += jump_av;
	}	while (--h);
	return TileUpdateState_TileEmpty;
create_8888_texture_2_av:
	// Build permute vector for storing high/lo 565 pixels into RgbxA
	// - results in R = r565+ggg, G = gggbbbbb, B = 0, A = glAlpha
	// -> good for comparison, green and blue are converted afterwards
	const vector unsigned char permute_hi_av = {0x00, 0x01, 0x12, 0x13, 0x02, 0x03, 0x16, 0x17, 0x04, 0x05, 0x1a, 0x1b, 0x06, 0x07, 0x1e, 0x1f};
 	// Computing the permute table just takes 2 instructions instead of 1 instruction + 4 memory reads
	const vector unsigned char permute_lo_av = vec_or(permute_hi_av, vec_splat_u8(8));

//	const vector unsigned long alpha_8888_av = {m_glAlpha, m_glAlpha, m_glAlpha, m_glAlpha};
	const vector unsigned long alpha_8888_av = m_glAlpha.Vector;
	// Build chromakey and alpha RgbxA vector
	const vector unsigned long chromakey_RgbxA_av = vec_perm((const vector unsigned long) chromakey_565_av, alpha_8888_av, permute_lo_av);
	// Constants
	const vector unsigned long const_3_av = vec_splat_u32(3);
	const vector unsigned long const_5_av = vec_splat_u32(5);
	// R5G6B500AA color masks
	const vector unsigned long mask_8888_ra = {0xf80000ff, 0xf80000ff, 0xf80000ff, 0xf80000ff};
	const vector unsigned long mask_8888_g  = {0x07e00000, 0x07e00000, 0x07e00000, 0x07e00000};
 	// Computing the mask just takes 2 instructions instead of 1 instruction + 4 memory reads
 	const vector unsigned long mask_8888_b  = vec_sr(mask_8888_ra, vec_splat_u32(11));

	vector unsigned long pixels_8888_src_av;
	vector bool long mask;
	vector unsigned long p;
	vector unsigned long q;
	vector unsigned long pixels_8888_dst_av;
	// Checksum the tile
	vector unsigned long c = null_av;
	vector unsigned long d;
	// Continue the loop and convert pixels from 565 to 8888
	vec_dstt(src_av, src_control, 0);
	do
	{
		do
		{
			const vector unsigned long pixels_565_av = (const vector unsigned long) (*src_av);
			// tile checksum part 1
			d = vec_sr(c, const_5_av);
			c = vec_add(c, pixels_565_av);
			// restore chroma key for next update
			*src_av++ = chromakey_565_av;

			// hi-word pixels
			pixels_8888_src_av = vec_perm(pixels_565_av, alpha_8888_av, permute_hi_av);
			mask = vec_cmpeq(pixels_8888_src_av, chromakey_RgbxA_av);
			// Keep red and alpha component
			pixels_8888_dst_av = vec_and(pixels_8888_src_av, mask_8888_ra);
			// add green component
			p = vec_and(pixels_8888_src_av, mask_8888_g);
			q = vec_sr(p, const_3_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);
			// add blue component
			p = vec_and(pixels_8888_src_av, mask_8888_b);
			q = vec_sr(p, const_5_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);

			// We're just writing to dst and thus can clear the cacheline in order
			// to avoid the read-in from system memory (32 bytes = 2 altivec writes)
			// Note: This is a G4 hack, but on a G5 the code will be fast enough anyway			
	 		__dcbz(dst_av, 0);

			// Select between pixels and chromakey
			*dst_av++ = vec_sel(pixels_8888_dst_av, null_av, mask);

			// tile checksum part 2
			c = vec_xor(c, d);

			// lo-word pixels
			pixels_8888_src_av = vec_perm(pixels_565_av, alpha_8888_av, permute_lo_av);
			mask = vec_cmpeq(pixels_8888_src_av, chromakey_RgbxA_av);
			// Keep red and alpha component
			pixels_8888_dst_av = vec_and(pixels_8888_src_av, mask_8888_ra);
			// add green component
			p = vec_and(pixels_8888_src_av, mask_8888_g);
			q = vec_sr(p, const_3_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);
			// add blue component
			p = vec_and(pixels_8888_src_av, mask_8888_b);
			q = vec_sr(p, const_5_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);
			// Select between pixels and chromakey
			*dst_av++ = vec_sel(pixels_8888_dst_av, null_av, mask);

		}	while (src_av != stop_av);
		src_av += stride_av;
		vec_dstt(src_av, src_control, 0);
		stop_av += jump_av;
	}	while (--h);
	// Skip downloading tile data to the gpu if the content hasn't changed
	if (vec_all_eq(c, m_tileChecksums[checksumIndex])) return TileUpdateState_TileDrawOnly;
	// The tile has been converted, been changed and must be downloaded to the gpu
	m_tileChecksums[checksumIndex] = c;
	return TileUpdateState_TileDownloadToGPU;
}
#endif

// Non-Altivec-code
inline Framebuffer::TileUpdateState Framebuffer::Convert565Kto8888(FxU16* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride)
{
	// Process two pixels at once
	const register unsigned long chromakey1 = m_ChromaKey.Scalar << 16;
	const register unsigned long chromakey2 = m_ChromaKey.Scalar;
	const register unsigned long chromakey12 = chromakey1 | chromakey2;
	width = width >> 1;
	stride = stride >> 1;
	register unsigned long pixel;
	register unsigned long* stop;
	register unsigned long jump = width + stride;
	register unsigned long* src = reinterpret_cast<unsigned long*>(buffer1);
	// check if tile must be processed in advance
	// to avoid useless writes to main memory
	// The tile should at least fit into the second level cache
	// so reading it again wouldn't hurt as much as doing needless writes
	register unsigned long h = height;
	stop = &src[width];
	do
	{
		do
		{
			pixel = *src++;
			if (pixel != chromakey12) goto create_8888_texture;
		}	while (src != stop);
		src += stride;
		stop += jump;
	}	while (--h);
	return TileUpdateState_TileEmpty;
create_8888_texture:
	const register unsigned long alpha = m_glAlpha.Scalar;
	const register unsigned long null = 0x00000000;
	const register unsigned long mask_pixel1 = 0xffff0000;
	const register unsigned long mask_pixel2 = 0x0000ffff;
	const register unsigned long mask_pixel1_r = 0xf8000000;
	const register unsigned long mask_pixel1_g = 0x07e00000;
	const register unsigned long mask_pixel1_b = 0x001f0000;
	const register unsigned long mask_pixel2_r = 0x0000f800;
	const register unsigned long mask_pixel2_g = 0x000007e0;
	const register unsigned long mask_pixel2_b = 0x0000001f;
	src = reinterpret_cast<unsigned long*>(buffer1);
	stop = &src[width];
	do
	{
		do
		{
			// GL_RGBA
			pixel = *src;
			if (pixel == chromakey12)
			{
				*buffer2++ = null;
				*buffer2++ = null;
			}
			else
			{
				*src = chromakey12;
				if ( (pixel & mask_pixel1) == chromakey1)
				{
					*buffer2++ = null;
				}
				else
				{
					*buffer2++ = ( alpha                         | // A
					             ( pixel & mask_pixel1_b ) >>  5 | // B
					             ( pixel & mask_pixel1_g ) >>  3 | // G
					             ( pixel & mask_pixel1_r ));       // R
				}
				if ( (pixel & mask_pixel2) == chromakey2)
				{
					*buffer2++ = null;
				}
				else
				{
					*buffer2++ = ( alpha                           | // A
					             ( pixel & mask_pixel2_b ) << 11   | // B
					             ( pixel & mask_pixel2_g ) << 13   | // G
					             ( pixel & mask_pixel2_r ) << 16);   // R
				}
			}
			src++;
		} while (src != stop);
		src += stride;
		stop += jump;
	} while (--height);
	return TileUpdateState_TileDownloadToGPU; 
}

inline Framebuffer::TileUpdateState Framebuffer::Convert1555Kto8888(FxU16* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride)
{
	// Process two pixels at once
	register unsigned long pixel;
	register unsigned long x;
	register unsigned long* src = reinterpret_cast<unsigned long*>(buffer1);
	const unsigned long null = 0x00000000;
	register unsigned long dstpixel = null;
	const register unsigned long chromakey1 = m_ChromaKey.Scalar << 16;
	const register unsigned long chromakey2 = m_ChromaKey.Scalar;
	const register unsigned long chromakey12 = chromakey1 | chromakey2;
	const register unsigned long alpha = m_glAlpha.Scalar;
	const register unsigned long mask_pixel1 = 0xffff0000;
	const register unsigned long mask_pixel2 = 0x0000ffff;
	const register unsigned long mask_pixel1_r = 0x7c000000;
	const register unsigned long mask_pixel1_g = 0x03e00000;
	const register unsigned long mask_pixel1_b = 0x001f0000;
	const register unsigned long mask_pixel2_r = 0x00007c00;
	const register unsigned long mask_pixel2_g = 0x000003e0;
	const register unsigned long mask_pixel2_b = 0x0000001f;
	width >>= 1;
	stride >>= 1;
	do
	{
		x = width;
		do
		{
			// GL_RGBA
			pixel = *src;
			if (pixel == chromakey12)
			{
				*buffer2++ = null;
				*buffer2++ = null;
			}
			else
			{
				*src = chromakey12;
				if ( (pixel & mask_pixel1) == chromakey1)
				{
					*buffer2++ = null;
				}
				else
				{
					dstpixel = ( alpha                           | // A
					           ( pixel & mask_pixel1_b ) >>  5   | // B
					           ( pixel & mask_pixel1_g ) >>  2   | // G
					           ( pixel & mask_pixel1_r ) <<  1);   // R
					*buffer2++ = dstpixel;
				}
				if ( (pixel & mask_pixel2) == chromakey2)
				{
					*buffer2++ = null;
				}
				else
				{
					dstpixel = ( alpha                           | // A
					           ( pixel & mask_pixel2_b ) << 11   | // B
					           ( pixel & mask_pixel2_g ) << 14   | // G
					           ( pixel & mask_pixel2_r ) << 17);   // R
					*buffer2++ = dstpixel;
				}
			}
			src++;
		}	while (--x);
		src += stride;
	} while (--height);
	return dstpixel != null ? TileUpdateState_TileDownloadToGPU : TileUpdateState_TileEmpty; 
}

inline Framebuffer::TileUpdateState Framebuffer::ConvertARGB8888Kto8888(FxU32* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride)
{
	// Process two pixels at once
	const register unsigned long chromakey = m_ChromaKey.Scalar || (m_ChromaKey.Scalar << 16);
	register unsigned long pixel;
	register unsigned long* stop;
	register unsigned jump = width + stride;
	register unsigned long* src = buffer1;
	// check if tile must be processed in advance
	// to avoid useless writes to main memory
	// The tile should at least fit into the second level cache
	// so reading it again wouldn't hurt as much as doing needless writes
	register unsigned long h = height;
	stop = &src[width];
	do
	{
		do
		{
			pixel = *src++;
			if (pixel != chromakey) goto create_8888_texture;
		}	while (src != stop);
		src += stride;
		stop += jump;
	}	while (--h);
	return TileUpdateState_TileEmpty;
create_8888_texture:
	const register unsigned long alpha = m_glAlpha.Scalar;
	src = buffer1;
	stop = &src[width];
	do
	{
		do
		{
			// GL_RGBA
			pixel = *src;
			if (pixel == chromakey)
			{
				*buffer2++ = 0;
			}
			else
			{
				*src = chromakey;
				*buffer2++ = (pixel << 8) | alpha;
			}
			src++;
		} while (src != stop);
		src += stride;
		stop += jump;
	}	while (--height);
	return TileUpdateState_TileDownloadToGPU; 
}

inline Framebuffer::TileUpdateState Framebuffer::createTextureData(FxU32* texbuffer, FxU32 x, FxU32 y, FxU32 x_step, FxU32 y_step, int checksumIndex)
{
	FxU32 stride = (m_width - x_step);
	FxU32 index = x + y * m_width;
	if (m_framebuffer->WriteMode == GR_LFBWRITEMODE_565)
	{
#ifdef __ALTIVEC__
		if (UserConfig.VectorUnitType == OpenGLideVectorUnitType_Altivec)
		{
#ifdef OGL_FRAMEBUFFER
			const vector unsigned long c = m_tileChecksums[checksumIndex];
#endif
		
			TileUpdateState state = Convert565Kto8888_AV(&m_framebuffer->Address[index], texbuffer, x_step, y_step, stride, checksumIndex);
#ifdef OGL_FRAMEBUFFER
			GlideMsg("Tile %d (%d,%d)-(%d,%d) update state is %s (%vlx)->(%vlx)\n",
			         checksumIndex,
			         x, y, x_step, y_step,
			         (state ==TileUpdateState_TileDownloadToGPU) ? "DownLoadToGPU"
			          : ((state == TileUpdateState_TileDrawOnly) ? "DrawOnly" : "TileEmpty"),
			          c,
			          m_tileChecksums[checksumIndex]);
#endif
			return state;
		}
		else
#endif
			return Convert565Kto8888(&m_framebuffer->Address[index], texbuffer, x_step, y_step, stride);
	}
	else if (m_framebuffer->WriteMode == GR_LFBWRITEMODE_1555)
	{
		return Convert1555Kto8888(&m_framebuffer->Address[index], texbuffer, x_step, y_step, stride);
	}
	else if (m_framebuffer->WriteMode == GR_LFBWRITEMODE_888)
	{
		FxU32* framebuffer = &reinterpret_cast<FxU32*>(m_framebuffer->Address)[index];
		return ConvertARGB8888Kto8888(framebuffer, texbuffer, x_step, y_step, stride);
	}
	else
	{
		return TileUpdateState_TileEmpty;
	}
}
\ No newline at end of file
+//**************************************************************
//*            OpenGLide - Glide to OpenGL Wrapper
//*             http://openglide.sourceforge.net
//*
//*                  framebuffer emulation
//*
//*         OpenGLide is OpenSource under LGPL license
//*  Mac version and additional features by Jens-Olaf Hemprich
//**************************************************************

#include "Framebuffer.h"
#include "Glide.h"
#include "GlideApplication.h"
#include "GlideSettings.h"
#include "GLRender.h"
#include "GLRenderUpdateState.h"
#include "GLColorAlphaCombineEnvTables.h"

// check if tile needs to be displayed
#define CHECK_RENDER_TILE

// Display small dots at opposite corners of rendered framebuffer tiles
//#define DEBUG_TILE_RENDERING

Framebuffer::Framebuffer()
: m_x_step_start(0)
, m_y_step_start(0)
, m_x_step_start_opaque(0)
, m_y_step_start_opaque(0)
, m_width(0)
, m_height(0)
, m_framebuffer(NULL)
, m_texbuffer(NULL)
, m_origin(GR_ORIGIN_UPPER_LEFT)
, m_glInternalFormat(-1)
, m_glFormat(-1)
, m_glType(-1)
, m_glDepth(1.0f)
, m_format_valid(false)
, m_use_client_storage(false)
, m_useRectangleARB(false)
, m_must_clear_buffer(true)
, m_custom_tilesizes(NULL)
{
	// Don't set the checksum to 0, as this would cause white screen in Carmageddon
	// because fully black tile also have a 0 checksum and no texture data would be
	// download at all (and the tile would be rendered as if TEXTURE_2D was disabled) 
	memset(m_tileChecksums, 0xff, sizeof(vector unsigned long) * m_tileCount);
}

Framebuffer::~Framebuffer()
{
}

bool Framebuffer::initialise_buffers(BufferStruct* framebuffer, BufferStruct* texbuffer, FxU32 width, FxU32 height, const tilesize* tilesizetable)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "GlideFrameBuffer::initialise_buffers(---, ---, %d, %d, ---)\n", width, height);
	#endif

	m_custom_tilesizes = tilesizetable;
	return initialise_buffers(framebuffer, texbuffer, width, height, 0, 0);
}

bool Framebuffer::initialise_buffers(BufferStruct* framebuffer, BufferStruct* texbuffer, FxU32 width, FxU32 height, FxU32 x_tile, FxU32 y_tile)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::initialise_buffers(---, ---, %d, %d, %d, %d)\n", width, height, x_tile, y_tile);
	#endif

	m_framebuffer = framebuffer;
	m_texbuffer = texbuffer;
	m_framebuffer->WriteMode = m_texbuffer->WriteMode = GR_LFBWRITEMODE_UNUSED;
	m_width = width;
	m_height = height;
	// find out largest texture size
	GLint tile_size;
	glGetIntegerv(GL_MAX_TEXTURE_SIZE, &tile_size);
	m_x_step_start_opaque = tile_size;
	m_y_step_start_opaque = tile_size;
	m_x_step_start = min(tile_size, x_tile);
	m_y_step_start = min(tile_size, y_tile);
	m_x_step_start = max(16, m_x_step_start);
	m_y_step_start = max(16, m_y_step_start);
	m_useRectangleARB = InternalConfig.ARB_texture_rectangle &&
	                    InternalConfig.EXT_compiled_vertex_array;
	// The texture priority is set to minimun because
	// frame buffer textures are never used a second time
	// @todo: This is not true anymore in all cases
	// because of the altivec checksum feature
	const GLfloat zero = 0.0f;
	glGenTextures(m_tileCount, &m_textureNames[0]);
	glPrioritizeTextures(m_tileCount, &m_textureNames[0], &zero);
	for(int i = 0; i < m_tileCount; i++)
	{
		const GLenum textureTarget = m_useRectangleARB ? GL_TEXTURE_RECTANGLE_ARB : GL_TEXTURE_2D;
		glBindTexture(textureTarget, m_textureNames[i]);
		if (m_useRectangleARB)
		{
			// This will not probably not work right now, because
			// we're using RGBA (the OS9 prefered texture format)
			// @todo: use ABGR??? in order to avoid byte swizzling
			// glTextureRangeAPPLE(GLenum target, GLsizei length, GLvoid *pointer);
			glTexParameteri(textureTarget,
			                GL_TEXTURE_STORAGE_HINT_APPLE,
			                GL_STORAGE_CACHED_APPLE);
			glReportError();
		}
		glTexParameteri(textureTarget, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
		glTexParameteri(textureTarget, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
		glTexParameteri(textureTarget, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
		glTexParameteri(textureTarget, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
	}
	// If a game has its own tilesize table, use
	// the largest tiles for opaque renderings
	GLint y_step = y_tile == 0 ? m_y_step_start_opaque : m_y_step_start;
	// init default/opaque tilesize table
	int w = 0;
	for(FxU32 y = 0; y < m_height && w < MaxTiles ; y += y_step, w++)
	{
		while (m_height - y < y_step)
		{
			y_step = y_step >> 1;
		}
		m_tilesizes[w].y = y_step;
		GLint x_step = x_tile == 0 ? m_x_step_start_opaque : m_x_step_start;
		int v = 0;
		for(FxU32 x = 0; x < m_width && v < MaxTiles; x += x_step, v++ )
		{
			while (m_width - x < x_step)
			{
				x_step = x_step >> 1;
			}
			m_tilesizes[w].x[v] = x_step;
		}
	}
	// Build compiled vertex arrays
	if (InternalConfig.EXT_compiled_vertex_array)
	{
		// Store various render buffers indices 
		m_tilesizesVertexArrayIndex = OGLRender.FrameBufferStartIndex;
		m_tilesizesCount = buildVertexArrays(&m_tilesizes[0], m_tilesizesVertexArrayIndex);
		if (m_custom_tilesizes)
		{
			m_customtilesizesVertexArrayIndex = m_tilesizesVertexArrayIndex + m_tilesizesCount * 2;
			m_customtilesizesCount = buildVertexArrays(m_custom_tilesizes, m_customtilesizesVertexArrayIndex);
		}
	}
	return m_width > 0 && m_height > 0 && m_x_step_start > 0 && m_y_step_start > 0 && m_format_valid;
}

void Framebuffer::free_buffers()
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::free_buffers()\n");
	#endif

	if (m_tilesizes) FreeObject(m_tilesizes);
	glDeleteTextures(m_tileCount, &m_textureNames[0]);
}

void Framebuffer::initialise_format(GrLfbWriteMode_t writemode)
{
	#if defined(OGL_PART_DONE) || defined(OGL_FRAMEBUFFER)
		GlideMsg("Framebuffer::initialise_format(0x%x)\n", writemode);
	#endif

	// Enlarge buffer?
	if (writemode >= GR_LFBWRITEMODE_888 &&
	    (m_framebuffer->WriteMode < GR_LFBWRITEMODE_888 || m_framebuffer->WriteMode == GR_LFBWRITEMODE_UNUSED) &&
	    m_framebuffer->Address)
	{
		// Delete existing buffer
		FreeFrameBuffer(m_framebuffer->Address);
		m_framebuffer->Address = NULL;
		m_texbuffer->Address = NULL;
	}
	// Allocate 32-bit buffer (16bit buffer has been allocated in grSstWinOpen()
	if (m_framebuffer->Address == NULL)
	{
		unsigned long openglpixels = OpenGL.WindowWidth * OpenGL.WindowHeight;
		// Framebuffer can be written to with 16bit or 32bit data
		unsigned long buffertypesize = (writemode >= GR_LFBWRITEMODE_888) ? sizeof(FxU32) : sizeof(FxU16);
		Glide.FrameBuffer.Address = (FxU16*) AllocFrameBuffer(Glide.WindowTotalPixels * buffertypesize + openglpixels * sizeof(FxU32), 1);
		Glide.TempBuffer.Address = &Glide.FrameBuffer.Address[Glide.WindowTotalPixels * buffertypesize >> 1];
		memset( Glide.FrameBuffer.Address, 0, Glide.WindowTotalPixels * buffertypesize);
		memset( Glide.TempBuffer.Address, 0, openglpixels * sizeof(FxU32));
	}
	m_framebuffer->WriteMode = writemode;
	m_glInternalFormat = 4;
	m_glFormat = GL_RGBA;
	m_glType = GL_UNSIGNED_BYTE;
	FxU16 chromakeyvalue;
	switch (writemode)
	{
	case GR_LFBWRITEMODE_565:
		chromakeyvalue = 	s_GlideApplication.GetType() == GlideApplication::Carmageddon
		                  ? 0x1f1f : 0x07ff;
		m_format_valid = true;
		break;
	case GR_LFBWRITEMODE_1555:
		chromakeyvalue = 0x03ff;
		m_format_valid = true;
		break;
	case GR_LFBWRITEMODE_888:
		chromakeyvalue = 0x7ffdfeff;
		m_format_valid = true;
		break;
	default:
		chromakeyvalue = 0x0;
		m_format_valid = false;
		break;
	}
	// When the chromakeyvalue changes, the buffer has to be cleared
	if (chromakeyvalue != m_ChromaKey.Scalar)
	{
		SetChromaKeyValue(chromakeyvalue);
		m_must_clear_buffer = true;
	}
}

bool Framebuffer::begin_write()
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg("Framebuffer::begin_write()\n");
#endif

	if (m_must_clear_buffer)
	{
		Clear();
		m_must_clear_buffer = false;
	}
	return true;
}

void Framebuffer::Clear()
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::Clear()\n");
	#endif

	const FxU16 chromakey = GetChromaKeyValue();
	const FxU32 count = m_width * m_height ;
	FxU16* framebuffer = m_framebuffer->Address;
	for ( int i = 0; i < count; i++)
	{
		framebuffer[i] = chromakey;
	}
}

bool Framebuffer::end_write(FxU32 alpha, GLfloat depth, bool pixelpipeline)
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg("Framebuffer::end_write(%d, %f, %d)\n", alpha, depth, pixelpipeline);
#endif

	m_glDepth = depth;
#ifdef __ALTIVEC__
	for(int i = 0; i < 4; i++)
	{
		(&m_glAlpha.Scalar)[i] = alpha;
	}
#else
	m_glAlpha.Scalar = alpha;
#endif
	// if all pixels are invisible, nothing must be rendered.
	// The pixel conversion functions assume alpha is != 0 in order
	// to determine if a tile contains any pixels to be rendered.
	if (m_glAlpha.Scalar == 0) return false;
	set_gl_state(pixelpipeline);
	if (InternalConfig.EXT_compiled_vertex_array)
	{
		if (m_custom_tilesizes)
		{
			drawCompiledVertexArrays(m_custom_tilesizes, m_customtilesizesVertexArrayIndex, m_customtilesizesCount, pixelpipeline);
		}
		else
		{
			drawCompiledVertexArrays(m_tilesizes, m_tilesizesVertexArrayIndex, m_tilesizesCount, pixelpipeline);
		}
	}
	else
	{
		const tilesize* tilesizes = m_custom_tilesizes ? m_custom_tilesizes : m_tilesizes;
		draw(tilesizes, pixelpipeline);
	}
	restore_gl_state(pixelpipeline);
	return true;
}

bool Framebuffer::end_write(FxU32 alpha)
{
#ifdef OGL_DONE
	GlideMsg("Framebuffer::end_write(%d)\n", alpha);
#endif

	// draw frame buffer
	// @todo: Depth should OpenGL.ZNear, but that breaks overlays in Myth
	FxBool result = end_write(alpha, 0.0, false);
	return result;
}

bool Framebuffer::end_write()
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg( "Framebuffer::end_write( )\n" );
#endif

	return end_write(0x000000ff);
}

bool Framebuffer::end_write_opaque()
{
#ifdef OGL_FRAMEBUFFER
	GlideMsg("Framebuffer::end_write_opaque()\n");
#endif

	// @todo: Depth should OpenGL.ZNear, but that breaks overlays in Myth
	return end_write(0x000000ff, 0.0, false);
}

void Framebuffer::set_gl_state(bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::set_gl_state(%d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::set_gl_state");
	VERIFY_ACTIVE_TEXTURE_UNIT(OpenGL.ColorAlphaUnit1);
	// Disable the cull mode
	glDisable(GL_CULL_FACE);
	// Disable clip volume hint manually to avoid recursion
	if (InternalConfig.EXT_clip_volume_hint && OpenGL.ClipVerticesEnabledState)
	{
		glHint(GL_CLIP_VOLUME_CLIPPING_HINT_EXT, GL_FASTEST);
	}
	
	if (pixelpipeline)
	{
		if (OpenGL.ColorAlphaUnit2)
		{
			// Pixelpipeline support for env cobine based rendering:
			// Framebuffer pixels must be routed through the coloralpha unit
			// as if they were produced by the vertex iterators without an 
			// additional GL texture unit -> source must be changed accordingly
			m_bRestoreColorCombine = false;
			if (Glide.State.ColorCombineLocal == GR_COMBINE_LOCAL_ITERATED)
			{
				Glide.State.ColorCombineLocal = GR_COMBINE_LOCAL_PIXELPIPELINE;
				m_bRestoreColorCombine = true;
			}
			if (Glide.State.ColorCombineOther == GR_COMBINE_OTHER_ITERATED)
			{
				Glide.State.ColorCombineOther = GR_COMBINE_OTHER_PIXELPIPELINE;
				m_bRestoreColorCombine = true;
			}
			if (m_bRestoreColorCombine)	SetColorCombineState();
			m_bRestoreAlphaCombine = false;
			if (Glide.State.AlphaLocal == GR_COMBINE_LOCAL_ITERATED)
			{
				Glide.State.AlphaLocal = GR_COMBINE_LOCAL_PIXELPIPELINE;
				m_bRestoreAlphaCombine = true;
			}
			if (Glide.State.AlphaOther == GR_COMBINE_OTHER_ITERATED)
			{
				Glide.State.AlphaOther = GR_COMBINE_OTHER_PIXELPIPELINE;
				m_bRestoreAlphaCombine = true;
			}
			if (m_bRestoreAlphaCombine)	SetAlphaCombineState();
			// Update the opengl state for the pixel pipeline
			RenderUpdateState();
			// If the write mode doesn't provide alpha then m_glAlpha is used
			// as the constant alpha value, and we can use the alpha test
			// to mask out chromakey pixels
			switch (m_framebuffer->WriteMode)
			{
			case GR_LFBWRITEMODE_565:
			case GR_LFBWRITEMODE_888:
				glEnable(GL_ALPHA_TEST);
				const GLenum alphaTestFunction = GL_EQUAL;
				const GLfloat alphaTestReferenceValue= m_glAlpha.Scalar * D1OVER255;
				OpenGL.AlphaTestFunction = alphaTestFunction;
				OpenGL.AlphaReferenceValue = alphaTestReferenceValue;
				glAlphaFunc(alphaTestFunction, alphaTestReferenceValue);
				glReportError();
				break;
			}
		}
		// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D
		if (m_useRectangleARB)
		{
			const GLenum textureTarget = GL_TEXTURE_RECTANGLE_ARB;
			const	bool enableColoralphaTextureUnit1 = OpenGL.ColorAlphaUnitColorEnabledState[0] || OpenGL.ColorAlphaUnitAlphaEnabledState[0];
			if (enableColoralphaTextureUnit1)
			{
				glEnable(textureTarget);
				glReportError();
			}
			if (OpenGL.ColorAlphaUnit2)
			{
				const	bool enableColoralphaTextureUnit2 = OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1];
				if (enableColoralphaTextureUnit2)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glReportError();
					glEnable(textureTarget);
					glReportError();
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
					glReportError();
				}
			}
			// The client texture state is already setup correctly since we just
			// have to adjust the texture rectangle state
			// (the texture_2d state is not changed when the pixelpipeline mode is active)
		}
		// Set the origin with clipping
		glMatrixMode(GL_PROJECTION);
		glLoadIdentity();
		if (m_origin == GR_ORIGIN_LOWER_LEFT)
		{
			glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
			        Glide.State.ClipMinY, Glide.State.ClipMaxY, 
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
			           OpenGL.OriginY + OpenGL.ClipMinY, 
			           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
			           OpenGL.ClipMaxY - OpenGL.ClipMinY); 
		}
		else
		{
			glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
			        Glide.State.ClipMaxY, Glide.State.ClipMinY, 
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
			           OpenGL.OriginY + OpenGL.WindowHeight - OpenGL.ClipMaxY, 
			           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
			           OpenGL.ClipMaxY - OpenGL.ClipMinY); 
		}
		// The scissor rectangle is not reset, because scissor mode
		// is only enabled when clearing the buffer
		glMatrixMode(GL_MODELVIEW);
		glReportError();
	}
	else
	{
		// disable blend
		glDisable(GL_BLEND);
		// disable depth buffer
		glDepthMask(false);
		// Enable colormask
		glColorMask( true, true, true, false);
		// Needed for displaying in-game menus
		if (Glide.State.DepthBufferMode != GR_DEPTHBUFFER_DISABLE)
		{
			glDisable(GL_DEPTH_TEST);
		}
		glEnable(GL_ALPHA_TEST);
		// Update state as we're calling update triggers on restore
		const GLenum alphaTestFunction = GL_GREATER;
		const GLfloat alphaTestReferenceValue= 0.0;
		OpenGL.AlphaTestFunction = alphaTestFunction;
		OpenGL.AlphaReferenceValue = alphaTestReferenceValue;
		glAlphaFunc(alphaTestFunction, alphaTestReferenceValue);
		glReportError();
		if (InternalConfig.EXT_secondary_color)
		{
			glDisable(GL_COLOR_SUM_EXT);
			glReportError();
		}
		// Reset the clipping window
		// and set the origin
		glMatrixMode(GL_PROJECTION);
		glLoadIdentity();
		if (m_origin == GR_ORIGIN_LOWER_LEFT)
		{
			glOrtho(0, Glide.WindowWidth,
			        0, Glide.WindowHeight,
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX,
			           OpenGL.OriginY,
			           OpenGL.WindowWidth,
			           OpenGL.WindowHeight); 
		}
		else
		{
			glOrtho(0, Glide.WindowWidth,
			        Glide.WindowHeight, 0,
			        OpenGL.ZNear, OpenGL.ZFar);
			glViewport(OpenGL.OriginX,
			           OpenGL.OriginY,
			           OpenGL.WindowWidth,
			           OpenGL.WindowHeight); 
		}
		// The scissor rectangle is not changed, because scissor mode
		// is only enabled when clearing the buffer
		glMatrixMode(GL_MODELVIEW);
		glReportError();
		
		// Disable fog
		bool disable_fog_texture_unit = OpenGL.FogTextureUnit;
		if (disable_fog_texture_unit)
		{
			glActiveTextureARB(OpenGL.FogTextureUnit);
			if (InternalConfig.EXT_compiled_vertex_array)
			{
				glClientActiveTextureARB(OpenGL.FogTextureUnit);
				glDisableClientState(GL_TEXTURE_COORD_ARRAY);
				glTexCoordPointer(4, GL_FLOAT, 0, NULL);
			}
			glDisable(GL_TEXTURE_2D);
		}
		if (OpenGL.Fog &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_None &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_EnvCombine)
		{
			glDisable(GL_FOG);
		}
		glReportError();
		
		// enable framebuffer texture unit
		if (OpenGL.ColorAlphaUnit2)
		{
			bool disable_coloralpha_texture_unit_2 = OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1];
			if (disable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit2);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
					// On MacOS9 (Classic?) the texcoord pointer needs to be reset
					// to the default value when glLockArrays/glUnlockArrays is used
					glTexCoordPointer( 4, GL_FLOAT, 0, NULL );
				}
				glDisable(GL_TEXTURE_2D);
			}
			if (disable_fog_texture_unit || disable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (!OpenGL.ColorAlphaUnitColorEnabledState[0] && !OpenGL.ColorAlphaUnitAlphaEnabledState[0])
			{
				glEnable(GL_TEXTURE_2D);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glEnableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer(4, GL_FLOAT, 0, &OGLRender.TTexture[0]);
				}
			}
		}
		else
		{
			if (disable_fog_texture_unit)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (OpenGL.Texture == false)
			{
				// GL_RECTANGLE_ARB overrides GL_RECTANGLE_2D
				if (!m_useRectangleARB) glEnable(GL_TEXTURE_2D);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glEnableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer(4, GL_FLOAT, 0, &OGLRender.TTexture[0]);
				}
			}
		}
		if (m_useRectangleARB)
		{
			glEnable(GL_TEXTURE_RECTANGLE_ARB);
			glReportError();
		}
		glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);
		glReportError();
	}
}

void Framebuffer::restore_gl_state(bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::restore_gl_state(%d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::restore_gl_state");
	// Restore the cull mode
	switch (Glide.State.CullMode)
	{
	case GR_CULL_DISABLE:
		break;
	case GR_CULL_NEGATIVE:
	case GR_CULL_POSITIVE:
		glEnable(GL_CULL_FACE);
		break;
	}
	if (InternalConfig.EXT_clip_volume_hint && OpenGL.ClipVerticesEnabledState)
	{
		glHint(GL_CLIP_VOLUME_CLIPPING_HINT_EXT, GL_NICEST);
	}
	
	// Restore the clipping window
	glMatrixMode(GL_PROJECTION);
	glLoadIdentity();
	if ( Glide.State.OriginInformation == GR_ORIGIN_LOWER_LEFT )
	{
		glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
							Glide.State.ClipMinY, Glide.State.ClipMaxY, 
							OpenGL.ZNear, OpenGL.ZFar);
		glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
		           OpenGL.OriginY + OpenGL.ClipMinY, 
		           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
		           OpenGL.ClipMaxY - OpenGL.ClipMinY); 
	}
	else
	{
		glOrtho(Glide.State.ClipMinX, Glide.State.ClipMaxX, 
							Glide.State.ClipMaxY, Glide.State.ClipMinY, 
							OpenGL.ZNear, OpenGL.ZFar);
		glViewport(OpenGL.OriginX + OpenGL.ClipMinX,
		           OpenGL.OriginY + OpenGL.WindowHeight - OpenGL.ClipMaxY, 
		           OpenGL.ClipMaxX - OpenGL.ClipMinX, 
		           OpenGL.ClipMaxY - OpenGL.ClipMinY);
	}
	// The scissor rectangle is not reset, because scissor mode
	// is only enabled when clearing the buffer
	glMatrixMode( GL_MODELVIEW );
	glReportError();
	
	if (pixelpipeline)
	{
		if (OpenGL.ColorAlphaUnit2)
		{
			// restore current values
			if (m_bRestoreColorCombine)
			{
				if (Glide.State.ColorCombineLocal == GR_COMBINE_LOCAL_PIXELPIPELINE)	Glide.State.ColorCombineLocal = GR_COMBINE_LOCAL_ITERATED;
				if (Glide.State.ColorCombineOther == GR_COMBINE_OTHER_PIXELPIPELINE)	Glide.State.ColorCombineOther = GR_COMBINE_OTHER_ITERATED;
				SetColorCombineState();
			}
			if(m_bRestoreAlphaCombine)
			{
				if (Glide.State.AlphaLocal == GR_COMBINE_LOCAL_PIXELPIPELINE)	Glide.State.AlphaLocal = GR_COMBINE_LOCAL_ITERATED;
				if (Glide.State.AlphaOther == GR_COMBINE_OTHER_PIXELPIPELINE)	Glide.State.AlphaOther = GR_COMBINE_LOCAL_ITERATED;
				SetAlphaCombineState();
			}
		}
		switch (m_framebuffer->WriteMode)
		{
		case GR_LFBWRITEMODE_565:
		case GR_LFBWRITEMODE_888:
			SetChromaKeyAndAlphaState();
			break;
		}
		// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D
		if (m_useRectangleARB)
		{
			const GLenum textureTarget = GL_TEXTURE_RECTANGLE_ARB;
			const	bool disableColoralphaTextureUnit1 = !(OpenGL.ColorAlphaUnitColorEnabledState[0] || OpenGL.ColorAlphaUnitAlphaEnabledState[0]);
			if (disableColoralphaTextureUnit1)
			{
				glDisable(textureTarget);
				glReportError();
			}
			if (OpenGL.ColorAlphaUnit2)
			{
				const	bool disableColoralphaTextureUnit2 = !(OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1]);
				if (disableColoralphaTextureUnit2)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glReportError();
					glDisable(textureTarget);
					glReportError();
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
					glReportError();
				}
			}
			// The client texture state is already setup correctly since we just
			// have to adjust the texture rectangle state to the texture_2d state
		}
	}
	else
	{
		// restore previous state
		if (OpenGL.DepthBufferWritting )
		{
			glDepthMask( true );
		}
		if (Glide.State.DepthBufferMode != GR_DEPTHBUFFER_DISABLE)
		{
			glEnable( GL_DEPTH_TEST );
		}
		// Restore colormask
		bool rgb = Glide.State.ColorMask;
		glColorMask(rgb, rgb, rgb, Glide.State.AlphaMask);

		if ( OpenGL.Blend )
		{
			glEnable( GL_BLEND );
		}
		if ( InternalConfig.EXT_secondary_color )
		{
			glEnable( GL_COLOR_SUM_EXT );
		}
		glReportError();
		// Enable fog?
		bool enable_fog_texture_unit = OpenGL.FogTextureUnit &&
		    ((OpenGL.Fog && InternalConfig.FogMode == OpenGLideFogEmulation_EnvCombine) ||
		     Glide.State.ColorCombineInvert ||
		     Glide.State.AlphaInvert);
		if (enable_fog_texture_unit)
		{
			glActiveTextureARB(OpenGL.FogTextureUnit);
			glEnable(GL_TEXTURE_2D);
			// We're not using glDrawArrays to render the frame buffer,
			// but without disabling the client state the next texture drawn
			// by RenderDrawTriangles would get the wrong coordinates.
			// Can be observed in Carmageddon: The sky texture is rendered "too high"
			if (InternalConfig.EXT_compiled_vertex_array)
			{
				glClientActiveTextureARB(OpenGL.FogTextureUnit);
				glEnableClientState(GL_TEXTURE_COORD_ARRAY);
				glTexCoordPointer(1, GL_FLOAT, 0, &OGLRender.TFog[0]);
			}
		}
		if (OpenGL.Fog &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_None &&
		    InternalConfig.FogMode != OpenGLideFogEmulation_EnvCombine)
		{
			glEnable(GL_FOG);
		}
		glReportError();
		
		if (OpenGL.ColorAlphaUnit2)
		{
			bool enable_coloralpha_texture_unit_2 = OpenGL.ColorAlphaUnitColorEnabledState[1] || OpenGL.ColorAlphaUnitAlphaEnabledState[1];
			if (enable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit2);
				glEnable(GL_TEXTURE_2D);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glEnableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer(4, GL_FLOAT, 0, &OGLRender.TTexture[0]);
				}
			}
			if (enable_fog_texture_unit || enable_coloralpha_texture_unit_2)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (!OpenGL.ColorAlphaUnitColorEnabledState[0] && !OpenGL.ColorAlphaUnitAlphaEnabledState[0])
			{
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer( 4, GL_FLOAT, 0, NULL );
				}
				glDisable(GL_TEXTURE_2D);
			}
			// Restore the previous texture environment
			glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_COMBINE_EXT);
		}
		else
		{
			if (enable_fog_texture_unit)
			{
				glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glClientActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
			}
			if (OpenGL.Texture == false)
			{
				if (InternalConfig.EXT_compiled_vertex_array)
				{
					glDisableClientState(GL_TEXTURE_COORD_ARRAY);
					glTexCoordPointer( 4, GL_FLOAT, 0, NULL );
				}
				// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D and this
				// has already been skipped in setState		
				if (!m_useRectangleARB) glDisable(GL_TEXTURE_2D);
			}
			// Restore the previous texture environment
			SetColorCombineState();
		}
		// GL_RECTANGLE_ARB overrides GL_TEXTURE_2D
		if (m_useRectangleARB)
		{
			glDisable(GL_TEXTURE_RECTANGLE_ARB);
			glReportError();
		}
		glReportError();
		// This must be a forced update because GlideState changes of ChromaKeyMode
		// that don't change the corresponding GL-state are filtered out
		ForceChromaKeyAndAlphaStateUpdate();
	}

	glReportError();
	VERIFY_ACTIVE_TEXTURE_UNIT(OpenGL.ColorAlphaUnit1);
	VERIFY_TEXTURE_ENABLED_STATE();
}

bool Framebuffer::draw(const tilesize* tilesizetable, bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::draw(---, %d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::draw()");
	bool init_second_textureunit = pixelpipeline && OpenGL.ColorAlphaUnit2;
	FxU32* texbuffer = reinterpret_cast<FxU32*>(m_texbuffer->Address);
	// Render the tiles
	GLint n = 0;
	GLint x;
	GLint y = 0;
	GLint y_step;
	for(int w = 0; y < m_height && w < MaxTiles; w++, y += y_step)
	{
		y_step = tilesizetable[w].y;
		x = 0;
		GLint x_step;
		for(int v = 0; x < m_width && v < MaxTiles; v++, x += x_step)
		{
			x_step = tilesizetable[w].x[v];
			// Use unique (but always the same) name for each texture in order
			// to maintain the size and avoid vram memory reallocation
			GLint texturename = m_textureNames[n];
			const TileUpdateState updateState = createTextureData(texbuffer, x, y, x_step, y_step, n);
			if (updateState != TileUpdateState_TileEmpty)
			{
#ifdef DEBUG_TILE_RENDERING
				unsigned int color;
				if (updateState == TileUpdateState_TileDownloadToGPU)
					// edges of downloaded tiles are red
					color = 0xff0000ff;
				else
					// edges of rendered tiles are cyan
					color = 0x00ffffff;
				((long*) texbuffer)[0] =
				((long*) texbuffer)[1] =
				((long*) texbuffer)[x_step] =
				((long*) texbuffer)[x_step -1] =
				((long*) texbuffer)[x_step -2] =
				((long*) texbuffer)[2 * x_step -1] =
				((long*) texbuffer)[(x_step - 1) * y_step] =
				((long*) texbuffer)[(x_step - 1) * y_step + 1] =
				((long*) texbuffer)[(x_step - 2) * y_step] =
				((long*) texbuffer)[x_step * y_step - 1] =
				((long*) texbuffer)[x_step * y_step - 2] =
				((long*) texbuffer)[x_step * (y_step - 1) - 1] = color;
#endif
				glBindTexture(GL_TEXTURE_2D, texturename);
				if (init_second_textureunit)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glBindTexture(GL_TEXTURE_2D, texturename);
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
				}
#ifndef DEBUG_TILE_RENDERING
				if (updateState == TileUpdateState_TileDownloadToGPU || InternalConfig.APPLE_client_storage == false)
				{
					glTexImage2D(GL_TEXTURE_2D, 0, m_glInternalFormat, x_step, y_step, 0, m_glFormat, m_glType, texbuffer);
					glReportError();
				}
#endif
				static struct
				{
					const GLfloat bl[4];
					const GLfloat br[4];
					const GLfloat tr[4];
					const GLfloat tl[4];
				}
				texcoords =
				{
					{0.0, 0.0, 1.0, 1.0},
					{1.0, 0.0, 1.0, 1.0},
					{1.0, 1.0, 1.0, 1.0},
					{0.0, 1.0, 1.0, 1.0}
				};
				glBegin(GL_QUADS);
					// counter clockwise
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.bl[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.bl[0]);
					}
					glVertex3f(x, y, m_glDepth);
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.br[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.br[0]);
					}
					glVertex3f(x + x_step, y, m_glDepth);
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.tr[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.tr[0]);
					}
					glVertex3f(x + x_step, y + y_step, m_glDepth);
					glColor3f(1.0, 1.0, 1.0);
					glTexCoord4fv(&texcoords.tl[0]);
					if (init_second_textureunit)
					{
						glMultiTexCoord4fvARB(OpenGL.ColorAlphaUnit2, &texcoords.tl[0]);
					}
					glVertex3f(x, y + y_step , m_glDepth);
				glEnd();
				glReportError();
				// Advance to the next texbuffer location
				texbuffer += x_step * y_step;
			}
			n ++;
		}
	}
	s_Framebuffer.SetRenderBufferChanged();
	return y == m_height && x == m_width; 
}

bool Framebuffer::drawCompiledVertexArrays(const tilesize* tilesizetable, int vertexarrayindex, int tilecount, bool pixelpipeline)
{
	#ifdef OGL_FRAMEBUFFER
		GlideMsg( "Framebuffer::draw(---, %d)\n", pixelpipeline);
	#endif

	glReportErrors("Framebuffer::drawCompiledVertexArrays()");

	// Finish rendering
	RenderUnlockArrays();
	// Transfer coords to VRAM
	glLockArraysEXT(vertexarrayindex * 3, tilecount * 6);
	OGLRender.BufferLocked = true;
	const bool init_second_textureunit = pixelpipeline &&
	                                     (OpenGL.ColorAlphaUnitColorEnabledState[1] ||
	                                      OpenGL.ColorAlphaUnitAlphaEnabledState[1]);
	FxU32* texbuffer = reinterpret_cast<FxU32*>(m_texbuffer->Address);
	// Render the tiles
	GLint n = 0;
	GLint x;
	GLint y = 0;
	GLint y_step;
	const GLenum textureTarget = m_useRectangleARB ? GL_TEXTURE_RECTANGLE_ARB : GL_TEXTURE_2D;
	for(int w = 0; y < m_height && w < MaxTiles; w++, y += y_step)
	{
		y_step = tilesizetable[w].y;
		x = 0;
		GLint x_step;
		for(int v = 0; x < m_width && v < MaxTiles; v++, x += x_step)
		{
			x_step = tilesizetable[w].x[v];
			const TileUpdateState updateState = createTextureData(texbuffer, x, y, x_step, y_step, n);
			if (updateState != TileUpdateState_TileEmpty)
			{
#ifdef DEBUG_TILE_RENDERING
				unsigned int color;
				if (updateState == TileUpdateState_TileDownloadToGPU)
					// edges of downloaded tiles are red
					color = 0xff0000ff;
				else
					// edges of rendered tiles are cyan
					color = 0x00ffffff;
				((long*) texbuffer)[0] =
				((long*) texbuffer)[1] =
				((long*) texbuffer)[x_step] =
				
				((long*) texbuffer)[x_step -1] =
				((long*) texbuffer)[x_step -2] =
				((long*) texbuffer)[2 * x_step -1] =

				((long*) texbuffer)[x_step * (y_step - 2)] =
				((long*) texbuffer)[x_step * (y_step - 1)] =
				((long*) texbuffer)[x_step * (y_step - 1) + 1] =

				((long*) texbuffer)[x_step * y_step - 1] =
				((long*) texbuffer)[x_step * y_step - 2] =
				((long*) texbuffer)[x_step * (y_step - 1) - 1] = color;
#endif
				// Use unique (but always the same) name for each
				// texture in order to be able to reuse tile data
				const GLint texturename = m_textureNames[n];
				// The texture rectangle is better suited for video,
				// which is close to a framebuffer
				if (init_second_textureunit)
				{
					glActiveTextureARB(OpenGL.ColorAlphaUnit2);
					glReportError();
					glBindTexture(textureTarget, texturename);
					glReportError();
					glActiveTextureARB(OpenGL.ColorAlphaUnit1);
					glReportError();
				}
				glBindTexture(textureTarget, texturename);
				glReportError();
#ifndef DEBUG_TILE_RENDERING
				if (updateState == TileUpdateState_TileDownloadToGPU || InternalConfig.APPLE_client_storage == false)
#endif
				{
					glTexImage2D(textureTarget, 0, m_glInternalFormat, x_step, y_step, 0, m_glFormat, m_glType, texbuffer);
					glReportError();
				}
				// Draw the tile
				glDrawArrays(GL_TRIANGLES, vertexarrayindex * 3 + n * 6,  6);
				// Advance to the next texbuffer location
				texbuffer += x_step * y_step;
				glReportError();
			}
			n++;
		}
	}
	s_Framebuffer.SetRenderBufferChanged();
	return y == m_height && x == m_width; 
}

int Framebuffer::buildVertexArrays(const tilesize* tilesizetable, int vertexarrayindex)
{
	// Compute coordinates for compiled vertex arrays
	TColorStruct* pC = &OGLRender.TColor[vertexarrayindex];
	TVertexStruct* pV = &OGLRender.TVertex[vertexarrayindex];
	TTextureStruct* pTS = &OGLRender.TTexture[vertexarrayindex];
	int n = 0;
	GLint y = 0;
	GLint y_step;
	for(int w = 0; y < m_height && w < MaxTiles; w++, y += y_step)
	{
		y_step = tilesizetable[w].y;
		GLint x = 0;
		GLint x_step;
		for(int v = 0; x < m_width && v < MaxTiles; v++, x += x_step)
		{
			x_step = tilesizetable[w].x[v];
			// Write coordinates counter clockwise into render buffers
			pC->ar = pC->ag = pC->ab =
			pC->br = pC->bg = pC->bb =
			pC->cr = pC->cg = pC->cb =
			pC->aa = pC->ba = pC->ca = 1.0f;
			pV->ax = x;	pV->ay = y;
			pV->bx = x + x_step; pV->by = y;
			pV->cx = x + x_step; pV->cy = y + y_step;
			pV->az = pV->bz = pV->cz = m_glDepth;

			pTS->as = 0.0;
			pTS->at = 0.0;
			pTS->bs = m_useRectangleARB ? x_step : 1.0f;
			pTS->bt = 0.0;
			pTS->cs = m_useRectangleARB ? x_step : 1.0f;
			pTS->ct = m_useRectangleARB ? y_step : 1.0f;

			pTS->aq = pTS->bq = pTS->cq = 0.0f;
			pTS->aoow = pTS->boow = pTS->coow = 1.0f;
			pC++; pV++; pTS++;

			pC->ar = pC->ag = pC->ab =
			pC->br = pC->bg = pC->bb =
			pC->cr = pC->cg = pC->cb =
			pC->aa = pC->ba = pC->ca = 1.0f;
			pV->ax = x + x_step; pV->ay = y + y_step;
			pV->bx = x;	pV->by = y + y_step;
			pV->cx = x;	pV->cy = y;
			pV->az = pV->bz = pV->cz = m_glDepth;

			pTS->as = m_useRectangleARB ? x_step : 1.0f;
			pTS->at = m_useRectangleARB ? y_step : 1.0f;
			pTS->bs = 0.0;
			pTS->bt = m_useRectangleARB ? y_step : 1.0f;
			pTS->cs = 0.0;
			pTS->ct = 0.0;
			
			pTS->aq = pTS->bq = pTS->cq = 0.0f;
			pTS->aoow = pTS->boow = pTS->coow = 1.0f;
			pC++; pV++; pTS++;
			
			n++;
		}
	}
	return n;
}

#ifdef __ALTIVEC__
// altivec code
inline Framebuffer::TileUpdateState Framebuffer::Convert565Kto8888_AV(FxU16* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride, int checksumIndex)
{
	const vector bool short chromakey_565_av = m_ChromaKey.Vector;
	const int width_av = width >> 3; // 8 16-bit words
	const int stride_av = stride >> 3; // 8 16-bit words
	const int jump_av = width_av + stride_av;
	vector bool short* src_av = (vector bool short*) buffer1;
	// Setup channel 0 for reading one row of 565 ushorts from src into the L1 cache
	// This isn't read again soon, just written back once so we can bypass L2 cache
	const int src_control = (((width_av >> 4) & 0x1f) << 3) + (1 << 8) + (stride_av << 16);
	vec_dstt(src_av, src_control, 0);
	int h = height;
	// loop through the src to check whether anything has to be copied at all
	vector bool short* stop_zero_av = &src_av[width_av];
	do
	{
		do
		{
			const vector bool short pixels_565_av = *src_av;
			if (!vec_all_eq(pixels_565_av, chromakey_565_av)) goto create_8888_texture_1_av;  // Test clear first before jumping to create_8888_texture_1_av
			src_av++;
		}	while (src_av != stop_zero_av);
		src_av += stride_av;
		// Update channel 0 to prefetch the next row into the L1 cache
		vec_dstt(src_av, src_control, 0);
		stop_zero_av += jump_av;
	}	while (--h);
	return TileUpdateState_TileEmpty;
 create_8888_texture_1_av:
	// Delete dst up to the last chromakey entry in src
	stop_zero_av = src_av;
	src_av = (vector bool short*) buffer1;
	vector unsigned long* dst_av = (vector unsigned long*) buffer2;
	const vector unsigned long null_av = vec_splat_u32(0);
	// We're just writing to dst, no reading must occur, and starting a prefetch is a bad idea
	h = height;
	vector bool short* stop_av = &src_av[width_av];
	do
	{
 		do
		{
			if (src_av == stop_zero_av) goto create_8888_texture_2_av;  // Test clear first
			// clear cacheline to prevent it from being read-in (32 bytes = 2 altivec writes)
			// we're just clearing the cache line since we're going to write zeros anyway
	 		__dcbz(dst_av, 0);
			dst_av += 2;
			src_av++;
		}	while (src_av != stop_av);
		src_av += stride_av;
		stop_av += jump_av;
	}	while (--h);
	return TileUpdateState_TileEmpty;
create_8888_texture_2_av:
	// Build permute vector for storing high/lo 565 pixels into RgbxA
	// - results in R = r565+ggg, G = gggbbbbb, B = 0, A = glAlpha
	// -> good for comparison, green and blue are converted afterwards
	const vector unsigned char permute_hi_av = {0x00, 0x01, 0x12, 0x13, 0x02, 0x03, 0x16, 0x17, 0x04, 0x05, 0x1a, 0x1b, 0x06, 0x07, 0x1e, 0x1f};
 	// Computing the permute table just takes 2 instructions instead of 1 instruction + 4 memory reads
	const vector unsigned char permute_lo_av = vec_or(permute_hi_av, vec_splat_u8(8));

//	const vector unsigned long alpha_8888_av = {m_glAlpha, m_glAlpha, m_glAlpha, m_glAlpha};
	const vector unsigned long alpha_8888_av = m_glAlpha.Vector;
	// Build chromakey and alpha RgbxA vector
	const vector unsigned long chromakey_RgbxA_av = vec_perm((const vector unsigned long) chromakey_565_av, alpha_8888_av, permute_lo_av);
	// Constants
	const vector unsigned long const_3_av = vec_splat_u32(3);
	const vector unsigned long const_5_av = vec_splat_u32(5);
	// R5G6B500AA color masks
	const vector unsigned long mask_8888_ra = {0xf80000ff, 0xf80000ff, 0xf80000ff, 0xf80000ff};
	const vector unsigned long mask_8888_g  = {0x07e00000, 0x07e00000, 0x07e00000, 0x07e00000};
 	// Computing the mask just takes 2 instructions instead of 1 instruction + 4 memory reads
 	const vector unsigned long mask_8888_b  = vec_sr(mask_8888_ra, vec_splat_u32(11));

	vector unsigned long pixels_8888_src_av;
	vector bool long mask;
	vector unsigned long p;
	vector unsigned long q;
	vector unsigned long pixels_8888_dst_av;
	// Checksum the tile
	vector unsigned long c = null_av;
	vector unsigned long d;
	// Continue the loop and convert pixels from 565 to 8888
	vec_dstt(src_av, src_control, 0);
	do
	{
		do
		{
			const vector unsigned long pixels_565_av = (const vector unsigned long) (*src_av);
			// tile checksum part 1
			d = vec_sr(c, const_5_av);
			c = vec_add(c, pixels_565_av);
			// restore chroma key for next update
			*src_av++ = chromakey_565_av;

			// hi-word pixels
			pixels_8888_src_av = vec_perm(pixels_565_av, alpha_8888_av, permute_hi_av);
			mask = vec_cmpeq(pixels_8888_src_av, chromakey_RgbxA_av);
			// Keep red and alpha component
			pixels_8888_dst_av = vec_and(pixels_8888_src_av, mask_8888_ra);
			// add green component
			p = vec_and(pixels_8888_src_av, mask_8888_g);
			q = vec_sr(p, const_3_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);
			// add blue component
			p = vec_and(pixels_8888_src_av, mask_8888_b);
			q = vec_sr(p, const_5_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);

			// We're just writing to dst and thus can clear the cacheline in order
			// to avoid the read-in from system memory (32 bytes = 2 altivec writes)
			// Note: This is a G4 hack, but on a G5 the code will be fast enough anyway			
	 		__dcbz(dst_av, 0);

			// Select between pixels and chromakey
			*dst_av++ = vec_sel(pixels_8888_dst_av, null_av, mask);

			// tile checksum part 2
			c = vec_xor(c, d);

			// lo-word pixels
			pixels_8888_src_av = vec_perm(pixels_565_av, alpha_8888_av, permute_lo_av);
			mask = vec_cmpeq(pixels_8888_src_av, chromakey_RgbxA_av);
			// Keep red and alpha component
			pixels_8888_dst_av = vec_and(pixels_8888_src_av, mask_8888_ra);
			// add green component
			p = vec_and(pixels_8888_src_av, mask_8888_g);
			q = vec_sr(p, const_3_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);
			// add blue component
			p = vec_and(pixels_8888_src_av, mask_8888_b);
			q = vec_sr(p, const_5_av);
			pixels_8888_dst_av = vec_or(pixels_8888_dst_av, q);
			// Select between pixels and chromakey
			*dst_av++ = vec_sel(pixels_8888_dst_av, null_av, mask);

		}	while (src_av != stop_av);
		src_av += stride_av;
		vec_dstt(src_av, src_control, 0);
		stop_av += jump_av;
	}	while (--h);
	// Skip downloading tile data to the gpu if the content hasn't changed
	if (vec_all_eq(c, m_tileChecksums[checksumIndex])) return TileUpdateState_TileDrawOnly;
	// The tile has been converted, been changed and must be downloaded to the gpu
	m_tileChecksums[checksumIndex] = c;
	return TileUpdateState_TileDownloadToGPU;
}
#endif

// Non-Altivec-code
inline Framebuffer::TileUpdateState Framebuffer::Convert565Kto8888(FxU16* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride)
{
	// Process two pixels at once
	const register unsigned long chromakey1 = m_ChromaKey.Scalar << 16;
	const register unsigned long chromakey2 = m_ChromaKey.Scalar;
	const register unsigned long chromakey12 = chromakey1 | chromakey2;
	width = width >> 1;
	stride = stride >> 1;
	register unsigned long pixel;
	register unsigned long* stop;
	register unsigned long jump = width + stride;
	register unsigned long* src = reinterpret_cast<unsigned long*>(buffer1);
	// check if tile must be processed in advance
	// to avoid useless writes to main memory
	// The tile should at least fit into the second level cache
	// so reading it again wouldn't hurt as much as doing needless writes
	register unsigned long h = height;
	stop = &src[width];
	do
	{
		do
		{
			pixel = *src++;
			if (pixel != chromakey12) goto create_8888_texture;
		}	while (src != stop);
		src += stride;
		stop += jump;
	}	while (--h);
	return TileUpdateState_TileEmpty;
create_8888_texture:
	const register unsigned long alpha = m_glAlpha.Scalar;
	const register unsigned long null = 0x00000000;
	const register unsigned long mask_pixel1 = 0xffff0000;
	const register unsigned long mask_pixel2 = 0x0000ffff;
	const register unsigned long mask_pixel1_r = 0xf8000000;
	const register unsigned long mask_pixel1_g = 0x07e00000;
	const register unsigned long mask_pixel1_b = 0x001f0000;
	const register unsigned long mask_pixel2_r = 0x0000f800;
	const register unsigned long mask_pixel2_g = 0x000007e0;
	const register unsigned long mask_pixel2_b = 0x0000001f;
	src = reinterpret_cast<unsigned long*>(buffer1);
	stop = &src[width];
	do
	{
		do
		{
			// GL_RGBA
			pixel = *src;
			if (pixel == chromakey12)
			{
				*buffer2++ = null;
				*buffer2++ = null;
			}
			else
			{
				*src = chromakey12;
				if ( (pixel & mask_pixel1) == chromakey1)
				{
					*buffer2++ = null;
				}
				else
				{
					*buffer2++ = ( alpha                         | // A
					             ( pixel & mask_pixel1_b ) >>  5 | // B
					             ( pixel & mask_pixel1_g ) >>  3 | // G
					             ( pixel & mask_pixel1_r ));       // R
				}
				if ( (pixel & mask_pixel2) == chromakey2)
				{
					*buffer2++ = null;
				}
				else
				{
					*buffer2++ = ( alpha                           | // A
					             ( pixel & mask_pixel2_b ) << 11   | // B
					             ( pixel & mask_pixel2_g ) << 13   | // G
					             ( pixel & mask_pixel2_r ) << 16);   // R
				}
			}
			src++;
		} while (src != stop);
		src += stride;
		stop += jump;
	} while (--height);
	return TileUpdateState_TileDownloadToGPU; 
}

inline Framebuffer::TileUpdateState Framebuffer::Convert1555Kto8888(FxU16* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride)
{
	// Process two pixels at once
	register unsigned long pixel;
	register unsigned long x;
	register unsigned long* src = reinterpret_cast<unsigned long*>(buffer1);
	const unsigned long null = 0x00000000;
	register unsigned long dstpixel = null;
	const register unsigned long chromakey1 = m_ChromaKey.Scalar << 16;
	const register unsigned long chromakey2 = m_ChromaKey.Scalar;
	const register unsigned long chromakey12 = chromakey1 | chromakey2;
	const register unsigned long alpha = m_glAlpha.Scalar;
	const register unsigned long mask_pixel1 = 0xffff0000;
	const register unsigned long mask_pixel2 = 0x0000ffff;
	const register unsigned long mask_pixel1_r = 0x7c000000;
	const register unsigned long mask_pixel1_g = 0x03e00000;
	const register unsigned long mask_pixel1_b = 0x001f0000;
	const register unsigned long mask_pixel2_r = 0x00007c00;
	const register unsigned long mask_pixel2_g = 0x000003e0;
	const register unsigned long mask_pixel2_b = 0x0000001f;
	width >>= 1;
	stride >>= 1;
	do
	{
		x = width;
		do
		{
			// GL_RGBA
			pixel = *src;
			if (pixel == chromakey12)
			{
				*buffer2++ = null;
				*buffer2++ = null;
			}
			else
			{
				*src = chromakey12;
				if ( (pixel & mask_pixel1) == chromakey1)
				{
					*buffer2++ = null;
				}
				else
				{
					dstpixel = ( alpha                           | // A
					           ( pixel & mask_pixel1_b ) >>  5   | // B
					           ( pixel & mask_pixel1_g ) >>  2   | // G
					           ( pixel & mask_pixel1_r ) <<  1);   // R
					*buffer2++ = dstpixel;
				}
				if ( (pixel & mask_pixel2) == chromakey2)
				{
					*buffer2++ = null;
				}
				else
				{
					dstpixel = ( alpha                           | // A
					           ( pixel & mask_pixel2_b ) << 11   | // B
					           ( pixel & mask_pixel2_g ) << 14   | // G
					           ( pixel & mask_pixel2_r ) << 17);   // R
					*buffer2++ = dstpixel;
				}
			}
			src++;
		}	while (--x);
		src += stride;
	} while (--height);
	return dstpixel != null ? TileUpdateState_TileDownloadToGPU : TileUpdateState_TileEmpty; 
}

inline Framebuffer::TileUpdateState Framebuffer::ConvertARGB8888Kto8888(FxU32* buffer1, register FxU32* buffer2, register FxU32 width, register FxU32 height, register FxU32 stride)
{
	// Process two pixels at once
	const register unsigned long chromakey = m_ChromaKey.Scalar || (m_ChromaKey.Scalar << 16);
	register unsigned long pixel;
	register unsigned long* stop;
	register unsigned jump = width + stride;
	register unsigned long* src = buffer1;
	// check if tile must be processed in advance
	// to avoid useless writes to main memory
	// The tile should at least fit into the second level cache
	// so reading it again wouldn't hurt as much as doing needless writes
	register unsigned long h = height;
	stop = &src[width];
	do
	{
		do
		{
			pixel = *src++;
			if (pixel != chromakey) goto create_8888_texture;
		}	while (src != stop);
		src += stride;
		stop += jump;
	}	while (--h);
	return TileUpdateState_TileEmpty;
create_8888_texture:
	const register unsigned long alpha = m_glAlpha.Scalar;
	src = buffer1;
	stop = &src[width];
	do
	{
		do
		{
			// GL_RGBA
			pixel = *src;
			if (pixel == chromakey)
			{
				*buffer2++ = 0;
			}
			else
			{
				*src = chromakey;
				*buffer2++ = (pixel << 8) | alpha;
			}
			src++;
		} while (src != stop);
		src += stride;
		stop += jump;
	}	while (--height);
	return TileUpdateState_TileDownloadToGPU; 
}

inline Framebuffer::TileUpdateState Framebuffer::createTextureData(FxU32* texbuffer, FxU32 x, FxU32 y, FxU32 x_step, FxU32 y_step, int checksumIndex)
{
	FxU32 stride = (m_width - x_step);
	FxU32 index = x + y * m_width;
	if (m_framebuffer->WriteMode == GR_LFBWRITEMODE_565)
	{
#ifdef __ALTIVEC__
		if (UserConfig.VectorUnitType == OpenGLideVectorUnitType_Altivec)
		{
#ifdef OGL_FRAMEBUFFER
			const vector unsigned long c = m_tileChecksums[checksumIndex];
#endif
		
			TileUpdateState state = Convert565Kto8888_AV(&m_framebuffer->Address[index], texbuffer, x_step, y_step, stride, checksumIndex);
#ifdef OGL_FRAMEBUFFER
			GlideMsg("Tile %d (%d,%d)-(%d,%d) update state is %s (%vlx)->(%vlx)\n",
			         checksumIndex,
			         x, y, x_step, y_step,
			         (state ==TileUpdateState_TileDownloadToGPU) ? "DownLoadToGPU"
			          : ((state == TileUpdateState_TileDrawOnly) ? "DrawOnly" : "TileEmpty"),
			          c,
			          m_tileChecksums[checksumIndex]);
#endif
			return state;
		}
		else
#endif
			return Convert565Kto8888(&m_framebuffer->Address[index], texbuffer, x_step, y_step, stride);
	}
	else if (m_framebuffer->WriteMode == GR_LFBWRITEMODE_1555)
	{
		return Convert1555Kto8888(&m_framebuffer->Address[index], texbuffer, x_step, y_step, stride);
	}
	else if (m_framebuffer->WriteMode == GR_LFBWRITEMODE_888)
	{
		FxU32* framebuffer = &reinterpret_cast<FxU32*>(m_framebuffer->Address)[index];
		return ConvertARGB8888Kto8888(framebuffer, texbuffer, x_step, y_step, stride);
	}
	else
	{
		return TileUpdateState_TileEmpty;
	}
}
\ No newline at end of file
diff --git a/MacGLide/OpenGLide/GLExtensions.cpp b/MacGLide/OpenGLide/GLExtensions.cpp
index d14519f..888411b 100644
--- a/MacGLide/OpenGLide/GLExtensions.cpp
+++ b/MacGLide/OpenGLide/GLExtensions.cpp
@@ -65,7 +65,7 @@ stExtensionSupport glNecessaryExt[] =
     { "GL_SGIS_texture_edge_clamp",       OGL_EXT_DESIRED,    &dummyExtVariable,                    &InternalConfig.EXT_SGIS_texture_edge_clamp },
     { "GL_EXT_paletted_texture",          OGL_EXT_DESIRED,    &UserConfig.EXT_paletted_texture,     &InternalConfig.EXT_paletted_texture },
     { "GL_APPLE_packed_pixels",           OGL_EXT_REQUIRED,   &dummyExtVariable,                    &dummyExtVariable2 },
-    { "GL_APPLE_client_storage",          OGL_EXT_DESIRED,    &dummyExtVariable,                    &InternalConfig.EXT_Client_Storage },
+    { "GL_APPLE_client_storage",          OGL_EXT_DESIRED,    &UserConfig.APPLE_client_storage,       &InternalConfig.APPLE_client_storage },
     { "GL_EXT_compiled_vertex_array",     OGL_EXT_DESIRED,    &UserConfig.EXT_compiled_vertex_array,&InternalConfig.EXT_compiled_vertex_array },
     { "GL_ARB_texture_rectangle",         OGL_EXT_DESIRED,    &UserConfig.ARB_texture_rectangle,    &InternalConfig.ARB_texture_rectangle },
 #ifdef OPENGLIDE_SYSTEM_HAS_FOGCOORD
@@ -249,7 +249,7 @@ void GLExtensions(void)
 	GlideMsg("Maximum level of anisotropy = %d\n", MaxAnisotropyLevel);
 
 	// Since this a global setting, texture data must not be uploaded from temp buffers
-	if (InternalConfig.EXT_Client_Storage)
+	if (InternalConfig.APPLE_client_storage)
 	{
 		glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, true);
 		glReportError();
diff --git a/MacGLide/OpenGLide/GlideSettings.cpp b/MacGLide/OpenGLide/GlideSettings.cpp
index 979b13f..5a53739 100644
--- a/MacGLide/OpenGLide/GlideSettings.cpp
+++ b/MacGLide/OpenGLide/GlideSettings.cpp
@@ -194,7 +194,7 @@ void GlideSettings::defaults()
   EXT_secondary_color            = true;
 	EXT_SGIS_generate_mipmap       = true;
 	EXT_SGIS_texture_edge_clamp    = true;
-  EXT_Client_Storage             = true;
+  APPLE_client_storage             = true;
 	EXT_compiled_vertex_array      = true;
 	ARB_texture_rectangle          = true;
 	EXT_texture_filter_anisotropic = true;
@@ -255,6 +255,7 @@ GlideSettings::IOErr GlideSettings::read_settings()
 	get("PedanticFrameBufferEmulation", &PedanticFrameBufferEmulation);
 	get("EXT_compiled_vertex_array", &EXT_compiled_vertex_array);
 	get("ARB_texture_rectangle", &ARB_texture_rectangle);
+	get("APPLE_client_storage", &APPLE_client_storage);
 	get("ARB_multitexture", &ARB_multitexture);
 	get("EXT_paletted_texture", &EXT_paletted_texture);
 	get("EXT_clip_volume_hint", &EXT_clip_volume_hint);
@@ -502,6 +503,7 @@ GlideSettings::IOErr GlideSettings::saveSettings()
 		put();
 		put("EXT_compiled_vertex_array", EXT_compiled_vertex_array);
 		put("ARB_texture_rectangle", ARB_texture_rectangle);
+		put("APPLE_client_storage", APPLE_client_storage);
 		put("ARB_multitexture", ARB_multitexture);
 		put("EXT_paletted_texture", EXT_paletted_texture);
 		put("EXT_clip_volume_hint", EXT_clip_volume_hint);
diff --git a/MacGLide/OpenGLide/GlideSettings.h b/MacGLide/OpenGLide/GlideSettings.h
index dc0f8c3..388e2e0 100644
--- a/MacGLide/OpenGLide/GlideSettings.h
+++ b/MacGLide/OpenGLide/GlideSettings.h
@@ -98,7 +98,7 @@ struct ConfigStruct
 	bool EXT_paletted_texture;
 	bool EXT_SGIS_generate_mipmap;
 	bool EXT_SGIS_texture_edge_clamp;
-	bool EXT_Client_Storage;
+	bool APPLE_client_storage;
 	bool EXT_compiled_vertex_array;
 	bool ARB_texture_rectangle;
 #ifdef OPENGLIDE_SYSTEM_HAS_FOGCOORD
diff --git a/MacGLide/OpenGLide/PGTexture.cpp b/MacGLide/OpenGLide/PGTexture.cpp
index 6e6aed3..f3a4352 100644
--- a/MacGLide/OpenGLide/PGTexture.cpp
+++ b/MacGLide/OpenGLide/PGTexture.cpp
@@ -60,7 +60,7 @@ void PGTexture::genPaletteMipmaps( FxU32 width, FxU32 height, const FxU8 *data )
     }
 }
 
-PGTexture::PGTexture( int mem_size )
+PGTexture::PGTexture(int mem_size)
 {
 	m_db = new TexDB( mem_size );
 	m_palette_dirty = true;
@@ -68,8 +68,17 @@ PGTexture::PGTexture( int mem_size )
 	m_chromakey_mode = GR_CHROMAKEY_DISABLE;
 	m_tex_memory_size = mem_size;
 	m_memory = (FxU8*) AllocBuffer(mem_size, sizeof(FxU8));
-	if (InternalConfig.EXT_Client_Storage)
+	// This is initialized before the rendering context is available and before
+	// the list of OpenGL extensions can be checked against APPLE_client_storage.
+	// As a result the texture buffer is allocated bases on the user setting.
+	// And it might be legal to download textures before grWinOpen() so
+	// using the user config setting is the best we can do
+	if (UserConfig.APPLE_client_storage)
 	{
+		// use client storage to avoid OpenGL-internal copy 
+		// OpenGL may still make a copy if the color format isn't supported natively
+		// by the graphics card but all xto8888 conversions should benefit from this
+		// @todo: Not true for OSX, but for now we stay compatible to native OS9
 		// Alloc 4-times the size of the Glide buffer for texture storage
 		m_textureCache = reinterpret_cast<FxU32*>(AllocBuffer(m_tex_memory_size, sizeof(FxU32)));
 		// No memory is wasted here because we save the internal OpenGL copy
@@ -623,14 +632,9 @@ bool PGTexture::MakeReady(TTextureStruct* tex_coords, unsigned long number_of_tr
 		}
 		glReportError();	
 
-		// use client storage to avoid OpenGL-internal copy 
-		// OpenGL may still make a copy if the color format isn't supported natively
-		// by the graphics card but all xto8888 conversions should benefit from this
-		// @todo: Not true for OSX, but for now we stay compatible to native OS9
-		const bool useClientStorage = InternalConfig.EXT_Client_Storage /* && !subtexcoords */;
 		if (subtexcoords)
 		{
-			// EXT_Client_Storage doesn't explicitely forbid to adjust pixel unpack :^)
+			// APPLE_client_storage doesn't explicitely forbid to adjust pixel unpack :^)
 			glPixelStorei(GL_UNPACK_SKIP_PIXELS, subtexcoords->left);
 			glPixelStorei(GL_UNPACK_SKIP_ROWS, subtexcoords->top);
 			glPixelStorei(GL_UNPACK_ROW_LENGTH, texVals.width);
@@ -640,11 +644,9 @@ bool PGTexture::MakeReady(TTextureStruct* tex_coords, unsigned long number_of_tr
 		}
 		FxU32* texBuffer;
 		// Which buffer
-		if (useClientStorage)
+		if (m_textureCache)
 		{
 			texBuffer = &m_textureCache[m_startAddress];
-			// glPixelStorei(GL_UNPACK_CLIENT_STORAGE_APPLE, true);
-			// glReportError();		
 		}
 		else
 		{
@@ -658,7 +660,7 @@ bool PGTexture::MakeReady(TTextureStruct* tex_coords, unsigned long number_of_tr
 			{
 				// Read about anisotropy and chromakey issues in macFormatConversions.cpp
 				Convert565Kto8888((FxU16*)data, m_chromakey_value_565, texBuffer, texVals.nPixels);
-				DownloadMipmapsToOpenGL(4, GL_RGBA, GL_UNSIGNED_BYTE, m_tex_temp, texVals, !use_mipmap_ext);          	
+				DownloadMipmapsToOpenGL(4, GL_RGBA, GL_UNSIGNED_BYTE, texBuffer, texVals, !use_mipmap_ext);          	
 			}
 			else
 			{