/*
===========================================================================
Copyright (C) 1999-2005 Id Software, Inc.

This file is part of Quake III Arena source code.

Quake III Arena source code is free software; you can redistribute it
and/or modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the License,
or (at your option) any later version.

Quake III Arena source code is distributed in the hope that it will be
useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with Quake III Arena source code; if not, write to the Free Software
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
===========================================================================
*/
// tr_map.c

#include "tr_local.h"

#define JSON_IMPLEMENTATION
#include "../qcommon/json.h"
#undef JSON_IMPLEMENTATION

/*

Loads and prepares a map file for scene rendering.

A single entry point:

void RE_LoadWorldMap( const char *name );

*/

static	world_t		s_worldData;
static	byte		*fileBase;

static int	c_gridVerts;

//===============================================================================

static void HSVtoRGB( float h, float s, float v, float rgb[3] )
{
	int i;
	float f;
	float p, q, t;

	h *= 5;

	i = floor( h );
	f = h - i;

	p = v * ( 1 - s );
	q = v * ( 1 - s * f );
	t = v * ( 1 - s * ( 1 - f ) );

	switch ( i )
	{
	case 0:
		rgb[0] = v;
		rgb[1] = t;
		rgb[2] = p;
		break;
	case 1:
		rgb[0] = q;
		rgb[1] = v;
		rgb[2] = p;
		break;
	case 2:
		rgb[0] = p;
		rgb[1] = v;
		rgb[2] = t;
		break;
	case 3:
		rgb[0] = p;
		rgb[1] = q;
		rgb[2] = v;
		break;
	case 4:
		rgb[0] = t;
		rgb[1] = p;
		rgb[2] = v;
		break;
	case 5:
		rgb[0] = v;
		rgb[1] = p;
		rgb[2] = q;
		break;
	}
}

/*
===============
R_ColorShiftLightingBytes

===============
*/
static	void R_ColorShiftLightingBytes( const byte in[4], byte out[4] ) {
	int		shift, r, g, b;

	// shift the color data based on overbright range
	shift = r_mapOverBrightBits->integer - tr.overbrightBits;

	// shift the data based on overbright range
	r = in[0] << shift;
	g = in[1] << shift;
	b = in[2] << shift;
	
	// normalize by color instead of saturating to white
	if ( ( r | g | b ) > 255 ) {
		int		max;

		max = r > g ? r : g;
		max = max > b ? max : b;
		r = r * 255 / max;
		g = g * 255 / max;
		b = b * 255 / max;
	}

	out[0] = r;
	out[1] = g;
	out[2] = b;
	out[3] = in[3];
}


/*
===============
R_ColorShiftLightingFloats

===============
*/
static void R_ColorShiftLightingFloats(const float in[4], float out[4])
{
	float	r, g, b;
	float   scale = (1 << (r_mapOverBrightBits->integer - tr.overbrightBits)) / 255.0f;

	r = in[0] * scale;
	g = in[1] * scale;
	b = in[2] * scale;

	// normalize by color instead of saturating to white
	if ( r > 1 || g > 1 || b > 1 ) {
		float	max;

		max = r > g ? r : g;
		max = max > b ? max : b;
		r = r / max;
		g = g / max;
		b = b / max;
	}

	out[0] = r;
	out[1] = g;
	out[2] = b;
	out[3] = in[3];
}

// Modified from http://graphicrants.blogspot.jp/2009/04/rgbm-color-encoding.html
void ColorToRGBM(const vec3_t color, unsigned char rgbm[4])
{
	vec3_t          sample;
	float			maxComponent;

	VectorCopy(color, sample);

	maxComponent = MAX(sample[0], sample[1]);
	maxComponent = MAX(maxComponent, sample[2]);
	maxComponent = CLAMP(maxComponent, 1.0f/255.0f, 1.0f);

	rgbm[3] = (unsigned char) ceil(maxComponent * 255.0f);
	maxComponent = 255.0f / rgbm[3];

	VectorScale(sample, maxComponent, sample);

	rgbm[0] = (unsigned char) (sample[0] * 255);
	rgbm[1] = (unsigned char) (sample[1] * 255);
	rgbm[2] = (unsigned char) (sample[2] * 255);
}

static void ColorToRGB16(const vec3_t color, uint16_t rgb16[3])
{
	rgb16[0] = color[0] * 65535.0f + 0.5f;
	rgb16[1] = color[1] * 65535.0f + 0.5f;
	rgb16[2] = color[2] * 65535.0f + 0.5f;
}


/*
===============
R_LoadLightmaps

===============
*/
#define	DEFAULT_LIGHTMAP_SIZE	128
static	void R_LoadLightmaps( const lump_t *l, const lump_t *surfs ) {
	imgFlags_t  imgFlags = IMGFLAG_NOLIGHTSCALE | IMGFLAG_NO_COMPRESSION | IMGFLAG_CLAMPTOEDGE;
	byte		*buf, *buf_p;
	dsurface_t  *surf;
	int			len;
	byte		*image;
	int			i, j, numLightmaps, textureInternalFormat = 0;
	int			numLightmapsPerPage = 16;
	float maxIntensity = 0;
	double sumIntensity = 0;

	// if we are in r_vertexLight mode, we don't need the lightmaps at all
	if ( ( r_vertexLight->integer && tr.vertexLightingAllowed ) || glConfig.hardwareType == GLHW_PERMEDIA2 ) {
		return;
	}

	len = l->filelen;
	if ( !len ) {
		return;
	}
	buf = fileBase + l->fileofs;

	// we are about to upload textures
	R_IssuePendingRenderCommands();

	tr.lightmapSize = DEFAULT_LIGHTMAP_SIZE;
	numLightmaps = len / (tr.lightmapSize * tr.lightmapSize * 3);

	// check for deluxe mapping
	if (numLightmaps <= 1)
	{
		tr.worldDeluxeMapping = qfalse;
	}
	else
	{
		tr.worldDeluxeMapping = qtrue;
		for( i = 0, surf = (dsurface_t *)(fileBase + surfs->fileofs);
			i < surfs->filelen / sizeof(dsurface_t); i++, surf++ ) {
			int lightmapNum = LittleLong( surf->lightmapNum );

			if ( lightmapNum >= 0 && (lightmapNum & 1) != 0 ) {
				tr.worldDeluxeMapping = qfalse;
				break;
			}
		}
	}

	image = ri.Malloc(tr.lightmapSize * tr.lightmapSize * 4 * 2);

	if (tr.worldDeluxeMapping)
		numLightmaps >>= 1;

	// Use fat lightmaps of an appropriate size.
	if (r_mergeLightmaps->integer)
	{
		int maxLightmapsPerAxis = glConfig.maxTextureSize / tr.lightmapSize;
		int lightmapCols = 4, lightmapRows = 4;

		// Increase width at first, then height.
		while (lightmapCols * lightmapRows < numLightmaps && lightmapCols != maxLightmapsPerAxis)
			lightmapCols <<= 1;

		while (lightmapCols * lightmapRows < numLightmaps && lightmapRows != maxLightmapsPerAxis)
			lightmapRows <<= 1;

		tr.fatLightmapCols  = lightmapCols;
		tr.fatLightmapRows  = lightmapRows;
		numLightmapsPerPage = lightmapCols * lightmapRows;

		tr.numLightmaps = (numLightmaps + (numLightmapsPerPage - 1)) / numLightmapsPerPage;
	}
	else
	{
		tr.numLightmaps = numLightmaps;
	}

	tr.lightmaps = ri.Hunk_Alloc( tr.numLightmaps * sizeof(image_t *), h_low );

	if (tr.worldDeluxeMapping)
		tr.deluxemaps = ri.Hunk_Alloc( tr.numLightmaps * sizeof(image_t *), h_low );

	textureInternalFormat = GL_RGBA8;
	if (r_hdr->integer)
	{
		// Check for the first hdr lightmap, if it exists, use GL_RGBA16 for textures.
		char filename[MAX_QPATH];

		Com_sprintf(filename, sizeof(filename), "maps/%s/lm_0000.hdr", s_worldData.baseName);
		if (ri.FS_FileExists(filename))
			textureInternalFormat = GL_RGBA16;
	}

	if (r_mergeLightmaps->integer)
	{
		int width  = tr.fatLightmapCols * tr.lightmapSize;
		int height = tr.fatLightmapRows * tr.lightmapSize;

		for (i = 0; i < tr.numLightmaps; i++)
		{
			tr.lightmaps[i] = R_CreateImage(va("_fatlightmap%d", i), NULL, width, height, IMGTYPE_COLORALPHA, imgFlags, textureInternalFormat);

			if (tr.worldDeluxeMapping)
				tr.deluxemaps[i] = R_CreateImage(va("_fatdeluxemap%d", i), NULL, width, height, IMGTYPE_DELUXE, imgFlags, 0);
		}
	}

	for(i = 0; i < numLightmaps; i++)
	{
		int xoff = 0, yoff = 0;
		int lightmapnum = i;
		// expand the 24 bit on-disk to 32 bit

		if (r_mergeLightmaps->integer)
		{
			int lightmaponpage = i % numLightmapsPerPage;
			xoff = (lightmaponpage % tr.fatLightmapCols) * tr.lightmapSize;
			yoff = (lightmaponpage / tr.fatLightmapCols) * tr.lightmapSize;

			lightmapnum /= numLightmapsPerPage;
		}

		// if (tr.worldLightmapping)
		{
			char filename[MAX_QPATH];
			byte *hdrLightmap = NULL;
			int size = 0;

			// look for hdr lightmaps
			if (textureInternalFormat == GL_RGBA16)
			{
				Com_sprintf( filename, sizeof( filename ), "maps/%s/lm_%04d.hdr", s_worldData.baseName, i * (tr.worldDeluxeMapping ? 2 : 1) );
				//ri.Printf(PRINT_ALL, "looking for %s\n", filename);

				size = ri.FS_ReadFile(filename, (void **)&hdrLightmap);
			}

			if (hdrLightmap)
			{
				byte *p = hdrLightmap, *end = hdrLightmap + size;
				//ri.Printf(PRINT_ALL, "found!\n");
				
				/* FIXME: don't just skip over this header and actually parse it */
				while (p < end && !(*p == '\n' && *(p+1) == '\n'))
					p++;

				p += 2;
				
				while (p < end && !(*p == '\n'))
					p++;

				p++;

				if (p >= end)
					ri.Error(ERR_DROP, "Bad header for %s!", filename);

				buf_p = p;

#if 0 // HDRFILE_RGBE
				if ((int)(end - hdrLightmap) != tr.lightmapSize * tr.lightmapSize * 4)
					ri.Error(ERR_DROP, "Bad size for %s (%i)!", filename, size);
#else // HDRFILE_FLOAT
				if ((int)(end - hdrLightmap) != tr.lightmapSize * tr.lightmapSize * 12)
					ri.Error(ERR_DROP, "Bad size for %s (%i)!", filename, size);
#endif
			}
			else
			{
				int imgOffset = tr.worldDeluxeMapping ? i * 2 : i;
				buf_p = buf + imgOffset * tr.lightmapSize * tr.lightmapSize * 3;
			}

			for ( j = 0 ; j < tr.lightmapSize * tr.lightmapSize; j++ ) 
			{
				if (hdrLightmap)
				{
					vec4_t color;

#if 0 // HDRFILE_RGBE
					float exponent = exp2(buf_p[j*4+3] - 128);

					color[0] = buf_p[j*4+0] * exponent;
					color[1] = buf_p[j*4+1] * exponent;
					color[2] = buf_p[j*4+2] * exponent;
#else // HDRFILE_FLOAT
					memcpy(color, &buf_p[j*12], 12);

					color[0] = LittleFloat(color[0]);
					color[1] = LittleFloat(color[1]);
					color[2] = LittleFloat(color[2]);
#endif
					color[3] = 1.0f;

					R_ColorShiftLightingFloats(color, color);

					ColorToRGB16(color, (uint16_t *)(&image[j * 8]));
					((uint16_t *)(&image[j * 8]))[3] = 65535;
				}
				else if (textureInternalFormat == GL_RGBA16)
				{
					vec4_t color;

					//hack: convert LDR lightmap to HDR one
					color[0] = MAX(buf_p[j*3+0], 0.499f);
					color[1] = MAX(buf_p[j*3+1], 0.499f);
					color[2] = MAX(buf_p[j*3+2], 0.499f);

					// if under an arbitrary value (say 12) grey it out
					// this prevents weird splotches in dimly lit areas
					if (color[0] + color[1] + color[2] < 12.0f)
					{
						float avg = (color[0] + color[1] + color[2]) * 0.3333f;
						color[0] = avg;
						color[1] = avg;
						color[2] = avg;
					}
					color[3] = 1.0f;

					R_ColorShiftLightingFloats(color, color);

					ColorToRGB16(color, (uint16_t *)(&image[j * 8]));
					((uint16_t *)(&image[j * 8]))[3] = 65535;
				}
				else
				{
					if ( r_lightmap->integer == 2 )
					{	// color code by intensity as development tool	(FIXME: check range)
						float r = buf_p[j*3+0];
						float g = buf_p[j*3+1];
						float b = buf_p[j*3+2];
						float intensity;
						float out[3] = {0.0, 0.0, 0.0};

						intensity = 0.33f * r + 0.685f * g + 0.063f * b;

						if ( intensity > 255 )
							intensity = 1.0f;
						else
							intensity /= 255.0f;

						if ( intensity > maxIntensity )
							maxIntensity = intensity;

						HSVtoRGB( intensity, 1.00, 0.50, out );

						image[j*4+0] = out[0] * 255;
						image[j*4+1] = out[1] * 255;
						image[j*4+2] = out[2] * 255;
						image[j*4+3] = 255;

						sumIntensity += intensity;
					}
					else
					{
						R_ColorShiftLightingBytes( &buf_p[j*3], &image[j*4] );
						image[j*4+3] = 255;
					}
				}
			}

			if (r_mergeLightmaps->integer)
				R_UpdateSubImage(tr.lightmaps[lightmapnum], image, xoff, yoff, tr.lightmapSize, tr.lightmapSize, textureInternalFormat);
			else
				tr.lightmaps[i] = R_CreateImage(va("*lightmap%d", i), image, tr.lightmapSize, tr.lightmapSize, IMGTYPE_COLORALPHA, imgFlags, textureInternalFormat );

			if (hdrLightmap)
				ri.FS_FreeFile(hdrLightmap);
		}

		if (tr.worldDeluxeMapping)
		{
			buf_p = buf + (i * 2 + 1) * tr.lightmapSize * tr.lightmapSize * 3;

			for ( j = 0 ; j < tr.lightmapSize * tr.lightmapSize; j++ ) {
				image[j*4+0] = buf_p[j*3+0];
				image[j*4+1] = buf_p[j*3+1];
				image[j*4+2] = buf_p[j*3+2];

				// make 0,0,0 into 127,127,127
				if ((image[j*4+0] == 0) && (image[j*4+1] == 0) && (image[j*4+2] == 0))
				{
					image[j*4+0] =
					image[j*4+1] =
					image[j*4+2] = 127;
				}

				image[j*4+3] = 255;
			}

			if (r_mergeLightmaps->integer)
				R_UpdateSubImage(tr.deluxemaps[lightmapnum], image, xoff, yoff, tr.lightmapSize, tr.lightmapSize, GL_RGBA8 );
			else
				tr.deluxemaps[i] = R_CreateImage(va("*deluxemap%d", i), image, tr.lightmapSize, tr.lightmapSize, IMGTYPE_DELUXE, imgFlags, 0 );
		}
	}

	if ( r_lightmap->integer == 2 )	{
		ri.Printf( PRINT_ALL, "Brightest lightmap value: %d\n", ( int ) ( maxIntensity * 255 ) );
	}

	ri.Free(image);
}


static float FatPackU(float input, int lightmapnum)
{
	if (lightmapnum < 0)
		return input;

	if (tr.worldDeluxeMapping)
		lightmapnum >>= 1;

	if (tr.fatLightmapCols > 0)
	{
		lightmapnum %= (tr.fatLightmapCols * tr.fatLightmapRows);
		return (input + (lightmapnum % tr.fatLightmapCols)) / (float)(tr.fatLightmapCols);
	}

	return input;
}

static float FatPackV(float input, int lightmapnum)
{
	if (lightmapnum < 0)
		return input;

	if (tr.worldDeluxeMapping)
		lightmapnum >>= 1;

	if (tr.fatLightmapCols > 0)
	{
		lightmapnum %= (tr.fatLightmapCols * tr.fatLightmapRows);
		return (input + (lightmapnum / tr.fatLightmapCols)) / (float)(tr.fatLightmapRows);
	}

	return input;
}


static int FatLightmap(int lightmapnum)
{
	if (lightmapnum < 0)
		return lightmapnum;

	if (tr.worldDeluxeMapping)
		lightmapnum >>= 1;

	if (tr.fatLightmapCols > 0)
		return lightmapnum / (tr.fatLightmapCols * tr.fatLightmapRows);
	
	return lightmapnum;
}

/*
=================
RE_SetWorldVisData

This is called by the clipmodel subsystem so we can share the 1.8 megs of
space in big maps...
=================
*/
void		RE_SetWorldVisData( const byte *vis ) {
	tr.externalVisData = vis;
}


/*
=================
R_LoadVisibility
=================
*/
static	void R_LoadVisibility( const lump_t *l ) {
	int		len;
	byte	*buf;

	len = l->filelen;
	if ( !len ) {
		return;
	}
	buf = fileBase + l->fileofs;

	s_worldData.numClusters = LittleLong( ((int *)buf)[0] );
	s_worldData.clusterBytes = LittleLong( ((int *)buf)[1] );

	// CM_Load should have given us the vis data to share, so
	// we don't need to allocate another copy
	if ( tr.externalVisData ) {
		s_worldData.vis = tr.externalVisData;
	} else {
		byte	*dest;

		dest = ri.Hunk_Alloc( len - 8, h_low );
		Com_Memcpy( dest, buf + 8, len - 8 );
		s_worldData.vis = dest;
	}
}

//===============================================================================


/*
===============
ShaderForShaderNum
===============
*/
static shader_t *ShaderForShaderNum( int shaderNum, int lightmapNum ) {
	shader_t	*shader;
	dshader_t	*dsh;

	int _shaderNum = LittleLong( shaderNum );
	if ( _shaderNum < 0 || _shaderNum >= s_worldData.numShaders ) {
		ri.Error( ERR_DROP, "ShaderForShaderNum: bad num %i", _shaderNum );
	}
	dsh = &s_worldData.shaders[ _shaderNum ];


	if ( ( r_vertexLight->integer && tr.vertexLightingAllowed ) || glConfig.hardwareType == GLHW_PERMEDIA2 ) {
		lightmapNum = LIGHTMAP_BY_VERTEX;
	}

	if ( r_fullbright->integer ) {
		lightmapNum = LIGHTMAP_WHITEIMAGE;
	}

	shader = R_FindShader( dsh->shader, lightmapNum, qtrue );

	// if the shader had errors, just use default shader
	if ( shader->defaultShader ) {
		return tr.defaultShader;
	}

	return shader;
}

static void LoadDrawVertToSrfVert(srfVert_t *s, const drawVert_t *d, int realLightmapNum, float hdrVertColors[3], vec3_t *bounds)
{
	vec4_t v;

	s->xyz[0] = LittleFloat(d->xyz[0]);
	s->xyz[1] = LittleFloat(d->xyz[1]);
	s->xyz[2] = LittleFloat(d->xyz[2]);

	if (bounds)
		AddPointToBounds(s->xyz, bounds[0], bounds[1]);

	s->st[0] = LittleFloat(d->st[0]);
	s->st[1] = LittleFloat(d->st[1]);

	if (realLightmapNum >= 0)
	{
		s->lightmap[0] = FatPackU(LittleFloat(d->lightmap[0]), realLightmapNum);
		s->lightmap[1] = FatPackV(LittleFloat(d->lightmap[1]), realLightmapNum);
	}
	else
	{
		s->lightmap[0] = LittleFloat(d->lightmap[0]);
		s->lightmap[1] = LittleFloat(d->lightmap[1]);
	}

	v[0] = LittleFloat(d->normal[0]);
	v[1] = LittleFloat(d->normal[1]);
	v[2] = LittleFloat(d->normal[2]);

	R_VaoPackNormal(s->normal, v);

	if (hdrVertColors)
	{
		v[0] = hdrVertColors[0];
		v[1] = hdrVertColors[1];
		v[2] = hdrVertColors[2];
	}
	else
	{
		//hack: convert LDR vertex colors to HDR
		if (r_hdr->integer)
		{
			v[0] = MAX(d->color.rgba[0], 0.499f);
			v[1] = MAX(d->color.rgba[1], 0.499f);
			v[2] = MAX(d->color.rgba[2], 0.499f);
		}
		else
		{
			v[0] = d->color.rgba[0];
			v[1] = d->color.rgba[1];
			v[2] = d->color.rgba[2];
		}

	}
	v[3] = d->color.rgba[3] / 255.0f;

	R_ColorShiftLightingFloats(v, v);
	R_VaoPackColor(s->color, v);
}


/*
===============
ParseFace
===============
*/
static void ParseFace( const dsurface_t *ds, const drawVert_t *verts, float *hdrVertColors, msurface_t *surf, int *indexes  ) {
	int			i, j;
	srfBspSurface_t	*cv;
	glIndex_t  *tri;
	int			numVerts, numIndexes, badTriangles;
	int realLightmapNum;

	realLightmapNum = LittleLong( ds->lightmapNum );

	// get fog volume
	surf->fogIndex = LittleLong( ds->fogNum ) + 1;

	// get shader value
	surf->shader = ShaderForShaderNum( ds->shaderNum, FatLightmap(realLightmapNum) );
	if ( r_singleShader->integer && !surf->shader->isSky ) {
		surf->shader = tr.defaultShader;
	}

	numVerts = LittleLong(ds->numVerts);
	if (numVerts > MAX_FACE_POINTS) {
		ri.Printf( PRINT_WARNING, "WARNING: MAX_FACE_POINTS exceeded: %i\n", numVerts);
		numVerts = MAX_FACE_POINTS;
		surf->shader = tr.defaultShader;
	}

	numIndexes = LittleLong(ds->numIndexes);

	//cv = ri.Hunk_Alloc(sizeof(*cv), h_low);
	cv = (void *)surf->data;
	cv->surfaceType = SF_FACE;

	cv->numIndexes = numIndexes;
	cv->indexes = ri.Hunk_Alloc(numIndexes * sizeof(cv->indexes[0]), h_low);

	cv->numVerts = numVerts;
	cv->verts = ri.Hunk_Alloc(numVerts * sizeof(cv->verts[0]), h_low);

	// copy vertexes
	surf->cullinfo.type = CULLINFO_PLANE | CULLINFO_BOX;
	ClearBounds(surf->cullinfo.bounds[0], surf->cullinfo.bounds[1]);
	verts += LittleLong(ds->firstVert);
	for(i = 0; i < numVerts; i++)
		LoadDrawVertToSrfVert(&cv->verts[i], &verts[i], realLightmapNum, hdrVertColors ? hdrVertColors + (ds->firstVert + i) * 3 : NULL, surf->cullinfo.bounds);

	// copy triangles
	badTriangles = 0;
	indexes += LittleLong(ds->firstIndex);
	for(i = 0, tri = cv->indexes; i < numIndexes; i += 3, tri += 3)
	{
		for(j = 0; j < 3; j++)
		{
			tri[j] = LittleLong(indexes[i + j]);

			if(tri[j] >= numVerts)
			{
				ri.Error(ERR_DROP, "Bad index in face surface");
			}
		}

		if ((tri[0] == tri[1]) || (tri[1] == tri[2]) || (tri[0] == tri[2]))
		{
			tri -= 3;
			badTriangles++;
		}
	}

	if (badTriangles)
	{
		ri.Printf(PRINT_WARNING, "Face has bad triangles, originally shader %s %d tris %d verts, now %d tris\n", surf->shader->name, numIndexes / 3, numVerts, numIndexes / 3 - badTriangles);
		cv->numIndexes -= badTriangles * 3;
	}

	// take the plane information from the lightmap vector
	for ( i = 0 ; i < 3 ; i++ ) {
		cv->cullPlane.normal[i] = LittleFloat( ds->lightmapVecs[2][i] );
	}
	cv->cullPlane.dist = DotProduct( cv->verts[0].xyz, cv->cullPlane.normal );
	SetPlaneSignbits( &cv->cullPlane );
	cv->cullPlane.type = PlaneTypeForNormal( cv->cullPlane.normal );
	surf->cullinfo.plane = cv->cullPlane;

	surf->data = (surfaceType_t *)cv;

	// Calculate tangent spaces
	{
		srfVert_t      *dv[3];

		for(i = 0, tri = cv->indexes; i < numIndexes; i += 3, tri += 3)
		{
			dv[0] = &cv->verts[tri[0]];
			dv[1] = &cv->verts[tri[1]];
			dv[2] = &cv->verts[tri[2]];

			R_CalcTangentVectors(dv);
		}
	}
}


/*
===============
ParseMesh
===============
*/
static void ParseMesh ( const dsurface_t *ds, const drawVert_t *verts, float *hdrVertColors, msurface_t *surf ) {
	srfBspSurface_t	*grid = (srfBspSurface_t *)surf->data;
	int				i;
	int				width, height, numPoints;
	srfVert_t points[MAX_PATCH_SIZE*MAX_PATCH_SIZE];
	vec3_t			bounds[2];
	vec3_t			tmpVec;
	static surfaceType_t	skipData = SF_SKIP;
	int realLightmapNum;

	realLightmapNum = LittleLong( ds->lightmapNum );

	// get fog volume
	surf->fogIndex = LittleLong( ds->fogNum ) + 1;

	// get shader value
	surf->shader = ShaderForShaderNum( ds->shaderNum, FatLightmap(realLightmapNum) );
	if ( r_singleShader->integer && !surf->shader->isSky ) {
		surf->shader = tr.defaultShader;
	}

	// we may have a nodraw surface, because they might still need to
	// be around for movement clipping
	if ( s_worldData.shaders[ LittleLong( ds->shaderNum ) ].surfaceFlags & SURF_NODRAW ) {
		surf->data = &skipData;
		return;
	}

	width = LittleLong( ds->patchWidth );
	height = LittleLong( ds->patchHeight );

	if(width < 0 || width > MAX_PATCH_SIZE || height < 0 || height > MAX_PATCH_SIZE)
		ri.Error(ERR_DROP, "ParseMesh: bad size");

	verts += LittleLong( ds->firstVert );
	numPoints = width * height;
	for(i = 0; i < numPoints; i++)
		LoadDrawVertToSrfVert(&points[i], &verts[i], realLightmapNum, hdrVertColors ? hdrVertColors + (ds->firstVert + i) * 3 : NULL, NULL);

	// pre-tesseleate
	R_SubdividePatchToGrid( grid, width, height, points );

	// copy the level of detail origin, which is the center
	// of the group of all curves that must subdivide the same
	// to avoid cracking
	for ( i = 0 ; i < 3 ; i++ ) {
		bounds[0][i] = LittleFloat( ds->lightmapVecs[0][i] );
		bounds[1][i] = LittleFloat( ds->lightmapVecs[1][i] );
	}
	VectorAdd( bounds[0], bounds[1], bounds[1] );
	VectorScale( bounds[1], 0.5f, grid->lodOrigin );
	VectorSubtract( bounds[0], grid->lodOrigin, tmpVec );
	grid->lodRadius = VectorLength( tmpVec );

	surf->cullinfo.type = CULLINFO_BOX | CULLINFO_SPHERE;
	VectorCopy(grid->cullBounds[0], surf->cullinfo.bounds[0]);
	VectorCopy(grid->cullBounds[1], surf->cullinfo.bounds[1]);
	VectorCopy(grid->cullOrigin, surf->cullinfo.localOrigin);
	surf->cullinfo.radius = grid->cullRadius;
}

/*
===============
ParseTriSurf
===============
*/
static void ParseTriSurf( const dsurface_t *ds, const drawVert_t *verts, float *hdrVertColors, msurface_t *surf, int *indexes ) {
	srfBspSurface_t *cv;
	glIndex_t  *tri;
	int             i, j;
	int             numVerts, numIndexes, badTriangles;

	// get fog volume
	surf->fogIndex = LittleLong( ds->fogNum ) + 1;

	// get shader
	surf->shader = ShaderForShaderNum( ds->shaderNum, LIGHTMAP_BY_VERTEX );
	if ( r_singleShader->integer && !surf->shader->isSky ) {
		surf->shader = tr.defaultShader;
	}

	numVerts = LittleLong(ds->numVerts);
	numIndexes = LittleLong(ds->numIndexes);

	//cv = ri.Hunk_Alloc(sizeof(*cv), h_low);
	cv = (void *)surf->data;
	cv->surfaceType = SF_TRIANGLES;

	cv->numIndexes = numIndexes;
	cv->indexes = ri.Hunk_Alloc(numIndexes * sizeof(cv->indexes[0]), h_low);

	cv->numVerts = numVerts;
	cv->verts = ri.Hunk_Alloc(numVerts * sizeof(cv->verts[0]), h_low);

	surf->data = (surfaceType_t *) cv;

	// copy vertexes
	surf->cullinfo.type = CULLINFO_BOX;
	ClearBounds(surf->cullinfo.bounds[0], surf->cullinfo.bounds[1]);
	verts += LittleLong(ds->firstVert);
	for(i = 0; i < numVerts; i++)
		LoadDrawVertToSrfVert(&cv->verts[i], &verts[i], -1, hdrVertColors ? hdrVertColors + (ds->firstVert + i) * 3 : NULL, surf->cullinfo.bounds);

	// copy triangles
	badTriangles = 0;
	indexes += LittleLong(ds->firstIndex);
	for(i = 0, tri = cv->indexes; i < numIndexes; i += 3, tri += 3)
	{
		for(j = 0; j < 3; j++)
		{
			tri[j] = LittleLong(indexes[i + j]);

			if(tri[j] >= numVerts)
			{
				ri.Error(ERR_DROP, "Bad index in face surface");
			}
		}

		if ((tri[0] == tri[1]) || (tri[1] == tri[2]) || (tri[0] == tri[2]))
		{
			tri -= 3;
			badTriangles++;
		}
	}

	if (badTriangles)
	{
		ri.Printf(PRINT_WARNING, "Trisurf has bad triangles, originally shader %s %d tris %d verts, now %d tris\n", surf->shader->name, numIndexes / 3, numVerts, numIndexes / 3 - badTriangles);
		cv->numIndexes -= badTriangles * 3;
	}

	// Calculate tangent spaces
	{
		srfVert_t      *dv[3];

		for(i = 0, tri = cv->indexes; i < numIndexes; i += 3, tri += 3)
		{
			dv[0] = &cv->verts[tri[0]];
			dv[1] = &cv->verts[tri[1]];
			dv[2] = &cv->verts[tri[2]];

			R_CalcTangentVectors(dv);
		}
	}
}

/*
===============
ParseFlare
===============
*/
static void ParseFlare( const dsurface_t *ds, const drawVert_t *verts, msurface_t *surf, int *indexes ) {
	srfFlare_t		*flare;
	int				i;

	// get fog volume
	surf->fogIndex = LittleLong( ds->fogNum ) + 1;

	// get shader
	surf->shader = ShaderForShaderNum( ds->shaderNum, LIGHTMAP_BY_VERTEX );
	if ( r_singleShader->integer && !surf->shader->isSky ) {
		surf->shader = tr.defaultShader;
	}

	//flare = ri.Hunk_Alloc( sizeof( *flare ), h_low );
	flare = (void *)surf->data;
	flare->surfaceType = SF_FLARE;

	surf->data = (surfaceType_t *)flare;

	for ( i = 0 ; i < 3 ; i++ ) {
		flare->origin[i] = LittleFloat( ds->lightmapOrigin[i] );
		flare->color[i] = LittleFloat( ds->lightmapVecs[0][i] );
		flare->normal[i] = LittleFloat( ds->lightmapVecs[2][i] );
	}

	surf->cullinfo.type = CULLINFO_NONE;
}


/*
=================
R_MergedWidthPoints

returns qtrue if there are grid points merged on a width edge
=================
*/
static int R_MergedWidthPoints( const srfBspSurface_t *grid, int offset) {
	int i, j;

	for (i = 1; i < grid->width-1; i++) {
		for (j = i + 1; j < grid->width-1; j++) {
			if ( fabs(grid->verts[i + offset].xyz[0] - grid->verts[j + offset].xyz[0]) > .1) continue;
			if ( fabs(grid->verts[i + offset].xyz[1] - grid->verts[j + offset].xyz[1]) > .1) continue;
			if ( fabs(grid->verts[i + offset].xyz[2] - grid->verts[j + offset].xyz[2]) > .1) continue;
			return qtrue;
		}
	}
	return qfalse;
}

/*
=================
R_MergedHeightPoints

returns qtrue if there are grid points merged on a height edge
=================
*/
static int R_MergedHeightPoints(const srfBspSurface_t *grid, int offset) {
	int i, j;

	for (i = 1; i < grid->height-1; i++) {
		for (j = i + 1; j < grid->height-1; j++) {
			if ( fabs(grid->verts[grid->width * i + offset].xyz[0] - grid->verts[grid->width * j + offset].xyz[0]) > .1) continue;
			if ( fabs(grid->verts[grid->width * i + offset].xyz[1] - grid->verts[grid->width * j + offset].xyz[1]) > .1) continue;
			if ( fabs(grid->verts[grid->width * i + offset].xyz[2] - grid->verts[grid->width * j + offset].xyz[2]) > .1) continue;
			return qtrue;
		}
	}
	return qfalse;
}

/*
=================
R_FixSharedVertexLodError_r

NOTE: never sync LoD through grid edges with merged points!

FIXME: write generalized version that also avoids cracks between a patch and one that meets half way?
=================
*/
static void R_FixSharedVertexLodError_r( int start, srfBspSurface_t *grid1 ) {
	int j, k, l, m, n, offset1, offset2, touch;
	srfBspSurface_t *grid2;

	for ( j = start; j < s_worldData.numsurfaces; j++ ) {
		//
		grid2 = (srfBspSurface_t *) s_worldData.surfaces[j].data;
		// if this surface is not a grid
		if ( grid2->surfaceType != SF_GRID ) continue;
		// if the LOD errors are already fixed for this patch
		if ( grid2->lodFixed == 2 ) continue;
		// grids in the same LOD group should have the exact same lod radius
		if ( grid1->lodRadius != grid2->lodRadius ) continue;
		// grids in the same LOD group should have the exact same lod origin
		if ( grid1->lodOrigin[0] != grid2->lodOrigin[0] ) continue;
		if ( grid1->lodOrigin[1] != grid2->lodOrigin[1] ) continue;
		if ( grid1->lodOrigin[2] != grid2->lodOrigin[2] ) continue;
		//
		touch = qfalse;
		for (n = 0; n < 2; n++) {
			//
			if (n) offset1 = (grid1->height-1) * grid1->width;
			else offset1 = 0;
			if (R_MergedWidthPoints(grid1, offset1)) continue;
			for (k = 1; k < grid1->width-1; k++) {
				for (m = 0; m < 2; m++) {

					if (m) offset2 = (grid2->height-1) * grid2->width;
					else offset2 = 0;
					if (R_MergedWidthPoints(grid2, offset2)) continue;
					for ( l = 1; l < grid2->width-1; l++) {
					//
						if ( fabs(grid1->verts[k + offset1].xyz[0] - grid2->verts[l + offset2].xyz[0]) > .1) continue;
						if ( fabs(grid1->verts[k + offset1].xyz[1] - grid2->verts[l + offset2].xyz[1]) > .1) continue;
						if ( fabs(grid1->verts[k + offset1].xyz[2] - grid2->verts[l + offset2].xyz[2]) > .1) continue;
						// ok the points are equal and should have the same lod error
						grid2->widthLodError[l] = grid1->widthLodError[k];
						touch = qtrue;
					}
				}
				for (m = 0; m < 2; m++) {

					if (m) offset2 = grid2->width-1;
					else offset2 = 0;
					if (R_MergedHeightPoints(grid2, offset2)) continue;
					for ( l = 1; l < grid2->height-1; l++) {
					//
						if ( fabs(grid1->verts[k + offset1].xyz[0] - grid2->verts[grid2->width * l + offset2].xyz[0]) > .1) continue;
						if ( fabs(grid1->verts[k + offset1].xyz[1] - grid2->verts[grid2->width * l + offset2].xyz[1]) > .1) continue;
						if ( fabs(grid1->verts[k + offset1].xyz[2] - grid2->verts[grid2->width * l + offset2].xyz[2]) > .1) continue;
						// ok the points are equal and should have the same lod error
						grid2->heightLodError[l] = grid1->widthLodError[k];
						touch = qtrue;
					}
				}
			}
		}
		for (n = 0; n < 2; n++) {
			//
			if (n) offset1 = grid1->width-1;
			else offset1 = 0;
			if (R_MergedHeightPoints(grid1, offset1)) continue;
			for (k = 1; k < grid1->height-1; k++) {
				for (m = 0; m < 2; m++) {

					if (m) offset2 = (grid2->height-1) * grid2->width;
					else offset2 = 0;
					if (R_MergedWidthPoints(grid2, offset2)) continue;
					for ( l = 1; l < grid2->width-1; l++) {
					//
						if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[0] - grid2->verts[l + offset2].xyz[0]) > .1) continue;
						if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[1] - grid2->verts[l + offset2].xyz[1]) > .1) continue;
						if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[2] - grid2->verts[l + offset2].xyz[2]) > .1) continue;
						// ok the points are equal and should have the same lod error
						grid2->widthLodError[l] = grid1->heightLodError[k];
						touch = qtrue;
					}
				}
				for (m = 0; m < 2; m++) {

					if (m) offset2 = grid2->width-1;
					else offset2 = 0;
					if (R_MergedHeightPoints(grid2, offset2)) continue;
					for ( l = 1; l < grid2->height-1; l++) {
					//
						if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[0] - grid2->verts[grid2->width * l + offset2].xyz[0]) > .1) continue;
						if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[1] - grid2->verts[grid2->width * l + offset2].xyz[1]) > .1) continue;
						if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[2] - grid2->verts[grid2->width * l + offset2].xyz[2]) > .1) continue;
						// ok the points are equal and should have the same lod error
						grid2->heightLodError[l] = grid1->heightLodError[k];
						touch = qtrue;
					}
				}
			}
		}
		if (touch) {
			grid2->lodFixed = 2;
			R_FixSharedVertexLodError_r ( start, grid2 );
			//NOTE: this would be correct but makes things really slow
			//grid2->lodFixed = 1;
		}
	}
}

/*
=================
R_FixSharedVertexLodError

This function assumes that all patches in one group are nicely stitched together for the highest LoD.
If this is not the case this function will still do its job but won't fix the highest LoD cracks.
=================
*/
static void R_FixSharedVertexLodError( void ) {
	int i;
	srfBspSurface_t *grid1;

	for ( i = 0; i < s_worldData.numsurfaces; i++ ) {
		//
		grid1 = (srfBspSurface_t *) s_worldData.surfaces[i].data;
		// if this surface is not a grid
		if ( grid1->surfaceType != SF_GRID )
			continue;
		//
		if ( grid1->lodFixed )
			continue;
		//
		grid1->lodFixed = 2;
		// recursively fix other patches in the same LOD group
		R_FixSharedVertexLodError_r( i + 1, grid1);
	}
}


/*
===============
R_StitchPatches
===============
*/
static int R_StitchPatches( int grid1num, int grid2num ) {
	float *v1, *v2;
	srfBspSurface_t *grid1, *grid2;
	int k, l, m, n, offset1, offset2, row, column;

	grid1 = (srfBspSurface_t *) s_worldData.surfaces[grid1num].data;
	grid2 = (srfBspSurface_t *) s_worldData.surfaces[grid2num].data;
	for (n = 0; n < 2; n++) {
		//
		if (n) offset1 = (grid1->height-1) * grid1->width;
		else offset1 = 0;
		if (R_MergedWidthPoints(grid1, offset1))
			continue;
		for (k = 0; k < grid1->width-2; k += 2) {

			for (m = 0; m < 2; m++) {

				if ( grid2->width >= MAX_GRID_SIZE )
					break;
				if (m) offset2 = (grid2->height-1) * grid2->width;
				else offset2 = 0;
				for ( l = 0; l < grid2->width-1; l++) {
				//
					v1 = grid1->verts[k + offset1].xyz;
					v2 = grid2->verts[l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[k + 2 + offset1].xyz;
					v2 = grid2->verts[l + 1 + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[l + offset2].xyz;
					v2 = grid2->verts[l + 1 + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert column into grid2 right after column l
					if (m) row = grid2->height-1;
					else row = 0;
					R_GridInsertColumn( grid2, l+1, row,
									grid1->verts[k + 1 + offset1].xyz, grid1->widthLodError[k+1]);
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
			for (m = 0; m < 2; m++) {

				if (grid2->height >= MAX_GRID_SIZE)
					break;
				if (m) offset2 = grid2->width-1;
				else offset2 = 0;
				for ( l = 0; l < grid2->height-1; l++) {
					//
					v1 = grid1->verts[k + offset1].xyz;
					v2 = grid2->verts[grid2->width * l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[k + 2 + offset1].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[grid2->width * l + offset2].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert row into grid2 right after row l
					if (m) column = grid2->width-1;
					else column = 0;
					R_GridInsertRow( grid2, l+1, column,
										grid1->verts[k + 1 + offset1].xyz, grid1->widthLodError[k+1]);
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
		}
	}
	for (n = 0; n < 2; n++) {
		//
		if (n) offset1 = grid1->width-1;
		else offset1 = 0;
		if (R_MergedHeightPoints(grid1, offset1))
			continue;
		for (k = 0; k < grid1->height-2; k += 2) {
			for (m = 0; m < 2; m++) {

				if ( grid2->width >= MAX_GRID_SIZE )
					break;
				if (m) offset2 = (grid2->height-1) * grid2->width;
				else offset2 = 0;
				for ( l = 0; l < grid2->width-1; l++) {
				//
					v1 = grid1->verts[grid1->width * k + offset1].xyz;
					v2 = grid2->verts[l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[grid1->width * (k + 2) + offset1].xyz;
					v2 = grid2->verts[l + 1 + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[l + offset2].xyz;
					v2 = grid2->verts[(l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert column into grid2 right after column l
					if (m) row = grid2->height-1;
					else row = 0;
					R_GridInsertColumn( grid2, l+1, row,
									grid1->verts[grid1->width * (k + 1) + offset1].xyz, grid1->heightLodError[k+1]);
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
			for (m = 0; m < 2; m++) {

				if (grid2->height >= MAX_GRID_SIZE)
					break;
				if (m) offset2 = grid2->width-1;
				else offset2 = 0;
				for ( l = 0; l < grid2->height-1; l++) {
				//
					v1 = grid1->verts[grid1->width * k + offset1].xyz;
					v2 = grid2->verts[grid2->width * l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[grid1->width * (k + 2) + offset1].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[grid2->width * l + offset2].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert row into grid2 right after row l
					if (m) column = grid2->width-1;
					else column = 0;
					R_GridInsertRow( grid2, l+1, column,
									grid1->verts[grid1->width * (k + 1) + offset1].xyz, grid1->heightLodError[k+1]);
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
		}
	}
	for (n = 0; n < 2; n++) {
		//
		if (n) offset1 = (grid1->height-1) * grid1->width;
		else offset1 = 0;
		if (R_MergedWidthPoints(grid1, offset1))
			continue;
		for (k = grid1->width-1; k > 1; k -= 2) {

			for (m = 0; m < 2; m++) {

				if ( !grid2 || grid2->width >= MAX_GRID_SIZE )
					break;
				if (m) offset2 = (grid2->height-1) * grid2->width;
				else offset2 = 0;
				for ( l = 0; l < grid2->width-1; l++) {
				//
					v1 = grid1->verts[k + offset1].xyz;
					v2 = grid2->verts[l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[k - 2 + offset1].xyz;
					v2 = grid2->verts[l + 1 + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[l + offset2].xyz;
					v2 = grid2->verts[(l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert column into grid2 right after column l
					if (m) row = grid2->height-1;
					else row = 0;
					R_GridInsertColumn( grid2, l+1, row,
										grid1->verts[k - 1 + offset1].xyz, grid1->widthLodError[k+1]);
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
			for (m = 0; m < 2; m++) {

				if (!grid2 || grid2->height >= MAX_GRID_SIZE)
					break;
				if (m) offset2 = grid2->width-1;
				else offset2 = 0;
				for ( l = 0; l < grid2->height-1; l++) {
				//
					v1 = grid1->verts[k + offset1].xyz;
					v2 = grid2->verts[grid2->width * l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[k - 2 + offset1].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[grid2->width * l + offset2].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert row into grid2 right after row l
					if (m) column = grid2->width-1;
					else column = 0;
					R_GridInsertRow( grid2, l+1, column,
										grid1->verts[k - 1 + offset1].xyz, grid1->widthLodError[k+1]);
					if (!grid2)
						break;
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
		}
	}
	for (n = 0; n < 2; n++) {
		//
		if (n) offset1 = grid1->width-1;
		else offset1 = 0;
		if (R_MergedHeightPoints(grid1, offset1))
			continue;
		for (k = grid1->height-1; k > 1; k -= 2) {
			for (m = 0; m < 2; m++) {

				if (!grid2 || grid2->width >= MAX_GRID_SIZE )
					break;
				if (m) offset2 = (grid2->height-1) * grid2->width;
				else offset2 = 0;
				for ( l = 0; l < grid2->width-1; l++) {
				//
					v1 = grid1->verts[grid1->width * k + offset1].xyz;
					v2 = grid2->verts[l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[grid1->width * (k - 2) + offset1].xyz;
					v2 = grid2->verts[l + 1 + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[l + offset2].xyz;
					v2 = grid2->verts[(l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert column into grid2 right after column l
					if (m) row = grid2->height-1;
					else row = 0;
					R_GridInsertColumn( grid2, l+1, row,
										grid1->verts[grid1->width * (k - 1) + offset1].xyz, grid1->heightLodError[k+1]);
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
			for (m = 0; m < 2; m++) {

				if (!grid2 || grid2->height >= MAX_GRID_SIZE)
					break;
				if (m) offset2 = grid2->width-1;
				else offset2 = 0;
				for ( l = 0; l < grid2->height-1; l++) {
				//
					v1 = grid1->verts[grid1->width * k + offset1].xyz;
					v2 = grid2->verts[grid2->width * l + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;

					v1 = grid1->verts[grid1->width * (k - 2) + offset1].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) > .1)
						continue;
					if ( fabs(v1[1] - v2[1]) > .1)
						continue;
					if ( fabs(v1[2] - v2[2]) > .1)
						continue;
					//
					v1 = grid2->verts[grid2->width * l + offset2].xyz;
					v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
					if ( fabs(v1[0] - v2[0]) < .01 &&
							fabs(v1[1] - v2[1]) < .01 &&
							fabs(v1[2] - v2[2]) < .01)
						continue;
					//
					//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
					// insert row into grid2 right after row l
					if (m) column = grid2->width-1;
					else column = 0;
					R_GridInsertRow( grid2, l+1, column,
										grid1->verts[grid1->width * (k - 1) + offset1].xyz, grid1->heightLodError[k+1]);
					grid2->lodStitched = qfalse;
					s_worldData.surfaces[grid2num].data = (void *) grid2;
					return qtrue;
				}
			}
		}
	}
	return qfalse;
}

/*
===============
R_TryStitchPatch

This function will try to stitch patches in the same LoD group together for the highest LoD.

Only single missing vertex cracks will be fixed.

Vertices will be joined at the patch side a crack is first found, at the other side
of the patch (on the same row or column) the vertices will not be joined and cracks
might still appear at that side.
===============
*/
static int R_TryStitchingPatch( int grid1num ) {
	int j, numstitches;
	srfBspSurface_t *grid1, *grid2;

	numstitches = 0;
	grid1 = (srfBspSurface_t *) s_worldData.surfaces[grid1num].data;
	for ( j = 0; j < s_worldData.numsurfaces; j++ ) {
		//
		grid2 = (srfBspSurface_t *) s_worldData.surfaces[j].data;
		// if this surface is not a grid
		if ( grid2->surfaceType != SF_GRID ) continue;
		// grids in the same LOD group should have the exact same lod radius
		if ( grid1->lodRadius != grid2->lodRadius ) continue;
		// grids in the same LOD group should have the exact same lod origin
		if ( grid1->lodOrigin[0] != grid2->lodOrigin[0] ) continue;
		if ( grid1->lodOrigin[1] != grid2->lodOrigin[1] ) continue;
		if ( grid1->lodOrigin[2] != grid2->lodOrigin[2] ) continue;
		//
		while (R_StitchPatches(grid1num, j))
		{
			numstitches++;
		}
	}
	return numstitches;
}

/*
===============
R_StitchAllPatches
===============
*/
static void R_StitchAllPatches( void ) {
	int i, stitched, numstitches;
	srfBspSurface_t *grid1;

	numstitches = 0;
	do
	{
		stitched = qfalse;
		for ( i = 0; i < s_worldData.numsurfaces; i++ ) {
			//
			grid1 = (srfBspSurface_t *) s_worldData.surfaces[i].data;
			// if this surface is not a grid
			if ( grid1->surfaceType != SF_GRID )
				continue;
			//
			if ( grid1->lodStitched )
				continue;
			//
			grid1->lodStitched = qtrue;
			stitched = qtrue;
			//
			numstitches += R_TryStitchingPatch( i );
		}
	}
	while (stitched);
	ri.Printf( PRINT_ALL, "stitched %d LoD cracks\n", numstitches );
}

/*
===============
R_MovePatchSurfacesToHunk
===============
*/
static void R_MovePatchSurfacesToHunk(void) {
	int i;
	srfBspSurface_t *grid;

	for ( i = 0; i < s_worldData.numsurfaces; i++ ) {
		void *copyFrom;
		//
		grid = (srfBspSurface_t *) s_worldData.surfaces[i].data;
		// if this surface is not a grid
		if ( grid->surfaceType != SF_GRID )
			continue;
		//

		copyFrom = grid->widthLodError;
		grid->widthLodError = ri.Hunk_Alloc( grid->width * 4, h_low );
		Com_Memcpy(grid->widthLodError, copyFrom, grid->width * 4);
		ri.Free(copyFrom);

		copyFrom = grid->heightLodError;
		grid->heightLodError = ri.Hunk_Alloc(grid->height * 4, h_low);
		Com_Memcpy(grid->heightLodError, copyFrom, grid->height * 4);
		ri.Free(copyFrom);

		copyFrom = grid->indexes;
		grid->indexes = ri.Hunk_Alloc(grid->numIndexes * sizeof(glIndex_t), h_low);
		Com_Memcpy(grid->indexes, copyFrom, grid->numIndexes * sizeof(glIndex_t));
		ri.Free(copyFrom);

		copyFrom = grid->verts;
		grid->verts = ri.Hunk_Alloc(grid->numVerts * sizeof(srfVert_t), h_low);
		Com_Memcpy(grid->verts, copyFrom, grid->numVerts * sizeof(srfVert_t));
		ri.Free(copyFrom);
	}
}


/*
===============
R_LoadSurfaces
===============
*/
static	void R_LoadSurfaces( const lump_t *surfs, const lump_t *verts, const lump_t *indexLump ) {
	const dsurface_t	*in;
	msurface_t	*out;
	const drawVert_t	*dv;
	int			*indexes;
	int			count;
	int			numFaces, numMeshes, numTriSurfs, numFlares;
	int			i;
	float *hdrVertColors = NULL;

	numFaces = 0;
	numMeshes = 0;
	numTriSurfs = 0;
	numFlares = 0;

	if (surfs->filelen % sizeof(*in))
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	count = surfs->filelen / sizeof(*in);

	dv = (void *)(fileBase + verts->fileofs);
	if (verts->filelen % sizeof(*dv))
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);

	indexes = (void *)(fileBase + indexLump->fileofs);
	if ( indexLump->filelen % sizeof(*indexes))
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);

	out = ri.Hunk_Alloc ( count * sizeof(*out), h_low );	

	s_worldData.surfaces = out;
	s_worldData.numsurfaces = count;
	s_worldData.surfacesViewCount = ri.Hunk_Alloc ( count * sizeof(*s_worldData.surfacesViewCount), h_low );
	s_worldData.surfacesDlightBits = ri.Hunk_Alloc ( count * sizeof(*s_worldData.surfacesDlightBits), h_low );
	s_worldData.surfacesPshadowBits = ri.Hunk_Alloc ( count * sizeof(*s_worldData.surfacesPshadowBits), h_low );

	// load hdr vertex colors
	if (r_hdr->integer)
	{
		char filename[MAX_QPATH];
		int size;

		Com_sprintf( filename, sizeof( filename ), "maps/%s/vertlight.raw", s_worldData.baseName);
		//ri.Printf(PRINT_ALL, "looking for %s\n", filename);

		size = ri.FS_ReadFile(filename, (void **)&hdrVertColors);

		if (hdrVertColors)
		{
			//ri.Printf(PRINT_ALL, "Found!\n");
			if (size != sizeof(float) * 3 * (verts->filelen / sizeof(*dv)))
				ri.Error(ERR_DROP, "Bad size for %s (%i, expected %i)!", filename, size, (int)((sizeof(float)) * 3 * (verts->filelen / sizeof(*dv))));
		}
	}


	// Two passes, allocate surfaces first, then load them full of data
	// This ensures surfaces are close together to reduce L2 cache misses when using VAOs,
	// which don't actually use the verts and indexes
	in = (void *)(fileBase + surfs->fileofs);
	out = s_worldData.surfaces;
	for ( i = 0 ; i < count ; i++, in++, out++ ) {
		switch ( LittleLong( in->surfaceType ) ) {
			case MST_PATCH:
				out->data = ri.Hunk_Alloc( sizeof(srfBspSurface_t), h_low);
				break;
			case MST_TRIANGLE_SOUP:
				out->data = ri.Hunk_Alloc( sizeof(srfBspSurface_t), h_low);
				break;
			case MST_PLANAR:
				out->data = ri.Hunk_Alloc( sizeof(srfBspSurface_t), h_low);
				break;
			case MST_FLARE:
				out->data = ri.Hunk_Alloc( sizeof(srfFlare_t), h_low);
				break;
			default:
				break;
		}
	}

	in = (void *)(fileBase + surfs->fileofs);
	out = s_worldData.surfaces;
	for ( i = 0 ; i < count ; i++, in++, out++ ) {
		switch ( LittleLong( in->surfaceType ) ) {
		case MST_PATCH:
			ParseMesh ( in, dv, hdrVertColors, out );
			numMeshes++;
			break;
		case MST_TRIANGLE_SOUP:
			ParseTriSurf( in, dv, hdrVertColors, out, indexes );
			numTriSurfs++;
			break;
		case MST_PLANAR:
			ParseFace( in, dv, hdrVertColors, out, indexes );
			numFaces++;
			break;
		case MST_FLARE:
			ParseFlare( in, dv, out, indexes );
			numFlares++;
			break;
		default:
			ri.Error( ERR_DROP, "Bad surfaceType" );
		}
	}

	if (hdrVertColors)
	{
		ri.FS_FreeFile(hdrVertColors);
	}

#ifdef PATCH_STITCHING
	R_StitchAllPatches();
#endif

	R_FixSharedVertexLodError();

#ifdef PATCH_STITCHING
	R_MovePatchSurfacesToHunk();
#endif

	ri.Printf( PRINT_ALL, "...loaded %d faces, %i meshes, %i trisurfs, %i flares\n", 
		numFaces, numMeshes, numTriSurfs, numFlares );
}



/*
=================
R_LoadSubmodels
=================
*/
static	void R_LoadSubmodels( const lump_t *l ) {
	const dmodel_t	*in;
	bmodel_t	*out;
	int			i, j, count;

	in = (void *)(fileBase + l->fileofs);
	if (l->filelen % sizeof(*in))
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	count = l->filelen / sizeof(*in);

	s_worldData.numBModels = count;
	s_worldData.bmodels = out = ri.Hunk_Alloc( count * sizeof(*out), h_low );

	for ( i=0 ; i<count ; i++, in++, out++ ) {
		model_t *model;

		model = R_AllocModel();

		assert( model != NULL );			// this should never happen
		if ( model == NULL ) {
			ri.Error(ERR_DROP, "R_LoadSubmodels: R_AllocModel() failed");
		}

		model->type = MOD_BRUSH;
		model->bmodel = out;
		Com_sprintf( model->name, sizeof( model->name ), "*%d", i );

		for (j=0 ; j<3 ; j++) {
			out->bounds[0][j] = LittleFloat (in->mins[j]);
			out->bounds[1][j] = LittleFloat (in->maxs[j]);
		}

		out->firstSurface = LittleLong( in->firstSurface );
		out->numSurfaces = LittleLong( in->numSurfaces );

		if(i == 0)
		{
			// Add this for limiting VAO surface creation
			s_worldData.numWorldSurfaces = out->numSurfaces;
		}
	}
}



//==================================================================

/*
=================
R_SetParent
=================
*/
static	void R_SetParent (mnode_t *node, mnode_t *parent)
{
	node->parent = parent;
	if (node->contents != -1)
		return;
	R_SetParent (node->children[0], node);
	R_SetParent (node->children[1], node);
}

/*
=================
R_LoadNodesAndLeafs
=================
*/
static	void R_LoadNodesAndLeafs (const lump_t *nodeLump, const lump_t *leafLump) {
	int			i, j, p;
	const dnode_t		*in;
	dleaf_t		*inLeaf;
	mnode_t 	*out;
	int			numNodes, numLeafs;

	in = (void *)(fileBase + nodeLump->fileofs);
	if (nodeLump->filelen % sizeof(dnode_t) ||
		leafLump->filelen % sizeof(dleaf_t) ) {
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	}
	numNodes = nodeLump->filelen / sizeof(dnode_t);
	numLeafs = leafLump->filelen / sizeof(dleaf_t);

	out = ri.Hunk_Alloc ( (numNodes + numLeafs) * sizeof(*out), h_low);	

	s_worldData.nodes = out;
	s_worldData.numnodes = numNodes + numLeafs;
	s_worldData.numDecisionNodes = numNodes;

	// load nodes
	for ( i=0 ; i<numNodes; i++, in++, out++)
	{
		for (j=0 ; j<3 ; j++)
		{
			out->mins[j] = LittleLong (in->mins[j]);
			out->maxs[j] = LittleLong (in->maxs[j]);
		}
	
		p = LittleLong(in->planeNum);
		out->plane = s_worldData.planes + p;

		out->contents = CONTENTS_NODE;	// differentiate from leafs

		for (j=0 ; j<2 ; j++)
		{
			p = LittleLong (in->children[j]);
			if (p >= 0)
				out->children[j] = s_worldData.nodes + p;
			else
				out->children[j] = s_worldData.nodes + numNodes + (-1 - p);
		}
	}
	
	// load leafs
	inLeaf = (void *)(fileBase + leafLump->fileofs);
	for ( i=0 ; i<numLeafs ; i++, inLeaf++, out++)
	{
		for (j=0 ; j<3 ; j++)
		{
			out->mins[j] = LittleLong (inLeaf->mins[j]);
			out->maxs[j] = LittleLong (inLeaf->maxs[j]);
		}

		out->cluster = LittleLong(inLeaf->cluster);
		out->area = LittleLong(inLeaf->area);

		if ( out->cluster >= s_worldData.numClusters ) {
			s_worldData.numClusters = out->cluster + 1;
		}

		out->firstmarksurface = LittleLong(inLeaf->firstLeafSurface);
		out->nummarksurfaces = LittleLong(inLeaf->numLeafSurfaces);
	}	

	// chain descendants
	R_SetParent (s_worldData.nodes, NULL);
}

//=============================================================================

/*
=================
R_LoadShaders
=================
*/
static	void R_LoadShaders( const lump_t *l ) {	
	int		i, count;
	dshader_t	*in, *out;
	
	in = (void *)(fileBase + l->fileofs);
	if (l->filelen % sizeof(*in))
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	count = l->filelen / sizeof(*in);
	out = ri.Hunk_Alloc ( count*sizeof(*out), h_low );

	s_worldData.shaders = out;
	s_worldData.numShaders = count;

	Com_Memcpy( out, in, count*sizeof(*out) );

	for ( i=0 ; i<count ; i++ ) {
		out[i].surfaceFlags = LittleLong( out[i].surfaceFlags );
		out[i].contentFlags = LittleLong( out[i].contentFlags );
	}
}


/*
=================
R_LoadMarksurfaces
=================
*/
static	void R_LoadMarksurfaces (const lump_t *l)
{	
	int		i, j, count;
	int		*in;
	int     *out;
	
	in = (void *)(fileBase + l->fileofs);
	if (l->filelen % sizeof(*in))
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	count = l->filelen / sizeof(*in);
	out = ri.Hunk_Alloc ( count*sizeof(*out), h_low);	

	s_worldData.marksurfaces = out;
	s_worldData.nummarksurfaces = count;

	for ( i=0 ; i<count ; i++)
	{
		j = LittleLong(in[i]);
		out[i] = j;
	}
}


/*
=================
R_LoadPlanes
=================
*/
static	void R_LoadPlanes( const lump_t *l ) {
	int			i, j;
	cplane_t	*out;
	const dplane_t 	*in;
	int			count;
	int			bits;
	
	in = (void *)(fileBase + l->fileofs);
	if (l->filelen % sizeof(*in))
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	count = l->filelen / sizeof(*in);
	out = ri.Hunk_Alloc ( count*2*sizeof(*out), h_low);	
	
	s_worldData.planes = out;
	s_worldData.numplanes = count;

	for ( i=0 ; i<count ; i++, in++, out++) {
		bits = 0;
		for (j=0 ; j<3 ; j++) {
			out->normal[j] = LittleFloat (in->normal[j]);
			if (out->normal[j] < 0) {
				bits |= 1<<j;
			}
		}

		out->dist = LittleFloat (in->dist);
		out->type = PlaneTypeForNormal( out->normal );
		out->signbits = bits;
	}
}

/*
=================
R_LoadFogs

=================
*/
static	void R_LoadFogs( const lump_t *l, const lump_t *brushesLump, const lump_t *sidesLump ) {
	int			i;
	fog_t		*out;
	const dfog_t		*fogs;
	const dbrush_t 	*brushes, *brush;
	const dbrushside_t	*sides;
	int			count, brushesCount, sidesCount;
	int			sideNum;
	int			planeNum;
	shader_t	*shader;
	float		d;
	int			firstSide;

	fogs = (void *)(fileBase + l->fileofs);
	if (l->filelen % sizeof(*fogs)) {
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	}
	count = l->filelen / sizeof(*fogs);

	// create fog structures for them
	s_worldData.numfogs = count + 1;
	s_worldData.fogs = ri.Hunk_Alloc ( s_worldData.numfogs*sizeof(*out), h_low);
	out = s_worldData.fogs + 1;

	if ( !count ) {
		return;
	}

	brushes = (void *)(fileBase + brushesLump->fileofs);
	if (brushesLump->filelen % sizeof(*brushes)) {
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	}
	brushesCount = brushesLump->filelen / sizeof(*brushes);

	sides = (void *)(fileBase + sidesLump->fileofs);
	if (sidesLump->filelen % sizeof(*sides)) {
		ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
	}
	sidesCount = sidesLump->filelen / sizeof(*sides);

	for ( i=0 ; i<count ; i++, fogs++) {
		out->originalBrushNumber = LittleLong( fogs->brushNum );

		if ( (unsigned)out->originalBrushNumber >= brushesCount ) {
			ri.Error( ERR_DROP, "fog brushNumber out of range" );
		}
		brush = brushes + out->originalBrushNumber;

		firstSide = LittleLong( brush->firstSide );

			if ( (unsigned)firstSide > sidesCount - 6 ) {
			ri.Error( ERR_DROP, "fog brush sideNumber out of range" );
		}

		// brushes are always sorted with the axial sides first
		sideNum = firstSide + 0;
		planeNum = LittleLong( sides[ sideNum ].planeNum );
		out->bounds[0][0] = -s_worldData.planes[ planeNum ].dist;

		sideNum = firstSide + 1;
		planeNum = LittleLong( sides[ sideNum ].planeNum );
		out->bounds[1][0] = s_worldData.planes[ planeNum ].dist;

		sideNum = firstSide + 2;
		planeNum = LittleLong( sides[ sideNum ].planeNum );
		out->bounds[0][1] = -s_worldData.planes[ planeNum ].dist;

		sideNum = firstSide + 3;
		planeNum = LittleLong( sides[ sideNum ].planeNum );
		out->bounds[1][1] = s_worldData.planes[ planeNum ].dist;

		sideNum = firstSide + 4;
		planeNum = LittleLong( sides[ sideNum ].planeNum );
		out->bounds[0][2] = -s_worldData.planes[ planeNum ].dist;

		sideNum = firstSide + 5;
		planeNum = LittleLong( sides[ sideNum ].planeNum );
		out->bounds[1][2] = s_worldData.planes[ planeNum ].dist;

		// get information from the shader for fog parameters
		shader = R_FindShader( fogs->shader, LIGHTMAP_NONE, qtrue );

		out->parms = shader->fogParms;

		out->colorInt = ColorBytes4 ( shader->fogParms.color[0],
			                          shader->fogParms.color[1],
			                          shader->fogParms.color[2], 1.0 );

		d = shader->fogParms.depthForOpaque < 1 ? 1 : shader->fogParms.depthForOpaque;
		out->tcScale = 1.0f / ( d * 8 );

		// set the gradient vector
		sideNum = LittleLong( fogs->visibleSide );

		if ( sideNum == -1 ) {
			out->hasSurface = qfalse;
		} else {
			out->hasSurface = qtrue;
			planeNum = LittleLong( sides[ firstSide + sideNum ].planeNum );
			VectorSubtract( vec3_origin, s_worldData.planes[ planeNum ].normal, out->surface );
			out->surface[3] = -s_worldData.planes[ planeNum ].dist;
		}

		out++;
	}

}


/*
================
R_LoadLightGrid

================
*/
static void R_LoadLightGrid( const lump_t *l ) {
	int		i;
	vec3_t	maxs;
	int		numGridPoints;
	world_t	*w;
	float	*wMins, *wMaxs;

	w = &s_worldData;

	w->lightGridInverseSize[0] = 1.0f / w->lightGridSize[0];
	w->lightGridInverseSize[1] = 1.0f / w->lightGridSize[1];
	w->lightGridInverseSize[2] = 1.0f / w->lightGridSize[2];

	wMins = w->bmodels[0].bounds[0];
	wMaxs = w->bmodels[0].bounds[1];

	for ( i = 0 ; i < 3 ; i++ ) {
		w->lightGridOrigin[i] = w->lightGridSize[i] * ceil( wMins[i] / w->lightGridSize[i] );
		maxs[i] = w->lightGridSize[i] * floor( wMaxs[i] / w->lightGridSize[i] );
		w->lightGridBounds[i] = (maxs[i] - w->lightGridOrigin[i])/w->lightGridSize[i] + 1;
	}

	numGridPoints = w->lightGridBounds[0] * w->lightGridBounds[1] * w->lightGridBounds[2];

	if ( l->filelen != numGridPoints * 8 ) {
		ri.Printf( PRINT_WARNING, "WARNING: light grid mismatch\n" );
		w->lightGridData = NULL;
		return;
	}

	w->lightGridData = ri.Hunk_Alloc( l->filelen, h_low );
	Com_Memcpy( w->lightGridData, (void *)(fileBase + l->fileofs), l->filelen );

	// deal with overbright bits
	for ( i = 0 ; i < numGridPoints ; i++ ) {
		R_ColorShiftLightingBytes( &w->lightGridData[i*8], &w->lightGridData[i*8] );
		R_ColorShiftLightingBytes( &w->lightGridData[i*8+3], &w->lightGridData[i*8+3] );
	}

	// load hdr lightgrid
	if (r_hdr->integer)
	{
		char filename[MAX_QPATH];
		float *hdrLightGrid;
		int size;

		Com_sprintf( filename, sizeof( filename ), "maps/%s/lightgrid.raw", s_worldData.baseName);
		//ri.Printf(PRINT_ALL, "looking for %s\n", filename);

		size = ri.FS_ReadFile(filename, (void **)&hdrLightGrid);

		if (hdrLightGrid)
		{
			//ri.Printf(PRINT_ALL, "found!\n");

			if (size != sizeof(float) * 6 * numGridPoints)
				ri.Error(ERR_DROP, "Bad size for %s (%i, expected %i)!", filename, size, (int)(sizeof(float)) * 6 * numGridPoints);

			w->lightGrid16 = ri.Hunk_Alloc(sizeof(w->lightGrid16) * 6 * numGridPoints, h_low);

			for (i = 0; i < numGridPoints ; i++)
			{
				vec4_t c;

				c[0] = hdrLightGrid[i * 6];
				c[1] = hdrLightGrid[i * 6 + 1];
				c[2] = hdrLightGrid[i * 6 + 2];
				c[3] = 1.0f;

				R_ColorShiftLightingFloats(c, c);
				ColorToRGB16(c, &w->lightGrid16[i * 6]);

				c[0] = hdrLightGrid[i * 6 + 3];
				c[1] = hdrLightGrid[i * 6 + 4];
				c[2] = hdrLightGrid[i * 6 + 5];
				c[3] = 1.0f;

				R_ColorShiftLightingFloats(c, c);
				ColorToRGB16(c, &w->lightGrid16[i * 6 + 3]);
			}
		}
		else if (0)
		{
			// promote 8-bit lightgrid to 16-bit
			w->lightGrid16 = ri.Hunk_Alloc(sizeof(w->lightGrid16) * 6 * numGridPoints, h_low);

			for (i = 0; i < numGridPoints; i++)
			{
				w->lightGrid16[i * 6]     = w->lightGridData[i * 8] * 257;
				w->lightGrid16[i * 6 + 1] = w->lightGridData[i * 8 + 1] * 257;
				w->lightGrid16[i * 6 + 2] = w->lightGridData[i * 8 + 2] * 257;
				w->lightGrid16[i * 6 + 3] = w->lightGridData[i * 8 + 3] * 257;
				w->lightGrid16[i * 6 + 4] = w->lightGridData[i * 8 + 4] * 257;
				w->lightGrid16[i * 6 + 5] = w->lightGridData[i * 8 + 5] * 257;
			}
		}

		if (hdrLightGrid)
			ri.FS_FreeFile(hdrLightGrid);
	}
}

/*
================
R_LoadEntities
================
*/
static void R_LoadEntities( const lump_t *l ) {
	const char *p, *token, *s;
	char keyname[MAX_TOKEN_CHARS];
	char value[MAX_TOKEN_CHARS];
	world_t	*w;

	w = &s_worldData;
	w->lightGridSize[0] = 64;
	w->lightGridSize[1] = 64;
	w->lightGridSize[2] = 128;

	p = (const char *)(fileBase + l->fileofs);

	// store for reference by the cgame
	w->entityString = ri.Hunk_Alloc( l->filelen + 1, h_low );
	strcpy( w->entityString, p );
	w->entityParsePoint = w->entityString;

	token = COM_ParseExt( &p, qtrue );
	if (!*token || *token != '{') {
		return;
	}

	// only parse the world spawn
	while ( 1 ) {	
		// parse key
		token = COM_ParseExt( &p, qtrue );

		if ( !*token || *token == '}' ) {
			break;
		}
		Q_strncpyz(keyname, token, sizeof(keyname));

		// parse value
		token = COM_ParseExt( &p, qtrue );

		if ( !*token || *token == '}' ) {
			break;
		}
		Q_strncpyz(value, token, sizeof(value));

		// check for remapping of shaders for vertex lighting
		s = "vertexremapshader";
		if (!Q_strncmp(keyname, s, strlen(s)) ) {
			char *vs = strchr(value, ';');
			if (!vs) {
				ri.Printf( PRINT_WARNING, "WARNING: no semi colon in vertexshaderremap '%s'\n", value );
				break;
			}
			*vs++ = 0;
			if ( r_vertexLight->integer && tr.vertexLightingAllowed ) {
				R_RemapShader(value, s, "0");
			}
			continue;
		}
		// check for remapping of shaders
		s = "remapshader";
		if (!Q_strncmp(keyname, s, strlen(s)) ) {
			char *vs = strchr(value, ';');
			if (!vs) {
				ri.Printf( PRINT_WARNING, "WARNING: no semi colon in shaderremap '%s'\n", value );
				break;
			}
			*vs++ = 0;
			R_RemapShader(value, s, "0");
			continue;
		}
		// check for a different grid size
		if (!Q_stricmp(keyname, "gridsize")) {
			sscanf(value, "%f %f %f", &w->lightGridSize[0], &w->lightGridSize[1], &w->lightGridSize[2] );
			continue;
		}

		// check for auto exposure
		if (!Q_stricmp(keyname, "autoExposureMinMax")) {
			sscanf(value, "%f %f", &tr.autoExposureMinMax[0], &tr.autoExposureMinMax[1]);
			continue;
		}
	}
}

/*
=================
R_GetEntityToken
=================
*/
qboolean R_GetEntityToken( char *buffer, int size ) {
	const char	*s;

	s = COM_Parse( &s_worldData.entityParsePoint );
	Q_strncpyz( buffer, s, size );
	if ( !s_worldData.entityParsePoint && !s[0] ) {
		s_worldData.entityParsePoint = s_worldData.entityString;
		return qfalse;
	} else {
		return qtrue;
	}
}

#ifndef MAX_SPAWN_VARS
#define MAX_SPAWN_VARS 64
#endif

// derived from G_ParseSpawnVars() in g_spawn.c
static qboolean R_ParseSpawnVars( char *spawnVarChars, int maxSpawnVarChars, int *numSpawnVars, const char *spawnVars[MAX_SPAWN_VARS][2] )
{
	char		keyname[MAX_TOKEN_CHARS];
	char		com_token[MAX_TOKEN_CHARS];
	int			numSpawnVarChars = 0;

	*numSpawnVars = 0;

	// parse the opening brace
	if ( !R_GetEntityToken( com_token, sizeof( com_token ) ) ) {
		// end of spawn string
		return qfalse;
	}
	if ( com_token[0] != '{' ) {
		ri.Printf( PRINT_ALL, "R_ParseSpawnVars: found %s when expecting {\n",com_token );
		return qfalse;
	}

	// go through all the key / value pairs
	while ( 1 ) {	
		int keyLength, tokenLength;

		// parse key
		if ( !R_GetEntityToken( keyname, sizeof( keyname ) ) ) {
			ri.Printf( PRINT_ALL, "R_ParseSpawnVars: EOF without closing brace\n" );
			return qfalse;
		}

		if ( keyname[0] == '}' ) {
			break;
		}
		
		// parse value	
		if ( !R_GetEntityToken( com_token, sizeof( com_token ) ) ) {
			ri.Printf( PRINT_ALL, "R_ParseSpawnVars: EOF without closing brace\n" );
			return qfalse;
		}

		if ( com_token[0] == '}' ) {
			ri.Printf( PRINT_ALL, "R_ParseSpawnVars: closing brace without data\n" );
			return qfalse;
		}

		if ( *numSpawnVars == MAX_SPAWN_VARS ) {
			ri.Printf( PRINT_ALL, "R_ParseSpawnVars: MAX_SPAWN_VARS\n" );
			return qfalse;
		}

		keyLength = strlen(keyname) + 1;
		tokenLength = strlen(com_token) + 1;

		if (numSpawnVarChars + keyLength + tokenLength > maxSpawnVarChars)
		{
			ri.Printf( PRINT_ALL, "R_ParseSpawnVars: MAX_SPAWN_VAR_CHARS\n" );
			return qfalse;
		}

		strcpy(spawnVarChars + numSpawnVarChars, keyname);
		spawnVars[ *numSpawnVars ][0] = spawnVarChars + numSpawnVarChars;
		numSpawnVarChars += keyLength;

		strcpy(spawnVarChars + numSpawnVarChars, com_token);
		spawnVars[ *numSpawnVars ][1] = spawnVarChars + numSpawnVarChars;
		numSpawnVarChars += tokenLength;

		(*numSpawnVars)++;
	}

	return qtrue;
}

static void R_LoadEnvironmentJson(const char *baseName)
{
	char filename[MAX_QPATH];

	union {
		char *c;
		void *v;
	} buffer;
	char *bufferEnd;

	const char *cubemapArrayJson;
	int filelen, i;

	Com_sprintf(filename, MAX_QPATH, "cubemaps/%s/env.json", baseName);

	filelen = ri.FS_ReadFile(filename, &buffer.v);
	if (!buffer.c)
		return;
	bufferEnd = buffer.c + filelen;

	if (JSON_ValueGetType(buffer.c, bufferEnd) != JSONTYPE_OBJECT)
	{
		ri.Printf(PRINT_ALL, "Bad %s: does not start with a object\n", filename);
		ri.FS_FreeFile(buffer.v);
		return;
	}

	cubemapArrayJson = JSON_ObjectGetNamedValue(buffer.c, bufferEnd, "Cubemaps");
	if (!cubemapArrayJson)
	{
		ri.Printf(PRINT_ALL, "Bad %s: no Cubemaps\n", filename);
		ri.FS_FreeFile(buffer.v);
		return;
	}

	if (JSON_ValueGetType(cubemapArrayJson, bufferEnd) != JSONTYPE_ARRAY)
	{
		ri.Printf(PRINT_ALL, "Bad %s: Cubemaps not an array\n", filename);
		ri.FS_FreeFile(buffer.v);
		return;
	}

	tr.numCubemaps = JSON_ArrayGetIndex(cubemapArrayJson, bufferEnd, NULL, 0);
	tr.cubemaps = ri.Hunk_Alloc(tr.numCubemaps * sizeof(*tr.cubemaps), h_low);
	memset(tr.cubemaps, 0, tr.numCubemaps * sizeof(*tr.cubemaps));

	for (i = 0; i < tr.numCubemaps; i++)
	{
		cubemap_t *cubemap = &tr.cubemaps[i];
		const char *cubemapJson, *keyValueJson, *indexes[3];
		int j;

		cubemapJson = JSON_ArrayGetValue(cubemapArrayJson, bufferEnd, i);

		keyValueJson = JSON_ObjectGetNamedValue(cubemapJson, bufferEnd, "Name");
		if (!JSON_ValueGetString(keyValueJson, bufferEnd, cubemap->name, MAX_QPATH))
			cubemap->name[0] = '\0';

		keyValueJson = JSON_ObjectGetNamedValue(cubemapJson, bufferEnd, "Position");
		JSON_ArrayGetIndex(keyValueJson, bufferEnd, indexes, 3);
		for (j = 0; j < 3; j++)
			cubemap->origin[j] = JSON_ValueGetFloat(indexes[j], bufferEnd);

		cubemap->parallaxRadius = 1000.0f;
		keyValueJson = JSON_ObjectGetNamedValue(cubemapJson, bufferEnd, "Radius");
		if (keyValueJson)
			cubemap->parallaxRadius = JSON_ValueGetFloat(keyValueJson, bufferEnd);
	}

	ri.FS_FreeFile(buffer.v);
}

static void R_LoadCubemapEntities(char *cubemapEntityName)
{
	char spawnVarChars[2048];
	int numSpawnVars;
	const char *spawnVars[MAX_SPAWN_VARS][2];
	int numCubemaps = 0;

	// count cubemaps
	numCubemaps = 0;
	while(R_ParseSpawnVars(spawnVarChars, sizeof(spawnVarChars), &numSpawnVars, spawnVars))
	{
		int i;

		for (i = 0; i < numSpawnVars; i++)
		{
			if (!Q_stricmp(spawnVars[i][0], "classname") && !Q_stricmp(spawnVars[i][1], cubemapEntityName))
				numCubemaps++;
		}
	}

	if (!numCubemaps)
		return;

	tr.numCubemaps = numCubemaps;
	tr.cubemaps = ri.Hunk_Alloc(tr.numCubemaps * sizeof(*tr.cubemaps), h_low);
	memset(tr.cubemaps, 0, tr.numCubemaps * sizeof(*tr.cubemaps));

	numCubemaps = 0;
	while(R_ParseSpawnVars(spawnVarChars, sizeof(spawnVarChars), &numSpawnVars, spawnVars))
	{
		int i;
		char name[MAX_QPATH];
		qboolean isCubemap = qfalse;
		qboolean originSet = qfalse;
		vec3_t origin;
		float parallaxRadius = 1000.0f;

		name[0] = '\0';
		for (i = 0; i < numSpawnVars; i++)
		{
			if (!Q_stricmp(spawnVars[i][0], "classname") && !Q_stricmp(spawnVars[i][1], cubemapEntityName))
				isCubemap = qtrue;

			if (!Q_stricmp(spawnVars[i][0], "name"))
				Q_strncpyz(name, spawnVars[i][1], sizeof(name));

			if (!Q_stricmp(spawnVars[i][0], "origin"))
			{
				sscanf(spawnVars[i][1], "%f %f %f", &origin[0], &origin[1], &origin[2]);
				originSet = qtrue;
			}
			else if (!Q_stricmp(spawnVars[i][0], "radius"))
			{
				sscanf(spawnVars[i][1], "%f", &parallaxRadius);
			}
		}

		if (isCubemap && originSet)
		{
			cubemap_t *cubemap = &tr.cubemaps[numCubemaps];
			Q_strncpyz(cubemap->name, name, sizeof(cubemap->name));
			VectorCopy(origin, cubemap->origin);
			cubemap->parallaxRadius = parallaxRadius;
			numCubemaps++;
		}
	}
}

static void R_AssignCubemapsToWorldSurfaces(void)
{
	world_t	*w;
	int i;

	w = &s_worldData;

	for (i = 0; i < w->numsurfaces; i++)
	{
		msurface_t *surf = &w->surfaces[i];
		vec3_t surfOrigin;

		if (surf->cullinfo.type & CULLINFO_SPHERE)
		{
			VectorCopy(surf->cullinfo.localOrigin, surfOrigin);
		}
		else if (surf->cullinfo.type & CULLINFO_BOX)
		{
			surfOrigin[0] = (surf->cullinfo.bounds[0][0] + surf->cullinfo.bounds[1][0]) * 0.5f;
			surfOrigin[1] = (surf->cullinfo.bounds[0][1] + surf->cullinfo.bounds[1][1]) * 0.5f;
			surfOrigin[2] = (surf->cullinfo.bounds[0][2] + surf->cullinfo.bounds[1][2]) * 0.5f;
		}
		else
		{
			//ri.Printf(PRINT_ALL, "surface %d has no cubemap\n", i);
			continue;
		}

		surf->cubemapIndex = R_CubemapForPoint(surfOrigin);
		//ri.Printf(PRINT_ALL, "surface %d has cubemap %d\n", i, surf->cubemapIndex);
	}
}


static void R_LoadCubemaps(void)
{
	int i;
	imgFlags_t flags = IMGFLAG_CLAMPTOEDGE | IMGFLAG_MIPMAP | IMGFLAG_NOLIGHTSCALE | IMGFLAG_CUBEMAP;

	for (i = 0; i < tr.numCubemaps; i++)
	{
		char filename[MAX_QPATH];
		cubemap_t *cubemap = &tr.cubemaps[i];

		Com_sprintf(filename, MAX_QPATH, "cubemaps/%s/%03d.dds", tr.world->baseName, i);

		cubemap->image = R_FindImageFile(filename, IMGTYPE_COLORALPHA, flags);
	}
}


static void R_RenderMissingCubemaps(void)
{
	int i, j;
	imgFlags_t flags = IMGFLAG_NO_COMPRESSION | IMGFLAG_CLAMPTOEDGE | IMGFLAG_MIPMAP | IMGFLAG_NOLIGHTSCALE | IMGFLAG_CUBEMAP;

	for (i = 0; i < tr.numCubemaps; i++)
	{
		if (!tr.cubemaps[i].image)
		{
			tr.cubemaps[i].image = R_CreateImage(va("*cubeMap%d", i), NULL, r_cubemapSize->integer, r_cubemapSize->integer, IMGTYPE_COLORALPHA, flags, GL_RGBA8);

			for (j = 0; j < 6; j++)
			{
				RE_ClearScene();
				R_RenderCubemapSide(i, j, qfalse);
				R_IssuePendingRenderCommands();
				R_InitNextFrame();
			}
		}
	}
}


static void R_CalcVertexLightDirs( void )
{
	int i, k;
	msurface_t *surface;

	for(k = 0, surface = &s_worldData.surfaces[0]; k < s_worldData.numsurfaces /* s_worldData.numWorldSurfaces */; k++, surface++)
	{
		srfBspSurface_t *bspSurf = (srfBspSurface_t *) surface->data;

		switch(bspSurf->surfaceType)
		{
			case SF_FACE:
			case SF_GRID:
			case SF_TRIANGLES:
				for(i = 0; i < bspSurf->numVerts; i++)
				{
					vec3_t lightDir;
					vec3_t normal;

					R_VaoUnpackNormal(normal, bspSurf->verts[i].normal);
					R_LightDirForPoint( bspSurf->verts[i].xyz, lightDir, normal, &s_worldData );
					R_VaoPackNormal(bspSurf->verts[i].lightdir, lightDir);
				}

				break;

			default:
				break;
		}
	}
}


/*
=================
RE_LoadWorldMap

Called directly from cgame
=================
*/
void RE_LoadWorldMap( const char *name ) {
	int			i;
	dheader_t	*header;
	union {
		byte *b;
		void *v;
	} buffer;
	byte		*startMarker;

	if ( tr.worldMapLoaded ) {
		ri.Error( ERR_DROP, "ERROR: attempted to redundantly load world map" );
	}

	// set default map light scale
	tr.sunShadowScale = 0.5f;

	// set default sun direction to be used if it isn't
	// overridden by a shader
	tr.sunDirection[0] = 0.45f;
	tr.sunDirection[1] = 0.3f;
	tr.sunDirection[2] = 0.9f;

	VectorNormalize( tr.sunDirection );

	// set default autoexposure settings
	tr.autoExposureMinMax[0] = -2.0f;
	tr.autoExposureMinMax[1] = 2.0f;

	// set default tone mapping settings
	tr.toneMinAvgMaxLevel[0] = -8.0f;
	tr.toneMinAvgMaxLevel[1] = -2.0f;
	tr.toneMinAvgMaxLevel[2] = 0.0f;

	// reset last cascade sun direction so last shadow cascade is rerendered
	VectorClear(tr.lastCascadeSunDirection);

	tr.worldMapLoaded = qtrue;

	// load it
    ri.FS_ReadFile( name, &buffer.v );
	if ( !buffer.b ) {
		ri.Error (ERR_DROP, "RE_LoadWorldMap: %s not found", name);
	}

	// clear tr.world so if the level fails to load, the next
	// try will not look at the partially loaded version
	tr.world = NULL;

	Com_Memset( &s_worldData, 0, sizeof( s_worldData ) );
	Q_strncpyz( s_worldData.name, name, sizeof( s_worldData.name ) );

	Q_strncpyz( s_worldData.baseName, COM_SkipPath( s_worldData.name ), sizeof( s_worldData.name ) );
	COM_StripExtension(s_worldData.baseName, s_worldData.baseName, sizeof(s_worldData.baseName));

	startMarker = ri.Hunk_Alloc(0, h_low);
	c_gridVerts = 0;

	header = (dheader_t *)buffer.b;
	fileBase = (byte *)header;

	i = LittleLong (header->version);
	/*if ( i != BSP_VERSION ) {
		ri.Error (ERR_DROP, "RE_LoadWorldMap: %s has wrong version number (%i should be %i)", 
			name, i, BSP_VERSION);
	}*/

	// swap all the lumps
	for (i=0 ; i<sizeof(dheader_t)/4 ; i++) {
		((int *)header)[i] = LittleLong ( ((int *)header)[i]);
	}

	// load into heap
	R_LoadEntities( &header->lumps[LUMP_ENTITIES] );
	R_LoadShaders( &header->lumps[LUMP_SHADERS] );
	R_LoadLightmaps( &header->lumps[LUMP_LIGHTMAPS], &header->lumps[LUMP_SURFACES] );
	R_LoadPlanes (&header->lumps[LUMP_PLANES]);
	R_LoadFogs( &header->lumps[LUMP_FOGS], &header->lumps[LUMP_BRUSHES], &header->lumps[LUMP_BRUSHSIDES] );
	R_LoadSurfaces( &header->lumps[LUMP_SURFACES], &header->lumps[LUMP_DRAWVERTS], &header->lumps[LUMP_DRAWINDEXES] );
	R_LoadMarksurfaces (&header->lumps[LUMP_LEAFSURFACES]);
	R_LoadNodesAndLeafs (&header->lumps[LUMP_NODES], &header->lumps[LUMP_LEAFS]);
	R_LoadSubmodels (&header->lumps[LUMP_MODELS]);
	R_LoadVisibility( &header->lumps[LUMP_VISIBILITY] );
	R_LoadLightGrid( &header->lumps[LUMP_LIGHTGRID] );

	// determine vertex light directions
	R_CalcVertexLightDirs();

	// determine which parts of the map are in sunlight
	if (0)
	{
		world_t	*w;
		uint8_t *primaryLightGrid, *data;
		int lightGridSize;
		int i;

		w = &s_worldData;

		lightGridSize = w->lightGridBounds[0] * w->lightGridBounds[1] * w->lightGridBounds[2];
		primaryLightGrid = ri.Malloc(lightGridSize * sizeof(*primaryLightGrid));

		memset(primaryLightGrid, 0, lightGridSize * sizeof(*primaryLightGrid));

		data = w->lightGridData;
		for (i = 0; i < lightGridSize; i++, data += 8)
		{
			int lat, lng;
			vec3_t gridLightDir, gridLightCol;

			// skip samples in wall
			if (!(data[0]+data[1]+data[2]+data[3]+data[4]+data[5]) )
				continue;

			gridLightCol[0] = ByteToFloat(data[3]);
			gridLightCol[1] = ByteToFloat(data[4]);
			gridLightCol[2] = ByteToFloat(data[5]);
			(void)gridLightCol; // Suppress unused-but-set-variable warning

			lat = data[7];
			lng = data[6];
			lat *= (FUNCTABLE_SIZE/256);
			lng *= (FUNCTABLE_SIZE/256);

			// decode X as cos( lat ) * sin( long )
			// decode Y as sin( lat ) * sin( long )
			// decode Z as cos( long )

			gridLightDir[0] = tr.sinTable[(lat+(FUNCTABLE_SIZE/4))&FUNCTABLE_MASK] * tr.sinTable[lng];
			gridLightDir[1] = tr.sinTable[lat] * tr.sinTable[lng];
			gridLightDir[2] = tr.sinTable[(lng+(FUNCTABLE_SIZE/4))&FUNCTABLE_MASK];

			// FIXME: magic number for determining if light direction is close enough to sunlight
			if (DotProduct(gridLightDir, tr.sunDirection) > 0.75f)
			{
				primaryLightGrid[i] = 1;
			}
			else
			{
				primaryLightGrid[i] = 255;
			}
		}

		if (0)
		{
			int i;
			byte *buffer = ri.Malloc(w->lightGridBounds[0] * w->lightGridBounds[1] * 3 + 18);
			byte *out;
			uint8_t *in;
			char fileName[MAX_QPATH];
			
			Com_Memset (buffer, 0, 18);
			buffer[2] = 2;		// uncompressed type
			buffer[12] = w->lightGridBounds[0] & 255;
			buffer[13] = w->lightGridBounds[0] >> 8;
			buffer[14] = w->lightGridBounds[1] & 255;
			buffer[15] = w->lightGridBounds[1] >> 8;
			buffer[16] = 24;	// pixel size

			in = primaryLightGrid;
			for (i = 0; i < w->lightGridBounds[2]; i++)
			{
				int j;

				sprintf(fileName, "primarylg%d.tga", i);

				out = buffer + 18;
				for (j = 0; j < w->lightGridBounds[0] * w->lightGridBounds[1]; j++)
				{
					if (*in == 1)
					{
						*out++ = 255;
						*out++ = 255;
						*out++ = 255;
					}
					else if (*in == 255)
					{
						*out++ = 64;
						*out++ = 64;
						*out++ = 64;
					}
					else
					{
						*out++ = 0;
						*out++ = 0;
						*out++ = 0;
					}
					in++;
				}

				ri.FS_WriteFile(fileName, buffer, w->lightGridBounds[0] * w->lightGridBounds[1] * 3 + 18);
			}

			ri.Free(buffer);
		}

		for (i = 0; i < w->numWorldSurfaces; i++)
		{
			msurface_t *surf = w->surfaces + i;
			cullinfo_t *ci = &surf->cullinfo;

			if(ci->type & CULLINFO_PLANE)
			{
				if (DotProduct(ci->plane.normal, tr.sunDirection) <= 0.0f)
				{
					//ri.Printf(PRINT_ALL, "surface %d is not oriented towards sunlight\n", i);
					continue;
				}
			}

			if(ci->type & CULLINFO_BOX)
			{
				int ibounds[2][3], x, y, z, goodSamples, numSamples;
				vec3_t lightOrigin;

				VectorSubtract( ci->bounds[0], w->lightGridOrigin, lightOrigin );

				ibounds[0][0] = floor(lightOrigin[0] * w->lightGridInverseSize[0]);
				ibounds[0][1] = floor(lightOrigin[1] * w->lightGridInverseSize[1]);
				ibounds[0][2] = floor(lightOrigin[2] * w->lightGridInverseSize[2]);

				VectorSubtract( ci->bounds[1], w->lightGridOrigin, lightOrigin );

				ibounds[1][0] = ceil(lightOrigin[0] * w->lightGridInverseSize[0]);
				ibounds[1][1] = ceil(lightOrigin[1] * w->lightGridInverseSize[1]);
				ibounds[1][2] = ceil(lightOrigin[2] * w->lightGridInverseSize[2]);

				ibounds[0][0] = CLAMP(ibounds[0][0], 0, w->lightGridSize[0]);
				ibounds[0][1] = CLAMP(ibounds[0][1], 0, w->lightGridSize[1]);
				ibounds[0][2] = CLAMP(ibounds[0][2], 0, w->lightGridSize[2]);

				ibounds[1][0] = CLAMP(ibounds[1][0], 0, w->lightGridSize[0]);
				ibounds[1][1] = CLAMP(ibounds[1][1], 0, w->lightGridSize[1]);
				ibounds[1][2] = CLAMP(ibounds[1][2], 0, w->lightGridSize[2]);

				/*
				ri.Printf(PRINT_ALL, "surf %d bounds (%f %f %f)-(%f %f %f) ibounds (%d %d %d)-(%d %d %d)\n", i,
					ci->bounds[0][0], ci->bounds[0][1], ci->bounds[0][2],
					ci->bounds[1][0], ci->bounds[1][1], ci->bounds[1][2],
					ibounds[0][0], ibounds[0][1], ibounds[0][2],
					ibounds[1][0], ibounds[1][1], ibounds[1][2]);
				*/

				goodSamples = 0;
				numSamples = 0;
				for (x = ibounds[0][0]; x <= ibounds[1][0]; x++)
				{
					for (y = ibounds[0][1]; y <= ibounds[1][1]; y++)
					{
						for (z = ibounds[0][2]; z <= ibounds[1][2]; z++)
						{
							uint8_t primaryLight = primaryLightGrid[x * 8 + y * 8 * w->lightGridBounds[0] + z * 8 * w->lightGridBounds[0] * w->lightGridBounds[2]];

							if (primaryLight == 0)
								continue;

							numSamples++;

							if (primaryLight == 1)
								goodSamples++;
						}
					}
				}

				// FIXME: magic number for determining whether object is mostly in sunlight
				if (goodSamples > numSamples * 0.75f)
				{
					//ri.Printf(PRINT_ALL, "surface %d is in sunlight\n", i);
					//surf->primaryLight = 1;
				}
			}
		}

		ri.Free(primaryLightGrid);
	}

	// load cubemaps
	if (r_cubeMapping->integer)
	{
		// Try loading an env.json file first
		R_LoadEnvironmentJson(s_worldData.baseName);

		if (!tr.numCubemaps)
		{
			R_LoadCubemapEntities("misc_cubemap");
		}

		if (!tr.numCubemaps)
		{
			// use deathmatch spawn points as cubemaps
			R_LoadCubemapEntities("info_player_deathmatch");
		}

		if (tr.numCubemaps)
		{
			R_AssignCubemapsToWorldSurfaces();
		}
	}

	s_worldData.dataSize = (byte *)ri.Hunk_Alloc(0, h_low) - startMarker;

	// only set tr.world now that we know the entire level has loaded properly
	tr.world = &s_worldData;

	// make sure the VAO glState entry is safe
	R_BindNullVao();

	// Render or load all cubemaps
	if (r_cubeMapping->integer && tr.numCubemaps && glRefConfig.framebufferObject)
	{
		R_LoadCubemaps();
		R_RenderMissingCubemaps();
	}

    ri.FS_FreeFile( buffer.v );
}